diff --git a/comm/server/adcclilib.py b/comm/server/adcclilib.py new file mode 100644 index 0000000000000000000000000000000000000000..fb03db34b7bf8148f0a26f4a642c203c20353d31 --- /dev/null +++ b/comm/server/adcclilib.py @@ -0,0 +1,59 @@ +#!/usr/bin/env python3 + +import subprocess + +PYTHON2 = "/usr/bin/python" +ADCCLI = "/home/pi/adc2018/conmgr/client/adccli" + +def _exec_adccli(cmd): + exec_cmd = "{} {} {}".format(PYTHON2, ADCCLI, cmd).strip() + print("ADCCLI: {}".format(exec_cmd)) + p = subprocess.run(exec_cmd, stdout=subprocess.PIPE, shell=True) + res = p.stdout.decode().strip() + print(res) + return res + +def login(url, username, password): + cmd = "--URL='{}' --username='{}' --password='{}' login".format(url, username, password) + return _exec_adccli(cmd) + +def logout(): + cmd = "logout" + return _exec_adccli(cmd) + +def whoami(): + cmd = "whoami" + return _exec_adccli(cmd) + +def put_message(message): + cmd = "put-user-alive '{}'" + return _exec_adccli(cmd) + +def post_user_q(qnum, filepath): + cmd = "post-user-q {} {}".format(qnum, filepath) + return _exec_adccli(cmd) + +def get_q_all(outpath): + cmd = "get-q" + question_list = _exec_adccli(cmd).split("\n") + + print("--- ADCCLI questions ---") + print(question_list) + + for v in question_list: + if v.startswith("Q"): + qnumber = int(v.replace("Q", "")) + out_file_path = "{}/NL_Q{:02d}.txt".format(outpath, qnumber) + cmd = "--output {} get-q {}".format(out_file_path, qnumber) + r = _exec_adccli(cmd) + + return question_list + +def put_a(qnum, filepath): + cmd = "put-a {} {}".format(qnum, filepath) + return _exec_adccli(cmd) + +def put_a_info(qnum, cpu, mem, misc): + cmd = "put-a-info {} {} {} '{}'".format(qnum, cpu, mem, misc) + return _exec_adccli(cmd) + diff --git a/comm/server/main.py b/comm/server/main.py index 4511d98918889a253d7493895878e66dd51b1cb4..68157a4bdcef71640336aba841d9c5378f451665 100644 --- a/comm/server/main.py +++ b/comm/server/main.py @@ -22,6 +22,8 @@ from queue import Queue sys.path.append(os.path.dirname(os.path.abspath(__file__)) + '/../../solver') import BoardStr +import adcclilib + app = Flask(__name__) app_args = {} questions = None @@ -440,6 +442,76 @@ def get_clients(): return json.dumps(res) +@app.route("/adccli-login") +def adccli_login(): + + global app_args + + with open(app_args['adccli'], 'r') as fp: + d = json.load(fp) + + r = adcclilib.login(d['url'], d['username'], d['password']) + res = {'message': r} + + return json.dumps(res) + +@app.route("/adccli-logout") +def adccli_logout(): + r = adcclilib.logout() + res = {'message': r} + return json.dumps(res) + +@app.route("/adccli-whoami") +def adccli_whoami(): + global app_args + + with open(app_args['adccli'], 'r') as fp: + d = json.load(fp) + + r = adcclilib.whoami() + + res = {'message': r, 'status': r == d['username']} + return json.dumps(res) + +@app.route("/adccli-get-all-q") +def adccli_get_all_q(): + global app_args + + out_abspath = os.path.abspath(app_args['question']) + r = adcclilib.get_q_all(out_abspath) + + load_questions() + + res = {'message': r} + return json.dumps(res) + +@app.route("/adccli-put-a", methods=["POST"]) +def adccli_put_a(): + global app_args + global questions + + qname = request.form["qname"] + + if not "best_json" in questions[qname]: + res = {'message': "Required to 'save' before submission"} + else: + out_path = "{}/{}".format(app_args['out'], "submit") + qnumber = qname.replace("NL_Q", "").replace(".txt", "") + ans_file_path = "{}/T01_A{}.txt".format(out_path, qnumber) + ans_file_abspath = os.path.abspath(ans_file_path) + + r = adcclilib.put_a(qnumber, ans_file_abspath) + mes = r + "\n" + + json_name = questions[qname]['best_json'] + data = questions[qname]['answers'][json_name] + r = adcclilib.put_a_info(qnumber, data['cputime'], "0", "Test") + mes += r + + res = {'message': mes} + + return json.dumps(res) + @app.route("/board-viewer", methods=["GET"]) def show_board_viewer(): @@ -524,6 +596,7 @@ if __name__ == "__main__": parser = argparse.ArgumentParser(description="PYNQ control panel.") parser.add_argument("-c", "--client", action="store", type=str, default=None, required=True, help="Client list.") + parser.add_argument("-a", "--adccli", action="store", type=str, default="./adccli.json", help="Path to the ADCCLI configuration json file.") parser.add_argument("-q", "--question", action="store", type=str, default="./problems", help="Path to the question folder.") parser.add_argument("-o", "--out", action="store", type=str, default="./answers", help="Path to the output folder.") parser.add_argument("--debug", action="store_true", default=False, help="Debug mode.") diff --git a/comm/server/static/js/pynq-manager.js b/comm/server/static/js/pynq-manager.js index a48aa8de3f611bb030d2fd48bbf4d8eee68578e5..995d93ca348f7431e90838ad183f351b16e38e39 100644 --- a/comm/server/static/js/pynq-manager.js +++ b/comm/server/static/js/pynq-manager.js @@ -145,6 +145,40 @@ $(function(){ }).done(function(d){ $("#client-control-pane").html(""); $("#client-control-pane").html(d); + + var button_action_with_ajax = function($obj, url){ + $obj.prop("disabled", "disabled"); + $.ajax({ + type: "GET", + url: url, + dataType: "json", + }).done((data)=>{ + $obj.prop("disabled", false); + alert(data['message']); + }); + }; + + $("#adccli-login-button").click(function(){ + button_action_with_ajax($(this), "/adccli-login"); + }); + $("#adccli-logout-button").click(function(){ + button_action_with_ajax($(this), "/adccli-logout"); + }); + $("#adccli-get-all-q").click(function(){ + button_action_with_ajax($(this), "/adccli-get-all-q"); + }); + + $.ajax({ + type: "GET", + url: "/adccli-whoami", + dataType: "json", + }).done((data)=>{ + if(data['status']) + $("#adccli-status").text("ログイン中"); + else + $("#adccli-status").text("ログアウト"); + }); + pm.getStatus(); }); } @@ -177,6 +211,17 @@ $(function(){ alert(data['status']); }); }); + $("#client-control-pane").find(".submit-button").eq(0).click(function(){ + var qname = $(this).data("qname"); + $.ajax({ + type: "POST", + dataType: "json", + url: "/adccli-put-a", + data: {qname: qname} + }).done((data) => { + alert(data['message']); + }); + }); $(".answer-detail-row td").click(function(){ var json_name = $(this).parent("tr").data("json"); diff --git a/comm/server/templates/part_client_table.html b/comm/server/templates/part_client_table.html index 6211db7a81feb30a171696ac0c605c2153922d12..7386b49fabcd3f66e5f4b18208e50f950b937557 100644 --- a/comm/server/templates/part_client_table.html +++ b/comm/server/templates/part_client_table.html @@ -9,6 +9,13 @@ Viewer Mode {% endif %}

+

自動運営システム

+ + + +
+ +

クライアント

diff --git a/comm/server/templates/part_question_status.html b/comm/server/templates/part_question_status.html index 225e00ec2829eb14717cbe0e2357dd759521938f..a6f634cd7df0708d1510b6629e52078ecfc9b7f8 100644 --- a/comm/server/templates/part_question_status.html +++ b/comm/server/templates/part_question_status.html @@ -3,8 +3,11 @@

【{{qname}}】

-
+ +

+

+

diff --git a/hls/ZU3EG_test/Makefile b/hls/ZU3EG_test/Makefile new file mode 100644 index 0000000000000000000000000000000000000000..8f79e7a62ede483ae7dcda810f306a02f915dfbb --- /dev/null +++ b/hls/ZU3EG_test/Makefile @@ -0,0 +1,20 @@ +TARGET = sim +OBJS = $(CPPS:.cpp=.o) +CPPS = $(wildcard *.cpp) +CXX = g++ +CXXFLAGS = -O3 -Wall -Wno-unknown-pragmas -Wno-unused-label -DSOFTWARE -DCALCTIME + +all: $(TARGET) + +$(TARGET): $(OBJS) + $(CXX) -O3 -o $@ $(OBJS) + +run: + python3 ../NLGenerator.py -x 20 -y 20 -z 6 -l 100;\ + python3 ./gen_boardstr.py Q-20x20x5_100_10.txt |\ + ./$(TARGET) - + + +clean: + rm *.o + rm $(TARGET) diff --git a/hls/ZU3EG_test/README.md b/hls/ZU3EG_test/README.md new file mode 100644 index 0000000000000000000000000000000000000000..662e2aba84de9b7126054d1d917eb00955b471ea --- /dev/null +++ b/hls/ZU3EG_test/README.md @@ -0,0 +1,62 @@ +# pynq-router (sw) + +Vivado HLS 用 pynq-router + + +## メモ + +* Vivado HLS 2016.1 +* Vivado 2016.1 + +高位合成オプション: +* Part: xc7z020clg400-1 +* Clock Period: 10.0 ns +* Clock Uncertainty: 3.0 ns, 3.5 ns // 2017/08/04 + +論理合成・配置配線のオプション: +* Synthesis strategy: ~~Vivado Synthesis Defaults~~ Flow_PerfOptimized_high +* Implementation strategy: ~~Vivado Implementation Defaults~~ Performance_Explore + +最適化記録： +* 2017/08/04: クリティカルパス遅延の最適化 + * (3.0) Estimated clock: 6.78, Max latency: 1,667,151,745 (1.00) + * (3.5) Estimated clock: 6.38, Max latency: 1,692,438,002 (1.00) +* 2017/08/05: キューpop処理の動作改善 + * (3.0) Estimated clock: 6.78, Max latency: 1,530,952,491 (0.92) + * (3.5) Estimated clock: 6.38, Max latency: 1,556,238,748 (0.92) +* 2017/08/07: 剰余演算の書き換え，隣接ノードの探索ループ展開 + * (3.0) Estimated clock: 6.78, Max latency: 1,423,592,713 (0.85) + * (3.5) Estimated clock: 6.38, Max latency: 1,428,235,716 (0.84) +* 2017/08/07: 高位合成パラメータの調整 + * (3.0) Estimated clock: 6.78, Max latency: 1,284,364,290 (0.77) + * (3.5) Estimated clock: 6.38, Max latency: 1,289,003,291 (0.76) +* 2017/08/11: 隣接ノードの探索ループ展開を戻した (そうしないと論理合成・配置配線でタイミング満たさない) + * (3.0) Estimated clock: 6.78, Max latency: 1,329,761,290 (0.80) + * (3.5) Estimated clock: 6.38, Max latency: 1,355,035,291 (0.80) +* 2017/08/11: starts と goals を FF パーティションした + * (3.0) Estimated clock: 6.78, Max latency: 1,329,245,034 (0.797) + * (3.5) Estimated clock: 6.45, Max latency: 1,354,519,035 (0.800) +* 2017/08/16: MAX_LINES を 256 に増やした + * (3.0) Estimated clock: 6.78, Max latency: 1,905,588,970 (1.143) + * (3.5) Estimated clock: 6.38, Max latency: 1,933,166,972 (1.142) +* 2017/08/16: MAX_LINES を 128 に戻した (そうしないと論理合成・配置配線でタイミング満たさない) + * (3.0) Estimated clock: 6.78, Max latency: 1,329,245,034 (0.797) + * (3.5) Estimated clock: 6.45, Max latency: 1,354,519,035 (0.800) +* 2017/08/16: コスト関数の波形をのこぎり型にした + * (3.0) Estimated clock: 6.78, Max latency: 1,329,245,034 (13.29 sec) (0.797) + * (3.25) Estimated clock: 6.52, Max latency: 1,329,249,035 (13.29 sec) + * (3.5) Estimated clock: 6.45, Max latency: 1,354,519,035 (13.55 sec) (0.800) +* 2017/08/18: 対象ライン選択に剰余演算を用いない方法を試してみた ※あまり変わらないからこれはやらずに元に戻す + * (3.0) Estimated clock: 6.52, Max latency: 1,329,101,007 (13.29 sec) (0.797) + * (3.5) Estimated clock: 6.45, Max latency: 1,354,371,007 (13.54 sec) (0.800) +* 2017/08/18: 対象ラインが連続して同じだったらルーティングスキップする + * (3.0) Estimated clock: 6.78, Max latency: 1,329,245,034 (13.29 sec) + * (3.25) Estimated clock: 6.52, Max latency: 1,329,249,035 (13.29 sec) **←今これ** + * (3.5) Estimated clock: 6.45, Max latency: 1,354,519,035 (13.55 sec) + + +## 入出力 + +入力は boardstr という問題ファイルをコンパクトに記述したようなワンラインの文字列。 + +出力は char 型の配列で、各要素に解答ファイルに相当する数値が入っている。 diff --git a/hls/ZU3EG_test/ap_int.h b/hls/ZU3EG_test/ap_int.h new file mode 100644 index 0000000000000000000000000000000000000000..b8d9fdc96ac3f7eb9c86cc55d0eac0a57bf670b5 --- /dev/null +++ b/hls/ZU3EG_test/ap_int.h @@ -0,0 +1,521 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_AP_SIM_H__ +#define __AESL_AP_SIM_H__ + +#ifndef __cplusplus +#error C++ is required to include this header file +#else + +#include "etc/ap_int_sim.h" +#include "etc/ap_fixed_sim.h" + +//Forward declaration +template class ap_fixed; +template class ap_ufixed; +template class ap_int; +template class ap_uint; + +//AP_INT +//-------------------------------------------------------- +template +class ap_int : public ap_private<_AP_W, true> { +#ifdef _MSC_VER +#pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + typedef ap_private<_AP_W, true> Base; + + //Constructor + INLINE ap_int(): Base() {} + template + INLINE ap_int(const volatile ap_int<_AP_W2> &op):Base((const ap_private<_AP_W2,true> &)(op)) {} + + template + INLINE ap_int(const ap_int<_AP_W2> &op):Base((const ap_private<_AP_W2,true> &)(op)) {} + + template + INLINE ap_int(const ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2,false> &)(op)) {} + + template + INLINE ap_int(const volatile ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2,false> &)(op)) {} + + template + INLINE ap_int(const ap_range_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_bit_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + template + INLINE ap_int(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_int(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_int(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op.to_ap_private()) {} + +#define CTOR(TYPE) \ + INLINE ap_int(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_int(const char* str, signed char rd):Base(str, rd) {} + //Assignment + //Another form of "write" + INLINE void operator = (const ap_int<_AP_W>& op2) volatile { + const_cast(this)->operator = (op2); + } + + INLINE void operator = (const volatile ap_int<_AP_W>& op2) volatile { + const_cast(this)->operator = (op2); + } + + INLINE ap_int<_AP_W>& operator = (const volatile ap_int<_AP_W>& op2) { + Base::operator = (const_cast& >(op2)); + return *this; + } + + INLINE ap_int<_AP_W>& operator = (const ap_int<_AP_W>& op2) { + Base::operator = ((const ap_private<_AP_W, true>&)op2); + return *this; + } +}; + +//AP_UINT +//--------------------------------------------------------------- +template +class ap_uint: public ap_private<_AP_W, false> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef ap_private<_AP_W, false> Base; + //Constructor + INLINE ap_uint(): Base() {} + INLINE ap_uint(const ap_uint<_AP_W>& op) :Base(dynamic_cast&>(op)) {} + INLINE ap_uint(const volatile ap_uint<_AP_W>& op):Base(dynamic_cast&>(op)){} + template + INLINE ap_uint(const volatile ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_uint(const ap_uint<_AP_W2> &op) : Base((const ap_private<_AP_W2, false>&)(op)){} + + template + INLINE ap_uint(const ap_int<_AP_W2> &op) : Base((const ap_private<_AP_W2, true>&)(op)) {} + + template + INLINE ap_uint(const volatile ap_int<_AP_W2> &op) : Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_uint(const ap_range_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_bit_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op) {} + + template + INLINE ap_uint(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + template + INLINE ap_uint(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_uint(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_uint(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op.to_ap_private()) {} + +#define CTOR(TYPE) \ + INLINE ap_uint(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_uint(const char* str, signed char rd):Base(str, rd) {} + //Assignment + //Another form of "write" + INLINE void operator = (const ap_uint<_AP_W>& op2) volatile { + Base::operator = (op2); + } + + INLINE void operator = (const volatile ap_uint<_AP_W>& op2) volatile { + Base::operator = (op2); + } + + INLINE ap_uint<_AP_W>& operator = (const volatile ap_uint<_AP_W>& op2) { + Base::operator = (op2); + return *this; + } + + INLINE ap_uint<_AP_W>& operator = (const ap_uint<_AP_W>& op2) { + Base::operator = ((const ap_private<_AP_W, false>&)(op2)); + return *this; + } +}; + +#define ap_bigint ap_int +#define ap_biguint ap_uint + +//AP_FIXED +//--------------------------------------------------------------------- +template +class ap_fixed: public ap_fixed_base<_AP_W, _AP_I, true, _AP_Q, _AP_O, _AP_N> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef ap_fixed_base<_AP_W, _AP_I, true, _AP_Q, _AP_O, _AP_N> Base; + //Constructor + INLINE ap_fixed():Base() {} + + template + INLINE ap_fixed(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_fixed(const ap_uint<_AP_W2>& op):Base(ap_private<_AP_W2, false>(op)) {} + + template + INLINE ap_fixed(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const volatile ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_fixed(const volatile ap_uint<_AP_W2>& op):Base(op) {} + + template + INLINE ap_fixed(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op) {} + + template + INLINE ap_fixed(const ap_bit_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_fixed(const ap_range_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_fixed(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op): + Base(op) {} + + template + INLINE ap_fixed(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + #define CTOR(TYPE) \ + INLINE ap_fixed(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_fixed(const char* str, signed char rd):Base(str, rd) {} + + //Assignment + INLINE ap_fixed& operator = (const ap_fixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } + + INLINE ap_fixed& operator = (const volatile ap_fixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } +}; + +//AP_ UFIXED +//--- ---------------------------------------------------------------- +template +class ap_ufixed : public ap_fixed_base<_AP_W, _AP_I, false, _AP_Q, _AP_O, _AP_N> { +#ifdef _MSC_VER +#pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + typedef ap_fixed_base<_AP_W, _AP_I, false, _AP_Q, _AP_O, _AP_N> Base; + + //Constructor + INLINE ap_ufixed():Base() {} + + template + INLINE ap_ufixed(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op) : Base(ap_fixed_base<_AP_W2, + _AP_I2, true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const ap_int<_AP_W2>& op): + Base((const ap_private<_AP_W2, true>&)(op)) {} + + template + INLINE ap_ufixed(const ap_uint<_AP_W2>& op): + Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_ufixed(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op) : Base(ap_fixed_base<_AP_W2, + _AP_I2, true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_uint<_AP_W2>& op): + Base(ap_private<_AP_W2, false>(op)) {} + + template + INLINE ap_ufixed(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2>& op):Base(op) {} + + template + INLINE ap_ufixed(const ap_bit_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_ufixed(const ap_range_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_ufixed(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op): + Base(op) {} + + template + INLINE ap_ufixed(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_ufixed(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_ufixed(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + #define CTOR(TYPE) \ + INLINE ap_ufixed(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_ufixed(const char* str, signed char rd):Base(str, rd) {} + + //Assignment + INLINE ap_ufixed& operator = (const ap_ufixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } + + INLINE ap_ufixed& operator = (const volatile ap_ufixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::V = const_cast(op); + return *this; + } +}; + +#if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_int<_AP_W> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_uint<_AP_W> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_fixed<_AP_W, _AP_I, _AP_Q, _AP_O, _AP_N >&op, const std::string &name) { + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_ufixed<_AP_W, _AP_I, _AP_Q, _AP_O, _AP_N >&op, const std::string &name) { + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} +#endif /* #if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) */ +#endif /* #ifndef __cplusplus */ +#endif /* #ifndef __AESL_AP_SIM_H__ */ \ No newline at end of file diff --git a/hls/ZU3EG_test/etc/ap_fixed_sim.h b/hls/ZU3EG_test/etc/ap_fixed_sim.h new file mode 100644 index 0000000000000000000000000000000000000000..5be571dd70e490d282d279a4eeed32db069703e9 --- /dev/null +++ b/hls/ZU3EG_test/etc/ap_fixed_sim.h @@ -0,0 +1,2451 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_GCC_AP_FIXED_H__ +#define __AESL_GCC_AP_FIXED_H__ + +#ifndef __cplusplus + #error C++ is required to include this header file +#endif /* #ifndef __cplusplus */ + + +#include +#ifndef __AESL_APDT_IN_SCFLOW__ + #include "etc/ap_int_sim.h" +#else + #include "../etc/ap_private.h" +#endif /* #ifndef __AESL_APDT_IN_SCFLOW__ */ + +#define FLOAT_MAN 23 +#define FLOAT_EXP 8 +#define DOUBLE_MAN 52 +#define DOUBLE_EXP 11 +// #define DOUBLE_MAN_MASK (~0ULL >> (64-DOUBLE_MAN-2)) +#define DOUBLE_MAN_MASK 0x3fffffffffffffULL +#define BIAS(e) ((1ULL<<(e-1))-1) +#define FLOAT_BIAS BIAS(FLOAT_EXP) +#define DOUBLE_BIAS BIAS(DOUBLE_EXP) + +/// Forward declaration. +template struct ap_fixed_base; + +///Proxy class, which allows bit selection to be used as both rvalue(for reading) and +//lvalue(for writing) +template +struct af_bit_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& d_bv; + int d_index; +public: + INLINE af_bit_ref(const af_bit_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& ref): + d_bv(ref.d_bv), d_index(ref.d_index) {} + + INLINE af_bit_ref(ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>* bv, int index=0): + d_bv(*bv),d_index(index) {} + + INLINE operator bool() const { + return d_bv.V[d_index]; + } + + INLINE af_bit_ref& operator=(unsigned long long val) { + if (val) + d_bv.V.set(d_index); + else + d_bv.V.clear(d_index); + return *this; + } + + template + INLINE af_bit_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + return operator=(val!=0); + } + + INLINE af_bit_ref& operator =(const af_bit_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& val) { + return operator=((unsigned long long)(bool)val); + } + + template + INLINE af_bit_ref operator=(const af_bit_ref<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)val); + } + + template + INLINE af_bit_ref& operator = ( const ap_bit_ref<_AP_W2, _AP_S2> &val) { + return operator =((unsigned long long) (bool) val); + } + + template + INLINE af_bit_ref& operator = ( const ap_range_ref<_AP_W2,_AP_S2>& val) { + return operator =((const ap_private<_AP_W2, false>) val); + } + + template + INLINE af_bit_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, false>)(val)); + } + + template + INLINE af_bit_ref& operator= (const ap_concat_ref<_AP_W2, _AP_T3, _AP_W3, _AP_T3>& val) { + return operator=((const ap_private<_AP_W2 + _AP_W3, false>)(val)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + ap_private<_AP_W2, _AP_S2> >(*this, op); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<1, af_bit_ref, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2> >(*this, const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<1, af_bit_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(op)); + } + + template + INLINE bool operator == (const af_bit_ref<_AP_W2, _AP_I2, + _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + return get() == op.get(); + } + + template + INLINE bool operator != (const af_bit_ref<_AP_W2, _AP_I2, + _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + return get() != op.get(); + } + + INLINE bool operator ~ () const { + bool bit = (d_bv.V)[d_index]; + return bit ? false : true; + } + + INLINE int length() const { + return 1; + } + + INLINE bool get() { + return d_bv.V[d_index]; + } + + INLINE bool get() const { + return d_bv.V[d_index]; + } + + INLINE std::string to_string() const { + return d_bv.V[d_index] ? "1" : "0"; + } +}; + +///Range(slice) reference +//------------------------------------------------------------ +template +struct af_range_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &d_bv; + int l_index; + int h_index; + +public: + INLINE af_range_ref(const af_range_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& ref): + d_bv(ref.d_bv), l_index(ref.l_index), h_index(ref.h_index) {} + + INLINE af_range_ref(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>* bv, int h, int l): + d_bv(*bv),l_index(l),h_index(h) { + //if (h < l) + // fprintf(stderr, + //"Warning! The bits selected will be returned in reverse order\n"); + } + + INLINE operator ap_private<_AP_W, false> () const { + if (h_index >= l_index) { + ap_private<_AP_W, false> val(d_bv.V); + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val>>=l_index; + return val&=mask; + } else { + ap_private<_AP_W, false> val = 0; + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if ((d_bv.V)[j]) val.set(i); + return val; + } + } + + INLINE operator unsigned long long() const { + return get().to_uint64(); + } + + template + INLINE af_range_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + ap_private<_AP_W, false> vval= ap_private<_AP_W, false>(val); + if (l_index > h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + vval[i]? d_bv.V.set(j):d_bv.V.clear(j); + } else { + ap_private<_AP_W,false> mask(-1); + if (l_index>0) { + mask<<=l_index; + vval<<=l_index; + } + if (h_index<_AP_W-1) { + ap_private<_AP_W,false> mask2(-1); + mask2>>=_AP_W-h_index-1; + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv.V &= mask; + d_bv.V |= vval; + } + return *this; + } + + INLINE af_range_ref& operator = (unsigned long long val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator =(const ap_bit_ref<_AP_W3, _AP_S3>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE af_range_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& val) { + const ap_private<_AP_W2, false> tmp= val.get(); + return operator = (tmp); + } + + INLINE af_range_ref& operator= (const af_range_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& val) { + const ap_private<_AP_W, false> tmp= val.get(); + return operator = (tmp); + } + + template + INLINE af_range_ref& operator= (const ap_fixed_base<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=(val.to_ap_private()); + } + + template + INLINE bool operator == (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + template + INLINE bool operator != (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + template + INLINE bool operator > (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + template + INLINE bool operator >= (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + template + INLINE bool operator < (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + template + INLINE bool operator == (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + template + INLINE bool operator != (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + template + INLINE bool operator > (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + template + INLINE bool operator >= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + template + INLINE bool operator < (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + template + INLINE void set(const ap_private<_AP_W2,false>& val) { + ap_private<_AP_W,_AP_S> vval=val; + if (l_index>h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + vval[i]? d_bv.V.set(j):d_bv.V.clear(j); + } else { + ap_private<_AP_W,_AP_S> mask(-1); + if (l_index>0) { + ap_private<_AP_W,false> mask1(-1); + mask1>>=_AP_W-l_index; + mask1.flip(); + mask=mask1; + //vval&=mask1; + vval<<=l_index; + } + if (h_index<_AP_W-1) { + ap_private<_AP_W,false> mask2(-1); + mask2<<=h_index+1; + mask2.flip(); + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } + + } + + INLINE ap_private<_AP_W,false> get() const { + if (h_index val(0); + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if ((d_bv.V)[j]) val.set(i); + return val; + } else { + ap_private<_AP_W, false> val = ap_private<_AP_W,false>(d_bv.V); + val>>= l_index; + if (h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + return val; + } + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + ap_private<_AP_W2, _AP_S2> >(*this, op); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& > (op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(op)); + } + + INLINE int length() const { + return h_index>=l_index?h_index-l_index+1:l_index-h_index+1; + } + + INLINE int to_int() const { + ap_private<_AP_W,false> val=get(); + return val.to_int(); + } + + INLINE unsigned int to_uint() const { + ap_private<_AP_W,false> val=get(); + return val.to_uint(); + } + + INLINE long to_long() const { + ap_private<_AP_W,false> val=get(); + return val.to_long(); + } + + INLINE unsigned long to_ulong() const { + ap_private<_AP_W,false> val=get(); + return val.to_ulong(); + } + + INLINE ap_slong to_int64() const { + ap_private<_AP_W,false> val=get(); + return val.to_int64(); + } + + INLINE ap_ulong to_uint64() const { + ap_private<_AP_W,false> val=get(); + return val.to_uint64(); + } + + INLINE std::string to_string(uint8_t radix) const { + return get().to_string(radix); + } + +}; + +//----------------------------------------------------------------------------- +///ap_fixed_base: AutoPilot fixed point +//----------------------------------------------------------------------------- +template +struct ap_fixed_base { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + template friend struct +ap_fixed_base; + template friend struct +af_bit_ref; + + INLINE void overflow_adjust(bool underflow, bool overflow, + bool lD, bool sign) { + if (!overflow && !underflow) return; + switch (_AP_O) { + case AP_WRAP: + if (_AP_N == 0) + return; + if (_AP_S) { + //signed SC_WRAP + //n_bits == 1; + if (_AP_N > 1) { + ap_private<_AP_W, _AP_S> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + if (sign) + V &= mask; + else + V |= ~mask; + } + sign ? V.set(_AP_W - 1) : V.clear(_AP_W - 1); + } else { + //unsigned SC_WRAP + ap_private<_AP_W, _AP_S> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + mask.flip(); + V |= mask; + } + break; + case AP_SAT_ZERO: + V.clear(); + break; + case AP_WRAP_SM: + { + bool Ro = ap_private_ops::get<_AP_W, _AP_S, _AP_W -1>(V); // V[_AP_W -1]; + if (_AP_N == 0) { + if (lD != Ro) { + V.flip(); + lD ? ap_private_ops::set<_AP_W, _AP_S, _AP_W - 1>(V) : + ap_private_ops::clear<_AP_W, _AP_S, _AP_W - 1>(V); + } + } else { + if (_AP_N == 1 && sign != Ro) { + V.flip(); + } else if (_AP_N > 1) { + bool lNo = ap_private_ops::get<_AP_W, _AP_S, _AP_W - _AP_N> (V); // V[_AP_W - _AP_N]; + if (lNo == sign) + V.flip(); + ap_private<_AP_W, false> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + if (sign) + V &= mask; + else + V |= mask.flip(); + sign ? ap_private_ops::set<_AP_W, _AP_S, _AP_W - 1>(V) : ap_private_ops::clear<_AP_W, _AP_S, _AP_W - 1>(V); + } + } + } + break; + default: + if (_AP_S) { + if (overflow) { + V.set(); ap_private_ops::clear<_AP_W, _AP_S, _AP_W-1>(V); + } else if (underflow) { + V.clear(); + ap_private_ops::set<_AP_W, _AP_S, _AP_W-1>(V); + if (_AP_O == AP_SAT_SYM) + ap_private_ops::set<_AP_W, _AP_S, 0>(V); + } + } else { + if (overflow) + V.set(); + else if (underflow) + V.clear(); + } + } + } + + INLINE bool quantization_adjust(bool qb, bool r, bool s) { + bool carry=ap_private_ops::get<_AP_W, _AP_S, _AP_W-1>(V); + switch (_AP_Q) { + case AP_TRN: + return false; + case AP_RND_ZERO: + qb &= s || r; + break; + case AP_RND_MIN_INF: + qb &= r; + break; + case AP_RND_INF: + qb &= !s || r; + break; + case AP_RND_CONV: + qb &= ap_private_ops::get<_AP_W, _AP_S, 0>(V) || r; + break; + case AP_TRN_ZERO: + qb = s && ( qb || r ); + break; + default:; + + } + if (qb) ++V; + //only when old V[_AP_W-1]==1 && new V[_AP_W-1]==0 + return carry && !(ap_private_ops::get<_AP_W, _AP_S, _AP_W-1>(V)); //(!V[_AP_W-1]); + } + + template + struct RType { + enum { + _AP_F=_AP_W-_AP_I, + F2=_AP_W2-_AP_I2, + mult_w = _AP_W+_AP_W2, + mult_i = _AP_I+_AP_I2, + mult_s = _AP_S||_AP_S2, + plus_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1+AP_MAX(_AP_F,F2), + plus_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1, + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1+AP_MAX(_AP_F,F2), + minus_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1, + minus_s = true, +#ifndef __SC_COMPATIBLE__ + div_w = _AP_W + AP_MAX(_AP_W2 - _AP_I2, 0) + _AP_S2, +#else + div_w = _AP_W + AP_MAX(_AP_W2 - _AP_I2, 0) + _AP_S2 + AP_MAX(_AP_I2, 0), +#endif /* #ifndef __SC_COMPATIBLE__ */ + div_i = _AP_I + (_AP_W2-_AP_I2) + _AP_S2, + div_s = _AP_S||_AP_S2, + logic_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+AP_MAX(_AP_F,F2), + logic_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + + typedef ap_fixed_base mult; + typedef ap_fixed_base plus; + typedef ap_fixed_base minus; + typedef ap_fixed_base logic; + typedef ap_fixed_base div; + typedef ap_fixed_base<_AP_W, _AP_I, _AP_S> arg1; + }; + INLINE void report() { +#if 0 + if (_AP_W > 1024 && _AP_W <= 4096) { + fprintf(stderr, "[W] W=%d is out of bound (1<=W<=1024):" + " for synthesis, please define macro AP_INT_TYPE_EXT(N) to" + " extend the valid range.\n", _AP_W); + } else +#endif /* #if 0 */ + if (_AP_W > MAX_MODE(AP_INT_MAX_W) * 1024) { + fprintf(stderr, "[E] ap_%sfixed<%d, ...>: Bitwidth exceeds the " + "default max value %d. Please use macro " + "AP_INT_MAX_W to set a larger max value.\n", + _AP_S?"":"u", _AP_W, + MAX_MODE(AP_INT_MAX_W) * 1024); + exit(1); + } + } + + /// Constructors. + // ------------------------------------------------------------------------- +#if 0 + #ifdef __SC_COMPATIBLE__ + INLINE ap_fixed_base():V(uint32_t(_AP_W), uint64_t(0)) {} + #else + INLINE ap_fixed_base():V(uint32_t(_AP_W)) {} + #endif /* #ifdef __SC_COMPATIBLE__ */ +#else + INLINE ap_fixed_base():V(0) {} +#endif /* #if 0 */ + // INLINE ap_fixed_base():V() {} + // INLINE explicit ap_fixed_base(const ap_private<_AP_W+_AP_I, _AP_S>& _V):V(_V) {} + INLINE ap_fixed_base(const ap_fixed_base& op):V(op.V) {} + template + INLINE ap_fixed_base(const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op):V(0) { + enum {N2=_AP_W2,_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2,QUAN_INC=F2>_AP_F && !(_AP_Q==AP_TRN || + (_AP_Q==AP_TRN_ZERO && !_AP_S2))}; + if (!op) return; + bool carry=false; + //handle quantization + enum { sh_amt =(F2>_AP_F)?F2-_AP_F:_AP_F-F2}; + const ap_private<_AP_W2, _AP_S2>& val = op.V; + bool neg_src=val.isNegative(); + if (F2==_AP_F) + V=val; + + else if (F2>_AP_F) { + if (sh_amt >= _AP_W2) + V = neg_src ? -1 : 0; + else + V = _AP_S2?val.ashr(sh_amt):val.lshr(sh_amt); + if (_AP_Q!=AP_TRN && !(_AP_Q==AP_TRN_ZERO && !_AP_S2)) { + bool qb = false; + if (F2-_AP_F>_AP_W2) + qb = neg_src; + else + qb = ap_private_ops::get<_AP_W2, _AP_S2, F2-_AP_F-1>(val); + + bool r=false; + enum { pos3 = F2-_AP_F-2}; + if (pos3>=_AP_W2-1) + r=val!=0; + else if (pos3>=0) + r = (val<<(_AP_W2-1-pos3))!=0; + carry = quantization_adjust(qb,r,neg_src); + } + } else { //no quantization + if (sh_amt < _AP_W) { + V=val; + V <<= sh_amt; + } + } + //hanle overflow/underflow + if ((_AP_O!=AP_WRAP || _AP_N != 0) && + ((!_AP_S && _AP_S2) || _AP_I-_AP_S < + _AP_I2 - _AP_S2 + (QUAN_INC|| (_AP_S2 && + _AP_O==AP_SAT_SYM)))) {//saturation + bool deleted_zeros = _AP_S2?true:!carry, + deleted_ones = true; + bool lD=(_AP_I2>_AP_I && _AP_W2-_AP_I2+_AP_I>=0) && + ap_private_ops::get<_AP_W2, _AP_S2, _AP_W2-_AP_I2+_AP_I>(val); + enum { pos1=F2-_AP_F+_AP_W, pos2=F2-_AP_F+_AP_W+1}; + if (pos1 < _AP_W2) { + bool Range1_all_ones= true; + bool Range1_all_zeros= true; + if (pos1 >= 0) { + enum { __W = (_AP_W2-pos1) > 0 ? (_AP_W2-pos1) : 1 }; + const ap_private<__W, _AP_S2> Range1=ap_private<__W, _AP_S2>(val.lshr(pos1)); + Range1_all_ones=Range1.isAllOnesValue(); + Range1_all_zeros=Range1.isMinValue(); + } else { + Range1_all_ones=false; + Range1_all_zeros=val.isMinValue(); + } + bool Range2_all_ones=true; + if (pos2<_AP_W2 && pos2>=0) { + enum { __W = (_AP_W2-pos2)>0 ? (_AP_W2-pos2) : 1}; + ap_private<__W, true> Range2=ap_private<__W, true>(val.lshr(pos2)); + Range2_all_ones=Range2.isAllOnesValue(); + } else if (pos2<0) + Range2_all_ones=false; + + deleted_zeros=deleted_zeros && (carry?Range1_all_ones:Range1_all_zeros); + deleted_ones=carry?Range2_all_ones&&(F2-_AP_F+_AP_W<0||!lD) + :Range1_all_ones; + neg_src= neg_src&&!(carry && Range1_all_ones); + } else + neg_src = neg_src && V[_AP_W-1]; + + bool neg_trg= V.isNegative(); + bool overflow=(neg_trg||!deleted_zeros) && !val.isNegative(); + bool underflow=(!neg_trg||!deleted_ones)&&neg_src; + //printf("neg_src = %d, neg_trg = %d, deleted_zeros = %d, + // deleted_ones = %d, overflow = %d, underflow = %d\n", + // neg_src, neg_trg, deleted_zeros, deleted_ones, + // overflow, underflow); + if (_AP_O==AP_SAT_SYM && _AP_S2 && _AP_S) + underflow |= neg_src && (_AP_W>1?V.isMinSignedValue():true); + overflow_adjust(underflow, overflow, lD, neg_src); + } + report(); + } + + template + INLINE ap_fixed_base(const volatile ap_fixed_base<_AP_W2,_AP_I2, + _AP_S2,_AP_Q2,_AP_O2, _AP_N2> &op) : V(op.V) { + *this = const_cast&>(op); + } + + template + INLINE ap_fixed_base(const ap_private<_AP_W2,_AP_S2>& op) { + ap_fixed_base<_AP_W2,_AP_W2,_AP_S2> f_op; + f_op.V=op; + *this = f_op; + } + + INLINE ap_fixed_base(bool b) { + *this=(ap_private<1,false>)b; + report(); + } + + INLINE ap_fixed_base(char b) { + *this=(ap_private<8,false>)b; + report(); + } + + INLINE ap_fixed_base(signed char b) { + *this=(ap_private<8,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned char b) { + *this=(ap_private<8,false>)b; + report(); + } + + INLINE ap_fixed_base(signed short b) { + *this=(ap_private<16,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned short b) { + *this=(ap_private<16,false>)b; + report(); + } + + INLINE ap_fixed_base(signed int b) { + *this=(ap_private<32,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned int b) { + *this=(ap_private<32,false>)b; + report(); + } +# if defined __x86_64__ + INLINE ap_fixed_base(signed long b) { + *this=(ap_private<64,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned long b) { + *this=(ap_private<64,false>)b; + report(); + } +# else + INLINE ap_fixed_base(signed long b) { + *this=(ap_private<32,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned long b) { + *this=(ap_private<32,false>)b; + report(); + } +# endif + + INLINE ap_fixed_base(ap_slong b) { + *this=(ap_private<64,true>)b; + report(); + } + + INLINE ap_fixed_base(ap_ulong b) { + *this=(ap_private<64,false>)b; + report(); + } + +#if 1 + INLINE ap_fixed_base(const char* val):V(0) { + ap_private<_AP_W, _AP_S> Tmp(val); + V = Tmp; + } + + INLINE ap_fixed_base(const char* val, signed char rd): V(0) { + ap_private<_AP_W, _AP_S> Tmp(val, rd); + V = Tmp; + } + +#endif + + INLINE ap_fixed_base(const std::string& val) { + ap_private<_AP_W, _AP_S> Tmp(val, 2); + V = Tmp; + report(); + } + + template + INLINE ap_fixed_base(const ap_bit_ref<_AP_W2, _AP_S2>& op) { + *this = ((bool)op); + report(); + } + + template + INLINE ap_fixed_base(const ap_range_ref<_AP_W2, _AP_S2>& op) { + *this = ap_private<_AP_W2, _AP_S2>(op); + report(); + } + + template + INLINE ap_fixed_base(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op) { + *this = ((const ap_private<_AP_W2 + _AP_W3, false>&)(op)); + report(); + } + + template + INLINE ap_fixed_base(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + *this = (bool(op)); + report(); + } + + template + INLINE ap_fixed_base(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + *this = (ap_private<_AP_W2, false>(op)); + report(); + } + + //helper function + INLINE unsigned long long doubleToRawBits(double pf)const { + union { + unsigned long long __L; + double __D; + }LD; + LD.__D=pf; + return LD.__L; + } + + + INLINE double rawBitsToDouble(unsigned long long pi) const { + union { + unsigned long long __L; + double __D; + }LD; + LD.__L=pi; + return LD.__D; + } + + INLINE float rawBitsToFloat(uint32_t pi) const { + union { + uint32_t __L; + float __D; + }LD; + LD.__L = pi; + return LD.__D; + } + + INLINE ap_fixed_base(double d):V(0) { + if (!d) return; + const bool isneg=d<0; + + const uint64_t ireg=doubleToRawBits(isneg?-d:d); + if ((ireg&0x7fffffffffffffffULL)!=0) { + const int32_t exp=(((ireg)>>DOUBLE_MAN)&0x07ff)-DOUBLE_BIAS; + ap_private man = ireg & DOUBLE_MAN_MASK; + man.clear(DOUBLE_MAN+1); + man.set(DOUBLE_MAN); + if (isneg) { + man.flip(); + man++; + } + + enum {_AP_S2=true, _AP_W2=DOUBLE_MAN+2,_AP_F=_AP_W -_AP_I }; + const int _AP_I2=exp+2; + const int F2=_AP_W2-_AP_I2; + const bool QUAN_INC=F2>_AP_F && !(_AP_Q==AP_TRN || (_AP_Q==AP_TRN_ZERO && + !_AP_S2)); + bool carry=false; + //handle quantization + const unsigned sh_amt=abs(F2-_AP_F); // sh_amt = F2>_AP_F ? F2 -_AP_F : _AP_F-F2; + if (F2==_AP_F ) + V=man; + else if (F2>_AP_F) { + if (sh_amt >= DOUBLE_MAN+2) + V=isneg?-1:0; + else + V=(man>>sh_amt) | ((man & 1ULL<<(DOUBLE_MAN+1)) ? (DOUBLE_MAN_MASK>>(DOUBLE_MAN+2-sh_amt) <<(DOUBLE_MAN+2-sh_amt)):0); + + if (_AP_Q!=AP_TRN && !(_AP_Q==AP_TRN_ZERO && !_AP_S2)) { + const bool qb=((F2-_AP_F > DOUBLE_MAN+2) ? isneg : (man & (1ULL<<(F2-_AP_F-1))) != 0); + const int pos3=F2-_AP_F-2; + const bool r = (pos3>= 0) ? (man << AP_MAX(0, _AP_W2-pos3-1)& DOUBLE_MAN_MASK)!=0 : false; + carry = quantization_adjust(qb,r,isneg); + } + } + else { //no quantization + // V=man; + if (sh_amt < _AP_W) { + V = man; + V <<= sh_amt; + } + } + //handle overflow/underflow + if ((_AP_O != AP_WRAP || _AP_N != 0) && + ((!_AP_S && _AP_S2) || _AP_I-_AP_S < + _AP_I2-_AP_S2+(QUAN_INC|| (_AP_S2 && + _AP_O==AP_SAT_SYM)) )) {// saturation + bool deleted_zeros = _AP_S2?true:!carry, + deleted_ones = true; + bool neg_src; + const bool lD=(_AP_I2>_AP_I) && (_AP_W2-_AP_I2+_AP_I>=0) && (man & (1ULL <<(DOUBLE_MAN+2-_AP_I2+_AP_I))); + int pos1=F2+_AP_W-_AP_F; + if (pos1 < _AP_W2) { + int pos2=pos1+1; + bool Range1_all_ones=true; + bool Range1_all_zeros=true; + if (pos1>=0) { + ap_private<_AP_W,_AP_S> Range1= + ap_private<_AP_W,_AP_S>((man >> pos1) | ((1ULL<<(DOUBLE_MAN+1)&man) ? (DOUBLE_MAN_MASK >> (DOUBLE_MAN+2-pos1) <<(DOUBLE_MAN+2-pos1)):0)); + Range1_all_ones = Range1.isAllOnesValue(); // Range1.isAllOnesValue(); + Range1_all_zeros = Range1.isMinValue(); // Range1.isMinValue(); + } else { + Range1_all_ones=false; + Range1_all_zeros = man==0; // man.isMinValue(); + } + bool Range2_all_ones=true; + if (pos2<_AP_W2 && pos2>=0) { + ap_private<_AP_W, _AP_S> Range2= + ap_private<_AP_W, _AP_S>((man >> pos2) | ((1ULL<<(DOUBLE_MAN+1)&man) ? (DOUBLE_MAN_MASK >> (DOUBLE_MAN+2-pos2) <<(DOUBLE_MAN+2-pos2)):0)); + Range2_all_ones=Range2.isAllOnesValue(); // Range2.isAllOnesValue(); + } else if (pos2<0) + Range2_all_ones=false; + deleted_zeros=deleted_zeros && (carry?Range1_all_ones:Range1_all_zeros); + deleted_ones=carry?Range2_all_ones&&(F2-_AP_F+_AP_W<0||!lD) : Range1_all_ones; + neg_src=isneg&&!(carry&Range1_all_ones); + } else + neg_src = isneg && V[_AP_W -1]; + + const bool neg_trg=V.isNegative(); + const bool overflow=(neg_trg||!deleted_zeros) && !isneg; + bool underflow=(!neg_trg||!deleted_ones)&&neg_src; + //printf("neg_src = %d, neg_trg = %d, deleted_zeros = %d, + // deleted_ones = %d, overflow = %d, underflow = %d\n", + // neg_src, neg_trg, deleted_zeros, deleted_ones, + // overflow, underflow); + if (_AP_O==AP_SAT_SYM && _AP_S2 && _AP_S) + underflow |= neg_src && (_AP_W>1?V.isMinSignedValue():true); + overflow_adjust(underflow,overflow,lD, neg_src); + } + } + report(); + } + + + ///assign operators + //------------------------------------------------------------------------- + + INLINE volatile ap_fixed_base& operator=(const ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) volatile { + V = op.V; + return *this; + } + + INLINE ap_fixed_base& operator=(const ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) { + V = op.V; + return *this; + } + + INLINE volatile ap_fixed_base& operator=(const volatile ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) volatile { + V = op.V; + return *this; + } + + INLINE ap_fixed_base& operator=(const volatile ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) { + V = op.V; + return *this; + } + + // Set this ap_fixed_base with a bits string. That means the ssdm_int::V + // inside this ap_fixed_base is assigned by bv. + // Note the input parameter should be a fixed-point formatted bit string. + INLINE ap_fixed_base& setBits(unsigned long long bv) { + V=bv; + return *this; + } + + // Return a ap_fixed_base object whose ssdm_int::V is assigned by bv. + // Note the input parameter should be a fixed-point formatted bit string. + static INLINE ap_fixed_base bitsToFixed(unsigned long long bv) { + ap_fixed_base Tmp=bv; + return Tmp; + } + + // Explicit conversion functions to ap_private that captures + // all integer bits (bits are truncated) + INLINE ap_private + to_ap_private(bool Cnative = true) const { + ap_private ret = ap_private ((_AP_I >= 1) ? (_AP_S==true ? V.ashr(AP_MAX(0,_AP_W - _AP_I)) : V.lshr(AP_MAX(0,_AP_W - _AP_I))) : ap_private<_AP_W, _AP_S>(0)); + + if (Cnative) { + bool r = false; + if (_AP_I < _AP_W) { + if (_AP_I > 0) r = !(V.getLoBits(_AP_W - _AP_I).isMinValue()); + else r = !(V.isMinValue()); + } + if (r && V.isNegative()) { // if this is negative integer + ++ret;//ap_private(1,_AP_S); + } + } else { + //Follow OSCI library, conversion from sc_fixed to sc_int + } + return ret; + } + + template + INLINE operator ap_private<_AP_W2,_AP_S2> () const { + return (ap_private<_AP_W2,_AP_S2>)to_ap_private(); + } + + template + INLINE operator ap_private<_AP_W2,_AP_S2,_AP_N2> () const { + return (ap_private<_AP_W2,_AP_S2,_AP_N2>)to_ap_private(); + } + + //Explict conversion function to C built-in integral type + INLINE int to_int() const { + return to_ap_private().to_int(); + } + + INLINE int to_uint() const { + return to_ap_private().to_uint(); + } + + INLINE ap_slong to_int64() const { + return to_ap_private().to_int64(); + } + + INLINE ap_ulong to_uint64() const { + return to_ap_private().to_uint64(); + } + + INLINE double to_double() const { + if (!V) + return 0; + if (_AP_W>64 || (_AP_W - _AP_I) > 0) { + bool isneg = _AP_S && V[_AP_W-1]; + uint64_t res = isneg ? 0x8000000000000000ULL : 0; + ap_private<_AP_W, false> tmp = V; + if (isneg) tmp = -tmp; + int i = _AP_W -1 - tmp.countLeadingZeros(); + int exp = _AP_I-(_AP_W-i); + res|=((uint64_t)(exp+DOUBLE_BIAS))<DOUBLE_MAN)?tmp.lshr(i-DOUBLE_MAN):tmp).to_uint64() & DOUBLE_MAN_MASK; + res |= i 0) { + /* This specialization is disabled. It is giving wrong results in some cases. + bool isneg=V.isNegative(); + double dp = V.get(); + dp /= (1<< (_AP_W - _AP_I)); + return dp;*/ + } else + return double(to_int64()); + } + + INLINE float to_float() const { + uint32_t res=0; + if (V==0) + return 0; + bool isneg=V.isNegative(); + ap_private<_AP_W, _AP_S> tmp=V; + if (isneg) tmp = -tmp; + if (_AP_W-_AP_I>0||_AP_W>64) { + if (isneg) + res=0x80000000; + int i=_AP_W-1; + i-=tmp.countLeadingZeros(); + int exp=_AP_I-(_AP_W-i); + res|=(exp+FLOAT_BIAS)< man = 0; + if (i!=0) { + tmp.clear(i); + if (i>FLOAT_MAN) + man=tmp.lshr(i-FLOAT_MAN); + else + man=tmp; + res |= i < FLOAT_MAN?man.getZExtValue()<<(FLOAT_MAN-i):man.getZExtValue(); + } + } else { + return float(to_int64()); + } + float dp=rawBitsToFloat(res); + return dp; + } + + INLINE operator double () const { + return to_double(); + } +#ifndef __SC_COMPATIBLE__ + INLINE operator float () const { + return to_float(); + } + + INLINE operator char () const { + return (char) to_int(); + } + + INLINE operator unsigned char () const { + return (unsigned char) to_uint(); + } + + INLINE operator short () const { + return (short) to_int(); + } + + INLINE operator unsigned short () const { + return (unsigned short) to_uint(); + } + + INLINE operator int () const { + return to_int(); + } + + INLINE operator unsigned int () const { + return to_uint(); + } +#if 1 +#ifdef __x86_64__ + INLINE operator long () const { + return (long)to_int64(); + } + + INLINE operator unsigned long () const { + return (unsigned long) to_uint64(); + } +#else + INLINE operator long () const { + return to_int64(); + } + + INLINE operator unsigned long () const { + return to_uint64(); + } + +#endif +#endif + INLINE operator unsigned long long () const { + return to_uint64(); + } + + INLINE operator long long () const { + return to_int64(); + } +#endif + + INLINE std::string to_string(uint8_t radix=2, bool sign=false) const; + + INLINE ap_slong bits_to_int64() const { + ap_private res(V); + return (ap_slong) res; + } + + INLINE ap_ulong bits_to_uint64() const { + ap_private res(V); + return (ap_ulong) res; + } + + INLINE int length() const {return _AP_W;} + + // Count the number of zeros from the most significant bit + // to the first one bit. Note this is only for ap_fixed_base whose + // _AP_W <= 64, otherwise will incur assertion. + INLINE int countLeadingZeros() { + return V.countLeadingZeros(); + } + + ///Arithmetic:Binary + //------------------------------------------------------------------------- + template + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::mult + operator * (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + typename RType<_AP_W2,_AP_I2,_AP_S2>::mult r; + r.V = V * op2.V; + return r; + } + + template + static INLINE ap_fixed_base multiply(const ap_fixed_base<_AP_W1,_AP_I1,_AP_S1>& op1, const + ap_fixed_base<_AP_W2,_AP_I2,_AP_S2>& op2) { + ap_private<_AP_W+_AP_W2, _AP_S> OP1=op1.V; + ap_private<_AP_W2,_AP_S2> OP2=op2.V; + return OP1*OP2; + } + + template + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::div + operator / (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {F2 = _AP_W2-_AP_I2, _W1=AP_MAX(_AP_W + AP_MAX(F2, 0), _AP_W2), + _W2=AP_MAX(_AP_W2,AP_MAX(_AP_W + AP_MAX(F2, 0), _AP_W2))}; + ap_private<_W1, _AP_S> dividend = (ap_private<_W1, _AP_S>(V)) << ((_W1>_AP_W)?F2:0); + ap_private<_W1, _AP_S2> divisior = ap_private<_W2, _AP_S2>(op2.V); + ap_private<_W1, _AP_S> ret = ap_private<_W1,_AP_S> ((_AP_S||_AP_S2) ? dividend.sdiv(divisior): dividend.udiv(divisior)); + typename RType<_AP_W2, _AP_I2, _AP_S2>::div r; + r.V = ret; + return r; + } +#define OP_BIN_AF(Sym, Rty, Width, Sign, Fun) \ + template \ + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty \ + operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + enum {_AP_F=_AP_W-_AP_I, F2=_AP_W2-_AP_I2}; \ + typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty r, lhs(*this), rhs(op2); \ + r.V = lhs.V.Fun(rhs.V); \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty \ + operator Sym (const ap_fixed_base& op2) const \ + { \ + typename RType<_AP_W,_AP_I,_AP_S>::Rty r; \ + r.V = V Sym op2.V; \ + return r; \ + } \ + + OP_BIN_AF(+, plus, plus_w, plus_s, Add) + OP_BIN_AF(-, minus, minus_w, minus_s, Sub) + +#define OP_LOGIC_BIN_AF(Sym, Rty, Width, Sign) \ + template \ + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty \ + operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty r, lhs(*this), rhs(op2); \ + r.V=lhs.V Sym rhs.V; \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty \ + operator Sym (const ap_fixed_base& op2) const \ + { \ + typename RType<_AP_W,_AP_I,_AP_S>::Rty r; \ + r.V = V Sym op2.V; \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty operator Sym(int op2) const \ + { \ + return V Sym (op2<<(_AP_W - _AP_I)); \ + } + OP_LOGIC_BIN_AF(&, logic, logic_w, logic_s) + OP_LOGIC_BIN_AF(|, logic, logic_w, logic_s) + OP_LOGIC_BIN_AF(^, logic, logic_w, logic_s) + + ///Arithmic : assign + //------------------------------------------------------------------------- +#define OP_ASSIGN_AF(Sym) \ + template \ + INLINE ap_fixed_base& operator Sym##= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) \ + { \ + *this=operator Sym (op2) ; \ + return *this; \ + } + + OP_ASSIGN_AF(+) + OP_ASSIGN_AF(-) + OP_ASSIGN_AF(&) + OP_ASSIGN_AF(|) + OP_ASSIGN_AF(^) + OP_ASSIGN_AF(*) + OP_ASSIGN_AF(/) + + ///Prefix increment, decrement + //------------------------------------------------------------------------- + INLINE ap_fixed_base& operator ++() { + operator+=(ap_fixed_base<1,1,false>(1)); //SystemC's semantics + return *this; + } + + INLINE ap_fixed_base& operator --() { + operator-=(ap_fixed_base<1,1,false>(1)); //SystemC's semantics + return *this; + } + + //Postfix increment, decrement + //------------------------------------------------------------------------- + INLINE const ap_fixed_base operator ++(int) { + ap_fixed_base t(*this); + operator++(); + return t; + } + + INLINE const ap_fixed_base operator --(int) { + ap_fixed_base t = *this; + operator--(); + return t; + } + + ///Unary arithmetic + //------------------------------------------------------------------------- + INLINE ap_fixed_base operator +() {return *this;} + + INLINE ap_fixed_base<_AP_W + 1, _AP_I + 1, true> operator -() const { + ap_fixed_base<_AP_W + 1, _AP_I + 1, true> Tmp(*this); + Tmp.V = - Tmp.V; + return Tmp; + } + + INLINE ap_fixed_base<_AP_W,_AP_I,true,_AP_Q,_AP_O, _AP_N> getNeg() { + ap_fixed_base<_AP_W,_AP_I,true,_AP_Q,_AP_O, _AP_N> Tmp(*this); + Tmp.V=-Tmp.V; + return Tmp; + } + + ///Not (!) + //------------------------------------------------------------------------- + INLINE bool operator !() const { + return !V; + } + + ///Bitwise complement + //------------------------------------------------------------------------- + INLINE ap_fixed_base<_AP_W, _AP_I, _AP_S> + operator ~() const { + ap_fixed_base<_AP_W, _AP_I, _AP_S> res(*this); + res.V.flip(); + return res; + } + + ///Shift + ///template argument as shift value + template + INLINE ap_fixed_base<_AP_W, _AP_I + _AP_SHIFT, _AP_S> lshift () const { + ap_fixed_base<_AP_W, _AP_I + _AP_SHIFT, _AP_S> r; + r.V = V; + return r; + } + + template + INLINE ap_fixed_base<_AP_W, _AP_I - _AP_SHIFT, _AP_S> rshift () const { + ap_fixed_base<_AP_W, _AP_I - _AP_SHIFT, _AP_S> r; + r.V = V; + return r; + } + + //Because the return type is the type of the the first operand, shift assign + //operators do not carry out any quantization or overflow + //While systemc, shift assigns for sc_fixed/sc_ufixed will result in + //quantization or overflow (depending on the mode of the first operand) + //------------------------------------------------------------------------- + INLINE ap_fixed_base operator << (int sh) const { + ap_fixed_base r; + bool isNeg=(sh&0x80000000) != 0; + sh=isNeg?-sh:sh; + bool shiftoverflow = sh >= _AP_W; + bool NegSrc = V.isNegative(); + if (isNeg) { + if (shiftoverflow) + NegSrc?r.V.set():r.V.clear(); + else + r.V=_AP_S?V.ashr(sh):V.lshr(sh); + } else { + if (shiftoverflow) + r.V.clear(); + else + r.V=V< 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } else if (isNeg == false && _AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh - 1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator<<(const ap_private<_AP_W2,true>& op2) const { + int sh = op2.to_int(); + return operator << (sh); + } + + INLINE ap_fixed_base operator << (unsigned int sh ) const { + ap_fixed_base r; + bool shiftoverflow = sh >= _AP_W; + r.V = shiftoverflow ? ap_private<_AP_W, _AP_S >(0) : V << sh; + if (sh == 0) return r; +#ifdef __SC_COMPATIBLE__ + bool NegSrc = V.isNegative(); + if (_AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh -1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator << (const ap_private<_AP_W2,false>& op2) const { + unsigned int sh = op2.to_uint(); + return operator << (sh); + } + + INLINE ap_fixed_base operator >> (int sh) const { + ap_fixed_base r; + bool isNeg=(sh&0x80000000) != 0; + bool NegSrc = V.isNegative(); + sh=isNeg?-sh:sh; + bool shiftoverflow = sh >= _AP_W; + if (isNeg && !shiftoverflow) r.V=V< 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } else if (isNeg == true && _AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh - 1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator >> (const ap_private<_AP_W2,true>& op2) const { + int sh = op2.to_int(); + return operator >> (sh); + } + + INLINE ap_fixed_base operator >> (unsigned int sh) const { + ap_fixed_base r; + bool NegSrc = V.isNegative(); + bool shiftoverflow = sh >= _AP_W; + if (shiftoverflow) + NegSrc?r.V.set():r.V.clear(); + else + r.V=_AP_S?V.ashr(sh):V.lshr(sh); +#ifdef __SC_COMPATIBLE__ + if (sh == 0) return r; + if (_AP_Q != AP_TRN) { + bool qb = false; + if (sh <= _AP_W) qb = V[sh - 1]; + bool rb = false; + if (sh > 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator >> (const ap_private<_AP_W2,false>& op2) const { + unsigned int sh = op2.to_uint(); + return operator >> (sh); + } + + ///shift assign + //------------------------------------------------------------------------- +#define OP_AP_SHIFT_AP_ASSIGN_AF(Sym) \ + template \ + INLINE ap_fixed_base& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op2) \ + { \ + *this=operator Sym (op2); \ + return *this; \ + } + + OP_AP_SHIFT_AP_ASSIGN_AF(<<) + OP_AP_SHIFT_AP_ASSIGN_AF(>>) + + ///Support shift(ap_fixed_base) +#define OP_AP_SHIFT_AF(Sym) \ + template \ + INLINE ap_fixed_base operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + return operator Sym (op2.to_ap_private()); \ + } \ + template \ + INLINE ap_fixed_base& operator Sym##= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) \ + { \ + *this=operator Sym (op2); \ + return *this; \ + } + + OP_AP_SHIFT_AF(<<) + OP_AP_SHIFT_AF(>>) + + INLINE ap_fixed_base& operator >>= (unsigned int sh) { + *this = operator >> (sh); + return *this; + } + + INLINE ap_fixed_base& operator <<= (unsigned int sh) { + *this = operator << (sh); + return *this; + } + + INLINE ap_fixed_base& operator >>= (int sh) { + *this = operator >> (sh); + return *this; + } + + INLINE ap_fixed_base& operator <<= (int sh) { + *this = operator << (sh); + return *this; + } + + ///Comparisons + //------------------------------------------------------------------------- + template + INLINE bool operator == (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator != (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this==op2); + } + + template + INLINE bool operator > (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator <= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this>op2); + } + + template + INLINE bool operator < (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator >= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this) + DOUBLE_CMP_AF(>=) + DOUBLE_CMP_AF(<) + DOUBLE_CMP_AF(<=) + + // Bit and Slice Select + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> operator [] (unsigned int index) { + assert(index<_AP_W&&"Attemping to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index); + } + + INLINE af_bit_ref<_AP_W, _AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> bit(unsigned int index) { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index); + } + + template + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> bit (const ap_private<_AP_W2,_AP_S2>& index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index.to_int()); + } + + INLINE bool bit (unsigned int index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index]; + } + + INLINE bool operator [] (unsigned int index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index]; + } + + template + INLINE bool bit (const ap_private<_AP_W2, _AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index.to_uint()]; + } + + template + INLINE bool operator [] (const ap_private<_AP_W2, _AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index.to_uint()]; + } + + INLINE af_bit_ref<_AP_W, _AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> get_bit(int index) { + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + assert(index >= _AP_I - _AP_W&& "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index + _AP_W - _AP_I); + } + + template + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> get_bit (const ap_private<_AP_W2, true>& index) { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index.to_int() + _AP_W - _AP_I); + } + + INLINE bool get_bit (int index) const { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return V[index + _AP_W - _AP_I]; + } + + template + INLINE bool get_bit (const ap_private<_AP_W2, true>& index) const { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return V[index.to_int() + _AP_W - _AP_I]; + } + + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range(int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W) &&"Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(const_cast(this), Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + operator () (int Hi, int Lo) const { + return this->range(Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(const_cast< + ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>*>(this), + Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range() { + return this->range(_AP_W - 1, 0); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range() const { + return this->range(_AP_W - 1, 0); + } + + INLINE bool is_zero () const { + return V.isMinValue(); + } + + INLINE bool is_neg () const { + if (V.isNegative()) + return true; + return false; + } + + INLINE int wl () const { + return _AP_W; + } + + INLINE int iwl () const { + return _AP_I; + } + + INLINE ap_q_mode q_mode () const { + return _AP_Q; + } + + INLINE ap_o_mode o_mode () const { + return _AP_O; + } + + INLINE int n_bits () const { + return 0; + } + + //private: +public: + ap_private<_AP_W, _AP_S> V; +}; + +template +std::string ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>::to_string( + uint8_t radix, bool sign) const { + std::string str; + str.clear(); + char step; + std::string prefix; + switch (radix) { + case 2 : prefix = "0b"; step = 1; break; + case 8 : prefix = "0o"; step = 3; break; + case 16 : prefix = "0x"; step = 4; break; + default : break; + } + if (_AP_W <= _AP_I) + str = this->to_ap_private().to_string(radix, + radix == 10 ? _AP_S : sign); + else { + if (radix == 10) { + bool isNeg = _AP_S && V.isNegative(); + if (_AP_I > 0) { + ap_private int_part(0); + int_part = this->to_ap_private(); + str += int_part.to_string(radix, false); + } else { + if (isNeg) str += '-'; + } + ap_fixed_base<_AP_W, _AP_I, _AP_S> tmp(*this); + if (isNeg && _AP_I <= 0) tmp = -tmp; + ap_fixed_base<_AP_W - AP_MIN(_AP_I, 0), 0, false> frac_part = tmp; + if (frac_part == 0) return str; + str += "."; + while (frac_part != 0) { + char digit = (frac_part * radix).to_ap_private(); + str += static_cast(digit + '0'); + frac_part *= radix; + } + } else { + if (_AP_I > 0) { + for (signed i = _AP_W - _AP_I; i < _AP_W; i += step) { + + char digit = (char)(this->range(AP_MIN(i + step - 1, _AP_W - 1), i)); + str = (digit < 10 ? static_cast(digit + '0') : + static_cast(digit - 10 + 'a')) + str; + } + } + str += '.'; + ap_fixed_base tmp(*this); + for (signed i = _AP_W - _AP_I - 1; i >= 0; i -= step) { + char digit = (char)(tmp.range(i, AP_MAX(0, i - step + 1))); + str += digit < 10 ? static_cast(digit + '0') : + static_cast(digit - 10 + 'a'); + } + } + } + str = prefix + str; + return str; +} + +template +INLINE void b_not(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op) { + ret.V = op.V; + ret.V.flip(); +} + +template +INLINE void b_and(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V & op2.V; +} + +template +INLINE void b_or(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V | op2.V; +} + +template +INLINE void b_xor(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V ^ op2.V; +} + +template +INLINE void neg(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + ap_fixed_base<_AP_W2+!_AP_S2, _AP_I2+!_AP_S2, true, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp.V = - op.V; + ret = Tmp; +} + +template +INLINE void neg(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op) { + ret.V = -op.V; +} + +template +INLINE void lshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op, + int i) { + ap_fixed_base<_AP_W2 - _AP_I2 + AP_MAX(_AP_I, _AP_I2), AP_MAX(_AP_I, _AP_I2), _AP_S2, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp = op; + Tmp.V <<= i; + ret = Tmp; +} + +template +INLINE void lshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op, + int i) { + ret.V = op.V << i; +} + +template +INLINE void rshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op, + int i) { + ap_fixed_base<_AP_I2 + AP_MAX(_AP_W - _AP_I, _AP_W2 - _AP_I2), _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp = op; + Tmp.V = _AP_S2 ? Tmp.V.ashr(i): Tmp.V.lshr(i); + ret = Tmp; +} + +template +INLINE void rshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op, + int i) { + ret.V = _AP_S ? op.V.ashr(i): op.V.lshr(i); +} + +#define AF_CTOR_SPEC_BASE(_AP_W,_AP_S,C_TYPE) \ + template<> INLINE ap_fixed_base<_AP_W,_AP_W,_AP_S,AP_TRN,AP_WRAP>::ap_fixed_base(C_TYPE i_op):V(i_op) \ + { \ + } + +#define AF_CTOR_SPEC(__W,C_TYPE) \ + AF_CTOR_SPEC_BASE(__W,true,C_TYPE) \ + AF_CTOR_SPEC_BASE(__W,false,C_TYPE) + +AF_CTOR_SPEC(1,bool) +AF_CTOR_SPEC(8, signed char) +AF_CTOR_SPEC(8, unsigned char) +AF_CTOR_SPEC(16, signed short) +AF_CTOR_SPEC(16, unsigned short) +AF_CTOR_SPEC(32, signed int) +AF_CTOR_SPEC(32, unsigned int) +AF_CTOR_SPEC(64, ap_slong) +AF_CTOR_SPEC(64, ap_ulong) + +///Output streaming +//----------------------------------------------------------------------------- +template +INLINE std::ostream& +operator <<(std::ostream& os, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + os << x.to_double(); + return os; +} + +///Input streaming +//----------------------------------------------------------------------------- +template +INLINE std::istream& +operator >> (std::istream& os, ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + double d; + os >> d; + x = ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(x); + return os; +} + +template +INLINE void print(const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + ap_private<_AP_W,_AP_S> data=x.V; + if (_AP_I>0) { + const ap_private<_AP_I,_AP_S> p1=data>>(_AP_W-_AP_I); + print(p1); + + } else + printf("0"); + printf("."); + if (_AP_I<_AP_W) { + const ap_private<_AP_W-_AP_I,false> p2=data; + print(p2,false); + } +} + +///Operators mixing Integers with ap_fixed_base +//----------------------------------------------------------------------------- +#if 1 +#define AF_BIN_OP_WITH_INT_SF(BIN_OP,C_TYPE,_AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator BIN_OP(ap_private<_AP_W2,_AP_S2>(i_op)); \ + } +#define AF_BIN_OP_WITH_INT(BIN_OP, C_TYPE, _AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator BIN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } + +#else +#define AF_BIN_OP_WITH_INT_SF(BIN_OP,C_TYPE,_AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op BIN_OP (i_op); \ + } +#define AF_BIN_OP_WITH_INT(BIN_OP, C_TYPE, _AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.V BIN_OP (i_op<<(_AP_W-_AP_I)); \ + } \ + \ + \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } + +#endif +#if 1 +#define AF_REL_OP_WITH_INT(REL_OP, C_TYPE, _AP_W2,_AP_S2) \ + template \ + INLINE bool operator REL_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator REL_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + \ + \ + template \ + INLINE bool operator REL_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator REL_OP (op); \ + } +#else +#define AF_REL_OP_WITH_INT(REL_OP, C_TYPE, _AP_W2,_AP_S2) \ + template \ + INLINE bool operator REL_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.V.operator REL_OP (i_op<<(_AP_W-_AP_I)); \ + } \ + \ + \ + template \ + INLINE bool operator REL_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return (i_op<<(_AP_W-_AP_I)) REL_OP (op.V.VAL); \ + } +#endif +#if 1 +#define AF_ASSIGN_OP_WITH_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.operator ASSIGN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } +#define AF_ASSIGN_OP_WITH_INT_SF(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.operator ASSIGN_OP (ap_private<_AP_W2,_AP_S2>(i_op)); \ + } +#else +#define AF_ASSIGN_OP_WITH_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.V.operator ASSIGN_OP (i_op); \ + } +#define AF_ASSIGN_OP_WITH_INT_SF(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.V.operator ASSIGN_OP (i_op); \ + } +#endif + +#define AF_OPS_WITH_INT(C_TYPE, WI, SI) \ + AF_BIN_OP_WITH_INT(+, C_TYPE, WI, SI, plus) \ + AF_BIN_OP_WITH_INT(-, C_TYPE, WI, SI, minus) \ + AF_BIN_OP_WITH_INT(*, C_TYPE, WI, SI, mult) \ + AF_BIN_OP_WITH_INT(/, C_TYPE, WI, SI, div) \ + AF_BIN_OP_WITH_INT_SF(>>, C_TYPE, WI, SI, arg1) \ + AF_BIN_OP_WITH_INT_SF(<<, C_TYPE, WI, SI, arg1) \ + AF_BIN_OP_WITH_INT(&, C_TYPE, WI, SI, logic) \ + AF_BIN_OP_WITH_INT(|, C_TYPE, WI, SI, logic) \ + AF_BIN_OP_WITH_INT(^, C_TYPE, WI, SI, logic) \ + \ + AF_REL_OP_WITH_INT(==, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(!=, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(>, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(>=, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(<, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(<=, C_TYPE, WI, SI) \ + \ + AF_ASSIGN_OP_WITH_INT(+=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(-=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(*=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(/=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT_SF(>>=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT_SF(<<=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(&=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(|=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(^=, C_TYPE, WI, SI) + +AF_OPS_WITH_INT(bool, 1, false) +AF_OPS_WITH_INT(char, 8, true) +AF_OPS_WITH_INT(signed char, 8, true) +AF_OPS_WITH_INT(unsigned char, 8, false) +AF_OPS_WITH_INT(short, 16, true) +AF_OPS_WITH_INT(unsigned short, 16, false) +AF_OPS_WITH_INT(int, 32, true) +AF_OPS_WITH_INT(unsigned int, 32, false) +# if defined __x86_64__ +AF_OPS_WITH_INT(long, 64, true) +AF_OPS_WITH_INT(unsigned long, 64, false) +# else +AF_OPS_WITH_INT(long, 32, true) +AF_OPS_WITH_INT(unsigned long, 32, false) +# endif +AF_OPS_WITH_INT(ap_slong, 64, true) +AF_OPS_WITH_INT(ap_ulong, 64, false) + +#define AF_BIN_OP_WITH_AP_INT(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>::template RType<_AP_W,_AP_I,_AP_S>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator BIN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } + +#define AF_REL_OP_WITH_AP_INT(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator REL_OP ( ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator REL_OP (op); \ + } + +#define AF_ASSIGN_OP_WITH_AP_INT(ASSIGN_OP) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator ASSIGN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + template \ + INLINE ap_private<_AP_W2,_AP_S2>& operator ASSIGN_OP ( ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return i_op.operator ASSIGN_OP (op.to_ap_private()); \ + } + +AF_BIN_OP_WITH_AP_INT(+, plus) +AF_BIN_OP_WITH_AP_INT(-, minus) +AF_BIN_OP_WITH_AP_INT(*, mult) +AF_BIN_OP_WITH_AP_INT(/, div) +AF_BIN_OP_WITH_AP_INT(&, logic) +AF_BIN_OP_WITH_AP_INT(|, logic) +AF_BIN_OP_WITH_AP_INT(^, logic) + +AF_REL_OP_WITH_AP_INT(==) +AF_REL_OP_WITH_AP_INT(!=) +AF_REL_OP_WITH_AP_INT(>) +AF_REL_OP_WITH_AP_INT(>=) +AF_REL_OP_WITH_AP_INT(<) +AF_REL_OP_WITH_AP_INT(<=) + +AF_ASSIGN_OP_WITH_AP_INT(+=) +AF_ASSIGN_OP_WITH_AP_INT(-=) +AF_ASSIGN_OP_WITH_AP_INT(*=) +AF_ASSIGN_OP_WITH_AP_INT(/=) +AF_ASSIGN_OP_WITH_AP_INT(&=) +AF_ASSIGN_OP_WITH_AP_INT(|=) +AF_ASSIGN_OP_WITH_AP_INT(^=) + +#define AF_REF_REL_OP_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ +template \ + INLINE bool operator REL_OP ( const af_range_ref<_AP_W,_AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ +template \ + INLINE bool operator REL_OP ( C_TYPE op2, const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ +template \ + INLINE bool operator REL_OP ( const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, C_TYPE op2) { \ + return (bool(op)) REL_OP op2; \ + } \ +template \ + INLINE bool operator REL_OP ( C_TYPE op2, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2 REL_OP (bool(op)); \ + } + +#define AF_REF_REL_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(>, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(<, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(>=, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(<=, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(==, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(!=, C_TYPE, _AP_WI, _AP_SI) + +AF_REF_REL_MIX_INT(bool, 1, false) +AF_REF_REL_MIX_INT(char, 8, true) +AF_REF_REL_MIX_INT(signed char, 8, true) +AF_REF_REL_MIX_INT(unsigned char, 8, false) +AF_REF_REL_MIX_INT(short, 16, true) +AF_REF_REL_MIX_INT(unsigned short, 16, false) +AF_REF_REL_MIX_INT(int, 32, true) +AF_REF_REL_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +AF_REF_REL_MIX_INT(long, 64, true) +AF_REF_REL_MIX_INT(unsigned long, 64, false) +# else +AF_REF_REL_MIX_INT(long, 32, true) +AF_REF_REL_MIX_INT(unsigned long, 32, false) +# endif +AF_REF_REL_MIX_INT(ap_slong, 64, true) +AF_REF_REL_MIX_INT(ap_ulong, 64, false) + +#define AF_REF_REL_OP_AP_INT(REL_OP) \ +template \ + INLINE bool operator REL_OP ( const af_range_ref<_AP_W,_AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, const ap_private<_AP_W2, _AP_S> &op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (op2); \ + } \ +template \ + INLINE bool operator REL_OP (const ap_private<_AP_W2, _AP_S2> &op2, const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2.operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ +template \ + INLINE bool operator REL_OP ( const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, const ap_private<_AP_W2, _AP_S2> &op2) { \ + return (ap_private<1, false>(op)).operator REL_OP (op2); \ + } \ +template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W2, _AP_S2> &op2, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2.operator REL_OP (ap_private<1,false>(op)); \ + } + +AF_REF_REL_OP_AP_INT(>) +AF_REF_REL_OP_AP_INT(<) +AF_REF_REL_OP_AP_INT(>=) +AF_REF_REL_OP_AP_INT(<=) +AF_REF_REL_OP_AP_INT(==) +AF_REF_REL_OP_AP_INT(!=) + +// Relational Operators with double +template +INLINE bool operator == ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator == (op1); +} + +template +INLINE bool operator != ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator != (op1); +} + +template +INLINE bool operator > ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator < (op1); +} + +template +INLINE bool operator >= ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator <= (op1); +} + +template +INLINE bool operator < ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator > (op1); +} + +template +INLINE bool operator <= ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator >= (op1); +} + +#endif /* #ifndef __AESL_GCC_AP_FIXED_H__ */ \ No newline at end of file diff --git a/hls/ZU3EG_test/etc/ap_int_sim.h b/hls/ZU3EG_test/etc/ap_int_sim.h new file mode 100644 index 0000000000000000000000000000000000000000..887ccd88bfd52d1777db8661d72c57703ccf736a --- /dev/null +++ b/hls/ZU3EG_test/etc/ap_int_sim.h @@ -0,0 +1,1629 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_GCC_AP_INT_H__ +#define __AESL_GCC_AP_INT_H__ + +#ifndef __cplusplus +#error C++ is required to include this header file +#endif /* #ifndef __cplusplus */ + +#undef _AP_DEBUG_ +#include +#include + +// for safety +#if (defined(_AP_N)|| defined(_AP_C)) +#error One or more of the following is defined: _AP_N, _AP_C. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_N)|| defined(_AP_C)) */ + +// for safety +#if (defined(_AP_W) || defined(_AP_I) || defined(_AP_S) || defined(_AP_Q) || defined(_AP_O) || defined(_AP_W2) || defined(_AP_I2) || defined(_AP_S2) || defined(_AP_Q2) || defined(_AP_O2)) +#error One or more of the following is defined: _AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W) || defined(_AP_I) || defined(_AP_S) || defined(_AP_Q) || defined(_AP_O) || defined(_AP_W2) || defined(_AP_I2) || defined(_AP_S2) || defined(_AP_Q2) || defined(_AP_O2)) */ + +//for safety +#if (defined(_AP_W3) || defined(_AP_S3) || defined(_AP_W4) || defined(_AP_S4)) +#error One or more of the following is defined: _AP_W3, _AP_S3, _AP_W4,_AP_S4. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W3) || defined(_AP_S3) || defined(_AP_W4) || defined(_AP_S4)) */ + +//for safety +#if (defined(_AP_W1) || defined(_AP_S1) || defined(_AP_I1) || defined(_AP_T) || defined(_AP_T1) || defined(_AP_T2) || defined(_AP_T3) || defined(_AP_T4)) +#error One or more of the following is defined: _AP_W1, _AP_S1, _AP_I1, _AP_T, _AP_T1, _AP_T2, _AP_T3, _AP_T4. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W1) || defined(_AP_S1) || defined(_AP_I1) || defined(_AP_T) || defined(_AP_T1) || defined(_AP_T2) || defined(_AP_T3) || defined(_AP_T4)) */ + +#define __AESL_APDT_IN_SCFLOW__ +#ifndef __AESL_APDT_IN_SCFLOW__ + #include "etc/ap_private.h" +#else + #include "../etc/ap_private.h" +#endif /* #ifndef __AESL_APDT_IN_SCFLOW__ */ + +#ifdef _AP_DEBUG_ + #define AP_DEBUG(s) s +#else + #define AP_DEBUG(s) +#endif /* #ifdef _AP_DEBUG_ */ + +#ifndef __SIMULATION__ + #define __SIMULATION__ +#endif /* #ifndef __SIMULATION__ */ + +#if !(defined SYSTEMC_H) && !(defined SYSTEMC_INCLUDED) + #ifndef SC_TRN + #define SC_TRN AP_TRN + #endif /* #ifndef SC_TRN */ + #ifndef SC_RND + #define SC_RND AP_RND + #endif /* #ifndef SC_RND */ + #ifndef SC_TRN_ZERO + #define SC_TRN_ZERO AP_TRN_ZERO + #endif /* #ifndef SC_TRN_ZERO */ + #ifndef SC_RND_ZERO + #define SC_RND_ZERO AP_RND_ZERO + #endif /* #ifndef SC_RND_ZERO */ + #ifndef SC_RND_INF + #define SC_RND_INF AP_RND_INF + #endif /* #ifndef SC_RND_INF */ + #ifndef SC_RND_MIN_INF + #define SC_RND_MIN_INF AP_RND_MIN_INF + #endif /* #ifndef SC_RND_MIN_INF */ + #ifndef SC_RND_CONV + #define SC_RND_CONV AP_RND_CONV + #endif /* #ifndef SC_RND_CONV */ + #ifndef SC_WRAP + #define SC_WRAP AP_WRAP + #endif /* #ifndef SC_WRAP */ + #ifndef SC_SAT + #define SC_SAT AP_SAT + #endif /* #ifndef SC_SAT */ + #ifndef SC_SAT_ZERO + #define SC_SAT_ZERO AP_SAT_ZERO + #endif /* #ifndef SC_SAT_ZERO */ + #ifndef SC_SAT_SYM + #define SC_SAT_SYM AP_SAT_SYM + #endif /* #ifndef SC_SAT_SYM */ + #ifndef SC_WRAP_SM + #define SC_WRAP_SM AP_WRAP_SM + #endif /* #ifndef SC_WRAP_SM */ + #ifndef SC_BIN + #define SC_BIN AP_BIN + #endif /* #ifndef SC_BIN */ + #ifndef SC_OCT + #define SC_OCT AP_OCT + #endif /* #ifndef SC_OCT */ + #ifndef SC_DEC + #define SC_DEC AP_DEC + #endif /* #ifndef SC_DEC */ + #ifndef SC_HEX + #define SC_HEX AP_HEX + #endif /* #ifndef SC_HEX */ +#endif /* #if !(defined SYSTEMC_H) && !(defined SYSTEMC_INCLUDED) */ +#ifndef AP_INT_MAX_W +#define AP_INT_MAX_W 1024 +#endif +#define BIT_WIDTH_UPPER_LIMIT (1 << 15) +#if AP_INT_MAX_W > BIT_WIDTH_UPPER_LIMIT +#error "Bitwidth exceeds 32768 (1 << 15), the maximum allowed value" +#endif +#define MAX_MODE(BITS) ((BITS + 1023) / 1024) + +///Forward declaration +template struct ap_range_ref; +template struct ap_bit_ref; + +template struct ap_fixed_base; +template struct af_range_ref; +template struct af_bit_ref; +template class ap_uint; + +enum { + AP_BIN = 2, + AP_OCT = 8, + AP_DEC = 10, + AP_HEX = 16 +}; + +///Why to use reference? +///Because we will operate the original object indirectly by operating the +///result object directly after concating or part selecting + +///Proxy class which allows concatination to be used as rvalue(for reading) and +//lvalue(for writing) + +/// Concatination reference. +// ---------------------------------------------------------------- +template +struct ap_concat_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + enum {_AP_WR=_AP_W1+_AP_W2,}; + _AP_T1& mbv1; + _AP_T2& mbv2; + + INLINE ap_concat_ref(const ap_concat_ref<_AP_W1, _AP_T1, + _AP_W2, _AP_T2>& ref): + mbv1(ref.mbv1), mbv2(ref.mbv2) {} + + INLINE ap_concat_ref(_AP_T1& bv1, _AP_T2& bv2):mbv1(bv1),mbv2(bv2) {} + + template + INLINE ap_concat_ref& operator = (const ap_private<_AP_W3,_AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> vval(val); + int W_ref1=mbv1.length(); + int W_ref2=mbv2.length(); + ap_private<_AP_W1,false> mask1(-1); + mask1>>=_AP_W1-W_ref1; + ap_private<_AP_W2,false> mask2(-1); + mask2>>=_AP_W2-W_ref2; + mbv1.set(ap_private<_AP_W1,false>((vval>>W_ref2)&mask1)); + mbv2.set(ap_private<_AP_W2,false>(vval&mask2)); + return *this; + } + + INLINE ap_concat_ref& operator = (unsigned long long val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + INLINE ap_concat_ref& operator = + (const ap_concat_ref <_AP_W1, _AP_T1, _AP_W2, _AP_T2>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator =(const ap_bit_ref<_AP_W3, _AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE ap_concat_ref& operator= (const af_range_ref<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator = ((const ap_private<_AP_W3, false>)(val)); + } + + template + INLINE ap_concat_ref& operator= (const ap_fixed_base<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator = (val.to_ap_private()); + } + + template + INLINE ap_concat_ref& operator= (const af_bit_ref<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + + INLINE operator ap_private<_AP_WR, false> () const { + return get(); + } + + INLINE operator unsigned long long () const { + return get().to_uint64(); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_range_ref<_AP_W3, _AP_S3> > + operator, (const ap_range_ref<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_range_ref<_AP_W3, _AP_S3> >(*this, + const_cast &>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_private<_AP_W3, _AP_S3> > + operator, (ap_private<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_private<_AP_W3, _AP_S3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_private<_AP_W3, _AP_S3> > + operator, (const ap_private<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_private<_AP_W3, _AP_S3> >(*this, + const_cast&>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, 1, ap_bit_ref<_AP_W3, _AP_S3> > + operator, (const ap_bit_ref<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + 1, ap_bit_ref<_AP_W3, _AP_S3> >(*this, + const_cast &>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3+_AP_W4, ap_concat_ref<_AP_W3,_AP_T3,_AP_W4,_AP_T4> > + operator, (const ap_concat_ref<_AP_W3,_AP_T3,_AP_W4,_AP_T4> &a2) + { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3+_AP_W4, ap_concat_ref<_AP_W3,_AP_T3,_AP_W4, + _AP_T4> >(*this, const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, af_range_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> > + operator, (const af_range_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, + _AP_O3, _AP_N3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, af_range_ref<_AP_W3, + _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_WR, ap_concat_ref, 1, af_bit_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> > + operator, (const af_bit_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, + _AP_O3, _AP_N3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, 1, af_bit_ref<_AP_W3, + _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> >(*this, + const_cast& >(a2)); + } + + template + INLINE ap_private + operator & (const ap_private<_AP_W3,_AP_S3>& a2) { + return get() & a2; + } + + + template + INLINE ap_private + operator | (const ap_private<_AP_W3,_AP_S3>& a2) { + return get() | a2; + } + + + template + INLINE ap_private + operator ^ (const ap_private<_AP_W3,_AP_S3>& a2) { + return ap_private(get() ^ a2); + } + + INLINE const ap_private<_AP_WR, false> get() const + { + ap_private<_AP_W1+_AP_W2, false> tmpVal = ap_private<_AP_W1+_AP_W2, false> (mbv1.get()); + ap_private<_AP_W1+_AP_W2, false> tmpVal2 = ap_private<_AP_W1+_AP_W2, false> (mbv2.get()); + int W_ref2 = mbv2.length(); + tmpVal <<= W_ref2; + tmpVal |= tmpVal2; + return tmpVal; + } + + INLINE const ap_private<_AP_WR, false> get() { + ap_private<_AP_W1+_AP_W2, false> tmpVal = ap_private<_AP_W1+_AP_W2, false> ( mbv1.get()); + ap_private<_AP_W1+_AP_W2, false> tmpVal2 = ap_private<_AP_W1+_AP_W2, false> (mbv2.get()); + int W_ref2 = mbv2.length(); + tmpVal <<= W_ref2; + tmpVal |= tmpVal2; + return tmpVal; + } + + template + INLINE void set(const ap_private<_AP_W3,false> & val) { + ap_private<_AP_W1+_AP_W2, false> vval(val); + int W_ref1=mbv1.length(); + int W_ref2=mbv2.length(); + ap_private<_AP_W1,false> mask1(-1); + mask1>>=_AP_W1-W_ref1; + ap_private<_AP_W2,false> mask2(-1); + mask2>>=_AP_W2-W_ref2; + mbv1.set(ap_private<_AP_W1,false>((vval>>W_ref2)&mask1)); + mbv2.set(ap_private<_AP_W2,false>(vval&mask2)); + } + + INLINE int length() const { + return mbv1.length()+mbv2.length(); + } + + INLINE std::string to_string(uint8_t radix=2) const { + return get().to_string(radix); + } +}; + +///Proxy class, which allows part selection to be used as rvalue(for reading) and +//lvalue(for writing) + +///Range(slice) reference +//------------------------------------------------------------ +template +struct ap_range_ref { +#ifdef _MSC_VER + #pragma warning( disable : 4521 4522 ) +#endif /* #ifdef _MSC_VER */ + ap_private<_AP_W,_AP_S> &d_bv; + int l_index; + int h_index; + +public: + INLINE ap_range_ref(const ap_range_ref<_AP_W, _AP_S>& ref): + d_bv(ref.d_bv), l_index(ref.l_index), h_index(ref.h_index) {} + + INLINE ap_range_ref(ap_private<_AP_W,_AP_S>* bv, int h, int l): + d_bv(*bv),l_index(l),h_index(h) { + //if (h < l) + //fprintf(stderr, "Warning! The bits selected will be returned in reverse order\n"); + } + + INLINE operator ap_private<_AP_W, false> () const { + ap_private<_AP_W, false> val(0); + if(h_index>=l_index) { + if (_AP_W > 64) { + val=d_bv; + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val>>=l_index; + val&=mask; + } else { + const static uint64_t mask = (~0ULL>> (64>_AP_W ? (64-_AP_W):0)); + val = (d_bv >> l_index) & (mask >>(_AP_W-(h_index-l_index+1))); + } + } else { + for(int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } + return val; + } + + INLINE operator unsigned long long () const { + return to_uint64(); + } + + template + INLINE ap_range_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + ap_private<_AP_W,false> vval=ap_private<_AP_W,false>(val); + if (l_index>h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + (vval)[i]? d_bv.set(j):d_bv.clear(j); + } else { + if (_AP_W > 64) { + ap_private<_AP_W,false> mask(-1); + if (l_index>0) { + mask<<=l_index; + vval<<=l_index; + } + if(h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask2(-1); + mask2>>=_AP_W-h_index-1; + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } else { + unsigned shift = 64-_AP_W; + uint64_t mask = ~0ULL>>(shift); + if(l_index>0) + { + vval = mask & vval << l_index; + mask = mask & mask << l_index; + } + if(h_index<_AP_W-1) + { + uint64_t mask2 = mask; + mask2 >>= (_AP_W-h_index-1); + mask&=mask2; + vval&=mask2; + } + mask = ~mask; + d_bv&=mask; + d_bv|=vval; + } + } + return *this; + } + + INLINE ap_range_ref& operator = (unsigned long long val) + { + const ap_private<_AP_W,_AP_S> vval=val; + return operator = (vval); + } + + + INLINE ap_range_ref& operator =(const ap_range_ref<_AP_W, _AP_S>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + + + template + INLINE ap_range_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_range_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE ap_range_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, _AP_S2>)(val)); + } + + template + INLINE ap_range_ref& operator= (const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=(val.to_ap_private()); + } + + template + INLINE ap_range_ref& operator= (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE ap_range_ref& operator= (const ap_bit_ref<_AP_W2, _AP_S2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE + ap_concat_ref<_AP_W,ap_range_ref,_AP_W2,ap_range_ref<_AP_W2,_AP_S2> > + operator, (const ap_range_ref<_AP_W2,_AP_S2> &a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, + ap_range_ref<_AP_W2,_AP_S2> >(*this, + const_cast& >(a2)); + } + + + template + INLINE ap_concat_ref<_AP_W,ap_range_ref,_AP_W2,ap_private<_AP_W2,_AP_S2> > + operator , (ap_private<_AP_W2,_AP_S2>& a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + INLINE ap_concat_ref<_AP_W,ap_range_ref,_AP_W,ap_private<_AP_W,_AP_S> > + operator , (ap_private<_AP_W, _AP_S>& a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W, + ap_private<_AP_W,_AP_S> >(*this, a2); + } + + + + template + INLINE + ap_concat_ref<_AP_W,ap_range_ref,1,ap_bit_ref<_AP_W2,_AP_S2> > + operator, (const ap_bit_ref<_AP_W2,_AP_S2> &a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, 1, + ap_bit_ref<_AP_W2,_AP_S2> >(*this, const_cast& >(a2)); + } + + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) + { + return ap_concat_ref<_AP_W, ap_range_ref, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_range_ref, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE bool operator == (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + + template + INLINE bool operator != (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + + template + INLINE bool operator > (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + + template + INLINE bool operator >= (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + + template + INLINE bool operator < (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + + template + INLINE void set(const ap_private<_AP_W2,false>& val) + { + ap_private<_AP_W,_AP_S> vval=val; + if(l_index>h_index) + { + for(int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + (vval)[i]? d_bv.set(j):d_bv.clear(j); + } else { + if (_AP_W>64 ) { + ap_private<_AP_W,_AP_S> mask(-1); + if(l_index>0) + { + ap_private<_AP_W,false> mask1(-1); + mask1>>=_AP_W-l_index; + mask1.flip(); + mask=mask1; + //vval&=mask1; + vval<<=l_index; + } + if(h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask2(-1); + mask2<<=h_index+1; + mask2.flip(); + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } else { + uint64_t mask = ~0ULL >> (64-_AP_W); + if(l_index>0) + { + uint64_t mask1 = mask; + mask1=mask & (mask1>>(_AP_W-l_index)); + vval =mask&( vval <> (64-_AP_W); + mask2 = mask &(mask2<<(h_index+1)); + mask&=~mask2; + vval&=~mask2; + } + d_bv&=(~mask&(~0ULL >> (64-_AP_W))); + d_bv|=vval; + } + } + } + + + INLINE ap_private<_AP_W,false> get() const + { + ap_private<_AP_W,false> val(0); + if(h_index=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } else { + val=d_bv; + val>>=l_index; + if(h_index<_AP_W-1) + { + if (_AP_W <= 64) { + const static uint64_t mask = (~0ULL>> (64>_AP_W ? (64-_AP_W):0)); + val &= (mask>> (_AP_W-(h_index-l_index+1))); + } else { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + } + } + return val; + } + + + INLINE ap_private<_AP_W,false> get() + { + ap_private<_AP_W,false> val(0); + if(h_index=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } else { + val=d_bv; + val>>=l_index; + if(h_index<_AP_W-1) + { + if (_AP_W <= 64 ) { + static const uint64_t mask = ~0ULL>> (64>_AP_W ? (64-_AP_W):0); + return val &= ( (mask) >> (_AP_W - (h_index-l_index+1))); + } else { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + } + } + return val; + } + + + INLINE int length() const + { + return h_index>=l_index?h_index-l_index+1:l_index-h_index+1; + } + + + INLINE int to_int() const + { + ap_private<_AP_W,false> val=get(); + return val.to_int(); + } + + + INLINE unsigned int to_uint() const + { + ap_private<_AP_W,false> val=get(); + return val.to_uint(); + } + + + INLINE long to_long() const + { + ap_private<_AP_W,false> val=get(); + return val.to_long(); + } + + + INLINE unsigned long to_ulong() const + { + ap_private<_AP_W,false> val=get(); + return val.to_ulong(); + } + + + INLINE ap_slong to_int64() const + { + ap_private<_AP_W,false> val=get(); + return val.to_int64(); + } + + + INLINE ap_ulong to_uint64() const + { + ap_private<_AP_W,false> val=get(); + return val.to_uint64(); + } + + INLINE std::string to_string(uint8_t radix=2) const { + return get().to_string(radix); + } + +}; + +///Proxy class, which allows bit selection to be used as rvalue(for reading) and +//lvalue(for writing) + +///Bit reference +//-------------------------------------------------------------- +template +struct ap_bit_ref { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif + ap_private<_AP_W,_AP_S>& d_bv; + int d_index; + +public: + INLINE ap_bit_ref(const ap_bit_ref<_AP_W, _AP_S>& ref): + d_bv(ref.d_bv), d_index(ref.d_index) {} + + INLINE ap_bit_ref(ap_private<_AP_W,_AP_S>& bv, int index=0): + d_bv(bv),d_index(index) + { +#ifdef _AP_DEBUG_ + assert(d_index<_AP_W&&"index out of bound"); +#endif + } + + + INLINE operator bool () const + { + return d_bv.get_bit(d_index); + } + + + INLINE bool to_bool() const + { + return operator bool (); + } + + + INLINE ap_bit_ref& operator = (unsigned long long val) + { + if(val) + d_bv.set(d_index); + else + d_bv.clear(d_index); + return *this; + } + + +#if 0 + INLINE ap_bit_ref& operator = (bool val) + { + if(val) + d_bv.set(d_index); + else + d_bv.clear(d_index); + return *this; + } +#endif + template + INLINE ap_bit_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(val != 0)); + } + + + template + INLINE ap_bit_ref& operator =(const ap_bit_ref<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(bool)val); + } + + INLINE ap_bit_ref& operator =(const ap_bit_ref<_AP_W,_AP_S>& val) + { + return operator =((unsigned long long)(bool)val); + } + + template + INLINE ap_bit_ref& operator =(const ap_range_ref<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(bool) val); + } + + + template + INLINE ap_bit_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, false>)(val)); + } + + template + INLINE ap_bit_ref& operator= (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE ap_bit_ref& operator= (const ap_concat_ref<_AP_W2, _AP_T3, _AP_W3, _AP_T3>& val) { + return operator=((const ap_private<_AP_W2 + _AP_W3, false>)(val)); + } + + + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> > + operator , (ap_private<_AP_W2, _AP_S2>& a2) + { + return ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_range_ref<_AP_W2,_AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) + { + return + ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_range_ref<_AP_W2,_AP_S2> >(*this, + const_cast &>(a2)); + } + + + template + INLINE ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref<_AP_W2,_AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) + { + return + ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref<_AP_W2,_AP_S2> >(*this, + const_cast &>(a2)); + } + + + INLINE ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref > + operator, (const ap_bit_ref &a2) + { + return + ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref >(*this, + const_cast(a2)); + } + + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> > + operator, (const ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> &a2) + { + return + ap_concat_ref<1,ap_bit_ref,_AP_W2+_AP_W3, + ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<1, ap_bit_ref, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<1, ap_bit_ref, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<1, ap_bit_ref, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<1, ap_bit_ref, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE bool operator == (const ap_bit_ref<_AP_W2, _AP_S2>& op) { + return get() == op.get(); + } + + template + INLINE bool operator != (const ap_bit_ref<_AP_W2, _AP_S2>& op) { + return get() != op.get(); + } + + + INLINE bool get() const + { + return operator bool (); + } + + + INLINE bool get() + { + return operator bool (); + } + + + template + INLINE void set(const ap_private<_AP_W3, false>& val) + { + operator = (val); + } + + INLINE bool operator ~ () const { + bool bit = (d_bv)[d_index]; + return bit ? false : true; + } + + INLINE int length() const { return 1; } + + INLINE std::string to_string() const { + bool val = get(); + return val ? "1" : "0"; + } +}; + +/// Operators mixing Integers with AP_Int +// ---------------------------------------------------------------- +#if 1 +#define OP_BIN_MIX_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + INLINE typename ap_private<_AP_WI,_AP_SI>::template RType<_AP_W,_AP_S>::RTYPE \ + operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_WI,_AP_SI>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (ap_private<_AP_WI,_AP_SI>(i_op)); \ + } +#else +#define OP_BIN_MIX_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + INLINE typename ap_private<_AP_WI,_AP_SI>::template RType<_AP_W,_AP_S>::RTYPE \ + operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_WI,_AP_SI>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (ap_private<_AP_WI,_AP_SI>(i_op)); \ + } +#endif +#define OP_REL_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return op.operator REL_OP (ap_private<_AP_W2, _AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (op); \ + } +#define OP_ASSIGN_MIX_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return op.operator ASSIGN_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } + +#define OP_BIN_SHIFT_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + C_TYPE operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return i_op BIN_OP (op.getVal()); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (i_op); \ + } +#define OP_ASSIGN_RSHIFT_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + op = op.operator >> (op2); \ + return op; \ + } +#define OP_ASSIGN_LSHIFT_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + op = op.operator << (op2); \ + return op; \ + } + +#define OPS_MIX_INT(C_TYPE, WI, SI) \ + OP_BIN_MIX_INT(*, C_TYPE, WI, SI, mult) \ + OP_BIN_MIX_INT(+, C_TYPE, WI, SI, plus) \ + OP_BIN_MIX_INT(-, C_TYPE, WI, SI, minus) \ + OP_BIN_MIX_INT(/, C_TYPE, WI, SI, div) \ + OP_BIN_MIX_INT(%, C_TYPE, WI, SI, mod) \ + OP_BIN_MIX_INT(&, C_TYPE, WI, SI, logic) \ + OP_BIN_MIX_INT(|, C_TYPE, WI, SI, logic) \ + OP_BIN_MIX_INT(^, C_TYPE, WI, SI, logic) \ + OP_BIN_SHIFT_INT(>>, C_TYPE, WI, SI, arg1) \ + OP_BIN_SHIFT_INT(<<, C_TYPE, WI, SI, arg1) \ + \ + OP_REL_MIX_INT(==, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(!=, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(>, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(>=, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(<, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(<=, C_TYPE, WI, SI) \ + \ + OP_ASSIGN_MIX_INT(+=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(-=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(*=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(/=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(%=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(&=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(|=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(^=, C_TYPE, WI, SI) \ + OP_ASSIGN_RSHIFT_INT(>>=, C_TYPE, WI, SI) \ + OP_ASSIGN_LSHIFT_INT(<<=, C_TYPE, WI, SI) + + +OPS_MIX_INT(bool, 1, false) +OPS_MIX_INT(char, 8, true) +OPS_MIX_INT(signed char, 8, true) +OPS_MIX_INT(unsigned char, 8, false) +OPS_MIX_INT(short, 16, true) +OPS_MIX_INT(unsigned short, 16, false) +OPS_MIX_INT(int, 32, true) +OPS_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +OPS_MIX_INT(long, 64, true) +OPS_MIX_INT(unsigned long, 64, false) +# else +OPS_MIX_INT(long, 32, true) +OPS_MIX_INT(unsigned long, 32, false) +# endif +OPS_MIX_INT(ap_slong, 64, true) +OPS_MIX_INT(ap_ulong, 64, false) + +#define OP_BIN_MIX_RANGE(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_range_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1, false>(op1).operator BIN_OP (op2); \ + } \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator BIN_OP (ap_private<_AP_W2, false>(op2)); \ + } + +#define OP_REL_MIX_RANGE(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_range_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1,false>(op1).operator REL_OP (op2); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator REL_OP (op2.operator ap_private<_AP_W2, false>()); \ + } + +#define OP_ASSIGN_MIX_RANGE(ASSIGN_OP) \ + template \ + INLINE ap_private<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator ASSIGN_OP (ap_private<_AP_W2, false>(op2)); \ + } \ + template \ + INLINE ap_range_ref<_AP_W1,_AP_S1>& operator ASSIGN_OP (ap_range_ref<_AP_W1,_AP_S1>& op1, ap_private<_AP_W2,_AP_S2>& op2) { \ + ap_private<_AP_W1, false> tmp(op1); \ + tmp.operator ASSIGN_OP (op2); \ + op1 = tmp; \ + return op1; \ + } + + +OP_ASSIGN_MIX_RANGE(+=) +OP_ASSIGN_MIX_RANGE(-=) +OP_ASSIGN_MIX_RANGE(*=) +OP_ASSIGN_MIX_RANGE(/=) +OP_ASSIGN_MIX_RANGE(%=) +OP_ASSIGN_MIX_RANGE(>>=) +OP_ASSIGN_MIX_RANGE(<<=) +OP_ASSIGN_MIX_RANGE(&=) +OP_ASSIGN_MIX_RANGE(|=) +OP_ASSIGN_MIX_RANGE(^=) + +OP_REL_MIX_RANGE(==) +OP_REL_MIX_RANGE(!=) +OP_REL_MIX_RANGE(>) +OP_REL_MIX_RANGE(>=) +OP_REL_MIX_RANGE(<) +OP_REL_MIX_RANGE(<=) + +OP_BIN_MIX_RANGE(+, plus) +OP_BIN_MIX_RANGE(-, minus) +OP_BIN_MIX_RANGE(*, mult) +OP_BIN_MIX_RANGE(/, div) +OP_BIN_MIX_RANGE(%, mod) +OP_BIN_MIX_RANGE(>>, arg1) +OP_BIN_MIX_RANGE(<<, arg1) +OP_BIN_MIX_RANGE(&, logic) +OP_BIN_MIX_RANGE(|, logic) +OP_BIN_MIX_RANGE(^, logic) + +#define OP_BIN_MIX_BIT(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_private<1, false>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_bit_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<1, false>(op1).operator BIN_OP (op2); \ + } \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<1,false>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator BIN_OP (ap_private<1, false>(op2)); \ + } + +#define OP_REL_MIX_BIT(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_bit_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1,false>(op1).operator REL_OP (op2); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator REL_OP (ap_private<1, false>(op2)); \ + } + +#define OP_ASSIGN_MIX_BIT(ASSIGN_OP) \ + template \ + INLINE ap_private<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_private<_AP_W1,_AP_S1>& op1, ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator ASSIGN_OP (ap_private<1, false>(op2)); \ + } \ + template \ + INLINE ap_bit_ref<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_bit_ref<_AP_W1,_AP_S1>& op1, ap_private<_AP_W2,_AP_S2>& op2) { \ + ap_private<1, false> tmp(op1); \ + tmp.operator ASSIGN_OP (op2); \ + op1 = tmp; \ + return op1; \ + } + + +OP_ASSIGN_MIX_BIT(+=) +OP_ASSIGN_MIX_BIT(-=) +OP_ASSIGN_MIX_BIT(*=) +OP_ASSIGN_MIX_BIT(/=) +OP_ASSIGN_MIX_BIT(%=) +OP_ASSIGN_MIX_BIT(>>=) +OP_ASSIGN_MIX_BIT(<<=) +OP_ASSIGN_MIX_BIT(&=) +OP_ASSIGN_MIX_BIT(|=) +OP_ASSIGN_MIX_BIT(^=) + +OP_REL_MIX_BIT(==) +OP_REL_MIX_BIT(!=) +OP_REL_MIX_BIT(>) +OP_REL_MIX_BIT(>=) +OP_REL_MIX_BIT(<) +OP_REL_MIX_BIT(<=) + +OP_BIN_MIX_BIT(+, plus) +OP_BIN_MIX_BIT(-, minus) +OP_BIN_MIX_BIT(*, mult) +OP_BIN_MIX_BIT(/, div) +OP_BIN_MIX_BIT(%, mod) +OP_BIN_MIX_BIT(>>, arg1) +OP_BIN_MIX_BIT(<<, arg1) +OP_BIN_MIX_BIT(&, logic) +OP_BIN_MIX_BIT(|, logic) +OP_BIN_MIX_BIT(^, logic) + +#define REF_REL_OP_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE bool operator REL_OP ( const ap_range_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_range_ref<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_bit_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (bool(op)) REL_OP op2; \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_bit_ref<_AP_W,_AP_S> &op) { \ + return op2 REL_OP (bool(op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_concat_ref<_AP_W,_AP_T, _AP_W1, _AP_T1> &op, C_TYPE op2) { \ + return (ap_private<_AP_W + _AP_W1, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_concat_ref<_AP_W,_AP_T, _AP_W1, _AP_T1> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W + _AP_W1, false>(op)); \ + } + +#define REF_REL_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(>, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(<, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(>=, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(<=, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(==, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(!=, C_TYPE, _AP_WI, _AP_SI) + +REF_REL_MIX_INT(bool, 1, false) +REF_REL_MIX_INT(char, 8, true) +REF_REL_MIX_INT(signed char, 8, true) +REF_REL_MIX_INT(unsigned char, 8, false) +REF_REL_MIX_INT(short, 16, true) +REF_REL_MIX_INT(unsigned short, 16, false) +REF_REL_MIX_INT(int, 32, true) +REF_REL_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +REF_REL_MIX_INT(long, 64, true) +REF_REL_MIX_INT(unsigned long, 64, false) +# else +REF_REL_MIX_INT(long, 32, true) +REF_REL_MIX_INT(unsigned long, 32, false) +# endif +REF_REL_MIX_INT(ap_slong, 64, true) +REF_REL_MIX_INT(ap_ulong, 64, false) + +#define REF_BIN_OP_MIX_INT(BIN_OP, RTYPE, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE typename ap_private<_AP_W, false>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_range_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator BIN_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE typename ap_private<_AP_W2, _AP_S2>::template RType<_AP_W,false>::RTYPE \ + operator BIN_OP ( C_TYPE op2, const ap_range_ref<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator BIN_OP (ap_private<_AP_W, false>(op)); \ + } + +#define REF_BIN_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(+, plus, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(-, minus, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(*, mult, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(/, div, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(%, mod, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(>>, arg1, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(<<, arg1, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(&, logic, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(|, logic, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(^, logic, C_TYPE, _AP_WI, _AP_SI) + +REF_BIN_MIX_INT(bool, 1, false) +REF_BIN_MIX_INT(char, 8, true) +REF_BIN_MIX_INT(signed char, 8, true) +REF_BIN_MIX_INT(unsigned char, 8, false) +REF_BIN_MIX_INT(short, 16, true) +REF_BIN_MIX_INT(unsigned short, 16, false) +REF_BIN_MIX_INT(int, 32, true) +REF_BIN_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +REF_BIN_MIX_INT(long, 64, true) +REF_BIN_MIX_INT(unsigned long, 64, false) +#else +REF_BIN_MIX_INT(long, 32, true) +REF_BIN_MIX_INT(unsigned long, 32, false) +#endif +REF_BIN_MIX_INT(ap_slong, 64, true) +REF_BIN_MIX_INT(ap_ulong, 64, false) + +#define REF_BIN_OP(BIN_OP, RTYPE) \ +template \ +INLINE typename ap_private<_AP_W, false>::template RType<_AP_W2, false>::RTYPE \ +operator BIN_OP (const ap_range_ref<_AP_W,_AP_S> &lhs, const ap_range_ref<_AP_W2,_AP_S2> &rhs) { \ + return ap_private<_AP_W,false>(lhs).operator BIN_OP (ap_private<_AP_W2, false>(rhs)); \ +} + +REF_BIN_OP(+, plus) +REF_BIN_OP(-, minus) +REF_BIN_OP(*, mult) +REF_BIN_OP(/, div) +REF_BIN_OP(%, mod) +REF_BIN_OP(>>, arg1) +REF_BIN_OP(<<, arg1) +REF_BIN_OP(&, logic) +REF_BIN_OP(|, logic) +REF_BIN_OP(^, logic) + +#if 1 +#define CONCAT_OP_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const ap_private<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + ret <<= _AP_WI; \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + }\ + ret |= val; \ + return ret;\ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_private<_AP_W, _AP_S>& op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + if (_AP_S) { \ + ret <<= _AP_WI; ret >>= _AP_WI; \ + } \ + ret |= val << _AP_W; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const ap_range_ref<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + ret <<= _AP_WI; \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + } \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_range_ref<_AP_W, _AP_S> &op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + int len = op2.length(); \ + val <<= len; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private<_AP_WI + 1, false > \ + operator, (const ap_bit_ref<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + 1, false> val(op2); \ + val[_AP_WI] = op1; \ + return val; \ +} \ +template \ +INLINE \ +ap_private<_AP_WI + 1, false > \ + operator, (C_TYPE op1, const ap_bit_ref<_AP_W, _AP_S> &op2) { \ + ap_private<_AP_WI + 1, false> val(op1); \ + val <<= 1; \ + val[0] = op2; \ + return val; \ +} \ +template \ +INLINE \ +ap_private<_AP_W + _AP_W2 + _AP_WI, false > \ + operator, (const ap_concat_ref<_AP_W, _AP_T, _AP_W2, _AP_T2> &op1, C_TYPE op2) {\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> val(op2);\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> ret(op1);\ + if (_AP_SI) { \ + val <<= _AP_W + _AP_W2; val >>= _AP_W + _AP_W2; \ + } \ + ret <<= _AP_WI; \ + ret |= val; \ + return ret; \ +}\ +template \ +INLINE \ +ap_private<_AP_W + _AP_W2 + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_concat_ref<_AP_W, _AP_T, _AP_W2, _AP_T2> &op2) {\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> val(op1);\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> ret(op2);\ + int len = op2.length(); \ + val <<= len; \ + ret |= val;\ + return ret; \ +}\ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + }\ + ret <<= _AP_WI; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const af_range_ref<_AP_W, _AP_I, _AP_S, \ + _AP_Q, _AP_O, _AP_N> &op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + int len = op2.length(); \ + val <<= len; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< 1 + _AP_WI, false> \ + operator, (const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, \ + _AP_N> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + 1, _AP_SI> val(op2); \ + val[_AP_WI] = op1; \ + return val; \ +} \ +template \ +INLINE \ +ap_private< 1 + _AP_WI, false> \ + operator, (C_TYPE op1, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q,\ + _AP_O, _AP_N> &op2) { \ + ap_private<_AP_WI + 1, _AP_SI> val(op1); \ + val <<= 1; \ + val[0] = op2; \ + return val; \ +} + +CONCAT_OP_MIX_INT(bool, 1, false) +CONCAT_OP_MIX_INT(char, 8, true) +CONCAT_OP_MIX_INT(signed char, 8, true) +CONCAT_OP_MIX_INT(unsigned char, 8, false) +CONCAT_OP_MIX_INT(short, 16, true) +CONCAT_OP_MIX_INT(unsigned short, 16, false) +CONCAT_OP_MIX_INT(int, 32, true) +CONCAT_OP_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +CONCAT_OP_MIX_INT(long, 64, true) +CONCAT_OP_MIX_INT(unsigned long, 64, false) +# else +CONCAT_OP_MIX_INT(long, 32, true) +CONCAT_OP_MIX_INT(unsigned long, 32, false) +# endif +CONCAT_OP_MIX_INT(ap_slong, 64, true) +CONCAT_OP_MIX_INT(ap_ulong, 64, false) +#endif + +#if 1 +#define CONCAT_SHIFT_MIX_INT(C_TYPE, op) \ +template \ +INLINE ap_uint<_AP_W+_AP_W1> operator op (const ap_concat_ref<_AP_W, _AP_T, _AP_W1, _AP_T1> lhs, C_TYPE rhs) { \ + return ((ap_uint<_AP_W+_AP_W1>)lhs.get()) op ((int)rhs); \ +} + +CONCAT_SHIFT_MIX_INT(long, <<) +CONCAT_SHIFT_MIX_INT(unsigned long, <<) +CONCAT_SHIFT_MIX_INT(unsigned int, <<) +CONCAT_SHIFT_MIX_INT(ap_ulong, <<) +CONCAT_SHIFT_MIX_INT(ap_slong, <<) +CONCAT_SHIFT_MIX_INT(long, >>) +CONCAT_SHIFT_MIX_INT(unsigned long, >>) +CONCAT_SHIFT_MIX_INT(unsigned int, >>) +CONCAT_SHIFT_MIX_INT(ap_ulong, >>) +CONCAT_SHIFT_MIX_INT(ap_slong, >>) +#endif + +#if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_private<_AP_W, _AP_S> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} +#endif + +template +INLINE std::ostream& operator<<(std::ostream& out, const ap_private<_AP_W,_AP_S> &op) +{ + ap_private<_AP_W, _AP_S> v=op; + const std::ios_base::fmtflags basefield = out.flags() & std::ios_base::basefield; + unsigned radix = (basefield == std::ios_base::hex) ? 16 : + ((basefield == std::ios_base::oct) ? 8 : 10); + std::string str=v.toString(radix,_AP_S); + out< +INLINE std::istream& operator >> (std::istream& in, ap_private<_AP_W,_AP_S> &op) +{ + std::string str; + in >> str; + op = ap_private<_AP_W, _AP_S>(str.c_str()); + return in; + +} + +template +INLINE std::ostream& operator<<(std::ostream& out, const ap_range_ref<_AP_W,_AP_S> &op) +{ + return operator<<(out, ap_private<_AP_W, _AP_S>(op)); +} + +template +INLINE std::istream& operator >> (std::istream& in, ap_range_ref<_AP_W,_AP_S> &op) +{ + return operator>>(in, ap_private<_AP_W, _AP_S>(op));; +} + +template +INLINE void print(const ap_private<_AP_W,_AP_S> &op, bool fill=true ) +{ + ap_private<_AP_W, _AP_S> v=op; + uint32_t ws=v.getNumWords(); + const uint64_t *ptr=v.getRawData(); + int i=ws-1; +#if 0 + if(fill) + printf("%016llx",*(ptr+i)); + else + printf("%llx",*(ptr+i)); +#else +//match SystemC output + if(_AP_W%64 != 0) { + uint32_t offset=_AP_W%64; + uint32_t count=(offset+3)/4; + int64_t data=*(ptr+i); + if(_AP_S) + data=(data<<(64-offset))>>(64-offset); + else + count=(offset+4)/4; + while(count-->0) + printf("%llx",(data>>(count*4))&0xf); + } else { + if(_AP_S==false) + printf("0"); + printf("%016llx",*(ptr+i)); + } +#endif + for(--i;i>=0;i--) + printf("%016llx",*(ptr+i)); + printf("\n"); + +} +#endif /* #ifndef __AESL_GCC_AP_INT_H__ */ \ No newline at end of file diff --git a/hls/ZU3EG_test/etc/ap_private.h b/hls/ZU3EG_test/etc/ap_private.h new file mode 100644 index 0000000000000000000000000000000000000000..1a68a9e56e950d7108a3014149623c017023d809 --- /dev/null +++ b/hls/ZU3EG_test/etc/ap_private.h @@ -0,0 +1,5858 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef LLVM_SUPPORT_MATHEXTRAS_H +#define LLVM_SUPPORT_MATHEXTRAS_H + +#ifdef _MSC_VER +#if _MSC_VER <= 1500 +typedef __int8 int8_t; +typedef unsigned __int8 uint8_t; +typedef __int16 int16_t; +typedef unsigned __int16 uint16_t; +typedef __int32 int32_t; +typedef unsigned __int32 uint32_t; +typedef __int64 int64_t; +typedef unsigned __int64 uint64_t; +#else if +#include +#endif /* #if _MSC_VER <= 1500 */ +#else +#include +#endif /* #if _MSC_VER <= 1500 */ +#undef INLINE +#if 1 +#define INLINE inline +#else +//Enable to debug ap_int/ap_fixed +#define INLINE __attribute__((weak)) +#endif +#define AP_MAX(a,b) ((a) > (b) ? (a) : (b)) +#define AP_MIN(a,b) ((a) < (b) ? (a) : (b)) +#define AP_ABS(a) ((a)>=0 ? (a):-(a)) +#ifndef AP_INT_MAX_W +#define AP_INT_MAX_W 1024 +#endif +#define BIT_WIDTH_UPPER_LIMIT (1 << 15) +#if AP_INT_MAX_W > BIT_WIDTH_UPPER_LIMIT +#error "Bitwidth exceeds 32768 (1 << 15), the maximum allowed value" +#endif +#define MAX_MODE(BITS) ((BITS + 1023) / 1024) + +// NOTE: The following support functions use the _32/_64 extensions instead of +// type overloading so that signed and unsigned integers can be used without +// ambiguity. + +/// Hi_32 - This function returns the high 32 bits of a 64 bit value. +INLINE uint32_t Hi_32(uint64_t Value) { + return static_cast(Value >> 32); +} + +/// Lo_32 - This function returns the low 32 bits of a 64 bit value. +INLINE uint32_t Lo_32(uint64_t Value) { + return static_cast(Value); +} + +/// ByteSwap_16 - This function returns a byte-swapped representation of the +/// 16-bit argument, Value. +INLINE uint16_t ByteSwap_16(uint16_t Value) { +#if defined(_MSC_VER) && !defined(_DEBUG) + // The DLL version of the runtime lacks these functions (bug!?), but in a + // release build they're replaced with BSWAP instructions anyway. + return (uint16_t)(_byteswap_ushort(Value)); +#else + uint16_t Hi = (uint16_t)((Value) << 8); + uint16_t Lo = (uint16_t)((Value) >> 8); + return Hi | Lo; +#endif +} + +/// ByteSwap_32 - This function returns a byte-swapped representation of the +/// 32-bit argument, Value. +INLINE uint32_t ByteSwap_32(uint32_t Value) { + uint32_t Byte0 = Value & 0x000000FF; + uint32_t Byte1 = Value & 0x0000FF00; + uint32_t Byte2 = Value & 0x00FF0000; + uint32_t Byte3 = Value & 0xFF000000; + return ((Byte0) << 24) | ((Byte1) << 8) | ((Byte2) >> 8) | ((Byte3) >> 24); +} + +/// ByteSwap_64 - This function returns a byte-swapped representation of the +/// 64-bit argument, Value. +INLINE uint64_t ByteSwap_64(uint64_t Value) { + uint64_t Hi = ByteSwap_32(uint32_t(Value)); + uint32_t Lo = ByteSwap_32(uint32_t(Value >> 32)); + return ((Hi) << 32) | Lo; +} + +/// CountLeadingZeros_32 - this function performs the platform optimal form of +/// counting the number of zeros from the most significant bit to the first one +/// bit. Ex. CountLeadingZeros_32(0x00F000FF) == 8. +/// Returns 32 if the word is zero. +INLINE unsigned CountLeadingZeros_32(uint32_t Value) { + unsigned Count; // result +#if __GNUC__ >= 4 + // PowerPC is defined for __builtin_clz(0) +#if !defined(__ppc__) && !defined(__ppc64__) + if (Value == 0) return 32; +#endif + Count = __builtin_clz(Value); +#else + if (Value == 0) return 32; + Count = 0; + // bisecton method for count leading zeros + for (unsigned Shift = 32 >> 1; Shift; Shift >>= 1) { + uint32_t Tmp = (Value) >> (Shift); + if (Tmp) { + Value = Tmp; + } else { + Count |= Shift; + } + } +#endif + return Count; +} + +/// CountLeadingZeros_64 - This function performs the platform optimal form +/// of counting the number of zeros from the most significant bit to the first +/// one bit (64 bit edition.) +/// Returns 64 if the word is zero. +INLINE unsigned CountLeadingZeros_64(uint64_t Value) { + unsigned Count; // result +#if __GNUC__ >= 4 + // PowerPC is defined for __builtin_clzll(0) +#if !defined(__ppc__) && !defined(__ppc64__) + if (!Value) return 64; +#endif + Count = __builtin_clzll(Value); +#else + if (sizeof(long) == sizeof(int64_t)) { + if (!Value) return 64; + Count = 0; + // bisecton method for count leading zeros + for (unsigned Shift = 64 >> 1; Shift; Shift >>= 1) { + uint64_t Tmp = (Value) >> (Shift); + if (Tmp) { + Value = Tmp; + } else { + Count |= Shift; + } + } + } else { + // get hi portion + uint32_t Hi = Hi_32(Value); + + // if some bits in hi portion + if (Hi) { + // leading zeros in hi portion plus all bits in lo portion + Count = CountLeadingZeros_32(Hi); + } else { + // get lo portion + uint32_t Lo = Lo_32(Value); + // same as 32 bit value + Count = CountLeadingZeros_32(Lo)+32; + } + } +#endif + return Count; +} + +/// CountTrailingZeros_64 - This function performs the platform optimal form +/// of counting the number of zeros from the least significant bit to the first +/// one bit (64 bit edition.) +/// Returns 64 if the word is zero. +INLINE unsigned CountTrailingZeros_64(uint64_t Value) { +#if __GNUC__ >= 4 + return (Value != 0) ? __builtin_ctzll(Value) : 64; +#else + static const unsigned Mod67Position[] = { + 64, 0, 1, 39, 2, 15, 40, 23, 3, 12, 16, 59, 41, 19, 24, 54, + 4, 64, 13, 10, 17, 62, 60, 28, 42, 30, 20, 51, 25, 44, 55, + 47, 5, 32, 65, 38, 14, 22, 11, 58, 18, 53, 63, 9, 61, 27, + 29, 50, 43, 46, 31, 37, 21, 57, 52, 8, 26, 49, 45, 36, 56, + 7, 48, 35, 6, 34, 33, 0 + }; + return Mod67Position[(uint64_t)(-(int64_t)Value & (int64_t)Value) % 67]; +#endif +} + +/// CountPopulation_64 - this function counts the number of set bits in a value, +/// (64 bit edition.) +INLINE unsigned CountPopulation_64(uint64_t Value) { +#if __GNUC__ >= 4 + return __builtin_popcountll(Value); +#else + uint64_t v = Value - (((Value) >> 1) & 0x5555555555555555ULL); + v = (v & 0x3333333333333333ULL) + (((v) >> 2) & 0x3333333333333333ULL); + v = (v + ((v) >> 4)) & 0x0F0F0F0F0F0F0F0FULL; + return unsigned((uint64_t)(v * 0x0101010101010101ULL) >> 56); +#endif +} + +#endif // LLVM_SUPPORT_MATHEXTRAS_H + + +#ifndef AP_PRIVATE_H +#define AP_PRIVATE_H + +#include +#include +#include +#include +#include +#include +#include +#include + +namespace AESL_std { + template + DataType INLINE min(DataType a, DataType b) { + // if (a >= b) return b; + // else return a; + return (a>=b) ? b : a; + } + + template + DataType INLINE max(DataType a, DataType b) { + // if (a >= b) return a; + // else return b; + return (a>=b) ? a : b; + } +} +enum ap_q_mode { + AP_RND, // rounding to plus infinity + AP_RND_ZERO,// rounding to zero + AP_RND_MIN_INF,// rounding to minus infinity + AP_RND_INF,// rounding to infinity + AP_RND_CONV, // convergent rounding + AP_TRN, // truncation + AP_TRN_ZERO // truncation to zero + +}; +enum ap_o_mode { + AP_SAT, // saturation + AP_SAT_ZERO, // saturation to zero + AP_SAT_SYM, // symmetrical saturation + AP_WRAP, // wrap-around (*) + AP_WRAP_SM // sign magnitude wrap-around (*) +}; + +template struct ap_fixed_base; +template struct af_range_ref; +template struct af_bit_ref; + +template struct ap_range_ref; +template struct ap_bit_ref; +template struct ap_concat_ref; +static bool InvalidDigit(const char* str, unsigned len, unsigned start, unsigned radix) { + unsigned i; + for (i = start; i < len; ++i) + if ((radix == 2 && (str[i] == '0' || str[i] == '1')) || + (radix == 8 && str[i] >= '0' && str[i] <= '7') || + (radix == 10 && str[i] >= '0' && str[i] <= '9') || + (radix == 16 && ((str[i] >= '0' && str[i] <= '9') || + (str[i] >= 'a' && str[i] <= 'f') || + (str[i] >= 'A' && str[i] <= 'F')))) + continue; + else + return true; + return false; +} + +static void ap_parse_sign(const char* str, uint32_t &base, bool &neg) { + if (str[0] == '+' || str[0] == '-') base = 1; + if (str[0] == '-') neg = true; + else neg = false; + return; +} + +static void ap_parse_prefix(const char* str, uint32_t &offset, uint32_t &radix) { + if (str[0] == '0') { + switch (str[1]) { + case 'b': + case 'B': offset = 2; radix = 2; break; + case 'x': + case 'X': offset = 2; radix = 16; break; + case 'd': + case 'D': offset = 2; radix = 10; break; + case 'o': + case 'O': offset = 2; radix = 8; break; + default: break; + } + } + if (offset == 0) + for (int i=0, len = strlen(str); i= 'a') || (str[i] <= 'F' && str[i] >= 'A')) { + radix = 16; + break; + } + return; +} + +/// sub_1 - This function subtracts a single "digit" (64-bit word), y, from +/// the multi-digit integer array, x[], propagating the borrowed 1 value until +/// no further borrowing is neeeded or it runs out of "digits" in x. The result +/// is 1 if "borrowing" exhausted the digits in x, or 0 if x was not exhausted. +/// In other words, if y > x then this function returns 1, otherwise 0. +/// @returns the borrow out of the subtraction +static bool sub_1(uint64_t x[], uint32_t len, uint64_t y) { + for (uint32_t i = 0; i < len; ++i) { + uint64_t __X = x[i]; + x[i] -= y; + if (y > __X) + y = 1; // We have to "borrow 1" from next "digit" + else { + y = 0; // No need to borrow + break; // Remaining digits are unchanged so exit early + } + } + return (y != 0); +} + + /// This enumeration just provides for internal constants used in this + /// translation unit. + enum { + MIN_INT_BITS = 1, ///< Minimum number of bits that can be specified + ///< Note that this must remain synchronized with IntegerType::MIN_INT_BITS + MAX_INT_BITS = (1<<23)-1 ///< Maximum number of bits that can be specified + ///< Note that this must remain synchronized with IntegerType::MAX_INT_BITS + }; + + /// A utility function for allocating memory and checking for allocation + /// failure. The content is not zeroed. + static uint64_t* getMemory(uint32_t numWords) { + return (uint64_t*) malloc(numWords*sizeof(uint64_t)); + } + + //===----------------------------------------------------------------------===// + // ap_private Class + //===----------------------------------------------------------------------===// + + /// ap_private - This class represents arbitrary precision constant integral values. + /// It is a functional replacement for common case unsigned integer type like + /// "unsigned", "unsigned long" or "uint64_t", but also allows non-byte-width + /// integer sizes and large integer value types such as 3-bits, 15-bits, or more + /// than 64-bits of precision. ap_private provides a variety of arithmetic operators + /// and methods to manipulate integer values of any bit-width. It supports both + /// the typical integer arithmetic and comparison operations as well as bitwise + /// manipulation. + /// + /// The class has several invariants worth noting: + /// * All bit, byte, and word positions are zero-based. + /// * Once the bit width is set, it doesn't change except by the Truncate, + /// SignExtend, or ZeroExtend operations. + /// * All binary operators must be on ap_private instances of the same bit width. + /// Attempting to use these operators on instances with different bit + /// widths will yield an assertion. + /// * The value is stored canonically as an unsigned value. For operations + /// where it makes a difference, there are both signed and unsigned variants + /// of the operation. For example, sdiv and udiv. However, because the bit + /// widths must be the same, operations such as Mul and Add produce the same + /// results regardless of whether the values are interpreted as signed or + /// not. + /// * In general, the class tries to follow the style of computation that LLVM + /// uses in its IR. This simplifies its use for LLVM. + /// + /// @brief Class for arbitrary precision integers. + template class ap_private; + namespace ap_private_ops{ + template + INLINE ap_private<_AP_W, _AP_S, _AP_N> lshr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt); + template + INLINE ap_private<_AP_W, _AP_S, _AP_N> shl(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt); + } + +#if defined(_MSC_VER) +# if _MSC_VER < 1400 && !defined(for) +# define for if(0);else for +# endif + typedef unsigned __int64 ap_ulong; + typedef signed __int64 ap_slong; +#else + typedef unsigned long long ap_ulong; + typedef signed long long ap_slong; +#endif + template struct retval { + }; + template<> struct retval { + typedef ap_slong Type; + }; + template<> struct retval { + typedef ap_ulong Type; + }; + + template + class ap_private { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef typename retval<_AP_S>::Type ValType; + template friend struct ap_fixed_base; + ///return type of variety of operations + //---------------------------------------------------------- + template + struct RType { + enum { + mult_w = _AP_W+_AP_W2, + mult_s = _AP_S||_AP_S2, + plus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + minus_s = true, + div_w = _AP_W+_AP_S2, + div_s = _AP_S||_AP_S2, + mod_w = AP_MIN(_AP_W,_AP_W2+(!_AP_S2&&_AP_S)), + mod_s = _AP_S, + logic_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + typedef ap_private mult; + typedef ap_private plus; + typedef ap_private minus; + typedef ap_private logic; + typedef ap_private div; + typedef ap_private mod; + typedef ap_private<_AP_W, _AP_S> arg1; + typedef bool reduce; + }; + + INLINE void report() { +#if 0 + if (_AP_W > 1024 && _AP_W <= 4096) { + fprintf(stderr, "[W] W=%d is out of bound (1<=W<=1024): for" + " synthesis: please define macro AP_INT_TYPE_EXT(N)" + " to extend the valid range.\n", _AP_W); + } else +#endif + if (_AP_W > MAX_MODE(AP_INT_MAX_W) * 1024) { + fprintf(stderr, "[E] ap_%sint<%d>: Bitwidth exceeds the " + "default max value %d. Please use macro " + "AP_INT_MAX_W to set a larger max value.\n", + _AP_S?"":"u", _AP_W, + MAX_MODE(AP_INT_MAX_W) * 1024); + exit(1); + } + } + + enum { BitWidth = _AP_W }; + /// This union is used to store the integer value. When the + /// integer bit-width <= 64, it uses VAL, otherwise it uses pVal. + + /// This enum is used to hold the constants we needed for ap_private. + uint64_t VAL; ///< Used to store the <= 64 bits integer value. + uint64_t pVal[_AP_N]; ///< Used to store the >64 bits integer value. + + /// This enum is used to hold the constants we needed for ap_private. + enum { + APINT_BITS_PER_WORD = sizeof(uint64_t) * 8, ///< Bits in a word + APINT_WORD_SIZE = sizeof(uint64_t) ///< Byte size of a word + }; + + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -(_AP_W%APINT_BITS_PER_WORD) : 0}; + static const uint64_t mask = ((uint64_t)~0ULL >> (excess_bits)); + + /// This constructor is used only internally for speed of construction of + /// temporaries. It is unsafe for general use so it is not public. + /* Constructors */ + + ap_private(const char* val) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + bool neg = false; + uint32_t radix = 16; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + ap_private ap_private_val(str.c_str(), strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + report(); + } + + ap_private(const char* val, int rd) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + uint32_t radix = rd; + bool neg = false; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + // uint32_t bitsNeeded = ap_private<_AP_W, _AP_S>::getBitsNeeded(strp, strLen, radix); + // ap_private<_AP_W, _AP_S> ap_private_val(bitsNeeded, strp , strLen, radix, base, offset); + ap_private ap_private_val(strp , strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + report(); + } + + /// Note that numWords can be smaller or larger than the corresponding bit + /// width but any extraneous bits will be dropped. + /// @param numBits the bit width of the constructed ap_private + /// @param numWords the number of words in bigVal + /// @param bigVal a sequence of words to form the initial value of the ap_private + /// @brief Construct an ap_private of numBits width, initialized as bigVal[]. + ap_private(uint32_t numWords, const uint64_t bigVal[]): VAL(0) { + assert(bigVal && "Null pointer detected!"); + { + // Get memory, cleared to 0 + memset(pVal, 0, _AP_N * sizeof(uint64_t)); + + // Calculate the number of words to copy + uint32_t words = AESL_std::min(numWords, _AP_N); + // Copy the words from bigVal to pVal + memcpy(pVal, bigVal, words * APINT_WORD_SIZE); + if (words >= _AP_W) + clearUnusedBits(); + // Make sure unused high bits are cleared + } + } + + /// This constructor interprets Val as a string in the given radix. The + /// interpretation stops when the first charater that is not suitable for the + /// radix is encountered. Acceptable radix values are 2, 8, 10 and 16. It is + /// an error for the value implied by the string to require more bits than + /// numBits. + /// @param numBits the bit width of the constructed ap_private + /// @param val the string to be interpreted + /// @param radix the radix of Val to use for the intepretation + /// @brief Construct an ap_private from a string representation. + ap_private(const std::string& val, uint8_t radix=2, int base=0, int offset=0): VAL(0) { + assert(!val.empty() && "The input string is empty."); + const char *c_str = val.c_str(); + fromString(c_str+base+offset, val.size()-base-offset, radix); + } + + /// This constructor interprets the slen characters starting at StrStart as + /// a string in the given radix. The interpretation stops when the first + /// character that is not suitable for the radix is encountered. Acceptable + /// radix values are 2, 8, 10 and 16. It is an error for the value implied by + /// the string to require more bits than numBits. + /// @param numBits the bit width of the constructed ap_private + /// @param strStart the start of the string to be interpreted + /// @param slen the maximum number of characters to interpret + /// @param radix the radix to use for the conversion + /// @brief Construct an ap_private from a string representation. + /// This method does not consider whether it is negative or not. + ap_private(const char strStart[], uint32_t slen, uint8_t radix, int base=0, int offset=0) : VAL(0) { + fromString(strStart+base+offset, slen-base-offset, radix); + } + + template + INLINE ap_private(const ap_range_ref<_AP_W2,_AP_S2>& ref) { + *this=ref.get(); + report(); + } + + template + INLINE ap_private(const ap_bit_ref<_AP_W2,_AP_S2>& ref) { + *this = ((uint64_t)(bool)ref); + report(); + } + + template + INLINE ap_private(const ap_concat_ref<_AP_W2, _AP_T2,_AP_W3, _AP_T3>& ref) { + *this=ref.get(); + report(); + } + + template + INLINE ap_private(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) { + *this = ((val.operator ap_private<_AP_W2, false> ())); + report(); + } + + template + INLINE ap_private(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) { + *this = (uint64_t)(bool)val; + report(); + } + + /// Simply makes *this a copy of that. + /// @brief Copy Constructor. + template + ap_private(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& that): VAL(0) { + operator = (const_cast& >(that)); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that): VAL(0) { + operator = (that); + } + + template + explicit ap_private(const ap_private<_AP_W1, _AP_S1, 1>& that): VAL(0) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (that.isNegative()) { + pVal[0] = that.VAL|that_sign_ext_mask; + memset(pVal+1, ~0, sizeof(uint64_t)*(_AP_N-1)); + } else { + pVal[0] = that.VAL; + memset(pVal+1, 0, sizeof(uint64_t)*(_AP_N-1)); + } + clearUnusedBits(); + } + + ap_private(const ap_private& that): VAL(0) { + memcpy(pVal, that.pVal, _AP_N * APINT_WORD_SIZE); + clearUnusedBits(); + } + + /// @brief Destructor. + virtual ~ap_private() {} + + /// Default constructor that creates an uninitialized ap_private. This is useful + /// for object deserialization (pair this with the static method Read). + ap_private(){memset(pVal, 0, sizeof(uint64_t)*(_AP_N));} + + ap_private(uint64_t* val, uint32_t bits=_AP_W) {assert(0);} + ap_private(const uint64_t *const val, uint32_t bits) {assert(0);} + + /// @name Constructors + /// @{ + /// If isSigned is true then val is treated as if it were a signed value + /// (i.e. as an int64_t) and the appropriate sign extension to the bit width + /// will be done. Otherwise, no sign extension occurs (high order bits beyond + /// the range of val are zero filled). + /// @param numBits the bit width of the constructed ap_private + /// @param val the initial value of the ap_private + /// @param isSigned how to treat signedness of val + /// @brief Create a new ap_private of numBits width, initialized as val. +#define CTOR(TYPE, SIGNED) \ + ap_private(TYPE val, bool isSigned=SIGNED) { \ + pVal[0] = val; \ + if (isSigned && int64_t(pVal[0]) < 0) { \ + memset(pVal+1, ~0, sizeof(uint64_t)*(_AP_N-1)); \ + } else { \ + memset(pVal+1, 0, sizeof(uint64_t)*(_AP_N-1)); \ + } \ + clearUnusedBits(); \ + } +#if 1 + CTOR(int, true) + CTOR(bool, false) + CTOR(signed char, true) + CTOR(unsigned char, false) + CTOR(short, true) + CTOR(unsigned short, false) + CTOR(unsigned int, false) + CTOR(long, true) + CTOR(unsigned long, false) + CTOR(unsigned long long, false) + CTOR(long long, true) + CTOR(float, false) + CTOR(double, false) +#undef CTOR +#else + CTOR(uint64_t) +#undef CTOR +#endif + + + /// @returns true if the number of bits <= 64, false otherwise. + /// @brief Determine if this ap_private just has one word to store value. + INLINE bool isSingleWord() const { + return false; + } + + /// @returns the word position for the specified bit position. + /// @brief Determine which word a bit is in. + static uint32_t whichWord(uint32_t bitPosition) { + // return bitPosition / APINT_BITS_PER_WORD; + return (bitPosition) >> 6; + } + + /// @returns the bit position in a word for the specified bit position + /// in the ap_private. + /// @brief Determine which bit in a word a bit is in. + static uint32_t whichBit(uint32_t bitPosition) { + // return bitPosition % APINT_BITS_PER_WORD; + return bitPosition & 0x3f; + } + + /// bit at a specific bit position. This is used to mask the bit in the + /// corresponding word. + /// @returns a uint64_t with only bit at "whichBit(bitPosition)" set + /// @brief Get a single bit mask. + static uint64_t maskBit(uint32_t bitPosition) { + return 1ULL << (whichBit(bitPosition)); + } + + /// @returns the corresponding word for the specified bit position. + /// @brief Get the word corresponding to a bit position + INLINE uint64_t getWord(uint32_t bitPosition) const { + return isSingleWord() ? VAL : pVal[whichWord(bitPosition)]; + } + + /// This method is used internally to clear the to "N" bits in the high order + /// word that are not used by the ap_private. This is needed after the most + /// significant word is assigned a value to ensure that those bits are + /// zero'd out. + /// @brief Clear unused high order bits + INLINE void clearUnusedBits(void) { + pVal[_AP_N-1] = _AP_S ? ((((int64_t)pVal[_AP_N-1])<<(excess_bits))>> excess_bits) : (excess_bits ? ((pVal[_AP_N-1])<<(excess_bits))>>(excess_bits) : pVal[_AP_N-1]); + } + + INLINE void clearUnusedBitsToZero(void) { + pVal[_AP_N-1] &= mask; + } + + INLINE void clearUnusedBitsToOne(void) { + pVal[_AP_N-1] |= mask; + } + + /// This is used by the constructors that take string arguments. + /// @brief Convert a char array into an ap_private + INLINE void fromString(const char *strStart, uint32_t slen, + uint8_t radix) ; + + INLINE ap_private read() volatile { + return *this; + } + + INLINE void write(const ap_private& op2) volatile { + *this = (op2); + } + + //Explicit conversions to C interger types + //----------------------------------------------------------- + operator ValType() const { + return getVal(); + } + + INLINE ValType getVal() const{ + return *pVal; + } + + INLINE int to_int() const { + return int(*this); + } + + INLINE unsigned to_uint() const { + return (unsigned) getVal(); + } + + INLINE long to_long() const { + return (long) getVal(); + } + + INLINE unsigned long to_ulong() const { + return (unsigned long) getVal(); + } + + INLINE ap_slong to_int64() const { + return (ap_slong) getVal(); + } + + INLINE ap_ulong to_uint64() const { + return (ap_ulong) getVal(); + } + + INLINE double to_double() const { + if (isNegative()) + return roundToDouble(true); + else + return roundToDouble(false); + } + + INLINE unsigned length() const { return _AP_W; } + + /*Reverse the contents of ap_private instance. I.e. LSB becomes MSB and vise versa*/ + INLINE ap_private& reverse () { + for (int i = 0; i < _AP_W/2; ++i) { + bool tmp = operator[](i); + if (operator[](_AP_W - 1 - i)) + set(i); + else + clear(i); + if (tmp) + set(_AP_W - 1 - i); + else + clear(_AP_W - 1 - i); + } + clearUnusedBits(); + return *this; + } + + /*Return true if the value of ap_private instance is zero*/ + INLINE bool iszero () const { + return isMinValue(); + } + + /* x < 0 */ + INLINE bool sign () const { + if (isNegative()) + return true; + return false; + } + + /* x[i] = !x[i] */ + INLINE void invert (int i) { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + flip(i); + } + + /* x[i] */ + INLINE bool test (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator[](i); + } + + //Set the ith bit into v + INLINE void set (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Set the ith bit into v + INLINE void set_bit (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + INLINE ap_private& set(uint32_t bitPosition) { + pVal[whichWord(bitPosition)] |= maskBit(bitPosition); + clearUnusedBits(); + return *this; + } + + INLINE void set() { + for (uint32_t i = 0; i < _AP_N; ++i) + pVal[i] = ~0ULL; + clearUnusedBits(); + } + + //Get the value of ith bit + INLINE bool get (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //Get the value of ith bit + INLINE bool get_bit (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the left + INLINE void lrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (shl(n) | lshr(_AP_W - n)); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the right + INLINE void rrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (lshr(n) | shl(_AP_W - n)); + } + + /// Set the given bit to 0 whose position is given as "bitPosition". + /// @brief Set a given bit to 0. + ap_private& clear(uint32_t bitPosition) { + pVal[whichWord(bitPosition)] &= ~maskBit(bitPosition); + clearUnusedBits(); + return *this; + } + + /// @brief Set every bit to 0. + void clear() { + memset(pVal, 0, _AP_N * APINT_WORD_SIZE); + } + + /// @brief Toggle every bit to its opposite value. + ap_private& flip() { + for (uint32_t i = 0; i < _AP_N; ++i) + pVal[i] ^= ~0ULL; + clearUnusedBits(); + return *this; + } + + /// Toggle a given bit to its opposite value whose position is given + /// as "bitPosition". + /// @brief Toggles a given bit to its opposite value. + ap_private& flip(uint32_t bitPosition) { + assert(bitPosition < BitWidth && "Out of the bit-width range!"); + if ((*this)[bitPosition]) clear(bitPosition); + else set(bitPosition); + return *this; + } + + //complements every bit + INLINE void b_not() { + flip(); + } + + ap_private getLoBits(uint32_t numBits) const { + return ap_private_ops::lshr(ap_private_ops::shl(*this, _AP_W - numBits), + _AP_W - numBits); + } + + ap_private getHiBits(uint32_t numBits) const { + return ap_private_ops::lshr(*this, _AP_W - numBits); + } + + //Binary Arithmetic + //----------------------------------------------------------- + + template + INLINE ap_private + operator & (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this & a2.get(); + } + + template + INLINE ap_private + operator | (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this | a2.get(); + } + + template + INLINE ap_private + operator ^ (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this ^ a2.get(); + } + + + ///Arithmetic assign + //------------------------------------------------------------- + +#define OP_BIN_LOGIC_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { \ + uint32_t numWords = AESL_std::min(_AP_N, _AP_N1); \ + uint32_t i; \ + for (i = 0; i < numWords; ++i) \ + pVal[i] Sym RHS.pVal[i]; \ + if (_AP_N1 < _AP_N) { \ + uint64_t ext = RHS.isNegative()?~0ULL:0; \ + for (;i<_AP_N; i++) \ + pVal[i] Sym ext; \ + } \ + clearUnusedBits(); \ + return *this; \ + } + + OP_BIN_LOGIC_ASSIGN_AP(&=); + OP_BIN_LOGIC_ASSIGN_AP(|=); + OP_BIN_LOGIC_ASSIGN_AP(^=); +#undef OP_BIN_LOGIC_ASSIGN_AP + + /// Adds the RHS APint to this ap_private. + /// @returns this, after addition of RHS. + /// @brief Addition assignment operator. + template + INLINE ap_private& operator+=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + add(pVal, pVal, RHS.pVal, _AP_N, _AP_N, _AP_N1, _AP_S, _AP_S1); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator-=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + sub(pVal, pVal, RHS.pVal, _AP_N, _AP_N, _AP_N1, _AP_S, _AP_S1); + clearUnusedBits(); + return *this; + } + + template + ap_private& operator*=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + // Get some bit facts about LHS and check for zero + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + if (!lhsWords) { + // 0 * X ===> 0 + return *this; + } + + ap_private dupRHS = RHS; + // Get some bit facts about RHS and check for zero + uint32_t rhsBits = dupRHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + if (!rhsWords) { + // X * 0 ===> 0 + clear(); + return *this; + } + + // Allocate space for the result + uint32_t destWords = rhsWords + lhsWords; + uint64_t *dest = getMemory(destWords); + + // Perform the long multiply + mul(dest, pVal, lhsWords, dupRHS.pVal, rhsWords, destWords); + + // Copy result back into *this + clear(); + uint32_t wordsToCopy = destWords >= _AP_N ? _AP_N : destWords; + + memcpy(pVal, dest, wordsToCopy* APINT_WORD_SIZE); + + uint64_t ext = (isNegative() ^ RHS.isNegative()) ? ~0ULL : 0ULL; + for (int i=wordsToCopy; i<_AP_N; i++) + pVal[i]=ext; + clearUnusedBits(); + // delete dest array and return + free(dest); + return *this; + } + +#define OP_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op); \ + return *this; \ + } \ + + OP_ASSIGN_AP(/) + OP_ASSIGN_AP(%) +#undef OP_ASSIGN_AP + +#define OP_BIN_LOGIC_AP(Sym) \ + template \ + INLINE \ + typename RType<_AP_W1, _AP_S1>::logic \ + operator Sym (const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { \ + enum { numWords = (RType<_AP_W1, _AP_S1>::logic_w +APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; \ + typename RType<_AP_W1, _AP_S1>::logic Result; \ + uint64_t *val = Result.pVal; \ + uint32_t i; \ + uint32_t min_N = std::min(_AP_N, _AP_N1); \ + uint32_t max_N = std::max(_AP_N, _AP_N1); \ + for (i = 0; i < min_N; ++i) \ + val[i] = pVal[i] Sym RHS.pVal[i]; \ + if (numWords > i) { \ + const uint64_t* tmpVal = (_AP_N>_AP_N1 ? pVal : RHS.pVal)+i; \ + uint64_t ext = ((_AP_N<_AP_N1 && isNegative() )||(_AP_N1 < _AP_N && RHS.isNegative())) ? ~0ULL : 0; \ + for (;i i) { \ + uint64_t ext2 = ((_AP_N>_AP_N1 && isNegative() )||(_AP_N1 > _AP_N && RHS.isNegative())) ? ~0ULL : 0; \ + val[i] = ext Sym ext2; \ + } \ + } \ + Result.clearUnusedBits(); \ + return Result; \ + } + + OP_BIN_LOGIC_AP(|); + OP_BIN_LOGIC_AP(&); + OP_BIN_LOGIC_AP(^); + +#undef OP_BIN_LOGIC_AP + + template + INLINE typename RType<_AP_W1,_AP_S1>::plus operator+(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + // assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); + typename RType<_AP_W1,_AP_S1>::plus Result; + bool carry = add(Result.pVal, this->pVal, RHS.pVal, (RType<_AP_W1,_AP_S1>::plus_w + 63) / 64, _AP_N, _AP_N1, _AP_S, _AP_S1); + if ((RType<_AP_W1,_AP_S1>::plus_w + 63) / 64> std::max(_AP_W, _AP_W1) ) + Result.pVal[(RType<_AP_W1,_AP_S1>::plus_w + 63)/64 - 1] = carry; + Result.clearUnusedBits(); + return Result; + } + + template + INLINE typename RType<_AP_W1,_AP_S1>::minus operator-(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + typename RType<_AP_W1,_AP_S1>::minus Result; + bool borrow = sub(Result.pVal, this->pVal, RHS.pVal, (RType<_AP_W1,_AP_S1>::minus_w + 63) / 64, _AP_N, _AP_N1, _AP_S, _AP_S1); + if ((RType<_AP_W1,_AP_S1>::minus_w + 63) / 64 > AESL_std::max(_AP_W, _AP_W1) ) { + Result.pVal[(RType<_AP_W1,_AP_S1>::minus_w+63)/64 - 1] = borrow; + } + Result.clearUnusedBits(); + return Result; + } + + template + typename RType<_AP_W1, _AP_S1>::mult operator*(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + + // Get some bit facts about LHS and check for zero + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + if (!lhsWords) + // 0 * X ===> 0 + return typename RType<_AP_W1, _AP_S1>::mult(); + + // Get some bit facts about RHS and check for zero + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + if (!rhsWords) { + // X * 0 ===> 0 + return typename RType<_AP_W1, _AP_S1>::mult(); + } + + //extend size to avoid result loss + typename RType<_AP_W1, _AP_S1>::mult dupLHS = *this; + typename RType<_AP_W1, _AP_S1>::mult dupRHS = RHS; + lhsBits = dupLHS.getActiveBits(); + lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + rhsBits = dupRHS.getActiveBits(); + rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + + // Allocate space for the result + enum { destWords =(RType<_AP_W1, _AP_S1>::mult_w+APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; + int destw = destWords; + typename RType<_AP_W1, _AP_S1>::mult Result; + uint64_t *dest = Result.pVal; + uint64_t ext = (isNegative() ^ RHS.isNegative()) ? ~0ULL : 0; + + // Perform the long multiply + mul(dest, dupLHS.pVal, lhsWords, dupRHS.pVal, rhsWords, destWords); + + for (int i=lhsWords+rhsWords; i + INLINE typename RType<_AP_W2,_AP_S2>::div + operator / (const ap_private<_AP_W2,_AP_S2>& op) const { + ap_private lhs=ap_private(*this); + ap_private rhs=ap_private(op); + return typename RType<_AP_W2,_AP_S2>::div((_AP_S||_AP_S2)?lhs.sdiv(rhs):lhs.udiv(rhs)); + } + + template + INLINE typename RType<_AP_W2,_AP_S2>::mod + operator % (const ap_private<_AP_W2,_AP_S2>& op) const { + ap_private lhs=*this; + ap_private rhs= op; + typename RType<_AP_W2,_AP_S2>::mod res = typename RType<_AP_W2,_AP_S2>::mod(_AP_S?lhs.srem(rhs):lhs.urem(rhs)); + return res; + } + + template + INLINE ap_private + operator << (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh=op2.to_uint(); + return *this << sh; + } + + INLINE ap_private + operator << (uint32_t sh) const { + ap_private r(*this); + bool overflow=(sh>=length()); + if(overflow) + r.clear(); + else + r = ap_private(r.shl(sh)); + return r; + } + + template + INLINE ap_private + operator >> (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh = op2.to_uint(); + return *this >> sh; + } + + INLINE ap_private + operator >> (uint32_t sh) const { + ap_private r(*this); + bool overflow=(sh>=_AP_W); + bool neg_v=r.isNegative(); + if(_AP_S) { + if(overflow) + neg_v?r.set():r.clear(); + else + return r.ashr(sh); + } else { + if(overflow) + r.clear(); + else + return r.lshr(sh); + } + return r; + } + + ///Shift assign + //------------------------------------------------------------------ +#define OP_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym##=(int op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } \ + INLINE ap_private& operator Sym##=(unsigned int op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } \ + template \ + INLINE ap_private& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } + OP_ASSIGN_AP(>>) + OP_ASSIGN_AP(<<) +#undef OP_ASSIGN_AP + ///Comparisons + //----------------------------------------------------------------- + bool operator==(const ap_private& RHS) const { + // Get some facts about the number of bits used in the two operands. + uint32_t n1 = getActiveBits(); + uint32_t n2 = RHS.getActiveBits(); + + // If the number of bits isn't the same, they aren't equal + if (n1 != n2) + return false; + + // If the number of bits fits in a word, we only need to compare the low word. + if (n1 <= APINT_BITS_PER_WORD) + return pVal[0] == RHS.pVal[0]; + + // Otherwise, compare everything + for (int i = whichWord(n1 - 1); i >= 0; --i) + if (pVal[i] != RHS.pVal[i]) + return false; + return true; + } + + template + INLINE bool operator == (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S|_AP_S2> lhs(*this); + ap_private<_AP_MAX_W, _AP_S|_AP_S2> rhs(op); + return lhs==rhs; + } + + bool operator==(uint64_t Val) const { + uint32_t n = getActiveBits(); + if (n <= APINT_BITS_PER_WORD) + return pVal[0] == Val; + else + return false; + } + + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2>& op) const { + return !(*this==op); + } + + template + INLINE bool operator!=(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !((*this) == RHS); + } + + INLINE bool operator!=(uint64_t Val) const { + return !((*this) == Val); + } + + + template + INLINE bool operator <= (const ap_private<_AP_W2,_AP_S2>& op) const { + return !(*this>op); + } + + INLINE bool operator <(const ap_private& op) const { + return _AP_S ? slt(op):ult(op); + } + + template + INLINE bool operator < (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.slt(rhs):lhs.ult(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + } + + template + INLINE bool operator >=(const ap_private<_AP_W2,_AP_S2>& op) const { + return !(*this + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.sgt(rhs):lhs.ugt(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ugt(rhs); + else + return lhs.sgt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ugt(rhs); + else + return lhs.sgt(rhs); + } + + ///Bit and Part Select + //-------------------------------------------------------------- + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast*>(this), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>((const_cast*> (this)), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast(this), Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (uint32_t index) { + assert(index >= 0&&"Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + template + INLINE bool operator [] (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br =operator [] (index); + return br.to_bool(); + } + + INLINE bool operator [](uint32_t bitPosition) const { + return (maskBit(bitPosition) & (pVal[whichWord(bitPosition)])) != 0; + } + + INLINE ap_bit_ref<_AP_W,_AP_S> bit (int index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> bit (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W &&"Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + INLINE bool bit (int index) const { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br(const_cast*>(this), index); + return br.to_bool(); + } + + template + INLINE bool bit (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br = bit(index); + return br.to_bool(); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(ap_private<_AP_W2,_AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(const ap_private<_AP_W2,_AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (ap_range_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (ap_bit_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + INLINE ap_private<_AP_W,false> get() const { + ap_private<_AP_W,false> ret(*this); + return ret; + } + + template + INLINE void set(const ap_private<_AP_W3, false> & val) { + operator = (ap_private<_AP_W3, _AP_S>(val)); + } + + /// @} + /// @name Value Tests + /// @{ + /// This tests the high bit of this ap_private to determine if it is set. + /// @returns true if this ap_private is negative, false otherwise + /// @brief Determine sign of this ap_private. + INLINE bool isNegative() const { + //just for get rid of warnings + enum {shift = (_AP_W-APINT_BITS_PER_WORD*(_AP_N-1)-1)}; + static const uint64_t mask = 1ULL << (shift); + return _AP_S && (pVal[_AP_N-1]&mask); + } + + /// This tests the high bit of the ap_private to determine if it is unset. + /// @brief Determine if this ap_private Value is positive (not negative). + INLINE bool isPositive() const { + return !isNegative(); + } + + /// This tests if the value of this ap_private is strictly positive (> 0). + /// @returns true if this ap_private is Positive and not zero. + /// @brief Determine if this ap_private Value is strictly positive. + INLINE bool isStrictlyPositive() const { + return isPositive() && (*this) != 0; + } + + /// This checks to see if the value has all bits of the ap_private are set or not. + /// @brief Determine if all bits are set + INLINE bool isAllOnesValue() const { + return countPopulation() == _AP_W; + } + + /// This checks to see if the value of this ap_private is the maximum unsigned + /// value for the ap_private's bit width. + /// @brief Determine if this is the largest unsigned value. + INLINE bool isMaxValue() const { + return countPopulation() == _AP_W; + } + + /// This checks to see if the value of this ap_private is the maximum signed + /// value for the ap_private's bit width. + /// @brief Determine if this is the largest signed value. + INLINE bool isMaxSignedValue() const { + return BitWidth == 1 ? VAL == 0 : + !isNegative() && countPopulation() == _AP_W - 1; + } + + /// This checks to see if the value of this ap_private is the minimum unsigned + /// value for the ap_private's bit width. + /// @brief Determine if this is the smallest unsigned value. + INLINE bool isMinValue() const { + return countPopulation() == 0; + } + + /// This checks to see if the value of this ap_private is the minimum signed + /// value for the ap_private's bit width. + /// @brief Determine if this is the smallest signed value. + INLINE bool isMinSignedValue() const { + return BitWidth == 1 ? VAL == 1 : + isNegative() && countPopulation() == 1; + } + + /// This function returns a pointer to the internal storage of the ap_private. + /// This is useful for writing out the ap_private in binary form without any + /// conversions. + INLINE const uint64_t* getRawData() const { + if (isSingleWord()) + return &VAL; + return &pVal[0]; + } + + ap_private sqrt() const; + + /// @} + /// @Assignment Operators + /// @{ + /// @returns *this after assignment of RHS. + /// @brief Copy assignment operator. + INLINE ap_private& operator=(const ap_private& RHS) { + if (this != &RHS) + memcpy(pVal, RHS.pVal, _AP_N * APINT_WORD_SIZE); + return *this; + } + INLINE ap_private& operator=(const volatile ap_private& RHS) { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + INLINE volatile ap_private& operator=(const ap_private& RHS) volatile { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + INLINE volatile ap_private& operator=(const volatile ap_private& RHS) volatile { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + + template + INLINE ap_private& operator=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + if (_AP_S1) + cpSextOrTrunc(RHS); + else + cpZextOrTrunc(RHS); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + if (_AP_S1) + cpSextOrTrunc(RHS); + else + cpZextOrTrunc(RHS); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const ap_private<_AP_W1, _AP_S1, 1>& RHS) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (RHS.isNegative()) { + pVal[0] = RHS.VAL | that_sign_ext_mask; + memset(pVal+1,~0, APINT_WORD_SIZE*(_AP_N-1)); + } else { + pVal[0] = RHS.VAL; + memset(pVal+1, 0, APINT_WORD_SIZE*(_AP_N-1)); + } + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const volatile ap_private<_AP_W1, _AP_S1, 1>& RHS) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (RHS.isNegative()) { + pVal[0] = RHS.VAL | that_sign_ext_mask; + memset(pVal+1,~0, APINT_WORD_SIZE*(_AP_N-1)); + } else { + pVal[0] = RHS.VAL; + memset(pVal+1, 0, APINT_WORD_SIZE*(_AP_N-1)); + } + clearUnusedBits(); + return *this; + } + + /// @} + /// @name Unary Operators + /// @{ + /// @returns a new ap_private value representing *this incremented by one + /// @brief Postfix increment operator. + INLINE const ap_private operator++(int) { + ap_private API(*this); + ++(*this); + return API; + } + + /// @returns *this incremented by one + /// @brief Prefix increment operator. + INLINE ap_private& operator++() { + add_1(pVal, pVal, _AP_N, 1); + clearUnusedBits(); + return *this; + } + + /// @returns a new ap_private representing *this decremented by one. + /// @brief Postfix decrement operator. + INLINE const ap_private operator--(int) { + ap_private API(*this); + --(*this); + return API; + } + + /// @returns *this decremented by one. + /// @brief Prefix decrement operator. + INLINE ap_private& operator--() { + sub_1(pVal, _AP_N, 1); + clearUnusedBits(); + return *this; + } + + /// Performs a bitwise complement operation on this ap_private. + /// @returns an ap_private that is the bitwise complement of *this + /// @brief Unary bitwise complement operator. + INLINE ap_private operator~() const { + ap_private Result(*this); + Result.flip(); + return Result; + } + + /// Negates *this using two's complement logic. + /// @returns An ap_private value representing the negation of *this. + /// @brief Unary negation operator + INLINE typename RType<1,false>::minus operator-() const { + return ap_private<1,false>(0) - (*this); + } + + /// Performs logical negation operation on this ap_private. + /// @returns true if *this is zero, false otherwise. + /// @brief Logical negation operator. + INLINE bool operator !() const { + for (uint32_t i = 0; i < _AP_N; ++i) + if (pVal[i]) + return false; + return true; + } + + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> And(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator&(RHS); + } + template + INLINE ap_private Or(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator|(RHS); + } + template + ap_private Xor(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator^(RHS); + } + + ap_private Mul(const ap_private& RHS) const { + ap_private Result(*this); + Result *= RHS; + return Result; + } + + ap_private Add(const ap_private& RHS) const { + ap_private Result(0); + bool carry = add(Result.pVal, this->pVal, RHS.pVal, _AP_N, _AP_N, _AP_N, _AP_S, _AP_S); + Result.clearUnusedBits(); + return Result; + } + + ap_private Sub(const ap_private& RHS) const { + ap_private Result(0); + sub(Result.pVal, this->pVal, RHS.pVal, _AP_N, _AP_N, _AP_N, _AP_S, _AP_S); + Result.clearUnusedBits(); + return Result; + } + + /// Arithmetic right-shift this ap_private by shiftAmt. + /// @brief Arithmetic right-shift function. + ap_private ashr(uint32_t shiftAmt) const { + assert(shiftAmt <= BitWidth && "Invalid shift amount, too big"); + // Handle a degenerate case + if (shiftAmt == 0) + return *this; + + // Handle single word shifts with built-in ashr + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(/*BitWidth, 0*/); // undefined + else { + uint32_t SignBit = APINT_BITS_PER_WORD - BitWidth; + return ap_private(/*BitWidth,*/ + (((int64_t(VAL) << (SignBit)) >> (SignBit)) >> (shiftAmt))); + } + } + + // If all the bits were shifted out, the result is, technically, undefined. + // We return -1 if it was negative, 0 otherwise. We check this early to avoid + // issues in the algorithm below. + if (shiftAmt == BitWidth) { + if (isNegative()) + return ap_private(-1); + else + return ap_private(0); + } + + // Create some space for the result. + ap_private Retval(0); + uint64_t * val = Retval.pVal; + + // Compute some values needed by the following shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; // bits to shift per word + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; // word offset for shift + uint32_t breakWord = _AP_N - 1 - offset; // last word affected + uint32_t bitsInWord = whichBit(BitWidth); // how many bits in last word? + if (bitsInWord == 0) + bitsInWord = APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + // Move the words containing significant bits + for (uint32_t i = 0; i <= breakWord; ++i) + val[i] = pVal[i+offset]; // move whole word + + // Adjust the top significant word for sign bit fill, if negative + if (isNegative()) + if (bitsInWord < APINT_BITS_PER_WORD) + val[breakWord] |= ~0ULL << (bitsInWord); // set high bits + } else { + // Shift the low order words + for (uint32_t i = 0; i < breakWord; ++i) { + // This combines the shifted corresponding word with the low bits from + // the next word (shifted into this word's high bits). + val[i] = ((pVal[i+offset]) >> (wordShift)); + val[i] |= ((pVal[i+offset+1]) << (APINT_BITS_PER_WORD - wordShift)); + } + + // Shift the break word. In this case there are no bits from the next word + // to include in this word. + val[breakWord] = (pVal[breakWord+offset]) >> (wordShift); + + // Deal with sign extenstion in the break word, and possibly the word before + // it. + if (isNegative()) { + if (wordShift > bitsInWord) { + if (breakWord > 0) + val[breakWord-1] |= + ~0ULL << (APINT_BITS_PER_WORD - (wordShift - bitsInWord)); + val[breakWord] |= ~0ULL; + } else + val[breakWord] |= (~0ULL << (bitsInWord - wordShift)); + } + } + + // Remaining words are 0 or -1, just assign them. + uint64_t fillValue = (isNegative() ? ~0ULL : 0); + for (uint32_t i = breakWord+1; i < _AP_N; ++i) + val[i] = fillValue; + Retval.clearUnusedBits(); + return Retval; + } + + /// Logical right-shift this ap_private by shiftAmt. + /// @brief Logical right-shift function. + ap_private lshr(uint32_t shiftAmt) const { + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(0); + else + return ap_private((this->VAL) >> (shiftAmt)); + } + + // If all the bits were shifted out, the result is 0. This avoids issues + // with shifting by the size of the integer type, which produces undefined + // results. We define these "undefined results" to always be 0. + if (shiftAmt == BitWidth) + return ap_private(0); + + // If none of the bits are shifted out, the result is *this. This avoids + // issues with shifting byt he size of the integer type, which produces + // undefined results in the code below. This is also an optimization. + if (shiftAmt == 0) + return *this; + + // Create some space for the result. + ap_private Retval(0); + uint64_t * val = Retval.pVal; + + // If we are shifting less than a word, compute the shift with a simple carry + if (shiftAmt < APINT_BITS_PER_WORD) { + uint64_t carry = 0; + for (int i = _AP_N-1; i >= 0; --i) { + val[i] = ((pVal[i]) >> (shiftAmt)) | carry; + carry = (pVal[i]) << (APINT_BITS_PER_WORD - shiftAmt); + } + Retval.clearUnusedBits(); + return Retval; + } + + // Compute some values needed by the remaining shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + for (uint32_t i = 0; i < _AP_N - offset; ++i) + val[i] = pVal[i+offset]; + for (uint32_t i = _AP_N-offset; i < _AP_N; i++) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + // Shift the low order words + uint32_t breakWord = _AP_N - offset -1; + for (uint32_t i = 0; i < breakWord; ++i) + val[i] = ((pVal[i+offset]) >> (wordShift)) | + ((pVal[i+offset+1]) << (APINT_BITS_PER_WORD - wordShift)); + // Shift the break word. + val[breakWord] = (pVal[breakWord+offset]) >> (wordShift); + + // Remaining words are 0 + for (uint32_t i = breakWord+1; i < _AP_N; ++i) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + /// Left-shift this ap_private by shiftAmt. + /// @brief Left-shift function. + ap_private shl(uint32_t shiftAmt) const { + assert(shiftAmt <= BitWidth && "Invalid shift amount, too big"); + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(0); // avoid undefined shift results + return ap_private((VAL) << (shiftAmt)); + } + + // If all the bits were shifted out, the result is 0. This avoids issues + // with shifting by the size of the integer type, which produces undefined + // results. We define these "undefined results" to always be 0. + if (shiftAmt == BitWidth) + return ap_private(0); + + // If none of the bits are shifted out, the result is *this. This avoids a + // lshr by the words size in the loop below which can produce incorrect + // results. It also avoids the expensive computation below for a common case. + if (shiftAmt == 0) + return *this; + + // Create some space for the result. + ap_private Retval(0); + uint64_t* val = Retval.pVal; + // If we are shifting less than a word, do it the easy way + if (shiftAmt < APINT_BITS_PER_WORD) { + uint64_t carry = 0; + for (uint32_t i = 0; i < _AP_N; i++) { + val[i] = ((pVal[i]) << (shiftAmt)) | carry; + carry = (pVal[i]) >> (APINT_BITS_PER_WORD - shiftAmt); + } + Retval.clearUnusedBits(); + return Retval; + } + + // Compute some values needed by the remaining shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + for (uint32_t i = 0; i < offset; i++) + val[i] = 0; + for (uint32_t i = offset; i < _AP_N; i++) + val[i] = pVal[i-offset]; + Retval.clearUnusedBits(); + return Retval; + } + + // Copy whole words from this to Result. + uint32_t i = _AP_N - 1; + for (; i > offset; --i) + val[i] = (pVal[i-offset]) << (wordShift) | + (pVal[i-offset-1]) >> (APINT_BITS_PER_WORD - wordShift); + val[offset] = (pVal[0]) << (wordShift); + for (i = 0; i < offset; ++i) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + INLINE ap_private rotl(uint32_t rotateAmt) const { + if (rotateAmt == 0) + return *this; + // Don't get too fancy, just use existing shift/or facilities + ap_private hi(*this); + ap_private lo(*this); + hi.shl(rotateAmt); + lo.lshr(BitWidth - rotateAmt); + return hi | lo; + } + + INLINE ap_private rotr(uint32_t rotateAmt) const { + if (rotateAmt == 0) + return *this; + // Don't get too fancy, just use existing shift/or facilities + ap_private hi(*this); + ap_private lo(*this); + lo.lshr(rotateAmt); + hi.shl(BitWidth - rotateAmt); + return hi | lo; + } + + /// Perform an unsigned divide operation on this ap_private by RHS. Both this and + /// RHS are treated as unsigned quantities for purposes of this division. + /// @returns a new ap_private value containing the division result + /// @brief Unsigned division operation. + ap_private udiv(const ap_private& RHS) const { + assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); + + // First, deal with the easy case + if (isSingleWord()) { + assert(RHS.VAL != 0 && "Divide by zero?"); + return ap_private(VAL / RHS.VAL); + } + + // Get some facts about the LHS and RHS number of bits and words + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Divided by zero???"); + uint32_t lhsBits = this->getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + + // Deal with some degenerate cases + if (!lhsWords) + // 0 / X ===> 0 + return ap_private(0); + else if (lhsWords < rhsWords || this->ult(RHS)) { + // X / Y ===> 0, iff X < Y + return ap_private(0); + } else if (*this == RHS) { + // X / X ===> 1 + return ap_private(1); + } else if (lhsWords == 1 && rhsWords == 1) { + // All high words are zero, just use native divide + return ap_private(this->pVal[0] / RHS.pVal[0]); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Quotient(0); // to hold result. + divide(*this, lhsWords, RHS, rhsWords, &Quotient, (ap_private*)0); + return Quotient; + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + INLINE ap_private sdiv(const ap_private& RHS) const { + if (isNegative()) + if (RHS.isNegative()) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS.isNegative()) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + /// Perform an unsigned remainder operation on this ap_private with RHS being the + /// divisor. Both this and RHS are treated as unsigned quantities for purposes + /// of this operation. Note that this is a true remainder operation and not + /// a modulo operation because the sign follows the sign of the dividend + /// which is *this. + /// @returns a new ap_private value containing the remainder result + /// @brief Unsigned remainder operation. + ap_private urem(const ap_private& RHS) const { + if (isSingleWord()) { + assert(RHS.VAL != 0 && "Remainder by zero?"); + return ap_private(VAL % RHS.VAL); + } + + // Get some facts about the LHS + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + + // Get some facts about the RHS + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Performing remainder operation by zero ???"); + + // Check the degenerate cases + if (lhsWords == 0) { + // 0 % Y ===> 0 + return ap_private(0); + } else if (lhsWords < rhsWords || this->ult(RHS)) { + // X % Y ===> X, iff X < Y + return *this; + } else if (*this == RHS) { + // X % X == 0; + return ap_private(0); + } else if (lhsWords == 1) { + // All high words are zero, just use native remainder + return ap_private(pVal[0] % RHS.pVal[0]); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Remainder(0); + divide(*this, lhsWords, RHS, rhsWords, (ap_private*)(0), &Remainder); + return Remainder; + } + + ap_private urem(uint64_t RHS) const { + // Get some facts about the LHS + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + // Get some facts about the RHS + uint32_t rhsBits = 64 - CountLeadingZeros_64(RHS); // RHS.getActiveBits(); + uint32_t rhsWords = 1;//!rhsBits ? 0 : (ap_private<_AP_W, _AP_S, _AP_N>::whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Performing remainder operation by zero ???"); + // Check the degenerate cases + if (lhsWords == 0) { + // 0 % Y ===> 0 + return ap_private(0); + } else if (lhsWords < rhsWords || this->ult(RHS)) { + // X % Y ===> X, iff X < Y + return *this; + } else if (*this == RHS) { + // X % X == 0; + return ap_private(0); + } else if (lhsWords == 1) { + // All high words are zero, just use native remainder + return ap_private(pVal[0] % RHS); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Remainder(0); + divide(*this, lhsWords, RHS, (ap_private*)(0), &Remainder); + return Remainder; + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(const ap_private& RHS) const { + if (isNegative()) { + ap_private lhs = -(*this); + if (RHS.isNegative()) { + ap_private rhs = -RHS; + return -(lhs.urem(rhs)); + } else + return -(lhs.urem(RHS)); + } else if (RHS.isNegative()) { + ap_private rhs = -RHS; + return this->urem(rhs); + } + return this->urem(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return -((-(*this)).urem(-RHS)); + else + return -((-(*this)).urem(RHS)); + else if (RHS<0) + return this->urem(-RHS); + return this->urem(RHS); + } + + /// Sometimes it is convenient to divide two ap_private values and obtain both + /// the quotient and remainder. This function does both operations in the + /// same computation making it a little more efficient. + /// @brief Dual division/remainder interface. + static void udivrem(const ap_private& LHS, const ap_private& RHS, ap_private &Quotient, ap_private& Remainder) { + // Get some size facts about the dividend and divisor + uint32_t lhsBits = LHS.getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (ap_private::whichWord(lhsBits - 1) + 1); + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (ap_private::whichWord(rhsBits - 1) + 1); + + // Check the degenerate cases + if (lhsWords == 0) { + Quotient = 0; // 0 / Y ===> 0 + Remainder = 0; // 0 % Y ===> 0 + return; + } + + if (lhsWords < rhsWords || LHS.ult(RHS)) { + Quotient = 0; // X / Y ===> 0, iff X < Y + Remainder = LHS; // X % Y ===> X, iff X < Y + return; + } + + if (LHS == RHS) { + Quotient = 1; // X / X ===> 1 + Remainder = 0; // X % X ===> 0; + return; + } + + if (lhsWords == 1 && rhsWords == 1) { + // There is only one word to consider so use the native versions. + if (LHS.isSingleWord()) { + Quotient = ap_private(LHS.VAL / RHS.VAL); + Remainder = ap_private(LHS.VAL % RHS.VAL); + } else { + Quotient = ap_private(LHS.pVal[0] / RHS.pVal[0]); + Remainder = ap_private(LHS.pVal[0] % RHS.pVal[0]); + } + return; + } + + // Okay, lets do it the long way + divide(LHS, lhsWords, RHS, rhsWords, &Quotient, &Remainder); + } + + static void sdivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS.isNegative()) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS.isNegative()) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + /// Compares this ap_private with RHS for the validity of the equality + /// relationship. + /// @returns true if *this == Val + /// @brief Equality comparison. + template + INLINE bool eq(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return (*this) == RHS; + } + + /// Compares this ap_private with RHS for the validity of the inequality + /// relationship. + /// @returns true if *this != Val + /// @brief Inequality comparison + template + INLINE bool ne(const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) const { + return !((*this) == RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered unsigned. + /// @brief Unsigned less than comparison + template + INLINE bool ult(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + // Get active bit length of both operands + uint32_t n1 = getActiveBits(); + uint32_t n2 = RHS.getActiveBits(); + + // If magnitude of LHS is less than RHS, return true. + if (n1 < n2) + return true; + + // If magnitude of RHS is greather than LHS, return false. + if (n2 < n1) + return false; + + // If they bot fit in a word, just compare the low order word + if (n1 <= APINT_BITS_PER_WORD && n2 <= APINT_BITS_PER_WORD) + return pVal[0] < RHS.pVal[0]; + + // Otherwise, compare all words + uint32_t topWord = whichWord(AESL_std::max(n1,n2)-1); + for (int i = topWord; i >= 0; --i) { + if (pVal[i] > RHS.pVal[i]) + return false; + if (pVal[i] < RHS.pVal[i]) + return true; + } + return false; + } + + INLINE bool ult(uint64_t RHS) const { + // Get active bit length of both operands + uint32_t n1 = getActiveBits(); + uint32_t n2 = 64 - CountLeadingZeros_64(RHS); //RHS.getActiveBits(); + + // If magnitude of LHS is less than RHS, return true. + if (n1 < n2) + return true; + + // If magnitude of RHS is greather than LHS, return false. + if (n2 < n1) + return false; + + // If they bot fit in a word, just compare the low order word + if (n1 <= APINT_BITS_PER_WORD && n2 <= APINT_BITS_PER_WORD) + return pVal[0] < RHS; + assert(0); + } + + template + INLINE bool slt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + ap_private lhs(*this); + ap_private<_AP_W, _AP_S1, _AP_N> rhs(RHS); + bool lhsNeg = isNegative(); + bool rhsNeg = rhs.isNegative(); + if (lhsNeg) { + // Sign bit is set so perform two's complement to make it positive + lhs.flip(); + lhs++; + } + if (rhsNeg) { + // Sign bit is set so perform two's complement to make it positive + rhs.flip(); + rhs++; + } + + // Now we have unsigned values to compare so do the comparison if necessary + // based on the negativeness of the values. + if (lhsNeg) + if (rhsNeg) + return lhs.ugt(rhs); + else + return true; + else if (rhsNeg) + return false; + else + return lhs.ult(rhs); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered unsigned. + /// @brief Unsigned less or equal comparison + template + INLINE bool ule(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return ult(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered signed. + /// @brief Signed less or equal comparison + template + INLINE bool sle(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return slt(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered unsigned. + /// @brief Unsigned greather than comparison + template + INLINE bool ugt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !ult(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered signed. + /// @brief Signed greather than comparison + template + INLINE bool sgt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !slt(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered unsigned. + /// @brief Unsigned greater or equal comparison + template + INLINE bool uge(const ap_private<_AP_W, _AP_S, _AP_N>& RHS) const { + return !ult(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered signed. + /// @brief Signed greather or equal comparison + template + INLINE bool sge(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !slt(RHS); + } + + template + void cpTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 > BitWidth && "Invalid ap_private Truncate request"); + assert(_AP_W1 >= MIN_INT_BITS && "Can't truncate to 0 bits"); + memcpy(pVal, that.pVal, _AP_N*APINT_WORD_SIZE); + } + + // Sign extend to a new width. + template + void cpSext(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 < BitWidth && "Invalid ap_private SignExtend request"); + assert(_AP_W1 <= MAX_INT_BITS && "Too many bits"); + // If the sign bit isn't set, this is the same as zext. + if (!that.isNegative()) { + cpZext(that); + return; + } + + // The sign bit is set. First, get some facts + enum { wordBits = _AP_W1 % APINT_BITS_PER_WORD}; + + // Mask the high order word appropriately + if (_AP_N1 == _AP_N) { + enum { newWordBits = _AP_W % APINT_BITS_PER_WORD}; + // The extension is contained to the wordsBefore-1th word. + static const uint64_t mask = wordBits?(~0ULL<<(wordBits)):0ULL; + if (_AP_N1 == 1) { + assert(0); + } else { + for (uint32_t i = 0; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + pVal[_AP_N1-1] |= mask; + return; + } + } + + if (_AP_N1 == 1) { + assert(0);// newVal[0] = VAL | mask; + } else { + enum { newWordBits = _AP_W % APINT_BITS_PER_WORD}; + // The extension is contained to the wordsBefore-1th word. + static const uint64_t mask = wordBits?(~0ULL<<(wordBits)):0ULL; + for (uint32_t i = 0; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + pVal[_AP_N1-1] |= mask; + } + for (uint32_t i=_AP_N1; i < _AP_N-1; i++) + pVal[i] = ~0ULL; + pVal[_AP_N-1] = ~0ULL; + clearUnusedBits(); + return; + } + + // Zero extend to a new width. + template + void cpZext(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 < BitWidth && "Invalid ap_private ZeroExtend request"); + assert(_AP_W1 <= MAX_INT_BITS && "Too many bits"); + uint32_t wordsAfter = _AP_N; + if (wordsAfter==1) { + assert(0); // return ap_private<_AP_W1, _AP_S, _AP_N1> (_AP_W1, VAL, _AP_S); + } else { + if (_AP_N1 == 1) { + assert(0); + // newVal[0] = VAL; + } else { + uint32_t i = 0; + for (; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + for (; i < _AP_N; ++i) + pVal[i] = 0; + } + } + clearUnusedBits(); + } + + template + void cpZextOrTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + if (BitWidth > _AP_W1) + cpZext(that); + else if (BitWidth < _AP_W1) + cpTrunc(that); + else { + for (int i=0; i<_AP_N1; ++i) + pVal[i]=that.pVal[i]; + clearUnusedBits(); + } + } + + template + void cpSextOrTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + if (BitWidth > _AP_W1) + cpSext(that); + else if (BitWidth < _AP_W1) + cpTrunc(that); + else { + for (int i=0; i<_AP_N1; ++i) + pVal[i] = that.pVal[i]; + clearUnusedBits(); + } + } + + /// @name Value Characterization Functions + /// @{ + + /// @returns the total number of bits. + INLINE uint32_t getBitWidth() const { + return BitWidth; + } + + /// Here one word's bitwidth equals to that of uint64_t. + /// @returns the number of words to hold the integer value of this ap_private. + /// @brief Get the number of words. + INLINE uint32_t getNumWords() const { + return (BitWidth + APINT_BITS_PER_WORD - 1) / APINT_BITS_PER_WORD; + } + + /// This function returns the number of active bits which is defined as the + /// bit width minus the number of leading zeros. This is used in several + /// computations to see how "wide" the value is. + /// @brief Compute the number of active bits in the value + INLINE uint32_t getActiveBits() const { + uint32_t bits=BitWidth - countLeadingZeros(); + return bits?bits:1; + } + + + /// This method attempts to return the value of this ap_private as a zero extended + /// uint64_t. The bitwidth must be <= 64 or the value must fit within a + /// uint64_t. Otherwise an assertion will result. + /// @brief Get zero extended value + INLINE uint64_t getZExtValue() const { + assert(getActiveBits() <= 64 && "Too many bits for uint64_t"); + return *pVal; + } + + /// This method attempts to return the value of this ap_private as a sign extended + /// int64_t. The bit width must be <= 64 or the value must fit within an + /// int64_t. Otherwise an assertion will result. + /// @brief Get sign extended value + INLINE int64_t getSExtValue() const { + assert(getActiveBits() <= 64 && "Too many bits for int64_t"); + return int64_t(pVal[0]); + } + + /// This method determines how many bits are required to hold the ap_private + /// equivalent of the string given by \p str of length \p slen. + /// @brief Get bits required for string value. + static uint32_t getBitsNeeded(const char* str, uint32_t slen, uint8_t radix); + + /// countLeadingZeros - This function is an ap_private version of the + /// countLeadingZeros_{32,64} functions in MathExtras.h. It counts the number + /// of zeros from the most significant bit to the first one bit. + /// @returns BitWidth if the value is zero. + /// @returns the number of zeros from the most significant bit to the first + /// one bits. + INLINE uint32_t countLeadingZeros() const ; + + /// countLeadingOnes - This function counts the number of contiguous 1 bits + /// in the high order bits. The count stops when the first 0 bit is reached. + /// @returns 0 if the high order bit is not set + /// @returns the number of 1 bits from the most significant to the least + /// @brief Count the number of leading one bits. + INLINE uint32_t countLeadingOnes() const ; + + /// countTrailingZeros - This function is an ap_private version of the + /// countTrailingZoers_{32,64} functions in MathExtras.h. It counts + /// the number of zeros from the least significant bit to the first set bit. + /// @returns BitWidth if the value is zero. + /// @returns the number of zeros from the least significant bit to the first + /// one bit. + /// @brief Count the number of trailing zero bits. + INLINE uint32_t countTrailingZeros() const ; + + /// countPopulation - This function is an ap_private version of the + /// countPopulation_{32,64} functions in MathExtras.h. It counts the number + /// of 1 bits in the ap_private value. + /// @returns 0 if the value is zero. + /// @returns the number of set bits. + /// @brief Count the number of bits set. + INLINE uint32_t countPopulation() const { + uint32_t Count = 0; + for (uint32_t i = 0; i<_AP_N-1 ; ++i) + Count += CountPopulation_64(pVal[i]); + Count += CountPopulation_64(pVal[_AP_N-1]&mask); + return Count; + } + + /// @} + /// @name Conversion Functions + /// @{ + + /// This is used internally to convert an ap_private to a string. + /// @brief Converts an ap_private to a std::string + INLINE std::string toString(uint8_t radix, bool wantSigned) const + ; + + /// Considers the ap_private to be unsigned and converts it into a string in the + /// radix given. The radix can be 2, 8, 10 or 16. + /// @returns a character interpretation of the ap_private + /// @brief Convert unsigned ap_private to string representation. + INLINE std::string toStringUnsigned(uint8_t radix = 10) const { + return toString(radix, false); + } + + /// Considers the ap_private to be unsigned and converts it into a string in the + /// radix given. The radix can be 2, 8, 10 or 16. + /// @returns a character interpretation of the ap_private + /// @brief Convert unsigned ap_private to string representation. + INLINE std::string toStringSigned(uint8_t radix = 10) const { + return toString(radix, true); + } + + /// @returns a byte-swapped representation of this ap_private Value. + INLINE ap_private byteSwap() const ; + + /// @brief Converts this ap_private to a double value. + INLINE double roundToDouble(bool isSigned) const ; + + /// @brief Converts this unsigned ap_private to a double value. + INLINE double roundToDouble() const { + return roundToDouble(false); + } + + /// @brief Converts this signed ap_private to a double value. + INLINE double signedRoundToDouble() const { + return roundToDouble(true); + } + + /// The conversion does not do a translation from integer to double, it just + /// re-interprets the bits as a double. Note that it is valid to do this on + /// any bit width. Exactly 64 bits will be translated. + /// @brief Converts ap_private bits to a double + INLINE double bitsToDouble() const { + union { + uint64_t __I; + double __D; + } __T; + __T.__I = pVal[0]; + return __T.__D; + } + + /// The conversion does not do a translation from integer to float, it just + /// re-interprets the bits as a float. Note that it is valid to do this on + /// any bit width. Exactly 32 bits will be translated. + /// @brief Converts ap_private bits to a double + INLINE float bitsToFloat() const { + union { + uint32_t __I; + float __F; + } __T; + __T.__I = uint32_t(pVal[0]); + return __T.__F; + } + + /// The conversion does not do a translation from double to integer, it just + /// re-interprets the bits of the double. Note that it is valid to do this on + /// any bit width but bits from V may get truncated. + /// @brief Converts a double to ap_private bits. + INLINE ap_private& doubleToBits(double __V) { + union { + uint64_t __I; + double __D; + } __T; + __T.__D = __V; + pVal[0] = __T.__I; + return *this; + } + + /// The conversion does not do a translation from float to integer, it just + /// re-interprets the bits of the float. Note that it is valid to do this on + /// any bit width but bits from V may get truncated. + /// @brief Converts a float to ap_private bits. + INLINE ap_private& floatToBits(float __V) { + union { + uint32_t __I; + float __F; + } __T; + __T.__F = __V; + pVal[0] = __T.__I; + } + + //Reduce operation + //----------------------------------------------------------- + INLINE bool and_reduce() const { + return isMaxValue(); + } + + INLINE bool nand_reduce() const { + return isMinValue(); + } + + INLINE bool or_reduce() const { + return (bool)countPopulation(); + } + + INLINE bool nor_reduce() const { + return countPopulation()==0; + } + + INLINE bool xor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?true:false; + } + + INLINE bool xnor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?false:true; + } + INLINE std::string to_string(uint8_t radix=16, bool sign=false) const { + return toString(radix, radix==10?_AP_S:sign); + } +}; + +template +INLINE bool operator==(uint64_t V1, const ap_private<_AP_W, _AP_S, _AP_N>& V2) { + return V2 == V1; +} + +template +INLINE bool operator!=(uint64_t V1, const ap_private<_AP_W, _AP_S, _AP_N>& V2) { + return V2 != V1; +} + +namespace ap_private_ops { + enum {APINT_BITS_PER_WORD=64}; + /// @brief Determine the smaller of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> smin(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.slt(RHS) ? LHS : RHS; + } + + /// @brief Determine the larger of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> smax(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.sgt(RHS) ? LHS : RHS; + } + + /// @brief Determine the smaller of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> umin(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.ult(RHS) ? LHS : RHS; + } + + /// @brief Determine the larger of two ap_privates considered to be unsigned. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> umax(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.ugt(RHS) ? LHS : RHS; + } + + /// @brief Check if the specified ap_private has a N-bits integer value. + template + INLINE bool isIntN(uint32_t __N, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.isIntN(__N); + } + + /// @returns true if the argument ap_private value is a sequence of ones + /// starting at the least significant bit with the remainder zero. + template + INLINE bool isMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.getBoolValue() && ((APIVal + ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) & APIVal) == 0; + } + + /// @returns true if the argument ap_private value contains a sequence of ones + /// with the remainder zero. + template + INLINE bool isShiftedMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return isMask(numBits, (APIVal - ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) | APIVal); + } + + /// @returns a byte-swapped representation of the specified ap_private Value. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> byteSwap(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.byteSwap(); + } + + /// @returns the floor log base 2 of the specified ap_private value. + template INLINE uint32_t logBase2(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.logBase2(); + } + + /// GreatestCommonDivisor - This function returns the greatest common + /// divisor of the two ap_private values using Enclid's algorithm. + /// @returns the greatest common divisor of Val1 and Val2 + /// @brief Compute GCD of two ap_private values. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> GreatestCommonDivisor(const ap_private<_AP_W, _AP_S, _AP_N>& Val1, const ap_private<_AP_W, _AP_S, _AP_N>& Val2) + ; + + /// Treats the ap_private as an unsigned value for conversion purposes. + /// @brief Converts the given ap_private to a double value. + template INLINE double Roundap_privateToDouble(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.roundToDouble(); + } + + /// Treats the ap_private as a signed value for conversion purposes. + /// @brief Converts the given ap_private to a double value. + template INLINE double RoundSignedap_privateToDouble(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.signedRoundToDouble(); + } + + /// @brief Converts the given ap_private to a float vlalue. + template INLINE float Roundap_privateToFloat(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return float(Roundap_privateToDouble(APIVal)); + } + + /// Treast the ap_private as a signed value for conversion purposes. + /// @brief Converts the given ap_private to a float value. + template INLINE float RoundSignedap_privateToFloat(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return float(APIVal.signedRoundToDouble()); + } + + /// RoundDoubleToap_private - This function convert a double value to an ap_private value. + /// @brief Converts the given double value into a ap_private. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> RoundDoubleToap_private(double Double, uint32_t width) ; + + /// RoundFloatToap_private - Converts a float value into an ap_private value. + /// @brief Converts a float value into a ap_private. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> RoundFloatToap_private(float Float, uint32_t width) { + return RoundDoubleToap_private<_AP_W, _AP_S, _AP_N>(double(Float), width); + } + + /// Arithmetic right-shift the ap_private by shiftAmt. + /// @brief Arithmetic right-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> ashr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.ashr(shiftAmt); + } + + /// Logical right-shift the ap_private by shiftAmt. + /// @brief Logical right-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> lshr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.lshr(shiftAmt); + } + + /// Left-shift the ap_private by shiftAmt. + /// @brief Left-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> shl(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.shl(shiftAmt); + } + + /// Signed divide ap_private LHS by ap_private RHS. + /// @brief Signed division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> sdiv(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.sdiv(RHS); + } + + /// Unsigned divide ap_private LHS by ap_private RHS. + /// @brief Unsigned division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> udiv(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.udiv(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> srem(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.srem(RHS); + } + + /// Unsigned remainder operation on ap_private. + /// @brief Function for unsigned remainder operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> urem(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.urem(RHS); + } + + /// Performs multiplication on ap_private values. + /// @brief Function for multiplication operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> mul(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS * RHS; + } + + /// Performs addition on ap_private values. + /// @brief Function for addition operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> add(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS + RHS; + } + + /// Performs subtraction on ap_private values. + /// @brief Function for subtraction operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> sub(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS - RHS; + } + + /// Performs bitwise AND operation on ap_private LHS and + /// ap_private RHS. + /// @brief Bitwise AND function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> And(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS & RHS; + } + + /// Performs bitwise OR operation on ap_private LHS and ap_private RHS. + /// @brief Bitwise OR function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> Or(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS | RHS; + } + + /// Performs bitwise XOR operation on ap_private. + /// @brief Bitwise XOR function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> Xor(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS ^ RHS; + } + + /// Performs a bitwise complement operation on ap_private. + /// @brief Bitwise complement function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> Not(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return ~APIVal; + } + + template void clearUnusedBits(uint64_t& msw) { + // Compute how many bits are used in the final word + // uint32_t wordBits = APIVal.getBitWidth() & 0x3f; + if (wordBits == 0) + // If all bits are used, we want to leave the value alone. This also + // avoids the undefined behavior of >> when the shfit is the same size as + // the word size (64). + return; + + // Mask out the hight bits. + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- wordBits); + msw &= mask; + } + template <> INLINE void clearUnusedBits<1>(uint64_t& msw) { + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- 1); + msw &= mask; + } + template void clearUnusedBits(int64_t& msw) { + // Compute how many bits are used in the final word + // uint32_t wordBits = APIVal.getBitWidth() & 0x3f; + if (wordBits == 0) + // If all bits are used, we want to leave the value alone. This also + // avoids the undefined behavior of >> when the shfit is the same size as + // the word size (64). + return; + + // Mask out the hight bits. + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- wordBits); + msw &= mask; + } + template <> INLINE void clearUnusedBits<1>(int64_t& msw) { + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- 1); + msw &= mask; + } + // template + template + INLINE ap_private<_AP_W, _AP_S> ashr(const ap_private<_AP_W, _AP_S>& a) { + return ashr(a, shiftAmt); + } + + template + INLINE ap_private<_AP_W, _AP_S> lshr(const ap_private<_AP_W, _AP_S>& a) { + return lshr(a, shiftAmt); + } + + template + INLINE ap_private<_AP_W, true> ashr(const ap_private<_AP_W, true, 1>& a) { + enum {APINT_BITS_PER_WORD=64, excess_bits=APINT_BITS_PER_WORD-_AP_W}; + static const uint64_t sign_bit = (1ULL<<(_AP_W-1)); + static const uint64_t sign_ext_mask = (_AP_W-shiftAmt>0)?~0ULL<<(APINT_BITS_PER_WORD-_AP_W+shiftAmt):~0ULL; + return ap_private<_AP_W, true>((((int64_t)a.VAL) >> (shiftAmt)) | (a.VAL & sign_bit? sign_ext_mask : 0ULL)); + } + + template + INLINE ap_private<_AP_W, false> ashr(const ap_private<_AP_W, false, 1>& a) { + return ap_private<_AP_W, false>((a.VAL) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W, _AP_S> lshr(const ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask = ~0ULL<<_AP_W; + return ap_private<_AP_W, _AP_S>((a.VAL&mask) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W-shiftAmt, _AP_S> shr(const ap_private<_AP_W, _AP_S>& a) { + return ap_private<_AP_W-shiftAmt, _AP_S>((a.VAL) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W+shiftAmt, _AP_S> shl(const ap_private<_AP_W, _AP_S>& a) { + return ap_private<_AP_W+shiftAmt, _AP_S>((a.VAL) << (shiftAmt)); + } + + template + INLINE bool get(const ap_private<_AP_W, _AP_S, 1>& a) { + unsigned shift = (index%APINT_BITS_PER_WORD); + static const uint64_t mask=1ULL << (shift); + return ((mask & a.VAL) != 0); + } + + template + INLINE bool get(const ap_private<_AP_W, _AP_S>& a) { + static const uint64_t mask=1ULL << (index&0x3f); + return ((mask & a.pVal[(index)>>6]) != 0); + } + + template + INLINE void set(ap_private<_AP_W, _AP_S, 1>& a) { + const uint64_t mask = ~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.VAL |= mask; + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask = ~(~0ULL >> (lsb) <<(APINT_BITS_PER_WORD-msb+lsb-1) >> (APINT_BITS_PER_WORD-msb-1)); + a.VAL &= mask; + } + + template + INLINE void set(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, + lsb_word = lsb_index /APINT_BITS_PER_WORD, + msb_word = msb_index / APINT_BITS_PER_WORD, + msb = msb_index % APINT_BITS_PER_WORD, + lsb=lsb_index % APINT_BITS_PER_WORD}; + if (msb_word==lsb_word) { + const uint64_t mask = ~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.pVal[msb_word] |= mask; + } else { + const uint64_t lsb_mask = ~0ULL >> (lsb) << (lsb); + const uint64_t msb_mask = ~0ULL << (APINT_BITS_PER_WORD-msb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.pVal[lsb_word] |=lsb_mask; + for (int i=lsb_word+1; i + INLINE void clear(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, + lsb_word = lsb_index /APINT_BITS_PER_WORD, + msb_word = msb_index / APINT_BITS_PER_WORD, + msb = msb_index % APINT_BITS_PER_WORD, + lsb=lsb_index % APINT_BITS_PER_WORD}; + if (msb_word == lsb_word) { + const uint64_t mask = ~(~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1)); + a.pVal[msb_word] &= mask; + } else { + const uint64_t lsb_mask = ~(~0ULL >> (lsb) << (lsb)); + const uint64_t msb_mask = ~(~0ULL << (APINT_BITS_PER_WORD-msb-1)>>(APINT_BITS_PER_WORD-msb-1)); + a.pVal[lsb_word] &=lsb_mask; + for (int i=lsb_word+1; i + INLINE void set(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask=1ULL << (index); + a.VAL |= mask; + a.clearUnusedBits(); + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask=~(1ULL << (index)); + a.VAL &= mask; + a.clearUnusedBits(); + } + + template + INLINE void set(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, word = index/APINT_BITS_PER_WORD}; + static const uint64_t mask=1ULL << (index%APINT_BITS_PER_WORD); + a.pVal[word] |= mask; + a.clearUnusedBits(); + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, word = index/APINT_BITS_PER_WORD}; + static const uint64_t mask=~(1ULL << (index%APINT_BITS_PER_WORD)); + a.pVal[word] &= mask; + a.clearUnusedBits(); + } + + template + INLINE bool isNegative(const ap_private<_AP_W, false>& a) { + return false; + } + + template + INLINE bool isNegative(const ap_private<_AP_W, true, 1>& a) { + static const uint64_t sign_mask = (1ULL << (_AP_W-1)); + return ((sign_mask & a.VAL) != 0); + } + + template + INLINE bool isNegative(const ap_private<_AP_W, true>& a) { + enum {APINT_BITS_PER_WORD=64,_AP_N=(_AP_W+APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; + static const uint64_t sign_mask = (1ULL << (_AP_W%APINT_BITS_PER_WORD-1)); + return sign_mask & a.pVal[_AP_N-1]; + } +} // End of ap_private_ops namespace + +/// @brief Check if the specified ap_private has a N-bits integer value. +template +INLINE bool isIntN(uint32_t __N, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.isIntN(__N); +} + +/// @returns true if the argument ap_private value is a sequence of ones +/// starting at the least significant bit with the remainder zero. +template +INLINE bool isMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.getBoolValue() && ((APIVal + ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) & APIVal) == 0; +} + +/// @returns true if the argument ap_private value contains a sequence of ones +/// with the remainder zero. +template +INLINE bool isShiftedMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return isMask(numBits, (APIVal - ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) | APIVal); +} + +#if 0 +/// add_1 - This function adds a single "digit" integer, y, to the multiple +/// "digit" integer array, x[]. x[] is modified to reflect the addition and +/// 1 is returned if there is a carry out, otherwise 0 is returned. +/// @returns the carry of the addition. +static bool add_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { + for (uint32_t i = 0; i < len; ++i) { + dest[i] = y + x[i]; + if (dest[i] < y) + y = 1; // Carry one to next digit. + else { + y = 0; // No need to carry so exit early + break; + } + } + return (y != 0); +} +#endif + +#if 0 +/// add - This function adds the integer array x to the integer array Y and +/// places the result in dest. +/// @returns the carry out from the addition +/// @brief General addition of 64-bit integer arrays +static bool add(uint64_t *dest, const uint64_t *x, const uint64_t *y, + uint32_t destlen, uint32_t xlen, uint32_t ylen, bool xsigned, bool ysigned) { + bool carry = false; + uint32_t len = AESL_std::min(xlen, ylen); + uint32_t i; + for (i = 0; i< len && i < destlen; ++i) { + uint64_t limit = AESL_std::min(x[i],y[i]); // must come first in case dest == x + dest[i] = x[i] + y[i] + carry; + carry = dest[i] < limit || (carry && dest[i] == limit); + } + if (xlen > ylen) { + const uint64_t yext = xsigned && int64_t(y[ylen-1])<0 ? -1 : 0; + for (i=ylen; i< xlen && i < destlen; i++) { + uint64_t limit = AESL_std::min(x[i], yext); + dest[i] = x[i] + yext + carry; + carry = (dest[i] < limit)||(carry && dest[i] == x[i]); + } + } else if (ylen> xlen) { + const uint64_t xext = ysigned && int64_t(x[xlen-1])<0 ? -1 : 0; + for (i=xlen; i< ylen && i < destlen; i++) { + uint64_t limit = AESL_std::min(xext, y[i]); + dest[i] = xext + y[i] + carry; + carry = (dest[i] < limit)||(carry && dest[i] == y[i]); + } + } + return carry; +} +#endif + +#if 0 +/// @returns returns the borrow out. +/// @brief Generalized subtraction of 64-bit integer arrays. +static bool sub(uint64_t *dest, const uint64_t *x, const uint64_t *y, + uint32_t destlen, uint32_t xlen, uint32_t ylen, bool xsigned, bool ysigned) { + bool borrow = false; + uint32_t i; + uint32_t len = AESL_std::min(xlen, ylen); + for (i = 0; i < len && i < destlen; ++i) { + uint64_t x_tmp = borrow ? x[i] - 1 : x[i]; + borrow = y[i] > x_tmp || (borrow && x[i] == 0); + dest[i] = x_tmp - y[i]; + } + if (xlen > ylen) { + const uint64_t yext = ysigned && int64_t(y[ylen-1])<0 ? -1 : 0; + for (i=ylen; i< xlen && i < destlen; i++) { + uint64_t x_tmp = borrow ? x[i] - 1 : x[i]; + borrow = yext > x_tmp || (borrow && x[i] == 0); + dest[i] = x_tmp - yext; + } + } else if (ylen> xlen) { + const uint64_t xext = xsigned && int64_t(x[xlen-1])<0 ? -1 : 0; + for (i=xlen; i< ylen && i < destlen; i++) { + uint64_t x_tmp = borrow ? xext - 1 : xext; + borrow = y[i] > x_tmp || (borrow && xext==0); + dest[i] = x_tmp - y[i]; + } + } + return borrow; +} +#endif + +/// Subtracts the RHS ap_private from this ap_private +/// @returns this, after subtraction +/// @brief Subtraction assignment operator. + +#if 0 +/// Multiplies an integer array, x by a a uint64_t integer and places the result +/// into dest. +/// @returns the carry out of the multiplication. +/// @brief Multiply a multi-digit ap_private by a single digit (64-bit) integer. +static uint64_t mul_1(uint64_t dest[], const uint64_t x[], uint32_t len, uint64_t y) { + // Split y into high 32-bit part (hy) and low 32-bit part (ly) + uint64_t ly = y & 0xffffffffULL, hy = (y) >> 32; + uint64_t carry = 0; + static const uint64_t two_power_32 = 1ULL << 32; + // For each digit of x. + for (uint32_t i = 0; i < len; ++i) { + // Split x into high and low words + uint64_t lx = x[i] & 0xffffffffULL; + uint64_t hx = (x[i]) >> 32; + // hasCarry - A flag to indicate if there is a carry to the next digit. + // hasCarry == 0, no carry + // hasCarry == 1, has carry + // hasCarry == 2, no carry and the calculation result == 0. + uint8_t hasCarry = 0; + dest[i] = carry + lx * ly; + // Determine if the add above introduces carry. + hasCarry = (dest[i] < carry) ? 1 : 0; + carry = hx * ly + ((dest[i]) >> 32) + (hasCarry ? two_power_32 : 0); + // The upper limit of carry can be (2^32 - 1)(2^32 - 1) + + // (2^32 - 1) + 2^32 = 2^64. + hasCarry = (!carry && hasCarry) ? 1 : (!carry ? 2 : 0); + + carry += (lx * hy) & 0xffffffffULL; + dest[i] = ((carry) << 32) | (dest[i] & 0xffffffffULL); + carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? two_power_32 : 0) + + ((carry) >> 32) + ((lx * hy) >> 32) + hx * hy; + } + return carry; +} +#endif + +#if 0 +/// Multiplies integer array x by integer array y and stores the result into +/// the integer array dest. Note that dest's size must be >= xlen + ylen. +/// @brief Generalized multiplicate of integer arrays. +static void mul(uint64_t dest[], const uint64_t x[], uint32_t xlen, const uint64_t y[], + uint32_t ylen, uint32_t destlen) { + dest[xlen] = mul_1(dest, x, xlen, y[0]); + for (uint32_t i = 1; i < ylen; ++i) { + uint64_t ly = y[i] & 0xffffffffULL, hy = (y[i]) >> 32; + uint64_t carry = 0, lx = 0, hx = 0; + for (uint32_t j = 0; j < xlen; ++j) { + lx = x[j] & 0xffffffffULL; + hx = (x[j]) >> 32; + // hasCarry - A flag to indicate if has carry. + // hasCarry == 0, no carry + // hasCarry == 1, has carry + // hasCarry == 2, no carry and the calculation result == 0. + uint8_t hasCarry = 0; + uint64_t resul = carry + lx * ly; + hasCarry = (resul < carry) ? 1 : 0; + carry = (hasCarry ? (1ULL << 32) : 0) + hx * ly + ((resul) >> 32); + hasCarry = (!carry && hasCarry) ? 1 : (!carry ? 2 : 0); + carry += (lx * hy) & 0xffffffffULL; + resul = ((carry) << 32) | (resul & 0xffffffffULL); + dest[i+j] += resul; + carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? (1ULL << 32) : 0)+ + ((carry) >> 32) + (dest[i+j] < resul ? 1 : 0) + + ((lx * hy) >> 32) + hx * hy; + } + dest[i+xlen] = carry; + } +} +#endif + + + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { + assert(str != 0 && "Invalid value string"); + assert(slen > 0 && "Invalid string length"); + + // Each computation below needs to know if its negative + uint32_t isNegative = str[0] == '-'; + if (isNegative) { + slen--; + str++; + } + // For radixes of power-of-two values, the bits required is accurately and + // easily computed + if (radix == 2) + return slen + isNegative; + if (radix == 8) + return slen * 3 + isNegative; + if (radix == 16) + return slen * 4 + isNegative; + + // Otherwise it must be radix == 10, the hard case + assert(radix == 10 && "Invalid radix"); + + // Convert to the actual binary value. + //ap_private<_AP_W, _AP_S, _AP_N> tmp(sufficient, str, slen, radix); + + // Compute how many bits are required. + //return isNegative + tmp.logBase2() + 1; + return isNegative + slen * 4; +} + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countLeadingZeros() const { + enum { msw_bits = (BitWidth % APINT_BITS_PER_WORD)?(BitWidth % APINT_BITS_PER_WORD):APINT_BITS_PER_WORD, + excessBits = APINT_BITS_PER_WORD - msw_bits }; + uint32_t Count = CountLeadingZeros_64(pVal[_AP_N-1]); + if (Count>=excessBits) + Count -= excessBits; + if (!pVal[_AP_N-1]) { + for (uint32_t i = _AP_N-1 ; i ; --i) { + if (!pVal[i-1]) + Count += APINT_BITS_PER_WORD; + else { + Count += CountLeadingZeros_64(pVal[i-1]); + break; + } + } + } + return Count; +} + +static uint32_t countLeadingOnes_64(uint64_t __V, uint32_t skip) { + uint32_t Count = 0; + if (skip) + (__V) <<= (skip); + while (__V && (__V & (1ULL << 63))) { + Count++; + (__V) <<= 1; + } + return Count; +} + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countLeadingOnes() const { + if (isSingleWord()) + return countLeadingOnes_64(VAL, APINT_BITS_PER_WORD - BitWidth); + + uint32_t highWordBits = BitWidth % APINT_BITS_PER_WORD; + uint32_t shift = (highWordBits == 0 ? 0 : APINT_BITS_PER_WORD - highWordBits); + int i = _AP_N - 1; + uint32_t Count = countLeadingOnes_64(pVal[i], shift); + if (Count == highWordBits) { + for (i--; i >= 0; --i) { + if (pVal[i] == ~0ULL) + Count += APINT_BITS_PER_WORD; + else { + Count += countLeadingOnes_64(pVal[i], 0); + break; + } + } + } + return Count; +} + +template +INLINE uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countTrailingZeros() const { + if (isSingleWord()) + return AESL_std::min(uint32_t(CountTrailingZeros_64(VAL)), BitWidth); + uint32_t Count = 0; + uint32_t i = 0; + for (; i < _AP_N && pVal[i] == 0; ++i) + Count += APINT_BITS_PER_WORD; + if (i < _AP_N) + Count += CountTrailingZeros_64(pVal[i]); + return AESL_std::min(Count, BitWidth); +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private<_AP_W, _AP_S, _AP_N>::byteSwap() const { + assert(BitWidth >= 16 && BitWidth % 16 == 0 && "Cannot byteswap!"); + if (BitWidth == 16) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_16(uint16_t(VAL))); + else if (BitWidth == 32) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_32(uint32_t(VAL))); + else if (BitWidth == 48) { + uint32_t Tmp1 = uint32_t((VAL) >> 16); + Tmp1 = ByteSwap_32(Tmp1); + uint16_t Tmp2 = uint16_t(VAL); + Tmp2 = ByteSwap_16(Tmp2); + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ((uint64_t(Tmp2)) << 32) | Tmp1); + } else if (BitWidth == 64) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_64(VAL)); + else { + ap_private<_AP_W, _AP_S, _AP_N> Result(0); + char *pByte = (char*)Result.pVal; + for (uint32_t i = 0; i < BitWidth / APINT_WORD_SIZE / 2; ++i) { + char Tmp = pByte[i]; + pByte[i] = pByte[BitWidth / APINT_WORD_SIZE - 1 - i]; + pByte[BitWidth / APINT_WORD_SIZE - i - 1] = Tmp; + } + return Result; + } +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private_ops::GreatestCommonDivisor(const ap_private<_AP_W, _AP_S, _AP_N>& API1, const ap_private<_AP_W, _AP_S, _AP_N>& API2) { + ap_private<_AP_W, _AP_S, _AP_N> __A = API1, __B = API2; + while (!!__B) { + ap_private<_AP_W, _AP_S, _AP_N> __T = __B; + __B = ap_private_ops::urem(__A, __B); + __A = __T; + } + return __A; +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private_ops::RoundDoubleToap_private(double Double, uint32_t width) { + union { + double __D; + uint64_t __I; + } __T; + __T.__D = Double; + + // Get the sign bit from the highest order bit + bool isNeg = (__T.__I) >> 63; + + // Get the 11-bit exponent and adjust for the 1023 bit bias + int64_t exp = (((__T.__I) >> 52) & 0x7ffULL) - 1023; + + // If the exponent is negative, the value is < 0 so just return 0. + if (exp < 0) + return ap_private<_AP_W, _AP_S, _AP_N>(width, 0u); + + // Extract the mantissa by clearing the top 12 bits (sign + exponent). + uint64_t mantissa = (__T.__I & (~0ULL >> 12)) | 1ULL << 52; + + // If the exponent doesn't shift all bits out of the mantissa + if (exp < 52) + return isNeg ? -ap_private<_AP_W, _AP_S, _AP_N>(width, (mantissa) >> (52 - exp)) : + ap_private<_AP_W, _AP_S, _AP_N>((mantissa) >> (52 - exp)); + + // If the client didn't provide enough bits for us to shift the mantissa into + // then the result is undefined, just return 0 + if (width <= exp - 52) + return ap_private<_AP_W, _AP_S, _AP_N>(width, 0); + + // Otherwise, we have to shift the mantissa bits up to the right location + ap_private<_AP_W, _AP_S, _AP_N> Tmp(width, mantissa); + Tmp = Tmp.shl(exp - 52); + return isNeg ? -Tmp : Tmp; +} + +/// RoundToDouble - This function convert this ap_private to a double. +/// The layout for double is as following (IEEE Standard 754): +/// -------------------------------------- +/// | Sign Exponent Fraction Bias | +/// |-------------------------------------- | +/// | 1[63] 11[62-52] 52[51-00] 1023 | +/// -------------------------------------- +template +double ap_private<_AP_W, _AP_S, _AP_N>::roundToDouble(bool isSigned) const { + + // Handle the simple case where the value is contained in one uint64_t. + if (isSingleWord() || getActiveBits() <= APINT_BITS_PER_WORD) { + uint64_t val; + if (isSingleWord()) val = VAL; + else val = pVal[0]; + if (isSigned) { + int64_t sext = ((int64_t(val)) << (64-BitWidth)) >> (64-BitWidth); + return double(sext); + } else + return double(val); + } + + // Determine if the value is negative. + bool isNeg = isSigned ? (*this)[BitWidth-1] : false; + + // Construct the absolute value if we're negative. + ap_private<_AP_W, _AP_S, _AP_N> Tmp(isNeg ? -(*this) : (*this)); + + // Figure out how many bits we're using. + uint32_t n = Tmp.getActiveBits(); + + // The exponent (without bias normalization) is just the number of bits + // we are using. Note that the sign bit is gone since we constructed the + // absolute value. + uint64_t exp = n; + + // Return infinity for exponent overflow + if (exp > 1023) { + if (!isSigned || !isNeg) + return std::numeric_limits::infinity(); + else + return -std::numeric_limits::infinity(); + } + exp += 1023; // Increment for 1023 bias + + // Number of bits in mantissa is 52. To obtain the mantissa value, we must + // extract the high 52 bits from the correct words in pVal. + uint64_t mantissa; + unsigned hiWord = whichWord(n-1); + if (hiWord == 0) { + mantissa = Tmp.pVal[0]; + if (n > 52) + (mantissa) >>= (n - 52); // shift down, we want the top 52 bits. + } else { + assert(hiWord > 0 && "High word is negative?"); + uint64_t hibits = (Tmp.pVal[hiWord]) << (52 - n % APINT_BITS_PER_WORD); + uint64_t lobits = (Tmp.pVal[hiWord-1]) >> (11 + n % APINT_BITS_PER_WORD); + mantissa = hibits | lobits; + } + + // The leading bit of mantissa is implicit, so get rid of it. + uint64_t sign = isNeg ? (1ULL << (APINT_BITS_PER_WORD - 1)) : 0; + union { + double __D; + uint64_t __I; + } __T; + __T.__I = sign | ((exp) << 52) | mantissa; + return __T.__D; +} + +// Square Root - this method computes and returns the square root of "this". +// Three mechanisms are used for computation. For small values (<= 5 bits), +// a table lookup is done. This gets some performance for common cases. For +// values using less than 52 bits, the value is converted to double and then +// the libc sqrt function is called. The result is rounded and then converted +// back to a uint64_t which is then used to construct the result. Finally, +// the Babylonian method for computing square roots is used. +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private<_AP_W, _AP_S, _AP_N>::sqrt() const { + + // Determine the magnitude of the value. + uint32_t magnitude = getActiveBits(); + + // Use a fast table for some small values. This also gets rid of some + // rounding errors in libc sqrt for small values. + if (magnitude <= 5) { + static const uint8_t results[32] = { + /* 0 */ 0, + /* 1- 2 */ 1, 1, + /* 3- 6 */ 2, 2, 2, 2, + /* 7-12 */ 3, 3, 3, 3, 3, 3, + /* 13-20 */ 4, 4, 4, 4, 4, 4, 4, 4, + /* 21-30 */ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + /* 31 */ 6 + }; + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ results[ (isSingleWord() ? VAL : pVal[0]) ]); + } + + // If the magnitude of the value fits in less than 52 bits (the precision of + // an IEEE double precision floating point value), then we can use the + // libc sqrt function which will probably use a hardware sqrt computation. + // This should be faster than the algorithm below. + if (magnitude < 52) { +#ifdef _MSC_VER + // Amazingly, VC++ doesn't have round(). + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ + uint64_t(::sqrt(double(isSingleWord()?VAL:pVal[0]))) + 0.5); +#else + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ + uint64_t(::round(::sqrt(double(isSingleWord()?VAL:pVal[0]))))); +#endif + } + + // Okay, all the short cuts are exhausted. We must compute it. The following + // is a classical Babylonian method for computing the square root. This code + // was adapted to APINt from a wikipedia article on such computations. + // See http://www.wikipedia.org/ and go to the page named + // Calculate_an_integer_square_root. + uint32_t nbits = BitWidth, i = 4; + ap_private<_AP_W, _AP_S, _AP_N> testy(16); + ap_private<_AP_W, _AP_S, _AP_N> x_old(/*BitWidth,*/ 1); + ap_private<_AP_W, _AP_S, _AP_N> x_new(0); + ap_private<_AP_W, _AP_S, _AP_N> two(/*BitWidth,*/ 2); + + // Select a good starting value using binary logarithms. + for (;; i += 2, testy = testy.shl(2)) + if (i >= nbits || this->ule(testy)) { + x_old = x_old.shl(i / 2); + break; + } + + // Use the Babylonian method to arrive at the integer square root: + for (;;) { + x_new = (this->udiv(x_old) + x_old).udiv(two); + if (x_old.ule(x_new)) + break; + x_old = x_new; + } + + // Make sure we return the closest approximation + // NOTE: The rounding calculation below is correct. It will produce an + // off-by-one discrepancy with results from pari/gp. That discrepancy has been + // determined to be a rounding issue with pari/gp as it begins to use a + // floating point representation after 192 bits. There are no discrepancies + // between this algorithm and pari/gp for bit widths < 192 bits. + ap_private<_AP_W, _AP_S, _AP_N> square(x_old * x_old); + ap_private<_AP_W, _AP_S, _AP_N> nextSquare((x_old + 1) * (x_old +1)); + if (this->ult(square)) + return x_old; + else if (this->ule(nextSquare)) { + ap_private<_AP_W, _AP_S, _AP_N> midpoint((nextSquare - square).udiv(two)); + ap_private<_AP_W, _AP_S, _AP_N> offset(*this - square); + if (offset.ult(midpoint)) + return x_old; + else + return x_old + 1; + } else + assert(0 && "Error in ap_private<_AP_W, _AP_S, _AP_N>::sqrt computation"); + return x_old + 1; +} + +/// Implementation of Knuth's Algorithm D (Division of nonnegative integers) +/// from "Art of Computer Programming, Volume 2", section 4.3.1, p. 272. The +/// variables here have the same names as in the algorithm. Comments explain +/// the algorithm and any deviation from it. +static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, + uint32_t m, uint32_t n) { + assert(u && "Must provide dividend"); + assert(v && "Must provide divisor"); + assert(q && "Must provide quotient"); + assert(u != v && u != q && v != q && "Must us different memory"); + assert(n>1 && "n must be > 1"); + + // Knuth uses the value b as the base of the number system. In our case b + // is 2^31 so we just set it to -1u. + uint64_t b = uint64_t(1) << 32; + + //DEBUG(cerr << "KnuthDiv: m=" << m << " n=" << n << '\n'); + //DEBUG(cerr << "KnuthDiv: original:"); + //DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); + //DEBUG(cerr << " by"); + //DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); + //DEBUG(cerr << '\n'); + // D1. [Normalize.] Set d = b / (v[n-1] + 1) and multiply all the digits of + // u and v by d. Note that we have taken Knuth's advice here to use a power + // of 2 value for d such that d * v[n-1] >= b/2 (b is the base). A power of + // 2 allows us to shift instead of multiply and it is easy to determine the + // shift amount from the leading zeros. We are basically normalizing the u + // and v so that its high bits are shifted to the top of v's range without + // overflow. Note that this can require an extra word in u so that u must + // be of length m+n+1. + uint32_t shift = CountLeadingZeros_32(v[n-1]); + uint32_t v_carry = 0; + uint32_t u_carry = 0; + if (shift) { + for (uint32_t i = 0; i < m+n; ++i) { + uint32_t u_tmp = (u[i]) >> (32 - shift); + u[i] = ((u[i]) << (shift)) | u_carry; + u_carry = u_tmp; + } + for (uint32_t i = 0; i < n; ++i) { + uint32_t v_tmp = (v[i]) >> (32 - shift); + v[i] = ((v[i]) << (shift)) | v_carry; + v_carry = v_tmp; + } + } + u[m+n] = u_carry; + //DEBUG(cerr << "KnuthDiv: normal:"); + //DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); + //DEBUG(cerr << " by"); + //DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); + //DEBUG(cerr << '\n'); + + // D2. [Initialize j.] Set j to m. This is the loop counter over the places. + int j = m; + do { + //DEBUG(cerr << "KnuthDiv: quotient digit #" << j << '\n'); + // D3. [Calculate q'.]. + // Set qp = (u[j+n]*b + u[j+n-1]) / v[n-1]. (qp=qprime=q') + // Set rp = (u[j+n]*b + u[j+n-1]) % v[n-1]. (rp=rprime=r') + // Now test if qp == b or qp*v[n-2] > b*rp + u[j+n-2]; if so, decrease + // qp by 1, inrease rp by v[n-1], and repeat this test if rp < b. The test + // on v[n-2] determines at high speed most of the cases in which the trial + // value qp is one too large, and it eliminates all cases where qp is two + // too large. + uint64_t dividend = ((uint64_t(u[j+n]) << 32) + u[j+n-1]); + //DEBUG(cerr << "KnuthDiv: dividend == " << dividend << '\n'); + uint64_t qp = dividend / v[n-1]; + uint64_t rp = dividend % v[n-1]; + if (qp == b || qp*v[n-2] > b*rp + u[j+n-2]) { + qp--; + rp += v[n-1]; + if (rp < b && (qp == b || qp*v[n-2] > b*rp + u[j+n-2])) + qp--; + } + //DEBUG(cerr << "KnuthDiv: qp == " << qp << ", rp == " << rp << '\n'); + + // D4. [Multiply and subtract.] Replace (u[j+n]u[j+n-1]...u[j]) with + // (u[j+n]u[j+n-1]..u[j]) - qp * (v[n-1]...v[1]v[0]). This computation + // consists of a simple multiplication by a one-place number, combined with + // a subtraction. + bool isNeg = false; + for (uint32_t i = 0; i < n; ++i) { + uint64_t u_tmp = uint64_t(u[j+i]) | ((uint64_t(u[j+i+1])) << 32); + uint64_t subtrahend = uint64_t(qp) * uint64_t(v[i]); + bool borrow = subtrahend > u_tmp; + /*DEBUG(cerr << "KnuthDiv: u_tmp == " << u_tmp + << ", subtrahend == " << subtrahend + << ", borrow = " << borrow << '\n');*/ + + uint64_t result = u_tmp - subtrahend; + uint32_t k = j + i; + u[k++] = (uint32_t)(result & (b-1)); // subtract low word + u[k++] = (uint32_t)((result) >> 32); // subtract high word + while (borrow && k <= m+n) { // deal with borrow to the left + borrow = u[k] == 0; + u[k]--; + k++; + } + isNeg |= borrow; + /*DEBUG(cerr << "KnuthDiv: u[j+i] == " << u[j+i] << ", u[j+i+1] == " << + u[j+i+1] << '\n');*/ + } + /*DEBUG(cerr << "KnuthDiv: after subtraction:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); + DEBUG(cerr << '\n');*/ + // The digits (u[j+n]...u[j]) should be kept positive; if the result of + // this step is actually negative, (u[j+n]...u[j]) should be left as the + // true value plus b**(n+1), namely as the b's complement of + // the true value, and a "borrow" to the left should be remembered. + // + if (isNeg) { + bool carry = true; // true because b's complement is "complement + 1" + for (uint32_t i = 0; i <= m+n; ++i) { + u[i] = ~u[i] + carry; // b's complement + carry = carry && u[i] == 0; + } + } + /*DEBUG(cerr << "KnuthDiv: after complement:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); + DEBUG(cerr << '\n');*/ + + // D5. [Test remainder.] Set q[j] = qp. If the result of step D4 was + // negative, go to step D6; otherwise go on to step D7. + q[j] = (uint32_t)qp; + if (isNeg) { + // D6. [Add back]. The probability that this step is necessary is very + // small, on the order of only 2/b. Make sure that test data accounts for + // this possibility. Decrease q[j] by 1 + q[j]--; + // and add (0v[n-1]...v[1]v[0]) to (u[j+n]u[j+n-1]...u[j+1]u[j]). + // A carry will occur to the left of u[j+n], and it should be ignored + // since it cancels with the borrow that occurred in D4. + bool carry = false; + for (uint32_t i = 0; i < n; i++) { + uint32_t limit = AESL_std::min(u[j+i],v[i]); + u[j+i] += v[i] + carry; + carry = u[j+i] < limit || (carry && u[j+i] == limit); + } + u[j+n] += carry; + } + /*DEBUG(cerr << "KnuthDiv: after correction:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr <<" " << u[i]); + DEBUG(cerr << "\nKnuthDiv: digit result = " << q[j] << '\n');*/ + + // D7. [Loop on j.] Decrease j by one. Now if j >= 0, go back to D3. + } while (--j >= 0); + + /*DEBUG(cerr << "KnuthDiv: quotient:"); + DEBUG(for (int i = m; i >=0; i--) cerr <<" " << q[i]); + DEBUG(cerr << '\n');*/ + + // D8. [Unnormalize]. Now q[...] is the desired quotient, and the desired + // remainder may be obtained by dividing u[...] by d. If r is non-null we + // compute the remainder (urem uses this). + if (r) { + // The value d is expressed by the "shift" value above since we avoided + // multiplication by d by using a shift left. So, all we have to do is + // shift right here. In order to mak + if (shift) { + uint32_t carry = 0; + //DEBUG(cerr << "KnuthDiv: remainder:"); + for (int i = n-1; i >= 0; i--) { + r[i] = ((u[i]) >> (shift)) | carry; + carry = (u[i]) << (32 - shift); + //DEBUG(cerr << " " << r[i]); + } + } else { + for (int i = n-1; i >= 0; i--) { + r[i] = u[i]; + //DEBUG(cerr << " " << r[i]); + } + } + //DEBUG(cerr << '\n'); + } + //DEBUG(cerr << std::setbase(10) << '\n'); +} + +template +void divide(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t lhsWords, + const ap_private<_AP_W, _AP_S, _AP_N>& RHS, uint32_t rhsWords, + ap_private<_AP_W, _AP_S, _AP_N> *Quotient, ap_private<_AP_W, _AP_S, _AP_N> *Remainder) { + assert(lhsWords >= rhsWords && "Fractional result"); + enum {APINT_BITS_PER_WORD=64}; + // First, compose the values into an array of 32-bit words instead of + // 64-bit words. This is a necessity of both the "short division" algorithm + // and the the Knuth "classical algorithm" which requires there to be native + // operations for +, -, and * on an m bit value with an m*2 bit result. We + // can't use 64-bit operands here because we don't have native results of + // 128-bits. Furthremore, casting the 64-bit values to 32-bit values won't + // work on large-endian machines. + uint64_t mask = ~0ull >> (sizeof(uint32_t)*8); + uint32_t n = rhsWords * 2; + uint32_t m = (lhsWords * 2) - n; + + // Allocate space for the temporary values we need either on the stack, if + // it will fit, or on the heap if it won't. + uint32_t SPACE[128]; + uint32_t *__U = 0; + uint32_t *__V = 0; + uint32_t *__Q = 0; + uint32_t *__R = 0; + if ((Remainder?4:3)*n+2*m+1 <= 128) { + __U = &SPACE[0]; + __V = &SPACE[m+n+1]; + __Q = &SPACE[(m+n+1) + n]; + if (Remainder) + __R = &SPACE[(m+n+1) + n + (m+n)]; + } else { + __U = new uint32_t[m + n + 1]; + __V = new uint32_t[n]; + __Q = new uint32_t[m+n]; + if (Remainder) + __R = new uint32_t[n]; + } + + // Initialize the dividend + memset(__U, 0, (m+n+1)*sizeof(uint32_t)); + for (unsigned i = 0; i < lhsWords; ++i) { + uint64_t tmp = (LHS.getNumWords() == 1 ? LHS.VAL : LHS.pVal[i]); + __U[i * 2] = (uint32_t)(tmp & mask); + __U[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + __U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm. + + // Initialize the divisor + memset(__V, 0, (n)*sizeof(uint32_t)); + for (unsigned i = 0; i < rhsWords; ++i) { + uint64_t tmp = (RHS.getNumWords() == 1 ? RHS.VAL : RHS.pVal[i]); + __V[i * 2] = (uint32_t)(tmp & mask); + __V[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + + // initialize the quotient and remainder + memset(__Q, 0, (m+n) * sizeof(uint32_t)); + if (Remainder) + memset(__R, 0, n * sizeof(uint32_t)); + + // Now, adjust m and n for the Knuth division. n is the number of words in + // the divisor. m is the number of words by which the dividend exceeds the + // divisor (i.e. m+n is the length of the dividend). These sizes must not + // contain any zero words or the Knuth algorithm fails. + for (unsigned i = n; i > 0 && __V[i-1] == 0; i--) { + n--; + m++; + } + for (unsigned i = m+n; i > 0 && __U[i-1] == 0; i--) + m--; + + // If we're left with only a single word for the divisor, Knuth doesn't work + // so we implement the short division algorithm here. This is much simpler + // and faster because we are certain that we can divide a 64-bit quantity + // by a 32-bit quantity at hardware speed and short division is simply a + // series of such operations. This is just like doing short division but we + // are using base 2^32 instead of base 10. + assert(n != 0 && "Divide by zero?"); + if (n == 1) { + uint32_t divisor = __V[0]; + uint32_t remainder = 0; + for (int i = m+n-1; i >= 0; i--) { + uint64_t partial_dividend = (uint64_t(remainder)) << 32 | __U[i]; + if (partial_dividend == 0) { + __Q[i] = 0; + remainder = 0; + } else if (partial_dividend < divisor) { + __Q[i] = 0; + remainder = (uint32_t)partial_dividend; + } else if (partial_dividend == divisor) { + __Q[i] = 1; + remainder = 0; + } else { + __Q[i] = (uint32_t)(partial_dividend / divisor); + remainder = (uint32_t)(partial_dividend - (__Q[i] * divisor)); + } + } + if (__R) + __R[0] = remainder; + } else { + // Now we're ready to invoke the Knuth classical divide algorithm. In this + // case n > 1. + KnuthDiv(__U, __V, __Q, __R, m, n); + } + + // If the caller wants the quotient + if (Quotient) { + // Set up the Quotient value's memory. + if (Quotient->BitWidth != LHS.BitWidth) { + if (Quotient->isSingleWord()) + Quotient->VAL = 0; + } else + Quotient->clear(); + + // The quotient is in Q. Reconstitute the quotient into Quotient's low + // order words. + if (lhsWords == 1) { + uint64_t tmp = + uint64_t(__Q[0]) | ((uint64_t(__Q[1])) << (APINT_BITS_PER_WORD / 2)); + if (Quotient->isSingleWord()) + Quotient->VAL = tmp; + else + Quotient->pVal[0] = tmp; + } else { + assert(!Quotient->isSingleWord() && "Quotient ap_private not large enough"); + for (unsigned i = 0; i < lhsWords; ++i) + Quotient->pVal[i] = + uint64_t(__Q[i*2]) | ((uint64_t(__Q[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Quotient->clearUnusedBits(); + } + + // If the caller wants the remainder + if (Remainder) { + // Set up the Remainder value's memory. + if (Remainder->BitWidth != RHS.BitWidth) { + if (Remainder->isSingleWord()) + Remainder->VAL = 0; + } else + Remainder->clear(); + + // The remainder is in R. Reconstitute the remainder into Remainder's low + // order words. + if (rhsWords == 1) { + uint64_t tmp = + uint64_t(__R[0]) | ((uint64_t(__R[1])) << (APINT_BITS_PER_WORD / 2)); + if (Remainder->isSingleWord()) + Remainder->VAL = tmp; + else + Remainder->pVal[0] = tmp; + } else { + assert(!Remainder->isSingleWord() && "Remainder ap_private not large enough"); + for (unsigned i = 0; i < rhsWords; ++i) + Remainder->pVal[i] = + uint64_t(__R[i*2]) | ((uint64_t(__R[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Remainder->clearUnusedBits(); + } + + // Clean up the memory we allocated. + if (__U != &SPACE[0]) { + delete [] __U; + delete [] __V; + delete [] __Q; + delete [] __R; + } +} + + +template +void ap_private<_AP_W, _AP_S, _AP_N>::fromString(const char *str, uint32_t slen, uint8_t radix) { + enum { numbits=_AP_W}; + // Check our assumptions here + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + assert(str && "String is null?"); + bool isNeg = str[0] == '-'; + if (isNeg) + str++, slen--; + + //skip any leading zero + while(*str == '0' && *(str+1) != '\0') {str++; slen--;} + assert((slen <= numbits || radix != 2) && "Insufficient bit width"); + assert(((slen - 1)*3 <= numbits || radix != 8) && "Insufficient bit width"); + assert(((slen - 1)*4 <= numbits || radix != 16) && "Insufficient bit width"); + assert((((slen -1)*64)/22 <= numbits || radix != 10) && "Insufficient bit width"); + + memset(pVal, 0, _AP_N * sizeof(uint64_t)); + + // Figure out if we can shift instead of multiply + uint32_t shift = (radix == 16 ? 4 : radix == 8 ? 3 : radix == 2 ? 1 : 0); + + // Set up an ap_private for the digit to add outside the loop so we don't + // constantly construct/destruct it. + uint64_t bigVal[_AP_N]; + memset(bigVal, 0, _AP_N * sizeof(uint64_t)); + ap_private<_AP_W, _AP_S, _AP_N> apdigit(getBitWidth(), bigVal); + ap_private<_AP_W, _AP_S, _AP_N> apradix(radix); + + // Enter digit traversal loop + for (unsigned i = 0; i < slen; i++) { + // Get a digit + uint32_t digit = 0; + char cdigit = str[i]; + if (radix == 16) { +#define isxdigit(c) (((c) >= '0' && (c) <= '9') || ((c) >= 'a' && (c) <= 'f') || ((c) >= 'A' && (c) <= 'F')) +#define isdigit(c) ((c) >= '0' && (c) <= '9') + if (!isxdigit(cdigit)) + assert(0 && "Invalid hex digit in string"); + if (isdigit(cdigit)) + digit = cdigit - '0'; + else if (cdigit >= 'a') + digit = cdigit - 'a' + 10; + else if (cdigit >= 'A') + digit = cdigit - 'A' + 10; + else + assert(0 && "huh? we shouldn't get here"); + } else if (isdigit(cdigit)) { + digit = cdigit - '0'; + } else { + assert(0 && "Invalid character in digit string"); + } +#undef isxdigit +#undef isdigit + // Shift or multiply the value by the radix + if (shift) + *this <<= shift; + else + *this *= apradix; + + // Add in the digit we just interpreted + if (apdigit.isSingleWord()) + apdigit.VAL = digit; + else + apdigit.pVal[0] = digit; + *this += apdigit; + } + // If its negative, put it in two's complement form + if (isNeg) { + (*this)--; + this->flip(); + } + clearUnusedBits(); +} + +template +std::string ap_private<_AP_W, _AP_S, _AP_N>::toString(uint8_t radix, bool wantSigned) const { + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + static const char *digits[] = { + "0","1","2","3","4","5","6","7","8","9","A","B","C","D","E","F" + }; + std::string result; + uint32_t bits_used = getActiveBits(); + + if (radix != 10) { + // For the 2, 8 and 16 bit cases, we can just shift instead of divide + // because the number of bits per digit (1,3 and 4 respectively) divides + // equaly. We just shift until there value is zero. + + // First, check for a zero value and just short circuit the logic below. + if (*this == (uint64_t)(0)) + result = "0"; + else { + ap_private<_AP_W, false, _AP_N> tmp(*this); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + tmp.clearUnusedBitsToZero(); + result = "-"; + insert_at = 1; + } + // Just shift tmp right for each digit width until it becomes zero + uint32_t shift = (radix == 16 ? 4 : (radix == 8 ? 3 : 1)); + uint64_t mask = radix - 1; + ap_private<_AP_W, false, _AP_N> zero(0); + while (tmp.ne(zero)) { + unsigned digit = (tmp.isSingleWord() ? tmp.VAL : tmp.pVal[0]) & mask; + result.insert(insert_at, digits[digit]); + tmp = tmp.lshr(shift); + } + } + return result; + } + + ap_private<_AP_W, false, _AP_N> tmp(*this); + ap_private<_AP_W, false, _AP_N> divisor(radix); + ap_private<_AP_W, false, _AP_N> zero(0); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + tmp.clearUnusedBitsToZero(); + result = "-"; + insert_at = 1; + } + if (tmp == ap_private<_AP_W, false, _AP_N>(0)) + result = "0"; + else while (tmp.ne(zero)) { + ap_private<_AP_W, false, _AP_N> APdigit(0); + ap_private<_AP_W, false, _AP_N> tmp2(0); + divide(tmp, tmp.getNumWords(), divisor, divisor.getNumWords(), &tmp2, + &APdigit); + uint32_t digit = APdigit.getZExtValue(); + assert(digit < radix && "divide failed"); + result.insert(insert_at,digits[digit]); + tmp = tmp2; + } + + return result; +} + +// This implements a variety of operations on a representation of +// arbitrary precision, two's-complement, bignum integer values. + +/* Assumed by lowHalf, highHalf, partMSB and partLSB. A fairly safe + and unrestricting assumption. */ + +/* Some handy functions local to this file. */ + +template +void divide(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t lhsWords, + uint64_t RHS, + ap_private<_AP_W, _AP_S, _AP_N> *Quotient, ap_private<_AP_W, _AP_S, _AP_N> *Remainder) { + uint32_t rhsWords=1; + assert(lhsWords >= rhsWords && "Fractional result"); + enum {APINT_BITS_PER_WORD=64}; + // First, compose the values into an array of 32-bit words instead of + // 64-bit words. This is a necessity of both the "short division" algorithm + // and the the Knuth "classical algorithm" which requires there to be native + // operations for +, -, and * on an m bit value with an m*2 bit result. We + // can't use 64-bit operands here because we don't have native results of + // 128-bits. Furthremore, casting the 64-bit values to 32-bit values won't + // work on large-endian machines. + uint64_t mask = ~0ull >> (sizeof(uint32_t)*8); + uint32_t n = 2; + uint32_t m = (lhsWords * 2) - n; + + // Allocate space for the temporary values we need either on the stack, if + // it will fit, or on the heap if it won't. + uint32_t SPACE[128]; + uint32_t *__U = 0; + uint32_t *__V = 0; + uint32_t *__Q = 0; + uint32_t *__R = 0; + if ((Remainder?4:3)*n+2*m+1 <= 128) { + __U = &SPACE[0]; + __V = &SPACE[m+n+1]; + __Q = &SPACE[(m+n+1) + n]; + if (Remainder) + __R = &SPACE[(m+n+1) + n + (m+n)]; + } else { + __U = new uint32_t[m + n + 1]; + __V = new uint32_t[n]; + __Q = new uint32_t[m+n]; + if (Remainder) + __R = new uint32_t[n]; + } + + // Initialize the dividend + memset(__U, 0, (m+n+1)*sizeof(uint32_t)); + for (unsigned i = 0; i < lhsWords; ++i) { + uint64_t tmp = (LHS.getNumWords() == 1 ? LHS.VAL : LHS.pVal[i]); + __U[i * 2] = tmp & mask; + __U[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + __U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm. + + // Initialize the divisor + memset(__V, 0, (n)*sizeof(uint32_t)); + __V[0] = RHS & mask; + __V[1] = (RHS) >> (sizeof(uint32_t)*8); + + // initialize the quotient and remainder + memset(__Q, 0, (m+n) * sizeof(uint32_t)); + if (Remainder) + memset(__R, 0, n * sizeof(uint32_t)); + + // Now, adjust m and n for the Knuth division. n is the number of words in + // the divisor. m is the number of words by which the dividend exceeds the + // divisor (i.e. m+n is the length of the dividend). These sizes must not + // contain any zero words or the Knuth algorithm fails. + for (unsigned i = n; i > 0 && __V[i-1] == 0; i--) { + n--; + m++; + } + for (unsigned i = m+n; i > 0 && __U[i-1] == 0; i--) + m--; + + // If we're left with only a single word for the divisor, Knuth doesn't work + // so we implement the short division algorithm here. This is much simpler + // and faster because we are certain that we can divide a 64-bit quantity + // by a 32-bit quantity at hardware speed and short division is simply a + // series of such operations. This is just like doing short division but we + // are using base 2^32 instead of base 10. + assert(n != 0 && "Divide by zero?"); + if (n == 1) { + uint32_t divisor = __V[0]; + uint32_t remainder = 0; + for (int i = m+n-1; i >= 0; i--) { + uint64_t partial_dividend = (uint64_t(remainder)) << 32 | __U[i]; + if (partial_dividend == 0) { + __Q[i] = 0; + remainder = 0; + } else if (partial_dividend < divisor) { + __Q[i] = 0; + remainder = partial_dividend; + } else if (partial_dividend == divisor) { + __Q[i] = 1; + remainder = 0; + } else { + __Q[i] = partial_dividend / divisor; + remainder = partial_dividend - (__Q[i] * divisor); + } + } + if (__R) + __R[0] = remainder; + } else { + // Now we're ready to invoke the Knuth classical divide algorithm. In this + // case n > 1. + KnuthDiv(__U, __V, __Q, __R, m, n); + } + + // If the caller wants the quotient + if (Quotient) { + // Set up the Quotient value's memory. + if (Quotient->BitWidth != LHS.BitWidth) { + if (Quotient->isSingleWord()) + Quotient->VAL = 0; + else + delete [] Quotient->pVal; + } else + Quotient->clear(); + + // The quotient is in Q. Reconstitute the quotient into Quotient's low + // order words. + if (lhsWords == 1) { + uint64_t tmp = + uint64_t(__Q[0]) | ((uint64_t(__Q[1])) << (APINT_BITS_PER_WORD / 2)); + if (Quotient->isSingleWord()) + Quotient->VAL = tmp; + else + Quotient->pVal[0] = tmp; + } else { + assert(!Quotient->isSingleWord() && "Quotient ap_private not large enough"); + for (unsigned i = 0; i < lhsWords; ++i) + Quotient->pVal[i] = + uint64_t(__Q[i*2]) | ((uint64_t(__Q[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Quotient->clearUnusedBits(); + } + + // If the caller wants the remainder + if (Remainder) { + // Set up the Remainder value's memory. + if (Remainder->BitWidth != 64 /* RHS.BitWidth */) { + if (Remainder->isSingleWord()) + Remainder->VAL = 0; + } else + Remainder->clear(); + + // The remainder is in __R. Reconstitute the remainder into Remainder's low + // order words. + if (rhsWords == 1) { + uint64_t tmp = + uint64_t(__R[0]) | ((uint64_t(__R[1])) << (APINT_BITS_PER_WORD / 2)); + if (Remainder->isSingleWord()) + Remainder->VAL = tmp; + else + Remainder->pVal[0] = tmp; + } else { + assert(!Remainder->isSingleWord() && "Remainder ap_private not large enough"); + for (unsigned i = 0; i < rhsWords; ++i) + Remainder->pVal[i] = + uint64_t(__R[i*2]) | ((uint64_t(__R[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Remainder->clearUnusedBits(); + } + + // Clean up the memory we allocated. + if (__U != &SPACE[0]) { + delete [] __U; + delete [] __V; + delete [] __Q; + delete [] __R; + } +} + +//When bitwidth < 64 +template class ap_private <_AP_W, _AP_S, 1> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef typename retval<_AP_S>::Type ValType; + template + struct RType { + enum { + _AP_N =1, + mult_w = _AP_W+_AP_W2, + mult_s = _AP_S||_AP_S2, //?? why + plus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, //shouldn't it be AP_MAX(_AP_W,_AP_W2)+!(_AP_S^_AP_S2)+1 ???? + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + minus_s = true, + div_w = _AP_W+_AP_S2, + div_s = _AP_S||_AP_S2, + mod_w = AP_MIN(_AP_W,_AP_W2+(!_AP_S2&&_AP_S)), + mod_s = _AP_S, + logic_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + typedef ap_private mult; + typedef ap_private plus; + typedef ap_private minus; + typedef ap_private logic; + typedef ap_private div; + typedef ap_private mod; + typedef ap_private<_AP_W, _AP_S> arg1; + typedef bool reduce; + }; + enum { APINT_BITS_PER_WORD = 64}; + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -(_AP_W%APINT_BITS_PER_WORD) : 0}; + static const uint64_t mask = ((uint64_t)~0ULL >> (excess_bits)); + static const uint64_t not_mask = ~mask; + static const uint64_t sign_bit_mask = 1ULL << (APINT_BITS_PER_WORD-1); + template struct sign_ext_mask { static const uint64_t mask=~0ULL<<_AP_W1;}; + + enum { BitWidth=_AP_W}; + uint64_t VAL; ///< Used to store the <= 64 bits integer value. + const uint64_t *const pVal; + + INLINE uint32_t getBitWidth() const { + return BitWidth; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + VAL = RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + VAL = RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const ap_private<_AP_W1, _AP_S1, 1>& RHS) { + VAL = RHS.VAL; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const volatile ap_private<_AP_W1, _AP_S1, 1>& RHS) { + VAL = RHS.VAL; + clearUnusedBits(); + return *this; + } + + volatile ap_private& operator=(const ap_private& RHS) volatile { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + ap_private& operator=(const ap_private& RHS) { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + + volatile ap_private& operator=(const volatile ap_private& RHS) volatile { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + ap_private& operator=(const volatile ap_private& RHS) { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + + template + INLINE ap_private& operator = (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + *this = ap_private<_AP_W2, false>(op2); + return *this; + } + + explicit INLINE ap_private(uint64_t* val) : VAL(val[0]), pVal(&VAL){ + clearUnusedBits(); + } + + INLINE bool isSingleWord() const { return true; } + + INLINE void fromString(const char *strStart, uint32_t slen, + uint8_t radix, int offset=0) { + // Check our assumptions here + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + assert(strStart && "String is null?"); + strStart+=offset; + switch(radix) { + case 2: + // sscanf(strStart,"%b",&VAL); + VAL = *strStart =='1' ? ~0ULL : 0; + for (;*strStart; ++strStart) { + assert((*strStart=='0'|| *strStart=='1')&&("Wrong binary number") ); + VAL <<=1; + VAL |= (*strStart-'0'); + } + break; + case 8: +#if __WIN32__ + sscanf(strStart,"%I64o",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lo",&VAL); +#else + sscanf(strStart,"%llo",&VAL); +#endif + +#endif + break; + case 10: +#if __WIN32__ + sscanf(strStart,"%I64u",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lu",&VAL); +#else + sscanf(strStart,"%llu",&VAL); +#endif + +#endif + break; + case 16: +#if __WIN32__ + sscanf(strStart,"%I64x",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lx",&VAL); +#else + sscanf(strStart,"%llx",&VAL); +#endif + +#endif + break; + default: + assert(true && "Unknown radix"); + // error + } + clearUnusedBits(); + } + + INLINE ap_private() : pVal(&VAL){VAL = 0ULL;} + +#define CTOR(TYPE) \ + INLINE ap_private(TYPE v) : VAL((uint64_t)v), pVal(&VAL) { \ + clearUnusedBits(); \ + } + CTOR(int) + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) +#undef CTOR + ap_private(uint32_t numWords, const uint64_t bigVal[]): VAL(bigVal[0]), pVal(&VAL) {clearUnusedBits();} + + ap_private(const std::string& val, uint8_t radix=2, int base=0, int offset=0): VAL(0), pVal(&VAL) { + assert(!val.empty() && "String empty?"); + fromString(val.c_str()+base, val.size()-base, radix); + } + + ap_private(const char strStart[], uint32_t slen, uint8_t radix, int base=0, int offset=0) : VAL(0), pVal(&VAL) { + fromString(strStart+base, slen-base, radix, offset); + } + + ap_private(const ap_private& that) : VAL(that.VAL), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, 1>& that) : VAL(that.VAL), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) : VAL(that.pVal[0]), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& that) : VAL(that.pVal[0]), pVal(&VAL) { + clearUnusedBits(); + } + +#if 0 +template + explicit ap_private(const ap_private<_AP_W1, true, 1+_AP_W1/64>& that) + : VAL((_AP_W1>_AP_W) ? that.VAL & mask : ((1ULL<<(_AP_W1-1)&that.pVal[0]) ? sign_ext_mask<_AP_W1>::mask | that.VAL : that.pVal[0])), pVal(&VAL) {} + +template + explicit ap_private(const ap_private<_AP_W1, false, (_AP_W1+63)/64>& that) + : VAL(that.VAL & mask), pVal(&VAL) {} +#endif + + explicit ap_private(const char* val) : pVal(&VAL) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + bool neg = false; + uint32_t radix = 10; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + ap_private<_AP_W, _AP_S> ap_private_val(str.c_str(), strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + } + + ap_private(const char* val, signed char rd): pVal(&VAL) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + uint32_t radix = rd; + bool neg = false; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + uint32_t bitsNeeded = ap_private<_AP_W, _AP_S>::getBitsNeeded(strp, strLen, radix); + ap_private<_AP_W, _AP_S> ap_private_val(strp , strLen, radix, base, offset); + //ap_private<_AP_W, _AP_S> ap_private_val(bitsNeeded, strp , strLen, radix, base, offset); + if (strp[0] == '-') + ap_private_val = -ap_private_val; + operator = (ap_private_val); + } + + INLINE bool isNegative() const { + static const uint64_t sign_mask = 1ULL << (_AP_W-1); + return _AP_S && (sign_mask & VAL); + } + + INLINE bool isPositive() const { + return !isNegative(); + } + + INLINE bool isStrictlyPositive() const { + return !isNegative() && VAL!=0; + } + + INLINE bool isAllOnesValue() const { + return (mask & VAL) == mask; + } + + template + INLINE bool operator==(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + return (VAL == RHS.VAL); + } + + INLINE bool operator==(const ap_private<_AP_W, _AP_S>& RHS) const { return VAL == RHS.VAL; } + INLINE bool operator==(const ap_private<_AP_W, !_AP_S>& RHS) const { return getVal() == RHS.getVal(); } + INLINE bool operator==(uint64_t Val) const { return (VAL == Val); } + INLINE bool operator!=(uint64_t Val) const { return (VAL != Val); } + INLINE bool operator!=(const ap_private<_AP_W, _AP_S>& RHS) const { return VAL != RHS.VAL; } + INLINE bool operator!=(const ap_private<_AP_W, !_AP_S>& RHS) const { return getVal() != RHS.getVal(); } + const ap_private operator++() { ++VAL; clearUnusedBits(); return *this; } + const ap_private operator--(int) { + ap_private orig(*this); + --VAL; clearUnusedBits(); + return orig; + } + const ap_private operator--() { --VAL; clearUnusedBits(); return *this;} + INLINE bool operator !() const { return !VAL;} + + const ap_private operator++(int) { + ap_private orig(*this); + VAL++; clearUnusedBits(); + return orig; + } + + const ap_private operator~() {return ap_private(~VAL);} + INLINE typename RType<1,false>::minus operator-() const { + return ap_private<1,false>(0) - (*this); + } + + INLINE std::string toString(uint8_t radix, bool wantSigned) const ; + INLINE std::string toStringUnsigned(uint8_t radix = 10) const { + return toString(radix, false); + } + INLINE std::string toStringSigned(uint8_t radix = 10) const { + return toString(radix, true); + } + INLINE void clear() { + VAL=0; + } + INLINE ap_private& clear(uint32_t bitPosition) { VAL &= ~(1ULL<<(bitPosition)); clearUnusedBits(); return *this;} + + INLINE ap_private ashr(uint32_t shiftAmt) const { + enum {excess_bits = APINT_BITS_PER_WORD - BitWidth}; + if (_AP_S) + return ap_private((shiftAmt == BitWidth) ? 0 : ((int64_t)VAL) >> (shiftAmt)); + else + return ap_private((shiftAmt == BitWidth) ? 0 : (VAL) >> (shiftAmt)); + } + + INLINE ap_private lshr(uint32_t shiftAmt) const { + return ap_private((shiftAmt == BitWidth) ? ap_private(0) : ap_private((VAL&mask) >> (shiftAmt))); + } + + INLINE ap_private shl(uint32_t shiftAmt) const { + if (shiftAmt > BitWidth) { + if (!isNegative()) + return ap_private(0); + else return ap_private(-1); + } + if (shiftAmt == BitWidth) return ap_private(0); + else return ap_private((VAL) << (shiftAmt)); + //return ap_private((shiftAmt == BitWidth) ? ap_private(0ULL) : ap_private(VAL << shiftAmt)); + } + + INLINE int64_t getSExtValue() const { + return VAL; + } + + INLINE uint64_t getZExtValue() const { + return VAL & mask; + } + + template + INLINE ap_private(const ap_range_ref<_AP_W2,_AP_S2>& ref) : pVal(&VAL) { + *this=ref.get(); + } + + template + INLINE ap_private(const ap_bit_ref<_AP_W2,_AP_S2>& ref) : pVal(&VAL) { + *this = ((uint64_t)(bool)ref); + } + + template + INLINE ap_private(const ap_concat_ref<_AP_W2, _AP_T2,_AP_W3, _AP_T3>& ref) : pVal(&VAL) { + *this=ref.get(); + } + + template + INLINE ap_private(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) : pVal(&VAL) { + *this = ((val.operator ap_private<_AP_W2, false> ())); + } + + template + INLINE ap_private(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) : pVal(&VAL) { + *this = (uint64_t)(bool)val; + } + + INLINE void write(const ap_private<_AP_W, _AP_S>& op2) volatile { + *this = (op2); + } + + //Explicit conversions to C interger types + //----------------------------------------------------------- + ValType getVal() const { + return VAL; + } + operator ValType () const { + return getVal(); + } + INLINE int to_int() const { + // ap_private<64 /* _AP_W */, _AP_S> res(V); + return (int) getVal(); + } + + INLINE unsigned to_uint() const { + return (unsigned) getVal(); + } + + INLINE long to_long() const { + return (long) getVal(); + } + + INLINE unsigned long to_ulong() const { + return (unsigned long) getVal(); + } + + INLINE ap_slong to_int64() const { + return (ap_slong) getVal(); + } + + INLINE ap_ulong to_uint64() const { + return (ap_ulong) getVal(); + } + + INLINE double to_double() const { + if (isNegative()) + return roundToDouble(true); + else + return roundToDouble(false); + } + + INLINE bool isMinValue() const { return VAL == 0;} + template INLINE ap_private& operator&=(const ap_private<_AP_W1, _AP_S1>& RHS) { + VAL = VAL&RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator|=(const ap_private<_AP_W1, _AP_S1>& RHS) { + VAL = VAL|RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator^=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL^RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator*=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL*RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator+=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL+RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator-=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL-RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + INLINE const ap_private& operator<<=(uint32_t shiftAmt) { VAL<<=shiftAmt; clearUnusedBits(); return *this; } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator&(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL & RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret & RHS; + } + } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator^(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL ^ RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret ^ RHS; + } + } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator|(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL | RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret | RHS; + } + } + + INLINE ap_private<_AP_W, _AP_S> And(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL & RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Or(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL | RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Xor(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL ^ RHS.VAL); + } +#if 1 + template + INLINE typename RType<_AP_W1, _AP_S1>::mult operator*(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::mult_w <= 64) { + typename RType<_AP_W1, _AP_S1>::mult Result(VAL * RHS.VAL); + return Result; + } else { + typename RType<_AP_W1, _AP_S1>::mult Result = typename RType<_AP_W1, _AP_S1>::mult(*this); + Result *= RHS; + return Result; + } + } +#endif + INLINE ap_private<_AP_W, _AP_S> Mul(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL * RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Add(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL + RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Sub(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL - RHS.VAL); + } + +#if 1 + INLINE ap_private& operator&=(uint64_t RHS) { VAL &= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator|=(uint64_t RHS) { VAL |= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator^=(uint64_t RHS){ VAL ^= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator*=(uint64_t RHS){ VAL *= RHS; clearUnusedBits(); return *this; } + INLINE ap_private& operator+=(uint64_t RHS){ VAL += RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator-=(uint64_t RHS){ VAL -= RHS; clearUnusedBits(); return *this; } + INLINE ap_private operator&(uint64_t RHS) const { return ap_private(VAL & RHS); } + INLINE ap_private operator|(uint64_t RHS) const { return ap_private(VAL | RHS); } + INLINE ap_private operator^(uint64_t RHS) const { return ap_private(VAL ^ RHS); } + INLINE ap_private operator*(uint64_t RHS) const { return ap_private(VAL * RHS); } + INLINE ap_private operator/(uint64_t RHS) const { return ap_private(VAL / RHS); } + INLINE ap_private operator+(uint64_t RHS) const { return ap_private(VAL + RHS); } + INLINE ap_private operator-(uint64_t RHS) const { return ap_private(VAL - RHS); } +#endif + INLINE bool isMinSignedValue() const { + static const uint64_t min_mask = ~(~0ULL << (_AP_W-1)); + return BitWidth == 1 ? VAL == 1 : + (ap_private_ops::isNegative<_AP_W>(*this) && ((min_mask & VAL)==0)); + } + +#if 1 + + template INLINE + typename RType<_AP_W1,_AP_S1>::plus operator+(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1,_AP_S1>::plus_w <=64) + return typename RType<_AP_W1,_AP_S1>::plus(RType<_AP_W1,_AP_S1>::plus_s ? int64_t(VAL+RHS.VAL):uint64_t(VAL+RHS.VAL)); + typename RType<_AP_W1,_AP_S1>::plus Result=RHS; + Result += VAL; + return Result; + } + + template INLINE + typename RType<_AP_W1,_AP_S1>::minus operator-(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1,_AP_S1>::minus_w <=64) + return typename RType<_AP_W1,_AP_S1>::minus(int64_t(VAL-RHS.VAL)); + typename RType<_AP_W1,_AP_S1>::minus Result=*this; + Result -= RHS; + return Result; + } +#endif // #if 1 + + INLINE ap_private& flip() { + VAL = (~0ULL^VAL)&mask; + clearUnusedBits(); + return *this; + } + + uint32_t countPopulation() const { return CountPopulation_64(VAL);} + uint32_t countLeadingZeros() const { + int remainder = BitWidth % APINT_BITS_PER_WORD; + int excessBits = (APINT_BITS_PER_WORD - remainder) % APINT_BITS_PER_WORD; + //enum { remainder = BitWidth % APINT_BITS_PER_WORD, excessBits = APINT_BITS_PER_WORD - remainder}; + uint32_t Count = CountLeadingZeros_64(VAL); + if (Count) + Count-=excessBits; + return AESL_std::min(Count, (uint32_t)_AP_W); + } + + /// HiBits - This function returns the high "numBits" bits of this ap_private. + ap_private<_AP_W, _AP_S, 1> getHiBits(uint32_t numBits) const { + ap_private<_AP_W, _AP_S, 1> ret(*this); + ret = (ret)>>(BitWidth - numBits); + return ret; + } + + /// LoBits - This function returns the low "numBits" bits of this ap_private. + ap_private<_AP_W, _AP_S, 1> getLoBits(uint32_t numBits) const { + ap_private<_AP_W, _AP_S, 1> ret((VAL) << (BitWidth - numBits)); + ret = (ret)>>(BitWidth - numBits); + return ret; + //return ap_private(numBits, (VAL << (BitWidth - numBits))>> (BitWidth - numBits)); + } + + ap_private<_AP_W, _AP_S,1>& set(uint32_t bitPosition) { + VAL |= (1ULL << (bitPosition)); + clearUnusedBits(); + return *this; // clearUnusedBits(); + } + + void set() { + VAL = ~0ULL; + clearUnusedBits(); + } + + template + INLINE void set(const ap_private<_AP_W3, false> & val) { + operator = (ap_private<_AP_W3, _AP_S>(val)); + } + + INLINE void set(const ap_private & val) { + operator = (val); + } + + bool operator[](uint32_t bitPosition) const { + return (((1ULL << (bitPosition)) & VAL) != 0); + } + + INLINE void clearUnusedBits(void) { + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -_AP_W%APINT_BITS_PER_WORD : 0}; + VAL = _AP_S ? ((((int64_t)VAL)<<(excess_bits))>> (excess_bits)) : (excess_bits ? ((VAL)<<(excess_bits))>>(excess_bits) : VAL); + } + + INLINE void clearUnusedBitsToZero(void) { + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -_AP_W%APINT_BITS_PER_WORD : 0}; + static uint64_t mask = ~0ULL >> (excess_bits); + VAL &= mask; + } + + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1> udiv(const ap_private<_AP_W, _AP_S1>& RHS) const { + return ap_private<_AP_W, _AP_S||_AP_S1>(VAL / RHS.VAL); + } + + INLINE ap_private udiv(uint64_t RHS) const { + return ap_private(VAL / RHS); + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1> sdiv(const ap_private<_AP_W, _AP_S1> & RHS) const { + if (isNegative()) + if (RHS.isNegative()) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS.isNegative()) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + INLINE ap_private sdiv(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS<0) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + template + INLINE ap_private urem(const ap_private<_AP_W, _AP_S2>& RHS) const { + assert(RHS.VAL != 0 && "Divide by 0"); + return ap_private(VAL%RHS.VAL); + } + + INLINE ap_private urem(uint64_t RHS) const { + assert(RHS != 0 && "Divide by 0"); + return ap_private(VAL%RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + template + INLINE ap_private srem(const ap_private<_AP_W, _AP_S2>& RHS) const { + if (isNegative()) { + ap_private lhs = -(*this); + if (RHS.isNegative()) { + ap_private rhs = -RHS; + return -(lhs.urem(rhs)); + } else + return -(lhs.urem(RHS)); + } else if (RHS.isNegative()) { + ap_private rhs = -RHS; + return this->urem(rhs); + } + return this->urem(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return -((-(*this)).urem(-RHS)); + else + return -((-(*this)).urem(RHS)); + else if (RHS<0) + return this->urem(-RHS); + return this->urem(RHS); + } + + INLINE static void udivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder){ + assert(RHS!=0 && "Divide by 0"); + Quotient = LHS.VAl/RHS.VAl; + Remainder = LHS.VAL % RHS.VAL; + } + + INLINE static void udivrem(const ap_private &LHS, uint64_t RHS, + ap_private &Quotient, ap_private &Remainder){ + assert(RHS!=0 && "Divide by 0"); + Quotient = LHS.VAl/RHS; + Remainder = LHS.VAL % RHS; + } + + INLINE static void sdivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS.isNegative()) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS.isNegative()) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + INLINE static void sdivrem(const ap_private &LHS, int64_t RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS<0) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS<0) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + template INLINE bool eq(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return (*this) == RHS; + } + + template INLINE bool ne(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !((*this) == RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered unsigned. + /// @brief Unsigned less than comparison + template INLINE bool ult(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + uint64_t lhsZext = ((uint64_t(VAL)) << (64-_AP_W)) >> (64-_AP_W); + uint64_t rhsZext = ((uint64_t(RHS.VAL)) << (64-_AP_W1)) >> (64-_AP_W1); + return lhsZext < rhsZext; + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered signed. + /// @brief Signed less than comparison + template INLINE bool slt(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + int64_t lhsSext = ((int64_t(VAL)) << (64-_AP_W)) >> (64-_AP_W); + int64_t rhsSext = ((int64_t(RHS.VAL)) << (64-_AP_W1)) >> (64-_AP_W1); + return lhsSext < rhsSext; + } + + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered unsigned. + /// @brief Unsigned less or equal comparison + template INLINE bool ule(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return ult(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered signed. + /// @brief Signed less or equal comparison + template INLINE bool sle(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return slt(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered unsigned. + /// @brief Unsigned greather than comparison + template INLINE bool ugt(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !ult(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered signed. + /// @brief Signed greather than comparison + template INLINE bool sgt(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !slt(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered unsigned. + /// @brief Unsigned greater or equal comparison + template INLINE bool uge(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !ult(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered signed. + /// @brief Signed greather or equal comparison + template INLINE bool sge(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !slt(RHS); + } + + INLINE ap_private abs() const { + if (isNegative()) + return -(*this); + return *this; + } + + ap_private<_AP_W, false> get() const { + ap_private<_AP_W,false> ret(*this); + return ret; + } + + INLINE static uint32_t getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { + return _AP_W; + } + + INLINE uint32_t getActiveBits() const { + uint32_t bits=_AP_W - countLeadingZeros(); + return bits?bits:1; + } + + INLINE double roundToDouble(bool isSigned=false) const { + const static uint64_t mask = ~0ULL << (APINT_BITS_PER_WORD - _AP_W); + return double(VAL); + } + + INLINE unsigned length() const { return _AP_W; } + + /*Reverse the contents of ap_private instance. I.e. LSB becomes MSB and vise versa*/ + INLINE ap_private& reverse () { + for (int i = 0; i < _AP_W/2; ++i) { + bool tmp = operator[](i); + if (operator[](_AP_W - 1 - i)) + set(i); + else + clear(i); + if (tmp) + set(_AP_W - 1 - i); + else + clear(_AP_W - 1 - i); + } + clearUnusedBits(); + return *this; + } + + /*Return true if the value of ap_private instance is zero*/ + INLINE bool iszero () const { + return isMinValue(); + } + + /* x < 0 */ + INLINE bool sign () const { + if (isNegative()) + return true; + return false; + } + + /* x[i] = !x[i] */ + INLINE void invert (int i) { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + flip(i); + } + + /* x[i] */ + INLINE bool test (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator[](i); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the left + INLINE void lrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (shl(n) | lshr(_AP_W - n)); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the right + INLINE void rrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (lshr(n) | shl(_AP_W - n)); + } + + //Set the ith bit into v + INLINE void set (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Set the ith bit into v + INLINE void set_bit (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Get the value of ith bit + INLINE bool get_bit (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //complements every bit + INLINE void b_not() { + flip(); + } + + //Binary Arithmetic + //----------------------------------------------------------- +#define OP_BIN_AP(Sym,Rty, Fun) \ + template \ + INLINE \ + typename RType<_AP_W2,_AP_S2>::Rty \ + operator Sym (const ap_private<_AP_W2,_AP_S2>& op) const { \ + typename RType<_AP_W2,_AP_S2>::Rty lhs(*this); \ + typename RType<_AP_W2,_AP_S2>::Rty rhs(op); \ + return lhs.Fun(rhs); \ + } \ + + ///Bitwise and, or, xor + //OP_BIN_AP(&,logic, And) + //OP_BIN_AP(|,logic, Or) + //OP_BIN_AP(^,logic, Xor) + +#undef OP_BIN_AP + template + INLINE typename RType<_AP_W2,_AP_S2>::div + operator / (const ap_private<_AP_W2,_AP_S2>&op) const { + ap_private lhs=ap_private(*this); + ap_private rhs=ap_private(op); + return typename RType<_AP_W2,_AP_S2>::div((_AP_S||_AP_S2)?lhs.sdiv(rhs):lhs.udiv(rhs)); + } + + + template + INLINE typename RType<_AP_W2,_AP_S2>::mod + operator % (const ap_private<_AP_W2,_AP_S2>&op) const { + ap_private lhs=*this; + ap_private rhs=op; + typename RType<_AP_W2,_AP_S2>::mod res = typename RType<_AP_W2,_AP_S2>::mod (_AP_S?lhs.srem(rhs):lhs.urem(rhs)); + return res; + } + + +#define OP_ASSIGN_AP_2(Sym) \ + template \ + INLINE ap_private<_AP_W, _AP_S>& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op); \ + return *this; \ + } \ + + OP_ASSIGN_AP_2(/) + OP_ASSIGN_AP_2(%) +#undef OP_ASSIGN_AP_2 + + ///Bitwise assign: and, or, xor + //------------------------------------------------------------- + // OP_ASSIGN_AP(&) + // OP_ASSIGN_AP(^) + // OP_ASSIGN_AP(|) +#undef OP_ASSIGN_AP +#if 1 + + template + INLINE ap_private<_AP_W, _AP_S> + operator << (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh=op2.to_uint(); + return *this << sh; + } + + INLINE ap_private<_AP_W, _AP_S> + operator << (uint32_t sh) const { + return shl(sh); + } + +#endif + + template + INLINE ap_private<_AP_W, _AP_S> + operator >> (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh = op2.to_uint(); + return *this >> sh; + } + + INLINE ap_private<_AP_W, _AP_S> + operator >>(uint32_t sh) const { + ap_private<_AP_W, _AP_S> r(*this); + bool overflow=(sh>=_AP_W); + bool neg_v=r.isNegative(); + if(_AP_S) { + if(overflow) + neg_v?r.set():r.clear(); + else + return r.ashr(sh); + } else { + if(overflow) + r.clear(); + else + return r.lshr(sh); + } + return r; + } + + ///Shift assign + //------------------------------------------------------------------ +#define OP_ASSIGN_AP_3_SINGLE(Sym) \ + template \ + INLINE ap_private<_AP_W, _AP_S>& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op.getVal()); \ + return *this; \ + } + OP_ASSIGN_AP_3_SINGLE(>>) +#undef OP_ASSIGN_AP_3_SINGLE + + ///Comparisons + //----------------------------------------------------------------- + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this==op); + } + + template + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return op < *this; + } + + template + INLINE bool operator <= (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this>op); + } + + template + INLINE bool operator < (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.slt(rhs):lhs.ult(rhs); + else if (_AP_W < 32 && _AP_W2 < 32) + return lhs.slt(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + } + + template + INLINE bool operator >=(const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this + INLINE bool operator == (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op == *this; + } + + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return !(op==*this); + } + + template + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op < (*this); + } + + template + INLINE bool operator <= (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op >= *this; + } + + template + INLINE bool operator <(const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op > *this; + } + + template + INLINE bool operator >=(const ap_private<_AP_W2,_AP_S2,_AP_N2>& op) const { + return op <= *this; + } + ///Bit and Part Select + //-------------------------------------------------------------- + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast*>(this), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>((const_cast*> (this)), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast(this), Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (uint32_t index) { + assert(index >= 0&&"Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S> (*this, (int)index); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + template + INLINE bool operator [] (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br =operator [] (index); + return br.to_bool(); + } + + INLINE ap_bit_ref<_AP_W,_AP_S> bit (int index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> bit (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W &&"Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + INLINE bool bit (int index) const { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br(const_cast*>(this), index); + return br.to_bool(); + } + + template + INLINE bool bit (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br = bit(index); + return br.to_bool(); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(const ap_private<_AP_W2,_AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(ap_private<_AP_W2,_AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (ap_range_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (ap_bit_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_private + operator & (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this & a2.get(); + } + + template + INLINE ap_private + operator | (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this | a2.get(); + } + + template + INLINE ap_private + operator ^ (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this ^ a2.get(); + } + + + //Reduce operation + //----------------------------------------------------------- + INLINE bool and_reduce() const { + return (VAL & mask) == mask; + } + + INLINE bool nand_reduce() const { + return (VAL & mask) != mask; + } + + INLINE bool or_reduce() const { + return (bool)VAL; + } + + INLINE bool nor_reduce() const { + return VAL==0; + } + + INLINE bool xor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?true:false; + } + + INLINE bool xnor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?false:true; + } + + INLINE std::string to_string(uint8_t radix=2, bool sign=false) const { + return toString(radix, radix==10?_AP_S:sign); + } +}; +template +std::string ap_private<_AP_W, _AP_S, 1>::toString(uint8_t radix, bool wantSigned) const { + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + static const char *digits[] = { + "0","1","2","3","4","5","6","7","8","9","a","b","c","d","e","f" + }; + std::string result; + if (radix != 10) { + // For the 2, 8 and 16 bit cases, we can just shift instead of divide + // because the number of bits per digit (1,3 and 4 respectively) divides + // equaly. We just shift until there value is zero. + + // First, check for a zero value and just short circuit the logic below. + if (*this == (uint64_t)(0)) + result = "0"; + else { + ap_private<_AP_W, false, 1> tmp(*this); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + result = "-"; + insert_at = 1; + } + // Just shift tmp right for each digit width until it becomes zero + uint32_t shift = (radix == 16 ? 4 : (radix == 8 ? 3 : 1)); + uint64_t mask = radix - 1; + ap_private<_AP_W, false, 1> zero(0); + while (tmp.ne(zero)) { + unsigned digit = (unsigned)(tmp.VAL & mask); + result.insert(insert_at, digits[digit]); + tmp = tmp.lshr(shift); + } + } + return result; + } + + ap_private<_AP_W, false, 1> tmp(*this); + ap_private<6, false, 1> divisor(radix); + ap_private<_AP_W, _AP_S, 1> zero(0); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + result = "-"; + insert_at = 1; + } + if (tmp == ap_private<_AP_W, false, 1>(0ULL)) + result = "0"; + else while (tmp.ne(zero)) { + ap_private<_AP_W, false, 1> APdigit = tmp%divisor; + ap_private<_AP_W, false, 1> tmp2 = tmp/divisor; + uint32_t digit = (uint32_t)(APdigit.getZExtValue()); + assert(digit < radix && "divide failed"); + result.insert(insert_at,digits[digit]); + tmp = tmp2; + } + return result; + +} + +#endif /* #ifndef LLVM_SUPPORT_MATHEXTRAS_H */ \ No newline at end of file diff --git a/hls/ZU3EG_test/main.cpp b/hls/ZU3EG_test/main.cpp new file mode 100644 index 0000000000000000000000000000000000000000..15026a6461d31674bb99505f9dd3c53d8d424d65 --- /dev/null +++ b/hls/ZU3EG_test/main.cpp @@ -0,0 +1,91 @@ +/** + * main.cpp + * + * for Vivado HLS + */ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +#ifdef CALCTIME +#include +#include +#endif + +#include "router.hpp" + + +int main(int argc, char *argv[]) { + using namespace std; + + // テストデータ (文字列形式) + // NL_Q00.txt + //char boardstr[BOARDSTR_SIZE] = "X10Y05Z3L0000107041L0004107002L0102102021L0900100003"; + // NL_Q06.txt + char boardstr[BOARDSTR_SIZE] = "X10Y18Z2L0900109002L0901105012L0902103052L0903103062L0904100102L0905106012L0906109022L0717109102L0808109112L0017209172L0401200072L0912208152L0009201092L0709209092L0901206052L0309204092L0701209072L0101201022L0011202152L0016202162"; + // NL_Q08.txt + //char boardstr[BOARDSTR_SIZE] = "X17Y20Z2L0000103022L1603115052L0916107032L0302108012L1104111042L1002100002L0919116162L1616113182L1001115012L0500201182L1603213152L0600210022"; + + // 指定されてればコマンドラインから問題文字列を読み込む + if (1 < argc) { + //先頭がXではないならば標準入力から読み込む + if(argv[1][0]!='X') + { + char* c_p=fgets(boardstr, BOARDSTR_SIZE, stdin); + int length=strlen(c_p); + boardstr[length-1]=0; + } + else + { + strcpy(boardstr, argv[1]); + } + } + + // 指定されてればシード値を読み込む + int seed = 12345; + if (2 < argc) { + seed = atoi(argv[2]); + } + + int size_x = (boardstr[1] - '0') * 10 + (boardstr[2] - '0'); + int size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); + int size_z = (boardstr[7] - '0'); + + // ソルバ実行 + ap_int<32> status; + clock_t clock_start, clock_done; + clock_start = clock(); + bool result = pynqrouter_zu3eg_test(boardstr, seed, &status); + clock_done = clock(); + if (result) { + cout << endl << "Test Passed!" << endl; + } else { + cout << endl << "Test Failed!" << endl; + } + cout << "status = " << (int)status << endl; + cout << "elapsed = " << ((double)(clock_done - clock_start) / CLOCKS_PER_SEC) << endl << endl; + + // 解表示 + cout << "SOLUTION" << endl; + cout << "========" << endl; + cout << "SIZE " << size_x << "X" << size_y << "X" << size_z << endl; + for (int z = 0; z < size_z; z++) { + cout << "LAYER " << (z + 1) << endl; + for (int y = 0; y < size_y; y++) { + for (int x = 0; x < size_x; x++) { + if (x != 0) { + cout << ","; + } + int i = ((x * MAX_WIDTH + y) << BITWIDTH_Z) | z; + //cout << setfill('0') << setw(2) << right << (unsigned int)(unsigned char)(boardstr[i]); + cout << (unsigned int)(unsigned char)(boardstr[i]); + } + cout << endl; + } + } + + return 0; +} diff --git a/hls/ZU3EG_test/router.cpp b/hls/ZU3EG_test/router.cpp new file mode 100644 index 0000000000000000000000000000000000000000..193f0ce4a3c02838af1f905e3d1a30e511abb8e7 --- /dev/null +++ b/hls/ZU3EG_test/router.cpp @@ -0,0 +1,744 @@ +/** + * router.cpp + * + * for Vivado HLS + */ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +#include "./router.hpp" + + +// ================================ // +// LFST +// ================================ // + +// 参考 https://highlevel-synthesis.com/2017/02/10/lfsr-in-hls/ +static ap_uint<32> lfsr; + +void lfsr_random_init(ap_uint<32> seed) { +#pragma HLS INLINE + lfsr = seed; +} + +ap_uint<32> lfsr_random() { +#pragma HLS INLINE + bool b_32 = lfsr.get_bit(32-32); + bool b_22 = lfsr.get_bit(32-22); + bool b_2 = lfsr.get_bit(32-2); + bool b_1 = lfsr.get_bit(32-1); + bool new_bit = b_32 ^ b_22 ^ b_2 ^ b_1; + lfsr = lfsr >> 1; + lfsr.set_bit(31, new_bit); + + return lfsr.to_uint(); +} + +// AからBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +/*ap_uint<32> lfsr_random_uint32(ap_uint<32> a, ap_uint<32> b) { +#pragma HLS INLINE + return lfsr_random() % (b - a + 1) + a; +}*/ + +// 0からBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +/*ap_uint<32> lfsr_random_uint32_0(ap_uint<32> b) { +#pragma HLS INLINE + return lfsr_random() % (b + 1); +}*/ + + +// ================================ // +// メインモジュール +// ================================ // + +// 重みの更新 +// TODO 調整 +// min_uint8(r, MAX_WEIGHT) と同じ +ap_uint<8> new_weight(ap_uint<16> x) { +#pragma HLS INLINE +#if 1 + // 下位8ビット (最大 255) を抜き出して、1/8 をかけて最大 31 (32) にする + ap_uint<8> y = x & 255; + return (ap_uint<8>)(y / 8 + 1); +#endif +#if 0 + // 下位10ビット (最大 1023) を抜き出して、1/32 をかけて最大 31 (32) にする + ap_uint<10> y = x & 1023; + return (ap_uint<8>)(y / 32 + 1); +#endif +#if 0 + ap_uint<8> y = x / 8; + if (y < (ap_uint<16>)MAX_WEIGHT) { return y; } + else { return MAX_WEIGHT; } +#endif +} + +// ボードに関する変数 +static ap_uint<7> size_x; // ボードの X サイズ +static ap_uint<7> size_y; // ボードの Y サイズ +static ap_uint<4> size_z; // ボードの Z サイズ + +static ap_uint<9> line_num = 0; // ラインの総数 + +// グローバル変数で定義する +#ifdef GLOBALVARS + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト + + ap_uint<8> weights[MAX_CELLS]; // セルの重み + + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン +#endif + +bool pynqrouter_zu3eg_test(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *status) { +#pragma HLS INTERFACE s_axilite port=boardstr bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=seed bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=status bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=return bundle=AXI4LS + + *status = -1; + +// グローバル変数では定義しない +#ifndef GLOBALVARS + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト +#pragma HLS ARRAY_PARTITION variable=starts complete dim=1 + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト +#pragma HLS ARRAY_PARTITION variable=goals complete dim=1 + + ap_uint<8> weights[MAX_CELLS]; // セルの重み +//#pragma HLS ARRAY_PARTITION variable=weights cyclic factor=8 dim=1 partition +// Note: weights は様々な順番でアクセスされるからパーティションしても全然効果ない + + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ +//#pragma HLS ARRAY_PARTITION variable=paths_size complete dim=1 + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) +//#pragma HLS ARRAY_PARTITION variable=paths cyclic factor=16 dim=2 partition + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン +//#pragma HLS ARRAY_PARTITION variable=adjacents complete dim=1 +#endif + + // ================================ + // 初期化 BEGIN + // ================================ + + // ループカウンタは1ビット余分に用意しないと終了判定できない + INIT_ADJACENTS: + for (ap_uint<11> i = 0; i < (ap_uint<11>)(MAX_LINES); i++) { + adjacents[i] = false; + } + + INIT_WEIGHTS: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=8 + weights[i] = 1; + } + + // ボードストリングの解釈 + + size_x = (boardstr[1] - '0') * 10 + (boardstr[2] - '0'); + size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); + size_z = (boardstr[7] - '0'); + + INIT_BOARDS: + for (ap_uint<16> idx = 8; idx < BOARDSTR_SIZE; idx+=11) { +//#pragma HLS LOOP_TRIPCOUNT min=100 max=32768 avg=1000 + + // 終端 (null) 文字 + if (boardstr[idx] == 0) { + break; + } + + ap_uint<7> s_x = (boardstr[idx+1] - '0') * 10 + (boardstr[idx+2] - '0'); + ap_uint<7> s_y = (boardstr[idx+3] - '0') * 10 + (boardstr[idx+4] - '0'); + ap_uint<3> s_z = (boardstr[idx+5] - '0') - 1; + ap_uint<7> g_x = (boardstr[idx+6] - '0') * 10 + (boardstr[idx+7] - '0'); + ap_uint<7> g_y = (boardstr[idx+8] - '0') * 10 + (boardstr[idx+9] - '0'); + ap_uint<3> g_z = (boardstr[idx+10] - '0') - 1; + //cout << "L " << line_num << ": (" << s_x << ", " << s_y << ", " << s_z << ") " + // "(" << g_x << ", " << g_y << ", " << g_z << ")" << endl; + + // スタートとゴール + ap_uint<16> start_id = (((ap_uint<16>)s_x * MAX_WIDTH + (ap_uint<16>)s_y) << BITWIDTH_Z) | (ap_uint<16>)s_z; + ap_uint<16> goal_id = (((ap_uint<16>)g_x * MAX_WIDTH + (ap_uint<16>)g_y) << BITWIDTH_Z) | (ap_uint<16>)g_z; + starts[line_num] = start_id; + goals[line_num] = goal_id; + + // 初期状態で数字が隣接しているか判断 + ap_int<8> dx = (ap_int<8>)g_x - (ap_int<8>)s_x; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<8> dy = (ap_int<8>)g_y - (ap_int<8>)s_y; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<4> dz = (ap_int<4>)g_z - (ap_int<4>)s_z; // 最小-7 最大7 (-> 符号付き4ビット) + if ((dx == 0 && dy == 0 && (dz == 1 || dz == -1)) || (dx == 0 && (dy == 1 || dy == -1) && dz == 0) || ((dx == 1 || dx == -1) && dy == 0 && dz == 0)) { + adjacents[line_num] = true; + paths_size[line_num] = 0; + } else { + adjacents[line_num] = false; + } + + paths_size[line_num] = 0; + weights[start_id] = MAX_WEIGHT; + weights[goal_id] = MAX_WEIGHT; + + line_num++; + } + //cout << size_x << " " << size_y << " " << size_z << endl; + + // 乱数の初期化 + lfsr_random_init(seed); + + // TODO + // すべてのラインが隣接してたらソルバを終わりにする + + // ================================ + // 初期化 END + // ================================ + + // ================================ + // ルーティング BEGIN + // ================================ + + // [Step 1] 初期ルーティング + cout << "Initial Routing" << endl; + FIRST_ROUTING: + for (ap_uint<11> i = 0; i < (ap_uint<11>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=1023 avg=50 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[i] == false) { + + // 経路探索 +#ifdef DEBUG_PRINT + cout << "LINE #" << (int)(i + 1) << endl; +#endif + search(&(paths_size[i]), paths[i], starts[i], goals[i], weights); + + } + } + + ap_uint<1> overlap_checks[MAX_CELLS]; +#pragma HLS ARRAY_PARTITION variable=overlap_checks cyclic factor=16 dim=1 partition + bool has_overlap = false; + +#ifndef USE_MOD_CALC + // line_num_2: line_num 以上で最小の2のべき乗数 + ap_uint<8> line_num_2; + CALC_LINE_NUM_2: + for (line_num_2 = 1; line_num_2 < (ap_uint<10>)line_num; line_num_2 *= 2) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 + ; + } + //cout << "line_num: " << line_num << endl; + //cout << "line_num_2: " << line_num_2 << endl; +#endif + + ap_uint<8> last_target = 255; + + // [Step 2] Rip-up 再ルーティング + ROUTING: + for (ap_uint<16> round = 1; round <= 32768 /* = (2048 * 16) */; round++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=4000 avg=50 + +//#ifdef DEBUG_PRINT + //cout << "ITERATION " << round; +//#endif + + // 対象ラインを選択 +#ifdef USE_MOD_CALC + // (1) 剰余演算を用いる方法 + ap_uint<8> target = lfsr_random() % line_num; // i.e., lfsr_random_uint32(0, line_num - 1); +#else + // (2) 剰余演算を用いない方法 + ap_uint<8> target = lfsr_random() & (line_num_2 - 1); + if (line_num <= target) { + //cout << endl; + continue; + } +#endif + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[target] == true) { + //cout << endl; + continue; + } + + // 直前のイテレーション (ラウンド) と同じ対象ラインだったらルーティングスキップする + if (target == last_target) { + //cout << endl; + continue; + } + last_target = target; + + // (1) 引きはがすラインの重みをリセット + ROUTING_RESET: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[target]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + weights[paths[target][j]] = 1; + } + // 対象ラインのスタートの重みも一旦リセット あとで (*) で戻す + weights[starts[target]] = 1; + + // (2) 重みを更新 + ap_uint<8> current_round_weight = new_weight(round); + //cout << " weight " << current_round_weight << endl; + ROUTING_UPDATE: + for (ap_uint<11> i = 0; i < (ap_uint<11>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=1023 avg=50 + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[i] == false && i != target) { + ROUTING_UPDATE_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + weights[paths[i][j]] = current_round_weight; + } + } + } + + // 経路探索 +#ifdef DEBUG_PRINT + cout << "LINE #" << (int)(target + 1) << endl; +#endif + search(&(paths_size[target]), paths[target], starts[target], goals[target], weights); + + // (*) 対象ラインのスタートの重みを戻す + weights[starts[target]] = MAX_WEIGHT; + + // ルーティング後 + // オーバーラップのチェック + has_overlap = false; + OVERLAP_RESET: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { +#pragma HLS UNROLL factor=16 + overlap_checks[i] = 0; + } + OVERLAP_CHECK: + for (ap_uint<11> i = 0; i < (ap_uint<11>)(line_num); i++) { +#pragma HLS LOOP_FLATTEN off +#pragma HLS LOOP_TRIPCOUNT min=2 max=1023 avg=50 + overlap_checks[starts[i]] = 1; + overlap_checks[goals[i]] = 1; + + // 数字が隣接する場合スキップ、そうでない場合は実行 + //if (adjacents[i] == false) { + + OVERLAP_CHECK_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 +//#pragma HLS PIPELINE rewind II=33 +#pragma HLS PIPELINE II=17 +#pragma HLS UNROLL factor=8 + ap_uint<16> cell_id = paths[i][j]; + if (overlap_checks[cell_id] == 1) { + has_overlap = true; + break; + } + overlap_checks[cell_id] = 1; + } + //} + } + // オーバーラップなければ探索終了 + if (has_overlap == false) { + break; + } + + } + + // 解導出できなかった場合 + if (has_overlap == true) { + *status = 1; + return false; + } + + // ================================ + // ルーティング END + // ================================ + + // ================================ + // 解生成 BEGIN + // ================================ + + // 空白 + OUTPUT_INIT: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { + boardstr[i] = 0; + } + // ライン + // このソルバでのラインIDを+1して表示する + // なぜなら空白を 0 で表すことにするからラインIDは 1 以上にしたい + OUTPUT_LINE: + for (ap_uint<11> i = 0; i < (ap_uint<11>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=1023 avg=50 + boardstr[starts[i]] = (i + 1); + boardstr[goals[i]] = (i + 1); + OUTPUT_LINE_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + boardstr[paths[i][j]] = (i + 1); + } + } + + // ================================ + // 解生成 END + // ================================ + + *status = 0; + return true; +} + + +// ================================ // +// 探索 +// ================================ // + +#ifdef USE_ASTAR +// A* ヒューリスティック用 +// 最大71 最小0 +ap_uint<7> abs_uint7(ap_uint<7> a, ap_uint<7> b) { +#pragma HLS INLINE + if (a < b) { return b - a; } + else { return a - b; } +} +// 最大7 最小0 +ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b) { +#pragma HLS INLINE + if (a < b) { return b - a; } + else { return a - b; } +} +#endif + +// * Pythonでダイクストラアルゴリズムを実装した - フツーって言うなぁ！ +// http://lethe2211.hatenablog.com/entry/2014/12/30/011030 +// * Implementation of A* +// http://www.redblobgames.com/pathfinding/a-star/implementation.html +// をベース +void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start, ap_uint<16> goal, ap_uint<8> w[MAX_CELLS]) { +//#pragma HLS INLINE // search関数はインラインすると遅くなるしBRAM足りなくなる +//#pragma HLS FUNCTION_INSTANTIATE variable=start +//#pragma HLS FUNCTION_INSTANTIATE variable=goal + + ap_uint<16> dist[MAX_CELLS]; // 始点から各頂点までの最短距離を格納する +#pragma HLS ARRAY_PARTITION variable=dist cyclic factor=64 dim=1 partition +// Note: dist のパーティションの factor は 128 にするとBRAMが足りなくなる + ap_uint<16> prev[MAX_CELLS]; // 最短経路における，その頂点の前の頂点のIDを格納する + + SEARCH_INIT_DIST: + for (ap_uint<16> i = 0; i < MAX_CELLS; i++) { +#pragma HLS UNROLL factor=64 + dist[i] = 65535; // = (2^16 - 1) + } + + // プライオリティ・キュー + ap_uint<12> pq_len = 0; + ap_uint<32> pq_nodes[MAX_PQ]; +//#pragma HLS ARRAY_PARTITION variable=pq_nodes complete dim=1 +//#pragma HLS ARRAY_PARTITION variable=pq_nodes cyclic factor=2 dim=1 partition + +#ifdef DEBUG_PRINT + // キューの最大長さチェック用 + ap_uint<12> max_pq_len = 0; +#endif + +#ifdef USE_ASTAR + // ゴールの座標 + ap_uint<13> goal_xy = (ap_uint<13>)(goal >> BITWIDTH_Z); + ap_uint<7> goal_x = (ap_uint<7>)(goal_xy / MAX_WIDTH); + ap_uint<7> goal_y = (ap_uint<7>)(goal_xy - goal_x * MAX_WIDTH); + ap_uint<3> goal_z = (ap_uint<3>)(goal & BITMASK_Z); +#endif + + dist[start] = 0; + pq_push(0, start, &pq_len, pq_nodes); // 始点をpush +#ifdef DEBUG_PRINT + if (max_pq_len < pq_len) { max_pq_len = pq_len; } +#endif + + SEARCH_PQ: + while (0 < pq_len) { +#pragma HLS LOOP_FLATTEN off +#pragma HLS LOOP_TRIPCOUNT min=1 max=1000 avg=100 + + ap_uint<16> prov_cost; + ap_uint<16> src; + pq_pop(&prov_cost, &src, &pq_len, pq_nodes); +#ifdef DEBUG_PRINT +//ap_uint<13> _src_xy = (ap_uint<13>)(src >> BITWIDTH_Z); +//ap_uint<7> _src_x = (ap_uint<7>)(_src_xy / MAX_WIDTH); +//ap_uint<7> _src_y = (ap_uint<7>)(_src_xy % MAX_WIDTH); +//ap_uint<3> _src_z = (ap_uint<3>)(src & BITMASK_Z); +//cout << "Picked up " << (int)src << " (" << (int)_src_x << "," << (int)_src_y << "," << (int)_src_z << ") prov_cost=" << (int)prov_cost << " this_cost=" << w[src] << endl; +#endif + ap_uint<16> dist_src = dist[src]; + +#ifndef USE_ASTAR + // プライオリティキューに格納されている最短距離が，現在計算できている最短距離より大きければ，distの更新をする必要はない + if (dist_src < prov_cost) { + continue; + } +#endif + + // PQの先頭がゴールの場合にPQがまだ空じゃなくても探索終わらせしまう + if (src == goal) { + break; + } + + // 隣接する他の頂点の探索 + // (0) コスト + ap_uint<8> cost = w[src]; + // (1) ノードIDから3次元座標をマスクして抜き出す + ap_uint<13> src_xy = (ap_uint<13>)(src >> BITWIDTH_Z); + ap_uint<7> src_x = (ap_uint<7>)(src_xy / MAX_WIDTH); + ap_uint<7> src_y = (ap_uint<7>)(src_xy - src_x * MAX_WIDTH); + ap_uint<3> src_z = (ap_uint<3>)(src & BITMASK_Z); + //cout << src << " " << src_x << " " << src_y << " " << src_z << endl; + // (2) 3次元座標で隣接するノード (6個) を調べる // 手動ループ展開 +/*********************************************************** + if (src_x > 0) { // x-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x-1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x-1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_x < (size_x-1)) { // x+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x+1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x+1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_y > 0) { // y-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y-1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y-1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_y < (size_y-1)) { // y+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y+1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y+1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_z > 0) { // z-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z-1); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z-1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_z < (size_z-1)) { // y+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z+1); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z+1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } +***********************************************************/ + + SEARCH_ADJACENTS: + for (ap_uint<3> a = 0; a < 6; a++) { +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + ap_int<8> dest_x = (ap_int<8>)src_x; // 最小-1 最大72 (->符号付き8ビット) + ap_int<8> dest_y = (ap_int<8>)src_y; // 最小-1 最大72 (->符号付き8ビット) + ap_int<5> dest_z = (ap_int<5>)src_z; // 最小-1 最大8 (->符号付き5ビット) + if (a == 0) { dest_x -= 1; } + if (a == 1) { dest_x += 1; } + if (a == 2) { dest_y -= 1; } + if (a == 3) { dest_y += 1; } + if (a == 4) { dest_z -= 1; } + if (a == 5) { dest_z += 1; } + + if (0 <= dest_x && dest_x < (ap_int<8>)size_x && 0 <= dest_y && dest_y < (ap_int<8>)size_y && 0 <= dest_z && dest_z < (ap_int<5>)size_z) { + ap_uint<16> dest = (((ap_uint<16>)dest_x * MAX_WIDTH + (ap_uint<16>)dest_y) << BITWIDTH_Z) | (ap_uint<16>)dest_z; + ap_uint<16> dist_new = dist_src + cost; +#ifdef DEBUG_PRINT +//cout << " adjacent " << (int)dest << " (" << (int)dest_x << "," << (int)dest_y << "," << (int)dest_z << ") dist_new=" << (int)dist_new; +#endif + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 +#ifdef USE_ASTAR + dist_new += abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z); // A* ヒューリスティック +#endif + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush +#ifdef DEBUG_PRINT +//cout << " h=" << (int)(abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z)) << endl; +//cout << (int)dest_x << " " << (int)goal_x << " " << (int)abs_uint7(dest_x, goal_x) << endl; +//cout << (int)dest_y << " " << (int)goal_y << " " << (int)abs_uint7(dest_y, goal_y) << endl; +//cout << (int)dest_z << " " << (int)goal_z << " " << (int)abs_uint7(dest_z, goal_z) << endl; + if (max_pq_len < pq_len) { max_pq_len = pq_len; } +#endif + } +#ifdef DEBUG_PRINT +//else { cout << " -> skip pushing" << endl; } +#endif + } + } + } + + // 経路を出力 + // ゴールからスタートへの順番で表示される (ゴールとスタートは含まれない) + ap_uint<16> t = prev[goal]; + +#ifdef DEBUG_PRINT + int dbg_start_xy = start >> BITWIDTH_Z; + int dbg_start_x = dbg_start_xy / MAX_WIDTH; + int dbg_start_y = dbg_start_xy % MAX_WIDTH; + int dbg_start_z = start & BITMASK_Z; + + int dbg_goal_xy = goal >> BITWIDTH_Z; + int dbg_goal_x = dbg_goal_xy / MAX_WIDTH; + int dbg_goal_y = dbg_goal_xy % MAX_WIDTH; + int dbg_goal_z = goal & BITMASK_Z; + + cout << "(" << dbg_start_x << ", " << dbg_start_y << ", " << dbg_start_z << ") #" << start << " -> " + << "(" << dbg_goal_x << ", " << dbg_goal_y << ", " << dbg_goal_z << ") #" << goal << endl; +#endif + + // バックトラック + ap_uint<8> p = 0; + SEARCH_BACKTRACK: + while (t != start) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 +#pragma HLS PIPELINE II=2 + +#ifdef DEBUG_PRINT + int t_xy = prev[t] >> BITWIDTH_Z; + int t_x = t_xy / MAX_WIDTH; + int t_y = t_xy % MAX_WIDTH; + int t_z = prev[t] & BITMASK_Z; + cout << " via " << "(" << t_x << ", " << t_y << ", " << t_z << ") #" << prev[t] << " dist=" << dist[t] << endl; +#endif + + path[p] = t; // 記録 + p++; + + t = prev[t]; // 次に移動 + } + *path_size = p; + +#ifdef DEBUG_PRINT + cout << "max_path_len = " << p << endl; + cout << "max_pq_len = " << max_pq_len << endl; +#endif + +} + +// プライオリティ・キュー (ヒープで実装) +// 優先度の最小値がヒープのルートに来る +// 参考 +// * ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E3%83%92%E3%83%BC%E3%83%97 +// * 二分ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E4%BA%8C%E5%88%86%E3%83%92%E3%83%BC%E3%83%97 +// * ヒープの正体 - http://www.maroontress.com/Heap/ +// * Priority queue - Rosetta Code https://rosettacode.org/wiki/Priority_queue#C +// Note +// インデックスが0から始まるとき (0-origin index) +// --> 親: (n-1)/2, 左の子: 2n+1, 右の子: 2n+2 +// インデックスが1から始まるとき (1-origin index) +// --> 親: n/2, 左の子: 2n, 右の子: 2n+1 +// FPGA的にはどちらも遅延は同じだけど 1-origin の方がLUTリソース少なくて済む (ただし配列の0要素が無駄になる) + +// ノードの挿入は，末尾に追加してから優先度が正しい高さの位置までノードを上げていく +// 探索の都合上，同じ優先度では後から入れた方を先に出したいから， +// ループの終了条件は挿入ノードの優先度が比較対象の優先度よりも小さくなったとき +void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { +#pragma HLS INLINE + + (*pq_len)++; + ap_uint<12> i = (*pq_len); // target + ap_uint<12> p = (*pq_len) >> 1; // i.e., (*pq_len) / 2; // 親 + PQ_PUSH_LOOP: + while (i > 1 && (ap_uint<16>)(pq_nodes[p] & PQ_PRIORITY_MASK) >= priority) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + pq_nodes[i] = pq_nodes[p]; + i = p; + p = p >> 1; // i.e., p / 2; // 親 + } + pq_nodes[i] = ((ap_uint<32>)data << 16) | (ap_uint<32>)priority; +} + +// ノードの取り出しは，ルートを取ってくるだけ +// 次に最小の優先度をもつノードをルートに移動させるために， +// まず，末尾のノードをルートに移動する +// 両方の子で優先度が小さい方を上にもっていく (ルートを適切な高さまで下げる) +// これを再帰的に繰り返す +void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { +#pragma HLS INLINE + + *ret_priority = (ap_uint<16>)(pq_nodes[1] & PQ_PRIORITY_MASK); + *ret_data = (ap_uint<16>)(pq_nodes[1] >> PQ_PRIORITY_WIDTH); + + //pq_nodes[1] = pq_nodes[*pq_len]; + ap_uint<12> i = 1; // 親ノード + //ap_uint<12> t = 1; // 交換対象ノード + + ap_uint<16> last_priority = (ap_uint<16>)(pq_nodes[*pq_len] & PQ_PRIORITY_MASK); // 末尾ノードの優先度 + + PQ_POP_LOOP: + while (1) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + ap_uint<12> c1 = i << 1; // i.e., 2 * i; // 左の子 + ap_uint<12> c2 = c1 + 1; // i.e., 2 * i + 1; // 右の子 + if (c1 < *pq_len && (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK) <= last_priority) { + if (c2 < *pq_len && (ap_uint<16>)(pq_nodes[c2] & PQ_PRIORITY_MASK) <= (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK)) { + pq_nodes[i] = pq_nodes[c2]; + i = c2; + } + else { + pq_nodes[i] = pq_nodes[c1]; + i = c1; + } + } + else { + if (c2 < *pq_len && (ap_uint<16>)(pq_nodes[c2] & PQ_PRIORITY_MASK) <= last_priority) { + pq_nodes[i] = pq_nodes[c2]; + i = c2; + } + else { + break; + } + } + } + pq_nodes[i] = pq_nodes[*pq_len]; + (*pq_len)--; + +// For verification + //for (int k = 1;k<(*pq_len);k++){ + // cout << (ap_uint<16>)(pq_nodes[k] & PQ_PRIORITY_MASK) << " "; + //} + //cout << endl; +} diff --git a/hls/ZU3EG_test/router.hpp b/hls/ZU3EG_test/router.hpp new file mode 100644 index 0000000000000000000000000000000000000000..c6d7515bb2729be1cd464b8554b5f6a3372f6a4e --- /dev/null +++ b/hls/ZU3EG_test/router.hpp @@ -0,0 +1,62 @@ +/** + * router.hpp + * + * for Vivado HLS + */ + +#ifndef __ROUTER_HPP__ +#define __ROUTER_HPP__ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +//#define DEBUG_PRINT // いろいろ表示する +#define USE_ASTAR // A* 探索を使う +#define USE_MOD_CALC // ターゲットラインの選択に剰余演算を用いる + +using namespace std; + + +// 各種設定値 +#define MAX_WIDTH 72 // X, Y の最大値 (7ビットで収まる) +#define BITWIDTH_XY 13 +#define BITMASK_XY 65528 // 1111 1111 1111 1000 +#define MAX_LAYER 8 // Z の最大値 (3ビット) +#define BITWIDTH_Z 3 +#define BITMASK_Z 7 // 0000 0000 0000 0111 + +#define MAX_CELLS 41472 // セルの総数 =72*72*8 (16ビットで収まる) +#define MAX_LINES 1024 // ライン数の最大値 (7ビット) +#define MAX_PATH 256 // 1つのラインが対応するセル数の最大値 (8ビット) +#define MAX_PQ 4096 // 探索時のプライオリティ・キューの最大サイズ (12ビット) 足りないかも？ + +#define PQ_PRIORITY_WIDTH 16 +#define PQ_PRIORITY_MASK 65535 // 0000 0000 0000 0000 1111 1111 1111 1111 +#define PQ_DATA_WIDTH 16 +#define PQ_DATA_MASK 4294901760 // 1111 1111 1111 1111 0000 0000 0000 0000 + +#define MAX_WEIGHT 255 // 重みの最大値 (8ビットで収まる) +#define BOARDSTR_SIZE 41472 // ボードストリングの最大文字数 (セル数 72*72*8 あれば良い) + + +void lfsr_random_init(ap_uint<32> seed); +ap_uint<32> lfsr_random(); +//ap_uint<32> lfsr_random_uint32(ap_uint<32> a, ap_uint<32> b); +//ap_uint<32> lfsr_random_uint32_0(ap_uint<32> b); + +ap_uint<8> new_weight(ap_uint<16> x); +bool pynqrouter_zu3eg_test(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *status); + +#ifdef USE_ASTAR +ap_uint<7> abs_uint7(ap_uint<7> a, ap_uint<7> b); +ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b); +#endif + +void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start, ap_uint<16> goal, ap_uint<8> w[MAX_WEIGHT]); +void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]); +void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]); + +#endif /* __ROUTER_HPP__ */ diff --git a/hls/lines128_length256/main.cpp b/hls/lines128_length256/main.cpp index e7089f20710bbb0abcb0d0e130a6b2ea8bbc27e4..41f1f511e78858fb1d4090e24c4957b81b416060 100644 --- a/hls/lines128_length256/main.cpp +++ b/hls/lines128_length256/main.cpp @@ -21,7 +21,7 @@ int main(int argc, char *argv[]) { using namespace std; - // eXgf[^ (`) + // テストデータ (文字列形式) // NL_Q00.txt //char boardstr[BOARDSTR_SIZE] = "X10Y05Z3L0000107041L0004107002L0102102021L0900100003"; // NL_Q06.txt @@ -29,9 +29,9 @@ int main(int argc, char *argv[]) { // NL_Q08.txt //char boardstr[BOARDSTR_SIZE] = "X17Y20Z2L0000103022L1603115052L0916107032L0302108012L1104111042L1002100002L0919116162L1616113182L1001115012L0500201182L1603213152L0600210022"; - // w肳Ă΃R}hC蕶ǂݍ + // 指定されてればコマンドラインから問題文字列を読み込む if (1 < argc) { - //擪Xł͂ȂȂΕW͂ǂݍ + //先頭がXではないならば標準入力から読み込む if(argv[1][0]!='X') { char* c_p=fgets(boardstr, BOARDSTR_SIZE, stdin); @@ -44,7 +44,7 @@ int main(int argc, char *argv[]) { } } - // w肳Ă΃V[hlǂݍ + // 指定されてればシード値を読み込む int seed = 12345; if (2 < argc) { seed = atoi(argv[2]); @@ -54,7 +54,7 @@ int main(int argc, char *argv[]) { int size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); int size_z = (boardstr[7] - '0'); - // \os + // ソルバ実行 ap_int<32> status; clock_t clock_start, clock_done; clock_start = clock(); @@ -68,7 +68,7 @@ int main(int argc, char *argv[]) { cout << "status = " << (int)status << endl; cout << "elapsed = " << ((double)(clock_done - clock_start) / CLOCKS_PER_SEC) << endl << endl; - // \ + // 解表示 cout << "SOLUTION" << endl; cout << "========" << endl; cout << "SIZE " << size_x << "X" << size_y << "X" << size_z << endl; diff --git a/hls/lines128_length256/router.cpp b/hls/lines128_length256/router.cpp index 83d45f9440c0eb66effe738a6ccbe658dd384cca..a5a64cf4899690c627feeed57a302273b318b892 100644 --- a/hls/lines128_length256/router.cpp +++ b/hls/lines128_length256/router.cpp @@ -17,7 +17,7 @@ // LFST // ================================ // -// Ql https://highlevel-synthesis.com/2017/02/10/lfsr-in-hls/ +// 参考 https://highlevel-synthesis.com/2017/02/10/lfsr-in-hls/ static ap_uint<32> lfsr; void lfsr_random_init(ap_uint<32> seed) { @@ -38,15 +38,15 @@ ap_uint<32> lfsr_random() { return lfsr.to_uint(); } -// AB͈̔ (AB܂) ̗̐~Ƃ -// Ql http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +// AからBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html /*ap_uint<32> lfsr_random_uint32(ap_uint<32> a, ap_uint<32> b) { #pragma HLS INLINE return lfsr_random() % (b - a + 1) + a; }*/ -// 0B͈̔ (AB܂) ̗̐~Ƃ -// Ql http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +// 0からBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html /*ap_uint<32> lfsr_random_uint32_0(ap_uint<32> b) { #pragma HLS INLINE return lfsr_random() % (b + 1); @@ -54,21 +54,21 @@ ap_uint<32> lfsr_random() { // ================================ // -// CW[ +// メインモジュール // ================================ // -// d݂̍XV -// TODO -// min_uint8(r, MAX_WEIGHT) Ɠ +// 重みの更新 +// TODO 調整 +// min_uint8(r, MAX_WEIGHT) と同じ ap_uint<8> new_weight(ap_uint<16> x) { #pragma HLS INLINE #if 1 - // 8rbg (ő 255) 𔲂oāA1/8 čő 31 (32) ɂ + // 下位8ビット (最大 255) を抜き出して、1/8 をかけて最大 31 (32) にする ap_uint<8> y = x & 255; return (ap_uint<8>)(y / 8 + 1); #endif #if 0 - // 10rbg (ő 1023) 𔲂oāA1/32 čő 31 (32) ɂ + // 下位10ビット (最大 1023) を抜き出して、1/32 をかけて最大 31 (32) にする ap_uint<10> y = x & 1023; return (ap_uint<8>)(y / 32 + 1); #endif @@ -79,23 +79,23 @@ ap_uint<8> new_weight(ap_uint<16> x) { #endif } -// {[hɊւϐ -static ap_uint<7> size_x; // {[h X TCY -static ap_uint<7> size_y; // {[h Y TCY -static ap_uint<4> size_z; // {[h Z TCY +// ボードに関する変数 +static ap_uint<7> size_x; // ボードの X サイズ +static ap_uint<7> size_y; // ボードの Y サイズ +static ap_uint<4> size_z; // ボードの Z サイズ -static ap_uint<7> line_num = 0; // C̑ +static ap_uint<7> line_num = 0; // ラインの総数 -// O[oϐŒ` +// グローバル変数で定義する #ifdef GLOBALVARS - ap_uint<16> starts[MAX_LINES]; // C̃X^[gXg - ap_uint<16> goals[MAX_LINES]; // C̃S[Xg + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト - ap_uint<8> weights[MAX_CELLS]; // Z̏d + ap_uint<8> weights[MAX_CELLS]; // セルの重み - ap_uint<8> paths_size[MAX_LINES]; // CΉZID̃TCY - ap_uint<16> paths[MAX_LINES][MAX_PATH]; // CΉZID̏W (X^[gƃS[͏) - bool adjacents[MAX_LINES]; // X^[gƃS[אڂĂ郉C + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン #endif bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *status) { @@ -106,30 +106,30 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat *status = -1; -// O[oϐł͒`Ȃ +// グローバル変数では定義しない #ifndef GLOBALVARS - ap_uint<16> starts[MAX_LINES]; // C̃X^[gXg + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト #pragma HLS ARRAY_PARTITION variable=starts complete dim=1 - ap_uint<16> goals[MAX_LINES]; // C̃S[Xg + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト #pragma HLS ARRAY_PARTITION variable=goals complete dim=1 - ap_uint<8> weights[MAX_CELLS]; // Z̏d + ap_uint<8> weights[MAX_CELLS]; // セルの重み //#pragma HLS ARRAY_PARTITION variable=weights cyclic factor=8 dim=1 partition -// Note: weights ͗lXȏԂŃANZX邩p[eBVĂSRʂȂ +// Note: weights は様々な順番でアクセスされるからパーティションしても全然効果ない - ap_uint<8> paths_size[MAX_LINES]; // CΉZID̃TCY + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ //#pragma HLS ARRAY_PARTITION variable=paths_size complete dim=1 - ap_uint<16> paths[MAX_LINES][MAX_PATH]; // CΉZID̏W (X^[gƃS[͏) + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) //#pragma HLS ARRAY_PARTITION variable=paths cyclic factor=16 dim=2 partition - bool adjacents[MAX_LINES]; // X^[gƃS[אڂĂ郉C + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン //#pragma HLS ARRAY_PARTITION variable=adjacents complete dim=1 #endif // ================================ - // BEGIN + // 初期化 BEGIN // ================================ - // [vJE^1rbg]ɗpӂȂƏIłȂ + // ループカウンタは1ビット余分に用意しないと終了判定できない INIT_ADJACENTS: for (ap_uint<8> i = 0; i < (ap_uint<8>)(MAX_LINES); i++) { adjacents[i] = false; @@ -142,7 +142,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat weights[i] = 1; } - // {[hXgỎ + // ボードストリングの解釈 size_x = (boardstr[1] - '0') * 10 + (boardstr[2] - '0'); size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); @@ -152,7 +152,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat for (ap_uint<16> idx = 8; idx < BOARDSTR_SIZE; idx+=11) { //#pragma HLS LOOP_TRIPCOUNT min=100 max=32768 avg=1000 - // I[ (null) + // 終端 (null) 文字 if (boardstr[idx] == 0) { break; } @@ -166,16 +166,16 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat //cout << "L " << line_num << ": (" << s_x << ", " << s_y << ", " << s_z << ") " // "(" << g_x << ", " << g_y << ", " << g_z << ")" << endl; - // X^[gƃS[ + // スタートとゴール ap_uint<16> start_id = (((ap_uint<16>)s_x * MAX_WIDTH + (ap_uint<16>)s_y) << BITWIDTH_Z) | (ap_uint<16>)s_z; ap_uint<16> goal_id = (((ap_uint<16>)g_x * MAX_WIDTH + (ap_uint<16>)g_y) << BITWIDTH_Z) | (ap_uint<16>)g_z; starts[line_num] = start_id; goals[line_num] = goal_id; - // ԂŐאڂĂ邩f - ap_int<8> dx = (ap_int<8>)g_x - (ap_int<8>)s_x; // ŏ-71 ő71 (-> t8rbg) - ap_int<8> dy = (ap_int<8>)g_y - (ap_int<8>)s_y; // ŏ-71 ő71 (-> t8rbg) - ap_int<4> dz = (ap_int<4>)g_z - (ap_int<4>)s_z; // ŏ-7 ő7 (-> t4rbg) + // 初期状態で数字が隣接しているか判断 + ap_int<8> dx = (ap_int<8>)g_x - (ap_int<8>)s_x; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<8> dy = (ap_int<8>)g_y - (ap_int<8>)s_y; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<4> dz = (ap_int<4>)g_z - (ap_int<4>)s_z; // 最小-7 最大7 (-> 符号付き4ビット) if ((dx == 0 && dy == 0 && (dz == 1 || dz == -1)) || (dx == 0 && (dy == 1 || dy == -1) && dz == 0) || ((dx == 1 || dx == -1) && dy == 0 && dz == 0)) { adjacents[line_num] = true; paths_size[line_num] = 0; @@ -191,21 +191,21 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat } //cout << size_x << " " << size_y << " " << size_z << endl; - // ̏ + // 乱数の初期化 lfsr_random_init(seed); // TODO - // ׂẴCאڂĂ\oIɂ + // すべてのラインが隣接してたらソルバを終わりにする // ================================ - // END + // 初期化 END // ================================ // ================================ - // [eBO BEGIN + // ルーティング BEGIN // ================================ - // [Step 1] [eBO + // [Step 1] 初期ルーティング cout << "Initial Routing" << endl; FIRST_ROUTING: for (ap_uint<8> i = 0; i < (ap_uint<8>)(line_num); i++) { @@ -213,10 +213,10 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat //#pragma HLS PIPELINE //#pragma HLS UNROLL factor=2 - // אڂꍇXLbvAłȂꍇ͎s + // 数字が隣接する場合スキップ、そうでない場合は実行 if (adjacents[i] == false) { - // oHT + // 経路探索 #ifdef DEBUG_PRINT cout << "LINE #" << (int)(i + 1) << endl; #endif @@ -230,7 +230,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat bool has_overlap = false; #ifndef USE_MOD_CALC - // line_num_2: line_num ȏōŏ2ׂ̂搔 + // line_num_2: line_num 以上で最小の2のべき乗数 ap_uint<8> line_num_2; CALC_LINE_NUM_2: for (line_num_2 = 1; line_num_2 < (ap_uint<8>)line_num; line_num_2 *= 2) { @@ -243,7 +243,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat ap_uint<8> last_target = 255; - // [Step 2] Rip-up ă[eBO + // [Step 2] Rip-up 再ルーティング ROUTING: for (ap_uint<16> round = 1; round <= 32768 /* = (2048 * 16) */; round++) { #pragma HLS LOOP_TRIPCOUNT min=1 max=4000 avg=50 @@ -252,12 +252,12 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat //cout << "ITERATION " << round; //#endif - // ΏۃCI + // 対象ラインを選択 #ifdef USE_MOD_CALC - // (1) ]Zp@ + // (1) 剰余演算を用いる方法 ap_uint<8> target = lfsr_random() % line_num; // i.e., lfsr_random_uint32(0, line_num - 1); #else - // (2) ]ZpȂ@ + // (2) 剰余演算を用いない方法 ap_uint<8> target = lfsr_random() & (line_num_2 - 1); if (line_num <= target) { //cout << endl; @@ -265,36 +265,36 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat } #endif - // אڂꍇXLbvAłȂꍇ͎s + // 数字が隣接する場合スキップ、そうでない場合は実行 if (adjacents[target] == true) { //cout << endl; continue; } - // ÕCe[V (Eh) ƓΏۃC烋[eBOXLbv + // 直前のイテレーション (ラウンド) と同じ対象ラインだったらルーティングスキップする if (target == last_target) { //cout << endl; continue; } last_target = target; - // (1) ͂C̏d݂Zbg + // (1) 引きはがすラインの重みをリセット ROUTING_RESET: for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[target]); j++) { #pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 weights[paths[target][j]] = 1; } - // ΏۃC̃X^[g̏d݂UZbg Ƃ (*) Ŗ߂ + // 対象ラインのスタートの重みも一旦リセット あとで (*) で戻す weights[starts[target]] = 1; - // (2) d݂XV + // (2) 重みを更新 ap_uint<8> current_round_weight = new_weight(round); //cout << " weight " << current_round_weight << endl; ROUTING_UPDATE: for (ap_uint<8> i = 0; i < (ap_uint<8>)(line_num); i++) { #pragma HLS LOOP_TRIPCOUNT min=2 max=127 avg=50 - // אڂꍇXLbvAłȂꍇ͎s + // 数字が隣接する場合スキップ、そうでない場合は実行 if (adjacents[i] == false && i != target) { ROUTING_UPDATE_PATH: for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { @@ -304,17 +304,17 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat } } - // oHT + // 経路探索 #ifdef DEBUG_PRINT cout << "LINE #" << (int)(target + 1) << endl; #endif search(&(paths_size[target]), paths[target], starts[target], goals[target], weights); - // (*) ΏۃC̃X^[g̏d݂߂ + // (*) 対象ラインのスタートの重みを戻す weights[starts[target]] = MAX_WEIGHT; - // [eBO - // I[o[bṽ`FbN + // ルーティング後 + // オーバーラップのチェック has_overlap = false; OVERLAP_RESET: for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { @@ -328,7 +328,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat overlap_checks[starts[i]] = 1; overlap_checks[goals[i]] = 1; - // אڂꍇXLbvAłȂꍇ͎s + // 数字が隣接する場合スキップ、そうでない場合は実行 //if (adjacents[i] == false) { OVERLAP_CHECK_PATH: @@ -346,35 +346,35 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat } //} } - // I[o[bvȂΒTI + // オーバーラップなければ探索終了 if (has_overlap == false) { break; } } - // 𓱏ołȂꍇ + // 解導出できなかった場合 if (has_overlap == true) { *status = 1; return false; } // ================================ - // [eBO END + // ルーティング END // ================================ // ================================ - // 𐶐 BEGIN + // 解生成 BEGIN // ================================ - // + // 空白 OUTPUT_INIT: for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { boardstr[i] = 0; } - // C - // ̃\oł̃CID+1ĕ\ - // ȂȂ󔒂 0 ŕ\Ƃɂ邩烉CID 1 ȏɂ + // ライン + // このソルバでのラインIDを+1して表示する + // なぜなら空白を 0 で表すことにするからラインIDは 1 以上にしたい OUTPUT_LINE: for (ap_uint<8> i = 0; i < (ap_uint<8>)(line_num); i++) { #pragma HLS LOOP_TRIPCOUNT min=2 max=127 avg=50 @@ -388,7 +388,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat } // ================================ - // 𐶐 END + // 解生成 END // ================================ *status = 0; @@ -397,18 +397,18 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat // ================================ // -// T +// 探索 // ================================ // #ifdef USE_ASTAR -// A* q[XeBbNp -// ő71 ŏ0 +// A* ヒューリスティック用 +// 最大71 最小0 ap_uint<7> abs_uint7(ap_uint<7> a, ap_uint<7> b) { #pragma HLS INLINE if (a < b) { return b - a; } else { return a - b; } } -// ő7 ŏ0 +// 最大7 最小0 ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b) { #pragma HLS INLINE if (a < b) { return b - a; } @@ -416,20 +416,20 @@ ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b) { } #endif -// * PythonŃ_CNXgASY - tc[ČȂI +// * Pythonでダイクストラアルゴリズムを実装した - フツーって言うなぁ！ // http://lethe2211.hatenablog.com/entry/2014/12/30/011030 // * Implementation of A* // http://www.redblobgames.com/pathfinding/a-star/implementation.html -// x[X +// をベース void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start, ap_uint<16> goal, ap_uint<8> w[MAX_CELLS]) { -//#pragma HLS INLINE // search֐̓CCƒxȂ邵BRAMȂȂ +//#pragma HLS INLINE // search関数はインラインすると遅くなるしBRAM足りなくなる //#pragma HLS FUNCTION_INSTANTIATE variable=start //#pragma HLS FUNCTION_INSTANTIATE variable=goal - ap_uint<16> dist[MAX_CELLS]; // n_e_܂ł̍ŒZi[ + ap_uint<16> dist[MAX_CELLS]; // 始点から各頂点までの最短距離を格納する #pragma HLS ARRAY_PARTITION variable=dist cyclic factor=64 dim=1 partition -// Note: dist ̃p[eBV factor 128 ɂBRAMȂȂ - ap_uint<16> prev[MAX_CELLS]; // ŒZoHɂC̒_̑O̒_IDi[ +// Note: dist のパーティションの factor は 128 にするとBRAMが足りなくなる + ap_uint<16> prev[MAX_CELLS]; // 最短経路における，その頂点の前の頂点のIDを格納する SEARCH_INIT_DIST: for (ap_uint<16> i = 0; i < MAX_CELLS; i++) { @@ -437,19 +437,19 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start dist[i] = 65535; // = (2^16 - 1) } - // vCIeBEL[ + // プライオリティ・キュー ap_uint<12> pq_len = 0; ap_uint<32> pq_nodes[MAX_PQ]; //#pragma HLS ARRAY_PARTITION variable=pq_nodes complete dim=1 //#pragma HLS ARRAY_PARTITION variable=pq_nodes cyclic factor=2 dim=1 partition #ifdef DEBUG_PRINT - // L[̍ő咷`FbNp + // キューの最大長さチェック用 ap_uint<12> max_pq_len = 0; #endif #ifdef USE_ASTAR - // S[̍W + // ゴールの座標 ap_uint<13> goal_xy = (ap_uint<13>)(goal >> BITWIDTH_Z); ap_uint<7> goal_x = (ap_uint<7>)(goal_xy / MAX_WIDTH); ap_uint<7> goal_y = (ap_uint<7>)(goal_xy - goal_x * MAX_WIDTH); @@ -457,7 +457,7 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start #endif dist[start] = 0; - pq_push(0, start, &pq_len, pq_nodes); // n_push + pq_push(0, start, &pq_len, pq_nodes); // 始点をpush #ifdef DEBUG_PRINT if (max_pq_len < pq_len) { max_pq_len = pq_len; } #endif @@ -480,86 +480,86 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start ap_uint<16> dist_src = dist[src]; #ifndef USE_ASTAR - // vCIeBL[Ɋi[ĂŒZC݌vZłĂŒZ傫΁Cdist̍XVKv͂Ȃ + // プライオリティキューに格納されている最短距離が，現在計算できている最短距離より大きければ，distの更新をする必要はない if (dist_src < prov_cost) { continue; } #endif - // PQ̐擪S[̏ꍇPQ܂󂶂ȂĂTI点܂ + // PQの先頭がゴールの場合にPQがまだ空じゃなくても探索終わらせしまう if (src == goal) { break; } - // אڂ鑼̒_̒T - // (0) RXg + // 隣接する他の頂点の探索 + // (0) コスト ap_uint<8> cost = w[src]; - // (1) m[hID3W}XNĔo + // (1) ノードIDから3次元座標をマスクして抜き出す ap_uint<13> src_xy = (ap_uint<13>)(src >> BITWIDTH_Z); ap_uint<7> src_x = (ap_uint<7>)(src_xy / MAX_WIDTH); ap_uint<7> src_y = (ap_uint<7>)(src_xy - src_x * MAX_WIDTH); ap_uint<3> src_z = (ap_uint<3>)(src & BITMASK_Z); //cout << src << " " << src_x << " " << src_y << " " << src_z << endl; - // (2) 3Wŗאڂm[h (6) 𒲂ׂ // 蓮[vWJ + // (2) 3次元座標で隣接するノード (6個) を調べる // 手動ループ展開 /*********************************************************** - if (src_x > 0) { // x-𒲍 + if (src_x > 0) { // x-を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x-1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x-1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x-1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } - if (src_x < (size_x-1)) { // x+𒲍 + if (src_x < (size_x-1)) { // x+を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x+1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x+1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x+1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } - if (src_y > 0) { // y-𒲍 + if (src_y > 0) { // y-を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y-1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y-1, goal_y) + abs_uint3(src_z, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y-1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } - if (src_y < (size_y-1)) { // y+𒲍 + if (src_y < (size_y-1)) { // y+を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y+1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y+1, goal_y) + abs_uint3(src_z, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y+1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } - if (src_z > 0) { // z-𒲍 + if (src_z > 0) { // z-を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z-1); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z-1, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z-1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } - if (src_z < (size_z-1)) { // y+𒲍 + if (src_z < (size_z-1)) { // y+を調査 ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z+1); ap_uint<16> dist_new = dist_src + cost; if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ - dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z+1, goal_z); // A* q[XeBbN - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z+1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush } } ***********************************************************/ @@ -568,9 +568,9 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start for (ap_uint<3> a = 0; a < 6; a++) { //#pragma HLS PIPELINE //#pragma HLS UNROLL factor=2 - ap_int<8> dest_x = (ap_int<8>)src_x; // ŏ-1 ő72 (->t8rbg) - ap_int<8> dest_y = (ap_int<8>)src_y; // ŏ-1 ő72 (->t8rbg) - ap_int<5> dest_z = (ap_int<5>)src_z; // ŏ-1 ő8 (->t5rbg) + ap_int<8> dest_x = (ap_int<8>)src_x; // 最小-1 最大72 (->符号付き8ビット) + ap_int<8> dest_y = (ap_int<8>)src_y; // 最小-1 最大72 (->符号付き8ビット) + ap_int<5> dest_z = (ap_int<5>)src_z; // 最小-1 最大8 (->符号付き5ビット) if (a == 0) { dest_x -= 1; } if (a == 1) { dest_x += 1; } if (a == 2) { dest_y -= 1; } @@ -585,12 +585,12 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start //cout << " adjacent " << (int)dest << " (" << (int)dest_x << "," << (int)dest_y << "," << (int)dest_z << ") dist_new=" << (int)dist_new; #endif if (dist[dest] > dist_new) { - dist[dest] = dist_new; // dist̍XV - prev[dest] = src; // O̒_L^ + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 #ifdef USE_ASTAR - dist_new += abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z); // A* q[XeBbN + dist_new += abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z); // A* ヒューリスティック #endif - pq_push(dist_new, dest, &pq_len, pq_nodes); // L[ɐVȉ̋̏push + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush #ifdef DEBUG_PRINT //cout << " h=" << (int)(abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z)) << endl; //cout << (int)dest_x << " " << (int)goal_x << " " << (int)abs_uint7(dest_x, goal_x) << endl; @@ -606,8 +606,8 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start } } - // oHo - // S[X^[gւ̏Ԃŕ\ (S[ƃX^[g͊܂܂Ȃ) + // 経路を出力 + // ゴールからスタートへの順番で表示される (ゴールとスタートは含まれない) ap_uint<16> t = prev[goal]; #ifdef DEBUG_PRINT @@ -625,7 +625,7 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start << "(" << dbg_goal_x << ", " << dbg_goal_y << ", " << dbg_goal_z << ") #" << goal << endl; #endif - // obNgbN + // バックトラック ap_uint<8> p = 0; SEARCH_BACKTRACK: while (t != start) { @@ -640,10 +640,10 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start cout << " via " << "(" << t_x << ", " << t_y << ", " << t_z << ") #" << prev[t] << " dist=" << dist[t] << endl; #endif - path[p] = t; // L^ + path[p] = t; // 記録 p++; - t = prev[t]; // Ɉړ + t = prev[t]; // 次に移動 } *path_size = p; @@ -654,29 +654,29 @@ void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start } -// vCIeBEL[ (q[vŎ) -// Dx̍ŏlq[ṽ[gɗ -// Ql -// * q[v - Wikipedia https://ja.wikipedia.org/wiki/%E3%83%92%E3%83%BC%E3%83%97 -// * 񕪃q[v - Wikipedia https://ja.wikipedia.org/wiki/%E4%BA%8C%E5%88%86%E3%83%92%E3%83%BC%E3%83%97 -// * q[v̐ - http://www.maroontress.com/Heap/ +// プライオリティ・キュー (ヒープで実装) +// 優先度の最小値がヒープのルートに来る +// 参考 +// * ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E3%83%92%E3%83%BC%E3%83%97 +// * 二分ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E4%BA%8C%E5%88%86%E3%83%92%E3%83%BC%E3%83%97 +// * ヒープの正体 - http://www.maroontress.com/Heap/ // * Priority queue - Rosetta Code https://rosettacode.org/wiki/Priority_queue#C // Note -// CfbNX0n܂Ƃ (0-origin index) -// --> e: (n-1)/2, ̎q: 2n+1, E̎q: 2n+2 -// CfbNX1n܂Ƃ (1-origin index) -// --> e: n/2, ̎q: 2n, E̎q: 2n+1 -// FPGAIɂ͂ǂx͓ 1-origin ̕LUT\[XȂčς (z0vfʂɂȂ) - -// m[h̑}́CɒljĂDẍʒu܂Ńm[hグĂ -// T̓sCDxł͌ォꂽɏoC -// [v̏I͑}m[h̗DxrΏۂ̗DxȂƂ +// インデックスが0から始まるとき (0-origin index) +// --> 親: (n-1)/2, 左の子: 2n+1, 右の子: 2n+2 +// インデックスが1から始まるとき (1-origin index) +// --> 親: n/2, 左の子: 2n, 右の子: 2n+1 +// FPGA的にはどちらも遅延は同じだけど 1-origin の方がLUTリソース少なくて済む (ただし配列の0要素が無駄になる) + +// ノードの挿入は，末尾に追加してから優先度が正しい高さの位置までノードを上げていく +// 探索の都合上，同じ優先度では後から入れた方を先に出したいから， +// ループの終了条件は挿入ノードの優先度が比較対象の優先度よりも小さくなったとき void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { #pragma HLS INLINE (*pq_len)++; ap_uint<12> i = (*pq_len); // target - ap_uint<12> p = (*pq_len) >> 1; // i.e., (*pq_len) / 2; // e + ap_uint<12> p = (*pq_len) >> 1; // i.e., (*pq_len) / 2; // 親 PQ_PUSH_LOOP: while (i > 1 && (ap_uint<16>)(pq_nodes[p] & PQ_PRIORITY_MASK) >= priority) { #pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 @@ -684,16 +684,16 @@ void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<12> *pq_len, ap_uin //#pragma HLS UNROLL factor=2 pq_nodes[i] = pq_nodes[p]; i = p; - p = p >> 1; // i.e., p / 2; // e + p = p >> 1; // i.e., p / 2; // 親 } pq_nodes[i] = ((ap_uint<32>)data << 16) | (ap_uint<32>)priority; } -// m[h̎óC[gĂ邾 -// ɍŏ̗Dxƒm[h[gɈړ邽߂ɁC -// ܂C̃m[h[gɈړ -// ̎qŗDxɂĂ ([gK؂ȍ܂ʼn) -// ċAIɌJԂ +// ノードの取り出しは，ルートを取ってくるだけ +// 次に最小の優先度をもつノードをルートに移動させるために， +// まず，末尾のノードをルートに移動する +// 両方の子で優先度が小さい方を上にもっていく (ルートを適切な高さまで下げる) +// これを再帰的に繰り返す void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<12> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { #pragma HLS INLINE @@ -701,18 +701,18 @@ void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<12> *pq_le *ret_data = (ap_uint<16>)(pq_nodes[1] >> PQ_PRIORITY_WIDTH); //pq_nodes[1] = pq_nodes[*pq_len]; - ap_uint<12> i = 1; // em[h - //ap_uint<12> t = 1; // Ώۃm[h + ap_uint<12> i = 1; // 親ノード + //ap_uint<12> t = 1; // 交換対象ノード - ap_uint<16> last_priority = (ap_uint<16>)(pq_nodes[*pq_len] & PQ_PRIORITY_MASK); // m[h̗Dx + ap_uint<16> last_priority = (ap_uint<16>)(pq_nodes[*pq_len] & PQ_PRIORITY_MASK); // 末尾ノードの優先度 PQ_POP_LOOP: while (1) { #pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 //#pragma HLS PIPELINE //#pragma HLS UNROLL factor=2 - ap_uint<12> c1 = i << 1; // i.e., 2 * i; // ̎q - ap_uint<12> c2 = c1 + 1; // i.e., 2 * i + 1; // E̎q + ap_uint<12> c1 = i << 1; // i.e., 2 * i; // 左の子 + ap_uint<12> c2 = c1 + 1; // i.e., 2 * i + 1; // 右の子 if (c1 < *pq_len && (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK) <= last_priority) { if (c2 < *pq_len && (ap_uint<16>)(pq_nodes[c2] & PQ_PRIORITY_MASK) <= (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK)) { pq_nodes[i] = pq_nodes[c2]; diff --git a/hls/lines128_length256/router.hpp b/hls/lines128_length256/router.hpp index b582749f02dfce0717fcfe7b67c9fc8892f5eb9f..2a8772275ebcddc7ea9e127a85d9797db2c5a784 100644 --- a/hls/lines128_length256/router.hpp +++ b/hls/lines128_length256/router.hpp @@ -13,33 +13,33 @@ #include #endif -//#define DEBUG_PRINT // 낢\ -#define USE_ASTAR // A* Tg -#define USE_MOD_CALC // ^[QbgC̑Iɏ]Zp +//#define DEBUG_PRINT // いろいろ表示する +#define USE_ASTAR // A* 探索を使う +#define USE_MOD_CALC // ターゲットラインの選択に剰余演算を用いる using namespace std; -// eݒl -#define MAX_WIDTH 72 // X, Y ̍ől (7rbgŎ܂) +// 各種設定値 +#define MAX_WIDTH 72 // X, Y の最大値 (7ビットで収まる) #define BITWIDTH_XY 13 #define BITMASK_XY 65528 // 1111 1111 1111 1000 -#define MAX_LAYER 8 // Z ̍ől (3rbg) +#define MAX_LAYER 8 // Z の最大値 (3ビット) #define BITWIDTH_Z 3 #define BITMASK_Z 7 // 0000 0000 0000 0111 -#define MAX_CELLS 41472 // Z̑ =72*72*8 (16rbgŎ܂) -#define MAX_LINES 128 // C̍ől (7rbg) -#define MAX_PATH 256 // 1‚̃CΉZ̍ől (8rbg) -#define MAX_PQ 4096 // T̃vCIeBEL[̍őTCY (12rbg) ȂH +#define MAX_CELLS 41472 // セルの総数 =72*72*8 (16ビットで収まる) +#define MAX_LINES 128 // ライン数の最大値 (7ビット) +#define MAX_PATH 256 // 1つのラインが対応するセル数の最大値 (8ビット) +#define MAX_PQ 4096 // 探索時のプライオリティ・キューの最大サイズ (12ビット) 足りないかも？ #define PQ_PRIORITY_WIDTH 16 #define PQ_PRIORITY_MASK 65535 // 0000 0000 0000 0000 1111 1111 1111 1111 #define PQ_DATA_WIDTH 16 #define PQ_DATA_MASK 4294901760 // 1111 1111 1111 1111 0000 0000 0000 0000 -#define MAX_WEIGHT 255 // d݂̍ől (8rbgŎ܂) -#define BOARDSTR_SIZE 41472 // {[hXgO̍ő啶 (Z 72*72*8 Ηǂ) +#define MAX_WEIGHT 255 // 重みの最大値 (8ビットで収まる) +#define BOARDSTR_SIZE 41472 // ボードストリングの最大文字数 (セル数 72*72*8 あれば良い) void lfsr_random_init(ap_uint<32> seed); diff --git a/hls/lines256_length128/Makefile b/hls/lines256_length128/Makefile new file mode 100644 index 0000000000000000000000000000000000000000..8f79e7a62ede483ae7dcda810f306a02f915dfbb --- /dev/null +++ b/hls/lines256_length128/Makefile @@ -0,0 +1,20 @@ +TARGET = sim +OBJS = $(CPPS:.cpp=.o) +CPPS = $(wildcard *.cpp) +CXX = g++ +CXXFLAGS = -O3 -Wall -Wno-unknown-pragmas -Wno-unused-label -DSOFTWARE -DCALCTIME + +all: $(TARGET) + +$(TARGET): $(OBJS) + $(CXX) -O3 -o $@ $(OBJS) + +run: + python3 ../NLGenerator.py -x 20 -y 20 -z 6 -l 100;\ + python3 ./gen_boardstr.py Q-20x20x5_100_10.txt |\ + ./$(TARGET) - + + +clean: + rm *.o + rm $(TARGET) diff --git a/hls/lines256_length128/README.md b/hls/lines256_length128/README.md new file mode 100644 index 0000000000000000000000000000000000000000..b397aac9536c2feec276240f838fada74b50dbb6 --- /dev/null +++ b/hls/lines256_length128/README.md @@ -0,0 +1,30 @@ +# pynq-router (sw) + +Vivado HLS 用 pynq-router + ++ LINE数 + - 128 ++ 長さ/Line + - 256 + +## メモ + +* Vivado HLS 2016.1 +* Vivado 2016.1 + +高位合成オプション: +* Part: xc7z020clg400-1 +* Clock Period: 10.0 ns +* Clock Uncertainty: 3.0 ns, 3.5 ns // 2017/08/04 + +論理合成・配置配線のオプション: +* Synthesis strategy: ~~Vivado Synthesis Defaults~~ Flow_PerfOptimized_high +* Implementation strategy: ~~Vivado Implementation Defaults~~ Performance_Explore + +最適化記録： + +## 入出力 + +入力は boardstr という問題ファイルをコンパクトに記述したようなワンラインの文字列。 + +出力は char 型の配列で、各要素に解答ファイルに相当する数値が入っている。 diff --git a/hls/lines256_length128/ap_int.h b/hls/lines256_length128/ap_int.h new file mode 100644 index 0000000000000000000000000000000000000000..b8d9fdc96ac3f7eb9c86cc55d0eac0a57bf670b5 --- /dev/null +++ b/hls/lines256_length128/ap_int.h @@ -0,0 +1,521 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_AP_SIM_H__ +#define __AESL_AP_SIM_H__ + +#ifndef __cplusplus +#error C++ is required to include this header file +#else + +#include "etc/ap_int_sim.h" +#include "etc/ap_fixed_sim.h" + +//Forward declaration +template class ap_fixed; +template class ap_ufixed; +template class ap_int; +template class ap_uint; + +//AP_INT +//-------------------------------------------------------- +template +class ap_int : public ap_private<_AP_W, true> { +#ifdef _MSC_VER +#pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + typedef ap_private<_AP_W, true> Base; + + //Constructor + INLINE ap_int(): Base() {} + template + INLINE ap_int(const volatile ap_int<_AP_W2> &op):Base((const ap_private<_AP_W2,true> &)(op)) {} + + template + INLINE ap_int(const ap_int<_AP_W2> &op):Base((const ap_private<_AP_W2,true> &)(op)) {} + + template + INLINE ap_int(const ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2,false> &)(op)) {} + + template + INLINE ap_int(const volatile ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2,false> &)(op)) {} + + template + INLINE ap_int(const ap_range_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_bit_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& ref):Base(ref) {} + + template + INLINE ap_int(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_int(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + template + INLINE ap_int(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_int(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_int(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op.to_ap_private()) {} + +#define CTOR(TYPE) \ + INLINE ap_int(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_int(const char* str, signed char rd):Base(str, rd) {} + //Assignment + //Another form of "write" + INLINE void operator = (const ap_int<_AP_W>& op2) volatile { + const_cast(this)->operator = (op2); + } + + INLINE void operator = (const volatile ap_int<_AP_W>& op2) volatile { + const_cast(this)->operator = (op2); + } + + INLINE ap_int<_AP_W>& operator = (const volatile ap_int<_AP_W>& op2) { + Base::operator = (const_cast& >(op2)); + return *this; + } + + INLINE ap_int<_AP_W>& operator = (const ap_int<_AP_W>& op2) { + Base::operator = ((const ap_private<_AP_W, true>&)op2); + return *this; + } +}; + +//AP_UINT +//--------------------------------------------------------------- +template +class ap_uint: public ap_private<_AP_W, false> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef ap_private<_AP_W, false> Base; + //Constructor + INLINE ap_uint(): Base() {} + INLINE ap_uint(const ap_uint<_AP_W>& op) :Base(dynamic_cast&>(op)) {} + INLINE ap_uint(const volatile ap_uint<_AP_W>& op):Base(dynamic_cast&>(op)){} + template + INLINE ap_uint(const volatile ap_uint<_AP_W2> &op):Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_uint(const ap_uint<_AP_W2> &op) : Base((const ap_private<_AP_W2, false>&)(op)){} + + template + INLINE ap_uint(const ap_int<_AP_W2> &op) : Base((const ap_private<_AP_W2, true>&)(op)) {} + + template + INLINE ap_uint(const volatile ap_int<_AP_W2> &op) : Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_uint(const ap_range_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_bit_ref<_AP_W2, _AP_S2>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& ref):Base(ref) {} + + template + INLINE ap_uint(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op.to_ap_private()) {} + + template + INLINE ap_uint(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, _AP_N2>& op) + :Base(op) {} + + template + INLINE ap_uint(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + template + INLINE ap_uint(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_uint(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2>& op):Base(op) {} + + template + INLINE ap_uint(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op.to_ap_private()) {} + +#define CTOR(TYPE) \ + INLINE ap_uint(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_uint(const char* str, signed char rd):Base(str, rd) {} + //Assignment + //Another form of "write" + INLINE void operator = (const ap_uint<_AP_W>& op2) volatile { + Base::operator = (op2); + } + + INLINE void operator = (const volatile ap_uint<_AP_W>& op2) volatile { + Base::operator = (op2); + } + + INLINE ap_uint<_AP_W>& operator = (const volatile ap_uint<_AP_W>& op2) { + Base::operator = (op2); + return *this; + } + + INLINE ap_uint<_AP_W>& operator = (const ap_uint<_AP_W>& op2) { + Base::operator = ((const ap_private<_AP_W, false>&)(op2)); + return *this; + } +}; + +#define ap_bigint ap_int +#define ap_biguint ap_uint + +//AP_FIXED +//--------------------------------------------------------------------- +template +class ap_fixed: public ap_fixed_base<_AP_W, _AP_I, true, _AP_Q, _AP_O, _AP_N> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef ap_fixed_base<_AP_W, _AP_I, true, _AP_Q, _AP_O, _AP_N> Base; + //Constructor + INLINE ap_fixed():Base() {} + + template + INLINE ap_fixed(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_fixed(const ap_uint<_AP_W2>& op):Base(ap_private<_AP_W2, false>(op)) {} + + template + INLINE ap_fixed(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, _AP_O2, + _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_fixed(const volatile ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_fixed(const volatile ap_uint<_AP_W2>& op):Base(op) {} + + template + INLINE ap_fixed(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op):Base(op) {} + + template + INLINE ap_fixed(const ap_bit_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_fixed(const ap_range_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_fixed(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op): + Base(op) {} + + template + INLINE ap_fixed(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_fixed(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + #define CTOR(TYPE) \ + INLINE ap_fixed(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_fixed(const char* str, signed char rd):Base(str, rd) {} + + //Assignment + INLINE ap_fixed& operator = (const ap_fixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } + + INLINE ap_fixed& operator = (const volatile ap_fixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } +}; + +//AP_ UFIXED +//--- ---------------------------------------------------------------- +template +class ap_ufixed : public ap_fixed_base<_AP_W, _AP_I, false, _AP_Q, _AP_O, _AP_N> { +#ifdef _MSC_VER +#pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + typedef ap_fixed_base<_AP_W, _AP_I, false, _AP_Q, _AP_O, _AP_N> Base; + + //Constructor + INLINE ap_ufixed():Base() {} + + template + INLINE ap_ufixed(const ap_fixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op) : Base(ap_fixed_base<_AP_W2, + _AP_I2, true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const ap_int<_AP_W2>& op): + Base((const ap_private<_AP_W2, true>&)(op)) {} + + template + INLINE ap_ufixed(const ap_uint<_AP_W2>& op): + Base((const ap_private<_AP_W2, false>&)(op)) {} + + template + INLINE ap_ufixed(const volatile ap_fixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op) : Base(ap_fixed_base<_AP_W2, + _AP_I2, true, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_ufixed<_AP_W2, _AP_I2, _AP_Q2, + _AP_O2, _AP_N2>& op): Base(ap_fixed_base<_AP_W2, _AP_I2, + false, _AP_Q2, _AP_O2, _AP_N2>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_int<_AP_W2>& op): + Base(ap_private<_AP_W2, true>(op)) {} + + template + INLINE ap_ufixed(const volatile ap_uint<_AP_W2>& op): + Base(ap_private<_AP_W2, false>(op)) {} + + template + INLINE ap_ufixed(const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2>& op):Base(op) {} + + template + INLINE ap_ufixed(const ap_bit_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_ufixed(const ap_range_ref<_AP_W2, _AP_S2>& op): + Base(op) {} + + template + INLINE ap_ufixed(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op): + Base(op) {} + + template + INLINE ap_ufixed(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_ufixed(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& op): Base(op) {} + + template + INLINE ap_ufixed(const ap_private<_AP_W2, _AP_S2>& op):Base(op) {} + + #define CTOR(TYPE) \ + INLINE ap_ufixed(TYPE v):Base(v) {} + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(int) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) + CTOR(const char*) + CTOR(const std::string&) +#undef CTOR + INLINE ap_ufixed(const char* str, signed char rd):Base(str, rd) {} + + //Assignment + INLINE ap_ufixed& operator = (const ap_ufixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::operator = (op); + return *this; + } + + INLINE ap_ufixed& operator = (const volatile ap_ufixed<_AP_W, _AP_I, + _AP_Q, _AP_O, _AP_N>& op) { + Base::V = const_cast(op); + return *this; + } +}; + +#if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_int<_AP_W> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_uint<_AP_W> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_fixed<_AP_W, _AP_I, _AP_Q, _AP_O, _AP_N >&op, const std::string &name) { + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} + +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_ufixed<_AP_W, _AP_I, _AP_Q, _AP_O, _AP_N >&op, const std::string &name) { + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} +#endif /* #if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) */ +#endif /* #ifndef __cplusplus */ +#endif /* #ifndef __AESL_AP_SIM_H__ */ \ No newline at end of file diff --git a/hls/lines256_length128/etc/ap_fixed_sim.h b/hls/lines256_length128/etc/ap_fixed_sim.h new file mode 100644 index 0000000000000000000000000000000000000000..5be571dd70e490d282d279a4eeed32db069703e9 --- /dev/null +++ b/hls/lines256_length128/etc/ap_fixed_sim.h @@ -0,0 +1,2451 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_GCC_AP_FIXED_H__ +#define __AESL_GCC_AP_FIXED_H__ + +#ifndef __cplusplus + #error C++ is required to include this header file +#endif /* #ifndef __cplusplus */ + + +#include +#ifndef __AESL_APDT_IN_SCFLOW__ + #include "etc/ap_int_sim.h" +#else + #include "../etc/ap_private.h" +#endif /* #ifndef __AESL_APDT_IN_SCFLOW__ */ + +#define FLOAT_MAN 23 +#define FLOAT_EXP 8 +#define DOUBLE_MAN 52 +#define DOUBLE_EXP 11 +// #define DOUBLE_MAN_MASK (~0ULL >> (64-DOUBLE_MAN-2)) +#define DOUBLE_MAN_MASK 0x3fffffffffffffULL +#define BIAS(e) ((1ULL<<(e-1))-1) +#define FLOAT_BIAS BIAS(FLOAT_EXP) +#define DOUBLE_BIAS BIAS(DOUBLE_EXP) + +/// Forward declaration. +template struct ap_fixed_base; + +///Proxy class, which allows bit selection to be used as both rvalue(for reading) and +//lvalue(for writing) +template +struct af_bit_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& d_bv; + int d_index; +public: + INLINE af_bit_ref(const af_bit_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& ref): + d_bv(ref.d_bv), d_index(ref.d_index) {} + + INLINE af_bit_ref(ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>* bv, int index=0): + d_bv(*bv),d_index(index) {} + + INLINE operator bool() const { + return d_bv.V[d_index]; + } + + INLINE af_bit_ref& operator=(unsigned long long val) { + if (val) + d_bv.V.set(d_index); + else + d_bv.V.clear(d_index); + return *this; + } + + template + INLINE af_bit_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + return operator=(val!=0); + } + + INLINE af_bit_ref& operator =(const af_bit_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& val) { + return operator=((unsigned long long)(bool)val); + } + + template + INLINE af_bit_ref operator=(const af_bit_ref<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)val); + } + + template + INLINE af_bit_ref& operator = ( const ap_bit_ref<_AP_W2, _AP_S2> &val) { + return operator =((unsigned long long) (bool) val); + } + + template + INLINE af_bit_ref& operator = ( const ap_range_ref<_AP_W2,_AP_S2>& val) { + return operator =((const ap_private<_AP_W2, false>) val); + } + + template + INLINE af_bit_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, false>)(val)); + } + + template + INLINE af_bit_ref& operator= (const ap_concat_ref<_AP_W2, _AP_T3, _AP_W3, _AP_T3>& val) { + return operator=((const ap_private<_AP_W2 + _AP_W3, false>)(val)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + ap_private<_AP_W2, _AP_S2> >(*this, op); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<1, af_bit_ref, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<1, af_bit_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, + _AP_N2> >(*this, const_cast& >(op)); + } + + template + INLINE ap_concat_ref<1, af_bit_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<1, af_bit_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(op)); + } + + template + INLINE bool operator == (const af_bit_ref<_AP_W2, _AP_I2, + _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + return get() == op.get(); + } + + template + INLINE bool operator != (const af_bit_ref<_AP_W2, _AP_I2, + _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + return get() != op.get(); + } + + INLINE bool operator ~ () const { + bool bit = (d_bv.V)[d_index]; + return bit ? false : true; + } + + INLINE int length() const { + return 1; + } + + INLINE bool get() { + return d_bv.V[d_index]; + } + + INLINE bool get() const { + return d_bv.V[d_index]; + } + + INLINE std::string to_string() const { + return d_bv.V[d_index] ? "1" : "0"; + } +}; + +///Range(slice) reference +//------------------------------------------------------------ +template +struct af_range_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &d_bv; + int l_index; + int h_index; + +public: + INLINE af_range_ref(const af_range_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& ref): + d_bv(ref.d_bv), l_index(ref.l_index), h_index(ref.h_index) {} + + INLINE af_range_ref(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>* bv, int h, int l): + d_bv(*bv),l_index(l),h_index(h) { + //if (h < l) + // fprintf(stderr, + //"Warning! The bits selected will be returned in reverse order\n"); + } + + INLINE operator ap_private<_AP_W, false> () const { + if (h_index >= l_index) { + ap_private<_AP_W, false> val(d_bv.V); + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val>>=l_index; + return val&=mask; + } else { + ap_private<_AP_W, false> val = 0; + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if ((d_bv.V)[j]) val.set(i); + return val; + } + } + + INLINE operator unsigned long long() const { + return get().to_uint64(); + } + + template + INLINE af_range_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + ap_private<_AP_W, false> vval= ap_private<_AP_W, false>(val); + if (l_index > h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + vval[i]? d_bv.V.set(j):d_bv.V.clear(j); + } else { + ap_private<_AP_W,false> mask(-1); + if (l_index>0) { + mask<<=l_index; + vval<<=l_index; + } + if (h_index<_AP_W-1) { + ap_private<_AP_W,false> mask2(-1); + mask2>>=_AP_W-h_index-1; + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv.V &= mask; + d_bv.V |= vval; + } + return *this; + } + + INLINE af_range_ref& operator = (unsigned long long val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator =(const ap_bit_ref<_AP_W3, _AP_S3>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE af_range_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE af_range_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& val) { + const ap_private<_AP_W2, false> tmp= val.get(); + return operator = (tmp); + } + + INLINE af_range_ref& operator= (const af_range_ref<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& val) { + const ap_private<_AP_W, false> tmp= val.get(); + return operator = (tmp); + } + + template + INLINE af_range_ref& operator= (const ap_fixed_base<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=(val.to_ap_private()); + } + + template + INLINE bool operator == (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + template + INLINE bool operator != (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + template + INLINE bool operator > (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + template + INLINE bool operator >= (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + template + INLINE bool operator < (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + template + INLINE bool operator == (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + template + INLINE bool operator != (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + template + INLINE bool operator > (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + template + INLINE bool operator >= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + template + INLINE bool operator < (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op2) { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + template + INLINE void set(const ap_private<_AP_W2,false>& val) { + ap_private<_AP_W,_AP_S> vval=val; + if (l_index>h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + vval[i]? d_bv.V.set(j):d_bv.V.clear(j); + } else { + ap_private<_AP_W,_AP_S> mask(-1); + if (l_index>0) { + ap_private<_AP_W,false> mask1(-1); + mask1>>=_AP_W-l_index; + mask1.flip(); + mask=mask1; + //vval&=mask1; + vval<<=l_index; + } + if (h_index<_AP_W-1) { + ap_private<_AP_W,false> mask2(-1); + mask2<<=h_index+1; + mask2.flip(); + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } + + } + + INLINE ap_private<_AP_W,false> get() const { + if (h_index val(0); + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if ((d_bv.V)[j]) val.set(i); + return val; + } else { + ap_private<_AP_W, false> val = ap_private<_AP_W,false>(d_bv.V); + val>>= l_index; + if (h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + return val; + } + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + ap_private<_AP_W2, _AP_S2> >(*this, op); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2 + _AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, _AP_W2, + af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& > (op)); + } + + template + INLINE ap_concat_ref<_AP_W, af_range_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &op) { + return ap_concat_ref<_AP_W, af_range_ref, 1, + af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(op)); + } + + INLINE int length() const { + return h_index>=l_index?h_index-l_index+1:l_index-h_index+1; + } + + INLINE int to_int() const { + ap_private<_AP_W,false> val=get(); + return val.to_int(); + } + + INLINE unsigned int to_uint() const { + ap_private<_AP_W,false> val=get(); + return val.to_uint(); + } + + INLINE long to_long() const { + ap_private<_AP_W,false> val=get(); + return val.to_long(); + } + + INLINE unsigned long to_ulong() const { + ap_private<_AP_W,false> val=get(); + return val.to_ulong(); + } + + INLINE ap_slong to_int64() const { + ap_private<_AP_W,false> val=get(); + return val.to_int64(); + } + + INLINE ap_ulong to_uint64() const { + ap_private<_AP_W,false> val=get(); + return val.to_uint64(); + } + + INLINE std::string to_string(uint8_t radix) const { + return get().to_string(radix); + } + +}; + +//----------------------------------------------------------------------------- +///ap_fixed_base: AutoPilot fixed point +//----------------------------------------------------------------------------- +template +struct ap_fixed_base { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ +public: + template friend struct +ap_fixed_base; + template friend struct +af_bit_ref; + + INLINE void overflow_adjust(bool underflow, bool overflow, + bool lD, bool sign) { + if (!overflow && !underflow) return; + switch (_AP_O) { + case AP_WRAP: + if (_AP_N == 0) + return; + if (_AP_S) { + //signed SC_WRAP + //n_bits == 1; + if (_AP_N > 1) { + ap_private<_AP_W, _AP_S> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + if (sign) + V &= mask; + else + V |= ~mask; + } + sign ? V.set(_AP_W - 1) : V.clear(_AP_W - 1); + } else { + //unsigned SC_WRAP + ap_private<_AP_W, _AP_S> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + mask.flip(); + V |= mask; + } + break; + case AP_SAT_ZERO: + V.clear(); + break; + case AP_WRAP_SM: + { + bool Ro = ap_private_ops::get<_AP_W, _AP_S, _AP_W -1>(V); // V[_AP_W -1]; + if (_AP_N == 0) { + if (lD != Ro) { + V.flip(); + lD ? ap_private_ops::set<_AP_W, _AP_S, _AP_W - 1>(V) : + ap_private_ops::clear<_AP_W, _AP_S, _AP_W - 1>(V); + } + } else { + if (_AP_N == 1 && sign != Ro) { + V.flip(); + } else if (_AP_N > 1) { + bool lNo = ap_private_ops::get<_AP_W, _AP_S, _AP_W - _AP_N> (V); // V[_AP_W - _AP_N]; + if (lNo == sign) + V.flip(); + ap_private<_AP_W, false> mask(-1); + if (_AP_N >= _AP_W) mask = 0; + else mask.lshr(_AP_N); + if (sign) + V &= mask; + else + V |= mask.flip(); + sign ? ap_private_ops::set<_AP_W, _AP_S, _AP_W - 1>(V) : ap_private_ops::clear<_AP_W, _AP_S, _AP_W - 1>(V); + } + } + } + break; + default: + if (_AP_S) { + if (overflow) { + V.set(); ap_private_ops::clear<_AP_W, _AP_S, _AP_W-1>(V); + } else if (underflow) { + V.clear(); + ap_private_ops::set<_AP_W, _AP_S, _AP_W-1>(V); + if (_AP_O == AP_SAT_SYM) + ap_private_ops::set<_AP_W, _AP_S, 0>(V); + } + } else { + if (overflow) + V.set(); + else if (underflow) + V.clear(); + } + } + } + + INLINE bool quantization_adjust(bool qb, bool r, bool s) { + bool carry=ap_private_ops::get<_AP_W, _AP_S, _AP_W-1>(V); + switch (_AP_Q) { + case AP_TRN: + return false; + case AP_RND_ZERO: + qb &= s || r; + break; + case AP_RND_MIN_INF: + qb &= r; + break; + case AP_RND_INF: + qb &= !s || r; + break; + case AP_RND_CONV: + qb &= ap_private_ops::get<_AP_W, _AP_S, 0>(V) || r; + break; + case AP_TRN_ZERO: + qb = s && ( qb || r ); + break; + default:; + + } + if (qb) ++V; + //only when old V[_AP_W-1]==1 && new V[_AP_W-1]==0 + return carry && !(ap_private_ops::get<_AP_W, _AP_S, _AP_W-1>(V)); //(!V[_AP_W-1]); + } + + template + struct RType { + enum { + _AP_F=_AP_W-_AP_I, + F2=_AP_W2-_AP_I2, + mult_w = _AP_W+_AP_W2, + mult_i = _AP_I+_AP_I2, + mult_s = _AP_S||_AP_S2, + plus_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1+AP_MAX(_AP_F,F2), + plus_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1, + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1+AP_MAX(_AP_F,F2), + minus_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+1, + minus_s = true, +#ifndef __SC_COMPATIBLE__ + div_w = _AP_W + AP_MAX(_AP_W2 - _AP_I2, 0) + _AP_S2, +#else + div_w = _AP_W + AP_MAX(_AP_W2 - _AP_I2, 0) + _AP_S2 + AP_MAX(_AP_I2, 0), +#endif /* #ifndef __SC_COMPATIBLE__ */ + div_i = _AP_I + (_AP_W2-_AP_I2) + _AP_S2, + div_s = _AP_S||_AP_S2, + logic_w = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2))+AP_MAX(_AP_F,F2), + logic_i = AP_MAX(_AP_I+(_AP_S2&&!_AP_S),_AP_I2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + + typedef ap_fixed_base mult; + typedef ap_fixed_base plus; + typedef ap_fixed_base minus; + typedef ap_fixed_base logic; + typedef ap_fixed_base div; + typedef ap_fixed_base<_AP_W, _AP_I, _AP_S> arg1; + }; + INLINE void report() { +#if 0 + if (_AP_W > 1024 && _AP_W <= 4096) { + fprintf(stderr, "[W] W=%d is out of bound (1<=W<=1024):" + " for synthesis, please define macro AP_INT_TYPE_EXT(N) to" + " extend the valid range.\n", _AP_W); + } else +#endif /* #if 0 */ + if (_AP_W > MAX_MODE(AP_INT_MAX_W) * 1024) { + fprintf(stderr, "[E] ap_%sfixed<%d, ...>: Bitwidth exceeds the " + "default max value %d. Please use macro " + "AP_INT_MAX_W to set a larger max value.\n", + _AP_S?"":"u", _AP_W, + MAX_MODE(AP_INT_MAX_W) * 1024); + exit(1); + } + } + + /// Constructors. + // ------------------------------------------------------------------------- +#if 0 + #ifdef __SC_COMPATIBLE__ + INLINE ap_fixed_base():V(uint32_t(_AP_W), uint64_t(0)) {} + #else + INLINE ap_fixed_base():V(uint32_t(_AP_W)) {} + #endif /* #ifdef __SC_COMPATIBLE__ */ +#else + INLINE ap_fixed_base():V(0) {} +#endif /* #if 0 */ + // INLINE ap_fixed_base():V() {} + // INLINE explicit ap_fixed_base(const ap_private<_AP_W+_AP_I, _AP_S>& _V):V(_V) {} + INLINE ap_fixed_base(const ap_fixed_base& op):V(op.V) {} + template + INLINE ap_fixed_base(const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op):V(0) { + enum {N2=_AP_W2,_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2,QUAN_INC=F2>_AP_F && !(_AP_Q==AP_TRN || + (_AP_Q==AP_TRN_ZERO && !_AP_S2))}; + if (!op) return; + bool carry=false; + //handle quantization + enum { sh_amt =(F2>_AP_F)?F2-_AP_F:_AP_F-F2}; + const ap_private<_AP_W2, _AP_S2>& val = op.V; + bool neg_src=val.isNegative(); + if (F2==_AP_F) + V=val; + + else if (F2>_AP_F) { + if (sh_amt >= _AP_W2) + V = neg_src ? -1 : 0; + else + V = _AP_S2?val.ashr(sh_amt):val.lshr(sh_amt); + if (_AP_Q!=AP_TRN && !(_AP_Q==AP_TRN_ZERO && !_AP_S2)) { + bool qb = false; + if (F2-_AP_F>_AP_W2) + qb = neg_src; + else + qb = ap_private_ops::get<_AP_W2, _AP_S2, F2-_AP_F-1>(val); + + bool r=false; + enum { pos3 = F2-_AP_F-2}; + if (pos3>=_AP_W2-1) + r=val!=0; + else if (pos3>=0) + r = (val<<(_AP_W2-1-pos3))!=0; + carry = quantization_adjust(qb,r,neg_src); + } + } else { //no quantization + if (sh_amt < _AP_W) { + V=val; + V <<= sh_amt; + } + } + //hanle overflow/underflow + if ((_AP_O!=AP_WRAP || _AP_N != 0) && + ((!_AP_S && _AP_S2) || _AP_I-_AP_S < + _AP_I2 - _AP_S2 + (QUAN_INC|| (_AP_S2 && + _AP_O==AP_SAT_SYM)))) {//saturation + bool deleted_zeros = _AP_S2?true:!carry, + deleted_ones = true; + bool lD=(_AP_I2>_AP_I && _AP_W2-_AP_I2+_AP_I>=0) && + ap_private_ops::get<_AP_W2, _AP_S2, _AP_W2-_AP_I2+_AP_I>(val); + enum { pos1=F2-_AP_F+_AP_W, pos2=F2-_AP_F+_AP_W+1}; + if (pos1 < _AP_W2) { + bool Range1_all_ones= true; + bool Range1_all_zeros= true; + if (pos1 >= 0) { + enum { __W = (_AP_W2-pos1) > 0 ? (_AP_W2-pos1) : 1 }; + const ap_private<__W, _AP_S2> Range1=ap_private<__W, _AP_S2>(val.lshr(pos1)); + Range1_all_ones=Range1.isAllOnesValue(); + Range1_all_zeros=Range1.isMinValue(); + } else { + Range1_all_ones=false; + Range1_all_zeros=val.isMinValue(); + } + bool Range2_all_ones=true; + if (pos2<_AP_W2 && pos2>=0) { + enum { __W = (_AP_W2-pos2)>0 ? (_AP_W2-pos2) : 1}; + ap_private<__W, true> Range2=ap_private<__W, true>(val.lshr(pos2)); + Range2_all_ones=Range2.isAllOnesValue(); + } else if (pos2<0) + Range2_all_ones=false; + + deleted_zeros=deleted_zeros && (carry?Range1_all_ones:Range1_all_zeros); + deleted_ones=carry?Range2_all_ones&&(F2-_AP_F+_AP_W<0||!lD) + :Range1_all_ones; + neg_src= neg_src&&!(carry && Range1_all_ones); + } else + neg_src = neg_src && V[_AP_W-1]; + + bool neg_trg= V.isNegative(); + bool overflow=(neg_trg||!deleted_zeros) && !val.isNegative(); + bool underflow=(!neg_trg||!deleted_ones)&&neg_src; + //printf("neg_src = %d, neg_trg = %d, deleted_zeros = %d, + // deleted_ones = %d, overflow = %d, underflow = %d\n", + // neg_src, neg_trg, deleted_zeros, deleted_ones, + // overflow, underflow); + if (_AP_O==AP_SAT_SYM && _AP_S2 && _AP_S) + underflow |= neg_src && (_AP_W>1?V.isMinSignedValue():true); + overflow_adjust(underflow, overflow, lD, neg_src); + } + report(); + } + + template + INLINE ap_fixed_base(const volatile ap_fixed_base<_AP_W2,_AP_I2, + _AP_S2,_AP_Q2,_AP_O2, _AP_N2> &op) : V(op.V) { + *this = const_cast&>(op); + } + + template + INLINE ap_fixed_base(const ap_private<_AP_W2,_AP_S2>& op) { + ap_fixed_base<_AP_W2,_AP_W2,_AP_S2> f_op; + f_op.V=op; + *this = f_op; + } + + INLINE ap_fixed_base(bool b) { + *this=(ap_private<1,false>)b; + report(); + } + + INLINE ap_fixed_base(char b) { + *this=(ap_private<8,false>)b; + report(); + } + + INLINE ap_fixed_base(signed char b) { + *this=(ap_private<8,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned char b) { + *this=(ap_private<8,false>)b; + report(); + } + + INLINE ap_fixed_base(signed short b) { + *this=(ap_private<16,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned short b) { + *this=(ap_private<16,false>)b; + report(); + } + + INLINE ap_fixed_base(signed int b) { + *this=(ap_private<32,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned int b) { + *this=(ap_private<32,false>)b; + report(); + } +# if defined __x86_64__ + INLINE ap_fixed_base(signed long b) { + *this=(ap_private<64,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned long b) { + *this=(ap_private<64,false>)b; + report(); + } +# else + INLINE ap_fixed_base(signed long b) { + *this=(ap_private<32,true>)b; + report(); + } + + INLINE ap_fixed_base(unsigned long b) { + *this=(ap_private<32,false>)b; + report(); + } +# endif + + INLINE ap_fixed_base(ap_slong b) { + *this=(ap_private<64,true>)b; + report(); + } + + INLINE ap_fixed_base(ap_ulong b) { + *this=(ap_private<64,false>)b; + report(); + } + +#if 1 + INLINE ap_fixed_base(const char* val):V(0) { + ap_private<_AP_W, _AP_S> Tmp(val); + V = Tmp; + } + + INLINE ap_fixed_base(const char* val, signed char rd): V(0) { + ap_private<_AP_W, _AP_S> Tmp(val, rd); + V = Tmp; + } + +#endif + + INLINE ap_fixed_base(const std::string& val) { + ap_private<_AP_W, _AP_S> Tmp(val, 2); + V = Tmp; + report(); + } + + template + INLINE ap_fixed_base(const ap_bit_ref<_AP_W2, _AP_S2>& op) { + *this = ((bool)op); + report(); + } + + template + INLINE ap_fixed_base(const ap_range_ref<_AP_W2, _AP_S2>& op) { + *this = ap_private<_AP_W2, _AP_S2>(op); + report(); + } + + template + INLINE ap_fixed_base(const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& op) { + *this = ((const ap_private<_AP_W2 + _AP_W3, false>&)(op)); + report(); + } + + template + INLINE ap_fixed_base(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + *this = (bool(op)); + report(); + } + + template + INLINE ap_fixed_base(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + *this = (ap_private<_AP_W2, false>(op)); + report(); + } + + //helper function + INLINE unsigned long long doubleToRawBits(double pf)const { + union { + unsigned long long __L; + double __D; + }LD; + LD.__D=pf; + return LD.__L; + } + + + INLINE double rawBitsToDouble(unsigned long long pi) const { + union { + unsigned long long __L; + double __D; + }LD; + LD.__L=pi; + return LD.__D; + } + + INLINE float rawBitsToFloat(uint32_t pi) const { + union { + uint32_t __L; + float __D; + }LD; + LD.__L = pi; + return LD.__D; + } + + INLINE ap_fixed_base(double d):V(0) { + if (!d) return; + const bool isneg=d<0; + + const uint64_t ireg=doubleToRawBits(isneg?-d:d); + if ((ireg&0x7fffffffffffffffULL)!=0) { + const int32_t exp=(((ireg)>>DOUBLE_MAN)&0x07ff)-DOUBLE_BIAS; + ap_private man = ireg & DOUBLE_MAN_MASK; + man.clear(DOUBLE_MAN+1); + man.set(DOUBLE_MAN); + if (isneg) { + man.flip(); + man++; + } + + enum {_AP_S2=true, _AP_W2=DOUBLE_MAN+2,_AP_F=_AP_W -_AP_I }; + const int _AP_I2=exp+2; + const int F2=_AP_W2-_AP_I2; + const bool QUAN_INC=F2>_AP_F && !(_AP_Q==AP_TRN || (_AP_Q==AP_TRN_ZERO && + !_AP_S2)); + bool carry=false; + //handle quantization + const unsigned sh_amt=abs(F2-_AP_F); // sh_amt = F2>_AP_F ? F2 -_AP_F : _AP_F-F2; + if (F2==_AP_F ) + V=man; + else if (F2>_AP_F) { + if (sh_amt >= DOUBLE_MAN+2) + V=isneg?-1:0; + else + V=(man>>sh_amt) | ((man & 1ULL<<(DOUBLE_MAN+1)) ? (DOUBLE_MAN_MASK>>(DOUBLE_MAN+2-sh_amt) <<(DOUBLE_MAN+2-sh_amt)):0); + + if (_AP_Q!=AP_TRN && !(_AP_Q==AP_TRN_ZERO && !_AP_S2)) { + const bool qb=((F2-_AP_F > DOUBLE_MAN+2) ? isneg : (man & (1ULL<<(F2-_AP_F-1))) != 0); + const int pos3=F2-_AP_F-2; + const bool r = (pos3>= 0) ? (man << AP_MAX(0, _AP_W2-pos3-1)& DOUBLE_MAN_MASK)!=0 : false; + carry = quantization_adjust(qb,r,isneg); + } + } + else { //no quantization + // V=man; + if (sh_amt < _AP_W) { + V = man; + V <<= sh_amt; + } + } + //handle overflow/underflow + if ((_AP_O != AP_WRAP || _AP_N != 0) && + ((!_AP_S && _AP_S2) || _AP_I-_AP_S < + _AP_I2-_AP_S2+(QUAN_INC|| (_AP_S2 && + _AP_O==AP_SAT_SYM)) )) {// saturation + bool deleted_zeros = _AP_S2?true:!carry, + deleted_ones = true; + bool neg_src; + const bool lD=(_AP_I2>_AP_I) && (_AP_W2-_AP_I2+_AP_I>=0) && (man & (1ULL <<(DOUBLE_MAN+2-_AP_I2+_AP_I))); + int pos1=F2+_AP_W-_AP_F; + if (pos1 < _AP_W2) { + int pos2=pos1+1; + bool Range1_all_ones=true; + bool Range1_all_zeros=true; + if (pos1>=0) { + ap_private<_AP_W,_AP_S> Range1= + ap_private<_AP_W,_AP_S>((man >> pos1) | ((1ULL<<(DOUBLE_MAN+1)&man) ? (DOUBLE_MAN_MASK >> (DOUBLE_MAN+2-pos1) <<(DOUBLE_MAN+2-pos1)):0)); + Range1_all_ones = Range1.isAllOnesValue(); // Range1.isAllOnesValue(); + Range1_all_zeros = Range1.isMinValue(); // Range1.isMinValue(); + } else { + Range1_all_ones=false; + Range1_all_zeros = man==0; // man.isMinValue(); + } + bool Range2_all_ones=true; + if (pos2<_AP_W2 && pos2>=0) { + ap_private<_AP_W, _AP_S> Range2= + ap_private<_AP_W, _AP_S>((man >> pos2) | ((1ULL<<(DOUBLE_MAN+1)&man) ? (DOUBLE_MAN_MASK >> (DOUBLE_MAN+2-pos2) <<(DOUBLE_MAN+2-pos2)):0)); + Range2_all_ones=Range2.isAllOnesValue(); // Range2.isAllOnesValue(); + } else if (pos2<0) + Range2_all_ones=false; + deleted_zeros=deleted_zeros && (carry?Range1_all_ones:Range1_all_zeros); + deleted_ones=carry?Range2_all_ones&&(F2-_AP_F+_AP_W<0||!lD) : Range1_all_ones; + neg_src=isneg&&!(carry&Range1_all_ones); + } else + neg_src = isneg && V[_AP_W -1]; + + const bool neg_trg=V.isNegative(); + const bool overflow=(neg_trg||!deleted_zeros) && !isneg; + bool underflow=(!neg_trg||!deleted_ones)&&neg_src; + //printf("neg_src = %d, neg_trg = %d, deleted_zeros = %d, + // deleted_ones = %d, overflow = %d, underflow = %d\n", + // neg_src, neg_trg, deleted_zeros, deleted_ones, + // overflow, underflow); + if (_AP_O==AP_SAT_SYM && _AP_S2 && _AP_S) + underflow |= neg_src && (_AP_W>1?V.isMinSignedValue():true); + overflow_adjust(underflow,overflow,lD, neg_src); + } + } + report(); + } + + + ///assign operators + //------------------------------------------------------------------------- + + INLINE volatile ap_fixed_base& operator=(const ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) volatile { + V = op.V; + return *this; + } + + INLINE ap_fixed_base& operator=(const ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) { + V = op.V; + return *this; + } + + INLINE volatile ap_fixed_base& operator=(const volatile ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) volatile { + V = op.V; + return *this; + } + + INLINE ap_fixed_base& operator=(const volatile ap_fixed_base<_AP_W, _AP_I, _AP_S, + _AP_Q, _AP_O, _AP_N>& op) { + V = op.V; + return *this; + } + + // Set this ap_fixed_base with a bits string. That means the ssdm_int::V + // inside this ap_fixed_base is assigned by bv. + // Note the input parameter should be a fixed-point formatted bit string. + INLINE ap_fixed_base& setBits(unsigned long long bv) { + V=bv; + return *this; + } + + // Return a ap_fixed_base object whose ssdm_int::V is assigned by bv. + // Note the input parameter should be a fixed-point formatted bit string. + static INLINE ap_fixed_base bitsToFixed(unsigned long long bv) { + ap_fixed_base Tmp=bv; + return Tmp; + } + + // Explicit conversion functions to ap_private that captures + // all integer bits (bits are truncated) + INLINE ap_private + to_ap_private(bool Cnative = true) const { + ap_private ret = ap_private ((_AP_I >= 1) ? (_AP_S==true ? V.ashr(AP_MAX(0,_AP_W - _AP_I)) : V.lshr(AP_MAX(0,_AP_W - _AP_I))) : ap_private<_AP_W, _AP_S>(0)); + + if (Cnative) { + bool r = false; + if (_AP_I < _AP_W) { + if (_AP_I > 0) r = !(V.getLoBits(_AP_W - _AP_I).isMinValue()); + else r = !(V.isMinValue()); + } + if (r && V.isNegative()) { // if this is negative integer + ++ret;//ap_private(1,_AP_S); + } + } else { + //Follow OSCI library, conversion from sc_fixed to sc_int + } + return ret; + } + + template + INLINE operator ap_private<_AP_W2,_AP_S2> () const { + return (ap_private<_AP_W2,_AP_S2>)to_ap_private(); + } + + template + INLINE operator ap_private<_AP_W2,_AP_S2,_AP_N2> () const { + return (ap_private<_AP_W2,_AP_S2,_AP_N2>)to_ap_private(); + } + + //Explict conversion function to C built-in integral type + INLINE int to_int() const { + return to_ap_private().to_int(); + } + + INLINE int to_uint() const { + return to_ap_private().to_uint(); + } + + INLINE ap_slong to_int64() const { + return to_ap_private().to_int64(); + } + + INLINE ap_ulong to_uint64() const { + return to_ap_private().to_uint64(); + } + + INLINE double to_double() const { + if (!V) + return 0; + if (_AP_W>64 || (_AP_W - _AP_I) > 0) { + bool isneg = _AP_S && V[_AP_W-1]; + uint64_t res = isneg ? 0x8000000000000000ULL : 0; + ap_private<_AP_W, false> tmp = V; + if (isneg) tmp = -tmp; + int i = _AP_W -1 - tmp.countLeadingZeros(); + int exp = _AP_I-(_AP_W-i); + res|=((uint64_t)(exp+DOUBLE_BIAS))<DOUBLE_MAN)?tmp.lshr(i-DOUBLE_MAN):tmp).to_uint64() & DOUBLE_MAN_MASK; + res |= i 0) { + /* This specialization is disabled. It is giving wrong results in some cases. + bool isneg=V.isNegative(); + double dp = V.get(); + dp /= (1<< (_AP_W - _AP_I)); + return dp;*/ + } else + return double(to_int64()); + } + + INLINE float to_float() const { + uint32_t res=0; + if (V==0) + return 0; + bool isneg=V.isNegative(); + ap_private<_AP_W, _AP_S> tmp=V; + if (isneg) tmp = -tmp; + if (_AP_W-_AP_I>0||_AP_W>64) { + if (isneg) + res=0x80000000; + int i=_AP_W-1; + i-=tmp.countLeadingZeros(); + int exp=_AP_I-(_AP_W-i); + res|=(exp+FLOAT_BIAS)< man = 0; + if (i!=0) { + tmp.clear(i); + if (i>FLOAT_MAN) + man=tmp.lshr(i-FLOAT_MAN); + else + man=tmp; + res |= i < FLOAT_MAN?man.getZExtValue()<<(FLOAT_MAN-i):man.getZExtValue(); + } + } else { + return float(to_int64()); + } + float dp=rawBitsToFloat(res); + return dp; + } + + INLINE operator double () const { + return to_double(); + } +#ifndef __SC_COMPATIBLE__ + INLINE operator float () const { + return to_float(); + } + + INLINE operator char () const { + return (char) to_int(); + } + + INLINE operator unsigned char () const { + return (unsigned char) to_uint(); + } + + INLINE operator short () const { + return (short) to_int(); + } + + INLINE operator unsigned short () const { + return (unsigned short) to_uint(); + } + + INLINE operator int () const { + return to_int(); + } + + INLINE operator unsigned int () const { + return to_uint(); + } +#if 1 +#ifdef __x86_64__ + INLINE operator long () const { + return (long)to_int64(); + } + + INLINE operator unsigned long () const { + return (unsigned long) to_uint64(); + } +#else + INLINE operator long () const { + return to_int64(); + } + + INLINE operator unsigned long () const { + return to_uint64(); + } + +#endif +#endif + INLINE operator unsigned long long () const { + return to_uint64(); + } + + INLINE operator long long () const { + return to_int64(); + } +#endif + + INLINE std::string to_string(uint8_t radix=2, bool sign=false) const; + + INLINE ap_slong bits_to_int64() const { + ap_private res(V); + return (ap_slong) res; + } + + INLINE ap_ulong bits_to_uint64() const { + ap_private res(V); + return (ap_ulong) res; + } + + INLINE int length() const {return _AP_W;} + + // Count the number of zeros from the most significant bit + // to the first one bit. Note this is only for ap_fixed_base whose + // _AP_W <= 64, otherwise will incur assertion. + INLINE int countLeadingZeros() { + return V.countLeadingZeros(); + } + + ///Arithmetic:Binary + //------------------------------------------------------------------------- + template + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::mult + operator * (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + typename RType<_AP_W2,_AP_I2,_AP_S2>::mult r; + r.V = V * op2.V; + return r; + } + + template + static INLINE ap_fixed_base multiply(const ap_fixed_base<_AP_W1,_AP_I1,_AP_S1>& op1, const + ap_fixed_base<_AP_W2,_AP_I2,_AP_S2>& op2) { + ap_private<_AP_W+_AP_W2, _AP_S> OP1=op1.V; + ap_private<_AP_W2,_AP_S2> OP2=op2.V; + return OP1*OP2; + } + + template + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::div + operator / (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {F2 = _AP_W2-_AP_I2, _W1=AP_MAX(_AP_W + AP_MAX(F2, 0), _AP_W2), + _W2=AP_MAX(_AP_W2,AP_MAX(_AP_W + AP_MAX(F2, 0), _AP_W2))}; + ap_private<_W1, _AP_S> dividend = (ap_private<_W1, _AP_S>(V)) << ((_W1>_AP_W)?F2:0); + ap_private<_W1, _AP_S2> divisior = ap_private<_W2, _AP_S2>(op2.V); + ap_private<_W1, _AP_S> ret = ap_private<_W1,_AP_S> ((_AP_S||_AP_S2) ? dividend.sdiv(divisior): dividend.udiv(divisior)); + typename RType<_AP_W2, _AP_I2, _AP_S2>::div r; + r.V = ret; + return r; + } +#define OP_BIN_AF(Sym, Rty, Width, Sign, Fun) \ + template \ + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty \ + operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + enum {_AP_F=_AP_W-_AP_I, F2=_AP_W2-_AP_I2}; \ + typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty r, lhs(*this), rhs(op2); \ + r.V = lhs.V.Fun(rhs.V); \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty \ + operator Sym (const ap_fixed_base& op2) const \ + { \ + typename RType<_AP_W,_AP_I,_AP_S>::Rty r; \ + r.V = V Sym op2.V; \ + return r; \ + } \ + + OP_BIN_AF(+, plus, plus_w, plus_s, Add) + OP_BIN_AF(-, minus, minus_w, minus_s, Sub) + +#define OP_LOGIC_BIN_AF(Sym, Rty, Width, Sign) \ + template \ + INLINE typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty \ + operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + typename RType<_AP_W2,_AP_I2,_AP_S2>::Rty r, lhs(*this), rhs(op2); \ + r.V=lhs.V Sym rhs.V; \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty \ + operator Sym (const ap_fixed_base& op2) const \ + { \ + typename RType<_AP_W,_AP_I,_AP_S>::Rty r; \ + r.V = V Sym op2.V; \ + return r; \ + } \ + INLINE typename RType<_AP_W,_AP_I,_AP_S>::Rty operator Sym(int op2) const \ + { \ + return V Sym (op2<<(_AP_W - _AP_I)); \ + } + OP_LOGIC_BIN_AF(&, logic, logic_w, logic_s) + OP_LOGIC_BIN_AF(|, logic, logic_w, logic_s) + OP_LOGIC_BIN_AF(^, logic, logic_w, logic_s) + + ///Arithmic : assign + //------------------------------------------------------------------------- +#define OP_ASSIGN_AF(Sym) \ + template \ + INLINE ap_fixed_base& operator Sym##= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) \ + { \ + *this=operator Sym (op2) ; \ + return *this; \ + } + + OP_ASSIGN_AF(+) + OP_ASSIGN_AF(-) + OP_ASSIGN_AF(&) + OP_ASSIGN_AF(|) + OP_ASSIGN_AF(^) + OP_ASSIGN_AF(*) + OP_ASSIGN_AF(/) + + ///Prefix increment, decrement + //------------------------------------------------------------------------- + INLINE ap_fixed_base& operator ++() { + operator+=(ap_fixed_base<1,1,false>(1)); //SystemC's semantics + return *this; + } + + INLINE ap_fixed_base& operator --() { + operator-=(ap_fixed_base<1,1,false>(1)); //SystemC's semantics + return *this; + } + + //Postfix increment, decrement + //------------------------------------------------------------------------- + INLINE const ap_fixed_base operator ++(int) { + ap_fixed_base t(*this); + operator++(); + return t; + } + + INLINE const ap_fixed_base operator --(int) { + ap_fixed_base t = *this; + operator--(); + return t; + } + + ///Unary arithmetic + //------------------------------------------------------------------------- + INLINE ap_fixed_base operator +() {return *this;} + + INLINE ap_fixed_base<_AP_W + 1, _AP_I + 1, true> operator -() const { + ap_fixed_base<_AP_W + 1, _AP_I + 1, true> Tmp(*this); + Tmp.V = - Tmp.V; + return Tmp; + } + + INLINE ap_fixed_base<_AP_W,_AP_I,true,_AP_Q,_AP_O, _AP_N> getNeg() { + ap_fixed_base<_AP_W,_AP_I,true,_AP_Q,_AP_O, _AP_N> Tmp(*this); + Tmp.V=-Tmp.V; + return Tmp; + } + + ///Not (!) + //------------------------------------------------------------------------- + INLINE bool operator !() const { + return !V; + } + + ///Bitwise complement + //------------------------------------------------------------------------- + INLINE ap_fixed_base<_AP_W, _AP_I, _AP_S> + operator ~() const { + ap_fixed_base<_AP_W, _AP_I, _AP_S> res(*this); + res.V.flip(); + return res; + } + + ///Shift + ///template argument as shift value + template + INLINE ap_fixed_base<_AP_W, _AP_I + _AP_SHIFT, _AP_S> lshift () const { + ap_fixed_base<_AP_W, _AP_I + _AP_SHIFT, _AP_S> r; + r.V = V; + return r; + } + + template + INLINE ap_fixed_base<_AP_W, _AP_I - _AP_SHIFT, _AP_S> rshift () const { + ap_fixed_base<_AP_W, _AP_I - _AP_SHIFT, _AP_S> r; + r.V = V; + return r; + } + + //Because the return type is the type of the the first operand, shift assign + //operators do not carry out any quantization or overflow + //While systemc, shift assigns for sc_fixed/sc_ufixed will result in + //quantization or overflow (depending on the mode of the first operand) + //------------------------------------------------------------------------- + INLINE ap_fixed_base operator << (int sh) const { + ap_fixed_base r; + bool isNeg=(sh&0x80000000) != 0; + sh=isNeg?-sh:sh; + bool shiftoverflow = sh >= _AP_W; + bool NegSrc = V.isNegative(); + if (isNeg) { + if (shiftoverflow) + NegSrc?r.V.set():r.V.clear(); + else + r.V=_AP_S?V.ashr(sh):V.lshr(sh); + } else { + if (shiftoverflow) + r.V.clear(); + else + r.V=V< 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } else if (isNeg == false && _AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh - 1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator<<(const ap_private<_AP_W2,true>& op2) const { + int sh = op2.to_int(); + return operator << (sh); + } + + INLINE ap_fixed_base operator << (unsigned int sh ) const { + ap_fixed_base r; + bool shiftoverflow = sh >= _AP_W; + r.V = shiftoverflow ? ap_private<_AP_W, _AP_S >(0) : V << sh; + if (sh == 0) return r; +#ifdef __SC_COMPATIBLE__ + bool NegSrc = V.isNegative(); + if (_AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh -1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator << (const ap_private<_AP_W2,false>& op2) const { + unsigned int sh = op2.to_uint(); + return operator << (sh); + } + + INLINE ap_fixed_base operator >> (int sh) const { + ap_fixed_base r; + bool isNeg=(sh&0x80000000) != 0; + bool NegSrc = V.isNegative(); + sh=isNeg?-sh:sh; + bool shiftoverflow = sh >= _AP_W; + if (isNeg && !shiftoverflow) r.V=V< 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } else if (isNeg == true && _AP_O != AP_WRAP) { + bool allones, allzeros; + if (sh < _AP_W ) { + ap_private<_AP_W, _AP_S > range1 = V.lshr(_AP_W - sh - 1); + allones = range1.isAllOnesValue(); + allzeros = range1.isMinValue(); + } else { + allones = false; + allzeros = V.isMinValue(); + } + bool overflow = !allzeros && !NegSrc; + bool underflow = !allones && NegSrc; + if (_AP_O == AP_SAT_SYM && _AP_S) + underflow |= NegSrc && (_AP_W > 1 ? r.V.isMinSignedValue():true); + bool lD = false; + if ( sh < _AP_W ) lD = V[_AP_W - sh - 1]; + r.overflow_adjust(underflow, overflow, lD, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator >> (const ap_private<_AP_W2,true>& op2) const { + int sh = op2.to_int(); + return operator >> (sh); + } + + INLINE ap_fixed_base operator >> (unsigned int sh) const { + ap_fixed_base r; + bool NegSrc = V.isNegative(); + bool shiftoverflow = sh >= _AP_W; + if (shiftoverflow) + NegSrc?r.V.set():r.V.clear(); + else + r.V=_AP_S?V.ashr(sh):V.lshr(sh); +#ifdef __SC_COMPATIBLE__ + if (sh == 0) return r; + if (_AP_Q != AP_TRN) { + bool qb = false; + if (sh <= _AP_W) qb = V[sh - 1]; + bool rb = false; + if (sh > 1 && sh <= _AP_W) + rb = (V << (_AP_W - sh + 1 )) != 0; + else if (sh > _AP_W) + rb = V != 0; + r.quantization_adjust(qb, rb, NegSrc); + } +#endif + return r; + } + + template + INLINE ap_fixed_base operator >> (const ap_private<_AP_W2,false>& op2) const { + unsigned int sh = op2.to_uint(); + return operator >> (sh); + } + + ///shift assign + //------------------------------------------------------------------------- +#define OP_AP_SHIFT_AP_ASSIGN_AF(Sym) \ + template \ + INLINE ap_fixed_base& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op2) \ + { \ + *this=operator Sym (op2); \ + return *this; \ + } + + OP_AP_SHIFT_AP_ASSIGN_AF(<<) + OP_AP_SHIFT_AP_ASSIGN_AF(>>) + + ///Support shift(ap_fixed_base) +#define OP_AP_SHIFT_AF(Sym) \ + template \ + INLINE ap_fixed_base operator Sym (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const \ + { \ + return operator Sym (op2.to_ap_private()); \ + } \ + template \ + INLINE ap_fixed_base& operator Sym##= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) \ + { \ + *this=operator Sym (op2); \ + return *this; \ + } + + OP_AP_SHIFT_AF(<<) + OP_AP_SHIFT_AF(>>) + + INLINE ap_fixed_base& operator >>= (unsigned int sh) { + *this = operator >> (sh); + return *this; + } + + INLINE ap_fixed_base& operator <<= (unsigned int sh) { + *this = operator << (sh); + return *this; + } + + INLINE ap_fixed_base& operator >>= (int sh) { + *this = operator >> (sh); + return *this; + } + + INLINE ap_fixed_base& operator <<= (int sh) { + *this = operator << (sh); + return *this; + } + + ///Comparisons + //------------------------------------------------------------------------- + template + INLINE bool operator == (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator != (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this==op2); + } + + template + INLINE bool operator > (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator <= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this>op2); + } + + template + INLINE bool operator < (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + enum {_AP_F=_AP_W-_AP_I,F2=_AP_W2-_AP_I2, shAmt1 = AP_MAX(F2-_AP_F, 0), shAmt2 = AP_MAX(_AP_F-F2,0), _AP_W3 = (_AP_F==F2) ? AP_MAX(_AP_W,_AP_W2) : AP_MAX(_AP_W+shAmt1, _AP_W2+shAmt2)}; + ap_private<_AP_W3, _AP_S > OP1= ap_private<_AP_W3, _AP_S >(V)< OP2=ap_private<_AP_W3,_AP_S2 >(op2.V)< + INLINE bool operator >= (const ap_fixed_base<_AP_W2,_AP_I2,_AP_S2,_AP_Q2,_AP_O2, _AP_N2>& op2) const { + return !(*this) + DOUBLE_CMP_AF(>=) + DOUBLE_CMP_AF(<) + DOUBLE_CMP_AF(<=) + + // Bit and Slice Select + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> operator [] (unsigned int index) { + assert(index<_AP_W&&"Attemping to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index); + } + + INLINE af_bit_ref<_AP_W, _AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> bit(unsigned int index) { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index); + } + + template + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> bit (const ap_private<_AP_W2,_AP_S2>& index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index.to_int()); + } + + INLINE bool bit (unsigned int index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index]; + } + + INLINE bool operator [] (unsigned int index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index]; + } + + template + INLINE bool bit (const ap_private<_AP_W2, _AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index.to_uint()]; + } + + template + INLINE bool operator [] (const ap_private<_AP_W2, _AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return V[index.to_uint()]; + } + + INLINE af_bit_ref<_AP_W, _AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> get_bit(int index) { + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + assert(index >= _AP_I - _AP_W&& "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index + _AP_W - _AP_I); + } + + template + INLINE af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N> get_bit (const ap_private<_AP_W2, true>& index) { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return af_bit_ref<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(this, index.to_int() + _AP_W - _AP_I); + } + + INLINE bool get_bit (int index) const { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return V[index + _AP_W - _AP_I]; + } + + template + INLINE bool get_bit (const ap_private<_AP_W2, true>& index) const { + assert(index >= _AP_I - _AP_W && "Attempting to read bit with negative index"); + assert(index < _AP_I && "Attempting to read bit beyond MSB"); + return V[index.to_int() + _AP_W - _AP_I]; + } + + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range(int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W) &&"Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(const_cast(this), Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + operator () (int Hi, int Lo) const { + return this->range(Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(this, Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + range(const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>(const_cast< + ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>*>(this), + Hi, Lo); + } + + template + INLINE af_range_ref<_AP_W,_AP_I,_AP_S, _AP_Q, _AP_O, _AP_N> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range() { + return this->range(_AP_W - 1, 0); + } + + INLINE af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> + range() const { + return this->range(_AP_W - 1, 0); + } + + INLINE bool is_zero () const { + return V.isMinValue(); + } + + INLINE bool is_neg () const { + if (V.isNegative()) + return true; + return false; + } + + INLINE int wl () const { + return _AP_W; + } + + INLINE int iwl () const { + return _AP_I; + } + + INLINE ap_q_mode q_mode () const { + return _AP_Q; + } + + INLINE ap_o_mode o_mode () const { + return _AP_O; + } + + INLINE int n_bits () const { + return 0; + } + + //private: +public: + ap_private<_AP_W, _AP_S> V; +}; + +template +std::string ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>::to_string( + uint8_t radix, bool sign) const { + std::string str; + str.clear(); + char step; + std::string prefix; + switch (radix) { + case 2 : prefix = "0b"; step = 1; break; + case 8 : prefix = "0o"; step = 3; break; + case 16 : prefix = "0x"; step = 4; break; + default : break; + } + if (_AP_W <= _AP_I) + str = this->to_ap_private().to_string(radix, + radix == 10 ? _AP_S : sign); + else { + if (radix == 10) { + bool isNeg = _AP_S && V.isNegative(); + if (_AP_I > 0) { + ap_private int_part(0); + int_part = this->to_ap_private(); + str += int_part.to_string(radix, false); + } else { + if (isNeg) str += '-'; + } + ap_fixed_base<_AP_W, _AP_I, _AP_S> tmp(*this); + if (isNeg && _AP_I <= 0) tmp = -tmp; + ap_fixed_base<_AP_W - AP_MIN(_AP_I, 0), 0, false> frac_part = tmp; + if (frac_part == 0) return str; + str += "."; + while (frac_part != 0) { + char digit = (frac_part * radix).to_ap_private(); + str += static_cast(digit + '0'); + frac_part *= radix; + } + } else { + if (_AP_I > 0) { + for (signed i = _AP_W - _AP_I; i < _AP_W; i += step) { + + char digit = (char)(this->range(AP_MIN(i + step - 1, _AP_W - 1), i)); + str = (digit < 10 ? static_cast(digit + '0') : + static_cast(digit - 10 + 'a')) + str; + } + } + str += '.'; + ap_fixed_base tmp(*this); + for (signed i = _AP_W - _AP_I - 1; i >= 0; i -= step) { + char digit = (char)(tmp.range(i, AP_MAX(0, i - step + 1))); + str += digit < 10 ? static_cast(digit + '0') : + static_cast(digit - 10 + 'a'); + } + } + } + str = prefix + str; + return str; +} + +template +INLINE void b_not(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op) { + ret.V = op.V; + ret.V.flip(); +} + +template +INLINE void b_and(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V & op2.V; +} + +template +INLINE void b_or(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V | op2.V; +} + +template +INLINE void b_xor(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op1, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op2) { + ret.V = op1.V ^ op2.V; +} + +template +INLINE void neg(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op) { + ap_fixed_base<_AP_W2+!_AP_S2, _AP_I2+!_AP_S2, true, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp.V = - op.V; + ret = Tmp; +} + +template +INLINE void neg(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op) { + ret.V = -op.V; +} + +template +INLINE void lshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op, + int i) { + ap_fixed_base<_AP_W2 - _AP_I2 + AP_MAX(_AP_I, _AP_I2), AP_MAX(_AP_I, _AP_I2), _AP_S2, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp = op; + Tmp.V <<= i; + ret = Tmp; +} + +template +INLINE void lshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op, + int i) { + ret.V = op.V << i; +} + +template +INLINE void rshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2>& op, + int i) { + ap_fixed_base<_AP_I2 + AP_MAX(_AP_W - _AP_I, _AP_W2 - _AP_I2), _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> Tmp; + Tmp = op; + Tmp.V = _AP_S2 ? Tmp.V.ashr(i): Tmp.V.lshr(i); + ret = Tmp; +} + +template +INLINE void rshift(ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& ret, + const ap_fixed_base<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N>& op, + int i) { + ret.V = _AP_S ? op.V.ashr(i): op.V.lshr(i); +} + +#define AF_CTOR_SPEC_BASE(_AP_W,_AP_S,C_TYPE) \ + template<> INLINE ap_fixed_base<_AP_W,_AP_W,_AP_S,AP_TRN,AP_WRAP>::ap_fixed_base(C_TYPE i_op):V(i_op) \ + { \ + } + +#define AF_CTOR_SPEC(__W,C_TYPE) \ + AF_CTOR_SPEC_BASE(__W,true,C_TYPE) \ + AF_CTOR_SPEC_BASE(__W,false,C_TYPE) + +AF_CTOR_SPEC(1,bool) +AF_CTOR_SPEC(8, signed char) +AF_CTOR_SPEC(8, unsigned char) +AF_CTOR_SPEC(16, signed short) +AF_CTOR_SPEC(16, unsigned short) +AF_CTOR_SPEC(32, signed int) +AF_CTOR_SPEC(32, unsigned int) +AF_CTOR_SPEC(64, ap_slong) +AF_CTOR_SPEC(64, ap_ulong) + +///Output streaming +//----------------------------------------------------------------------------- +template +INLINE std::ostream& +operator <<(std::ostream& os, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + os << x.to_double(); + return os; +} + +///Input streaming +//----------------------------------------------------------------------------- +template +INLINE std::istream& +operator >> (std::istream& os, ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + double d; + os >> d; + x = ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>(x); + return os; +} + +template +INLINE void print(const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& x) { + ap_private<_AP_W,_AP_S> data=x.V; + if (_AP_I>0) { + const ap_private<_AP_I,_AP_S> p1=data>>(_AP_W-_AP_I); + print(p1); + + } else + printf("0"); + printf("."); + if (_AP_I<_AP_W) { + const ap_private<_AP_W-_AP_I,false> p2=data; + print(p2,false); + } +} + +///Operators mixing Integers with ap_fixed_base +//----------------------------------------------------------------------------- +#if 1 +#define AF_BIN_OP_WITH_INT_SF(BIN_OP,C_TYPE,_AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator BIN_OP(ap_private<_AP_W2,_AP_S2>(i_op)); \ + } +#define AF_BIN_OP_WITH_INT(BIN_OP, C_TYPE, _AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator BIN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } + +#else +#define AF_BIN_OP_WITH_INT_SF(BIN_OP,C_TYPE,_AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op BIN_OP (i_op); \ + } +#define AF_BIN_OP_WITH_INT(BIN_OP, C_TYPE, _AP_W2,_AP_S2,RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.V BIN_OP (i_op<<(_AP_W-_AP_I)); \ + } \ + \ + \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } + +#endif +#if 1 +#define AF_REL_OP_WITH_INT(REL_OP, C_TYPE, _AP_W2,_AP_S2) \ + template \ + INLINE bool operator REL_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.operator REL_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + \ + \ + template \ + INLINE bool operator REL_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator REL_OP (op); \ + } +#else +#define AF_REL_OP_WITH_INT(REL_OP, C_TYPE, _AP_W2,_AP_S2) \ + template \ + INLINE bool operator REL_OP (const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) \ + { \ + return op.V.operator REL_OP (i_op<<(_AP_W-_AP_I)); \ + } \ + \ + \ + template \ + INLINE bool operator REL_OP (C_TYPE i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) \ + { \ + return (i_op<<(_AP_W-_AP_I)) REL_OP (op.V.VAL); \ + } +#endif +#if 1 +#define AF_ASSIGN_OP_WITH_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.operator ASSIGN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } +#define AF_ASSIGN_OP_WITH_INT_SF(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.operator ASSIGN_OP (ap_private<_AP_W2,_AP_S2>(i_op)); \ + } +#else +#define AF_ASSIGN_OP_WITH_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.V.operator ASSIGN_OP (i_op); \ + } +#define AF_ASSIGN_OP_WITH_INT_SF(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, C_TYPE i_op) { \ + return op.V.operator ASSIGN_OP (i_op); \ + } +#endif + +#define AF_OPS_WITH_INT(C_TYPE, WI, SI) \ + AF_BIN_OP_WITH_INT(+, C_TYPE, WI, SI, plus) \ + AF_BIN_OP_WITH_INT(-, C_TYPE, WI, SI, minus) \ + AF_BIN_OP_WITH_INT(*, C_TYPE, WI, SI, mult) \ + AF_BIN_OP_WITH_INT(/, C_TYPE, WI, SI, div) \ + AF_BIN_OP_WITH_INT_SF(>>, C_TYPE, WI, SI, arg1) \ + AF_BIN_OP_WITH_INT_SF(<<, C_TYPE, WI, SI, arg1) \ + AF_BIN_OP_WITH_INT(&, C_TYPE, WI, SI, logic) \ + AF_BIN_OP_WITH_INT(|, C_TYPE, WI, SI, logic) \ + AF_BIN_OP_WITH_INT(^, C_TYPE, WI, SI, logic) \ + \ + AF_REL_OP_WITH_INT(==, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(!=, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(>, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(>=, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(<, C_TYPE, WI, SI) \ + AF_REL_OP_WITH_INT(<=, C_TYPE, WI, SI) \ + \ + AF_ASSIGN_OP_WITH_INT(+=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(-=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(*=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(/=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT_SF(>>=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT_SF(<<=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(&=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(|=, C_TYPE, WI, SI) \ + AF_ASSIGN_OP_WITH_INT(^=, C_TYPE, WI, SI) + +AF_OPS_WITH_INT(bool, 1, false) +AF_OPS_WITH_INT(char, 8, true) +AF_OPS_WITH_INT(signed char, 8, true) +AF_OPS_WITH_INT(unsigned char, 8, false) +AF_OPS_WITH_INT(short, 16, true) +AF_OPS_WITH_INT(unsigned short, 16, false) +AF_OPS_WITH_INT(int, 32, true) +AF_OPS_WITH_INT(unsigned int, 32, false) +# if defined __x86_64__ +AF_OPS_WITH_INT(long, 64, true) +AF_OPS_WITH_INT(unsigned long, 64, false) +# else +AF_OPS_WITH_INT(long, 32, true) +AF_OPS_WITH_INT(unsigned long, 32, false) +# endif +AF_OPS_WITH_INT(ap_slong, 64, true) +AF_OPS_WITH_INT(ap_ulong, 64, false) + +#define AF_BIN_OP_WITH_AP_INT(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>::template RType<_AP_W,_AP_I,_AP_S>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_fixed_base<_AP_W,_AP_I,_AP_S>::template RType<_AP_W2,_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator BIN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } + +#define AF_REL_OP_WITH_AP_INT(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator REL_OP ( ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op).operator REL_OP (op); \ + } + +#define AF_ASSIGN_OP_WITH_AP_INT(ASSIGN_OP) \ + template \ + INLINE ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& operator ASSIGN_OP ( ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op, const ap_private<_AP_W2,_AP_S2>& i_op) { \ + return op.operator ASSIGN_OP (ap_fixed_base<_AP_W2,_AP_W2,_AP_S2>(i_op)); \ + } \ + template \ + INLINE ap_private<_AP_W2,_AP_S2>& operator ASSIGN_OP ( ap_private<_AP_W2,_AP_S2>& i_op, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op) { \ + return i_op.operator ASSIGN_OP (op.to_ap_private()); \ + } + +AF_BIN_OP_WITH_AP_INT(+, plus) +AF_BIN_OP_WITH_AP_INT(-, minus) +AF_BIN_OP_WITH_AP_INT(*, mult) +AF_BIN_OP_WITH_AP_INT(/, div) +AF_BIN_OP_WITH_AP_INT(&, logic) +AF_BIN_OP_WITH_AP_INT(|, logic) +AF_BIN_OP_WITH_AP_INT(^, logic) + +AF_REL_OP_WITH_AP_INT(==) +AF_REL_OP_WITH_AP_INT(!=) +AF_REL_OP_WITH_AP_INT(>) +AF_REL_OP_WITH_AP_INT(>=) +AF_REL_OP_WITH_AP_INT(<) +AF_REL_OP_WITH_AP_INT(<=) + +AF_ASSIGN_OP_WITH_AP_INT(+=) +AF_ASSIGN_OP_WITH_AP_INT(-=) +AF_ASSIGN_OP_WITH_AP_INT(*=) +AF_ASSIGN_OP_WITH_AP_INT(/=) +AF_ASSIGN_OP_WITH_AP_INT(&=) +AF_ASSIGN_OP_WITH_AP_INT(|=) +AF_ASSIGN_OP_WITH_AP_INT(^=) + +#define AF_REF_REL_OP_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ +template \ + INLINE bool operator REL_OP ( const af_range_ref<_AP_W,_AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ +template \ + INLINE bool operator REL_OP ( C_TYPE op2, const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ +template \ + INLINE bool operator REL_OP ( const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, C_TYPE op2) { \ + return (bool(op)) REL_OP op2; \ + } \ +template \ + INLINE bool operator REL_OP ( C_TYPE op2, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2 REL_OP (bool(op)); \ + } + +#define AF_REF_REL_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(>, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(<, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(>=, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(<=, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(==, C_TYPE, _AP_WI, _AP_SI) \ +AF_REF_REL_OP_MIX_INT(!=, C_TYPE, _AP_WI, _AP_SI) + +AF_REF_REL_MIX_INT(bool, 1, false) +AF_REF_REL_MIX_INT(char, 8, true) +AF_REF_REL_MIX_INT(signed char, 8, true) +AF_REF_REL_MIX_INT(unsigned char, 8, false) +AF_REF_REL_MIX_INT(short, 16, true) +AF_REF_REL_MIX_INT(unsigned short, 16, false) +AF_REF_REL_MIX_INT(int, 32, true) +AF_REF_REL_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +AF_REF_REL_MIX_INT(long, 64, true) +AF_REF_REL_MIX_INT(unsigned long, 64, false) +# else +AF_REF_REL_MIX_INT(long, 32, true) +AF_REF_REL_MIX_INT(unsigned long, 32, false) +# endif +AF_REF_REL_MIX_INT(ap_slong, 64, true) +AF_REF_REL_MIX_INT(ap_ulong, 64, false) + +#define AF_REF_REL_OP_AP_INT(REL_OP) \ +template \ + INLINE bool operator REL_OP ( const af_range_ref<_AP_W,_AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, const ap_private<_AP_W2, _AP_S> &op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (op2); \ + } \ +template \ + INLINE bool operator REL_OP (const ap_private<_AP_W2, _AP_S2> &op2, const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2.operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ +template \ + INLINE bool operator REL_OP ( const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op, const ap_private<_AP_W2, _AP_S2> &op2) { \ + return (ap_private<1, false>(op)).operator REL_OP (op2); \ + } \ +template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W2, _AP_S2> &op2, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op) { \ + return op2.operator REL_OP (ap_private<1,false>(op)); \ + } + +AF_REF_REL_OP_AP_INT(>) +AF_REF_REL_OP_AP_INT(<) +AF_REF_REL_OP_AP_INT(>=) +AF_REF_REL_OP_AP_INT(<=) +AF_REF_REL_OP_AP_INT(==) +AF_REF_REL_OP_AP_INT(!=) + +// Relational Operators with double +template +INLINE bool operator == ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator == (op1); +} + +template +INLINE bool operator != ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator != (op1); +} + +template +INLINE bool operator > ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator < (op1); +} + +template +INLINE bool operator >= ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator <= (op1); +} + +template +INLINE bool operator < ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator > (op1); +} + +template +INLINE bool operator <= ( double op1, const ap_fixed_base<_AP_W,_AP_I,_AP_S,_AP_Q,_AP_O, _AP_N>& op2) { + return op2.operator >= (op1); +} + +#endif /* #ifndef __AESL_GCC_AP_FIXED_H__ */ \ No newline at end of file diff --git a/hls/lines256_length128/etc/ap_int_sim.h b/hls/lines256_length128/etc/ap_int_sim.h new file mode 100644 index 0000000000000000000000000000000000000000..887ccd88bfd52d1777db8661d72c57703ccf736a --- /dev/null +++ b/hls/lines256_length128/etc/ap_int_sim.h @@ -0,0 +1,1629 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef __AESL_GCC_AP_INT_H__ +#define __AESL_GCC_AP_INT_H__ + +#ifndef __cplusplus +#error C++ is required to include this header file +#endif /* #ifndef __cplusplus */ + +#undef _AP_DEBUG_ +#include +#include + +// for safety +#if (defined(_AP_N)|| defined(_AP_C)) +#error One or more of the following is defined: _AP_N, _AP_C. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_N)|| defined(_AP_C)) */ + +// for safety +#if (defined(_AP_W) || defined(_AP_I) || defined(_AP_S) || defined(_AP_Q) || defined(_AP_O) || defined(_AP_W2) || defined(_AP_I2) || defined(_AP_S2) || defined(_AP_Q2) || defined(_AP_O2)) +#error One or more of the following is defined: _AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W) || defined(_AP_I) || defined(_AP_S) || defined(_AP_Q) || defined(_AP_O) || defined(_AP_W2) || defined(_AP_I2) || defined(_AP_S2) || defined(_AP_Q2) || defined(_AP_O2)) */ + +//for safety +#if (defined(_AP_W3) || defined(_AP_S3) || defined(_AP_W4) || defined(_AP_S4)) +#error One or more of the following is defined: _AP_W3, _AP_S3, _AP_W4,_AP_S4. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W3) || defined(_AP_S3) || defined(_AP_W4) || defined(_AP_S4)) */ + +//for safety +#if (defined(_AP_W1) || defined(_AP_S1) || defined(_AP_I1) || defined(_AP_T) || defined(_AP_T1) || defined(_AP_T2) || defined(_AP_T3) || defined(_AP_T4)) +#error One or more of the following is defined: _AP_W1, _AP_S1, _AP_I1, _AP_T, _AP_T1, _AP_T2, _AP_T3, _AP_T4. Definition conflicts with their usage as template parameters. +#endif /* #if (defined(_AP_W1) || defined(_AP_S1) || defined(_AP_I1) || defined(_AP_T) || defined(_AP_T1) || defined(_AP_T2) || defined(_AP_T3) || defined(_AP_T4)) */ + +#define __AESL_APDT_IN_SCFLOW__ +#ifndef __AESL_APDT_IN_SCFLOW__ + #include "etc/ap_private.h" +#else + #include "../etc/ap_private.h" +#endif /* #ifndef __AESL_APDT_IN_SCFLOW__ */ + +#ifdef _AP_DEBUG_ + #define AP_DEBUG(s) s +#else + #define AP_DEBUG(s) +#endif /* #ifdef _AP_DEBUG_ */ + +#ifndef __SIMULATION__ + #define __SIMULATION__ +#endif /* #ifndef __SIMULATION__ */ + +#if !(defined SYSTEMC_H) && !(defined SYSTEMC_INCLUDED) + #ifndef SC_TRN + #define SC_TRN AP_TRN + #endif /* #ifndef SC_TRN */ + #ifndef SC_RND + #define SC_RND AP_RND + #endif /* #ifndef SC_RND */ + #ifndef SC_TRN_ZERO + #define SC_TRN_ZERO AP_TRN_ZERO + #endif /* #ifndef SC_TRN_ZERO */ + #ifndef SC_RND_ZERO + #define SC_RND_ZERO AP_RND_ZERO + #endif /* #ifndef SC_RND_ZERO */ + #ifndef SC_RND_INF + #define SC_RND_INF AP_RND_INF + #endif /* #ifndef SC_RND_INF */ + #ifndef SC_RND_MIN_INF + #define SC_RND_MIN_INF AP_RND_MIN_INF + #endif /* #ifndef SC_RND_MIN_INF */ + #ifndef SC_RND_CONV + #define SC_RND_CONV AP_RND_CONV + #endif /* #ifndef SC_RND_CONV */ + #ifndef SC_WRAP + #define SC_WRAP AP_WRAP + #endif /* #ifndef SC_WRAP */ + #ifndef SC_SAT + #define SC_SAT AP_SAT + #endif /* #ifndef SC_SAT */ + #ifndef SC_SAT_ZERO + #define SC_SAT_ZERO AP_SAT_ZERO + #endif /* #ifndef SC_SAT_ZERO */ + #ifndef SC_SAT_SYM + #define SC_SAT_SYM AP_SAT_SYM + #endif /* #ifndef SC_SAT_SYM */ + #ifndef SC_WRAP_SM + #define SC_WRAP_SM AP_WRAP_SM + #endif /* #ifndef SC_WRAP_SM */ + #ifndef SC_BIN + #define SC_BIN AP_BIN + #endif /* #ifndef SC_BIN */ + #ifndef SC_OCT + #define SC_OCT AP_OCT + #endif /* #ifndef SC_OCT */ + #ifndef SC_DEC + #define SC_DEC AP_DEC + #endif /* #ifndef SC_DEC */ + #ifndef SC_HEX + #define SC_HEX AP_HEX + #endif /* #ifndef SC_HEX */ +#endif /* #if !(defined SYSTEMC_H) && !(defined SYSTEMC_INCLUDED) */ +#ifndef AP_INT_MAX_W +#define AP_INT_MAX_W 1024 +#endif +#define BIT_WIDTH_UPPER_LIMIT (1 << 15) +#if AP_INT_MAX_W > BIT_WIDTH_UPPER_LIMIT +#error "Bitwidth exceeds 32768 (1 << 15), the maximum allowed value" +#endif +#define MAX_MODE(BITS) ((BITS + 1023) / 1024) + +///Forward declaration +template struct ap_range_ref; +template struct ap_bit_ref; + +template struct ap_fixed_base; +template struct af_range_ref; +template struct af_bit_ref; +template class ap_uint; + +enum { + AP_BIN = 2, + AP_OCT = 8, + AP_DEC = 10, + AP_HEX = 16 +}; + +///Why to use reference? +///Because we will operate the original object indirectly by operating the +///result object directly after concating or part selecting + +///Proxy class which allows concatination to be used as rvalue(for reading) and +//lvalue(for writing) + +/// Concatination reference. +// ---------------------------------------------------------------- +template +struct ap_concat_ref { +#ifdef _MSC_VER + #pragma warning(disable: 4521 4522) +#endif /* #ifdef _MSC_VER */ + enum {_AP_WR=_AP_W1+_AP_W2,}; + _AP_T1& mbv1; + _AP_T2& mbv2; + + INLINE ap_concat_ref(const ap_concat_ref<_AP_W1, _AP_T1, + _AP_W2, _AP_T2>& ref): + mbv1(ref.mbv1), mbv2(ref.mbv2) {} + + INLINE ap_concat_ref(_AP_T1& bv1, _AP_T2& bv2):mbv1(bv1),mbv2(bv2) {} + + template + INLINE ap_concat_ref& operator = (const ap_private<_AP_W3,_AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> vval(val); + int W_ref1=mbv1.length(); + int W_ref2=mbv2.length(); + ap_private<_AP_W1,false> mask1(-1); + mask1>>=_AP_W1-W_ref1; + ap_private<_AP_W2,false> mask2(-1); + mask2>>=_AP_W2-W_ref2; + mbv1.set(ap_private<_AP_W1,false>((vval>>W_ref2)&mask1)); + mbv2.set(ap_private<_AP_W2,false>(vval&mask2)); + return *this; + } + + INLINE ap_concat_ref& operator = (unsigned long long val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + INLINE ap_concat_ref& operator = + (const ap_concat_ref <_AP_W1, _AP_T1, _AP_W2, _AP_T2>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator =(const ap_bit_ref<_AP_W3, _AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_concat_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) { + ap_private<_AP_W1+_AP_W2, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE ap_concat_ref& operator= (const af_range_ref<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator = ((const ap_private<_AP_W3, false>)(val)); + } + + template + INLINE ap_concat_ref& operator= (const ap_fixed_base<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator = (val.to_ap_private()); + } + + template + INLINE ap_concat_ref& operator= (const af_bit_ref<_AP_W3, _AP_I3, _AP_S3, + _AP_Q3, _AP_O3, _AP_N3>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + + INLINE operator ap_private<_AP_WR, false> () const { + return get(); + } + + INLINE operator unsigned long long () const { + return get().to_uint64(); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_range_ref<_AP_W3, _AP_S3> > + operator, (const ap_range_ref<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_range_ref<_AP_W3, _AP_S3> >(*this, + const_cast &>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_private<_AP_W3, _AP_S3> > + operator, (ap_private<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_private<_AP_W3, _AP_S3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, ap_private<_AP_W3, _AP_S3> > + operator, (const ap_private<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3, ap_private<_AP_W3, _AP_S3> >(*this, + const_cast&>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, 1, ap_bit_ref<_AP_W3, _AP_S3> > + operator, (const ap_bit_ref<_AP_W3, _AP_S3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, + 1, ap_bit_ref<_AP_W3, _AP_S3> >(*this, + const_cast &>(a2)); + } + + template + INLINE ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3+_AP_W4, ap_concat_ref<_AP_W3,_AP_T3,_AP_W4,_AP_T4> > + operator, (const ap_concat_ref<_AP_W3,_AP_T3,_AP_W4,_AP_T4> &a2) + { + return ap_concat_ref<_AP_WR, ap_concat_ref, + _AP_W3+_AP_W4, ap_concat_ref<_AP_W3,_AP_T3,_AP_W4, + _AP_T4> >(*this, const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, af_range_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> > + operator, (const af_range_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, + _AP_O3, _AP_N3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, _AP_W3, af_range_ref<_AP_W3, + _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_WR, ap_concat_ref, 1, af_bit_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> > + operator, (const af_bit_ref<_AP_W3, _AP_I3, _AP_S3, _AP_Q3, + _AP_O3, _AP_N3> &a2) { + return ap_concat_ref<_AP_WR, ap_concat_ref, 1, af_bit_ref<_AP_W3, + _AP_I3, _AP_S3, _AP_Q3, _AP_O3, _AP_N3> >(*this, + const_cast& >(a2)); + } + + template + INLINE ap_private + operator & (const ap_private<_AP_W3,_AP_S3>& a2) { + return get() & a2; + } + + + template + INLINE ap_private + operator | (const ap_private<_AP_W3,_AP_S3>& a2) { + return get() | a2; + } + + + template + INLINE ap_private + operator ^ (const ap_private<_AP_W3,_AP_S3>& a2) { + return ap_private(get() ^ a2); + } + + INLINE const ap_private<_AP_WR, false> get() const + { + ap_private<_AP_W1+_AP_W2, false> tmpVal = ap_private<_AP_W1+_AP_W2, false> (mbv1.get()); + ap_private<_AP_W1+_AP_W2, false> tmpVal2 = ap_private<_AP_W1+_AP_W2, false> (mbv2.get()); + int W_ref2 = mbv2.length(); + tmpVal <<= W_ref2; + tmpVal |= tmpVal2; + return tmpVal; + } + + INLINE const ap_private<_AP_WR, false> get() { + ap_private<_AP_W1+_AP_W2, false> tmpVal = ap_private<_AP_W1+_AP_W2, false> ( mbv1.get()); + ap_private<_AP_W1+_AP_W2, false> tmpVal2 = ap_private<_AP_W1+_AP_W2, false> (mbv2.get()); + int W_ref2 = mbv2.length(); + tmpVal <<= W_ref2; + tmpVal |= tmpVal2; + return tmpVal; + } + + template + INLINE void set(const ap_private<_AP_W3,false> & val) { + ap_private<_AP_W1+_AP_W2, false> vval(val); + int W_ref1=mbv1.length(); + int W_ref2=mbv2.length(); + ap_private<_AP_W1,false> mask1(-1); + mask1>>=_AP_W1-W_ref1; + ap_private<_AP_W2,false> mask2(-1); + mask2>>=_AP_W2-W_ref2; + mbv1.set(ap_private<_AP_W1,false>((vval>>W_ref2)&mask1)); + mbv2.set(ap_private<_AP_W2,false>(vval&mask2)); + } + + INLINE int length() const { + return mbv1.length()+mbv2.length(); + } + + INLINE std::string to_string(uint8_t radix=2) const { + return get().to_string(radix); + } +}; + +///Proxy class, which allows part selection to be used as rvalue(for reading) and +//lvalue(for writing) + +///Range(slice) reference +//------------------------------------------------------------ +template +struct ap_range_ref { +#ifdef _MSC_VER + #pragma warning( disable : 4521 4522 ) +#endif /* #ifdef _MSC_VER */ + ap_private<_AP_W,_AP_S> &d_bv; + int l_index; + int h_index; + +public: + INLINE ap_range_ref(const ap_range_ref<_AP_W, _AP_S>& ref): + d_bv(ref.d_bv), l_index(ref.l_index), h_index(ref.h_index) {} + + INLINE ap_range_ref(ap_private<_AP_W,_AP_S>* bv, int h, int l): + d_bv(*bv),l_index(l),h_index(h) { + //if (h < l) + //fprintf(stderr, "Warning! The bits selected will be returned in reverse order\n"); + } + + INLINE operator ap_private<_AP_W, false> () const { + ap_private<_AP_W, false> val(0); + if(h_index>=l_index) { + if (_AP_W > 64) { + val=d_bv; + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val>>=l_index; + val&=mask; + } else { + const static uint64_t mask = (~0ULL>> (64>_AP_W ? (64-_AP_W):0)); + val = (d_bv >> l_index) & (mask >>(_AP_W-(h_index-l_index+1))); + } + } else { + for(int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } + return val; + } + + INLINE operator unsigned long long () const { + return to_uint64(); + } + + template + INLINE ap_range_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) { + ap_private<_AP_W,false> vval=ap_private<_AP_W,false>(val); + if (l_index>h_index) { + for (int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + (vval)[i]? d_bv.set(j):d_bv.clear(j); + } else { + if (_AP_W > 64) { + ap_private<_AP_W,false> mask(-1); + if (l_index>0) { + mask<<=l_index; + vval<<=l_index; + } + if(h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask2(-1); + mask2>>=_AP_W-h_index-1; + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } else { + unsigned shift = 64-_AP_W; + uint64_t mask = ~0ULL>>(shift); + if(l_index>0) + { + vval = mask & vval << l_index; + mask = mask & mask << l_index; + } + if(h_index<_AP_W-1) + { + uint64_t mask2 = mask; + mask2 >>= (_AP_W-h_index-1); + mask&=mask2; + vval&=mask2; + } + mask = ~mask; + d_bv&=mask; + d_bv|=vval; + } + } + return *this; + } + + INLINE ap_range_ref& operator = (unsigned long long val) + { + const ap_private<_AP_W,_AP_S> vval=val; + return operator = (vval); + } + + + INLINE ap_range_ref& operator =(const ap_range_ref<_AP_W, _AP_S>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + + + template + INLINE ap_range_ref& operator = + (const ap_concat_ref <_AP_W3, _AP_T3, _AP_W4, _AP_T4>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator = (tmpVal); + } + + template + INLINE ap_range_ref& operator =(const ap_range_ref<_AP_W3,_AP_S3>& val) + { + const ap_private<_AP_W, false> tmpVal(val); + return operator =(tmpVal); + } + + template + INLINE ap_range_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, _AP_S2>)(val)); + } + + template + INLINE ap_range_ref& operator= (const ap_fixed_base<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=(val.to_ap_private()); + } + + template + INLINE ap_range_ref& operator= (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE ap_range_ref& operator= (const ap_bit_ref<_AP_W2, _AP_S2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE + ap_concat_ref<_AP_W,ap_range_ref,_AP_W2,ap_range_ref<_AP_W2,_AP_S2> > + operator, (const ap_range_ref<_AP_W2,_AP_S2> &a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, + ap_range_ref<_AP_W2,_AP_S2> >(*this, + const_cast& >(a2)); + } + + + template + INLINE ap_concat_ref<_AP_W,ap_range_ref,_AP_W2,ap_private<_AP_W2,_AP_S2> > + operator , (ap_private<_AP_W2,_AP_S2>& a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + INLINE ap_concat_ref<_AP_W,ap_range_ref,_AP_W,ap_private<_AP_W,_AP_S> > + operator , (ap_private<_AP_W, _AP_S>& a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, _AP_W, + ap_private<_AP_W,_AP_S> >(*this, a2); + } + + + + template + INLINE + ap_concat_ref<_AP_W,ap_range_ref,1,ap_bit_ref<_AP_W2,_AP_S2> > + operator, (const ap_bit_ref<_AP_W2,_AP_S2> &a2) + { + return + ap_concat_ref<_AP_W, ap_range_ref, 1, + ap_bit_ref<_AP_W2,_AP_S2> >(*this, const_cast& >(a2)); + } + + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) + { + return ap_concat_ref<_AP_W, ap_range_ref, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_range_ref, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_range_ref, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_range_ref, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE bool operator == (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs==rhs; + } + + + template + INLINE bool operator != (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs!=rhs; + } + + + template + INLINE bool operator > (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>rhs; + } + + + template + INLINE bool operator >= (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs>=rhs; + } + + + template + INLINE bool operator < (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs + INLINE bool operator <= (const ap_range_ref<_AP_W2, _AP_S2>& op2) + { + ap_private<_AP_W,false> lhs=get(); + ap_private<_AP_W2,false> rhs=op2.get(); + return lhs<=rhs; + } + + + template + INLINE void set(const ap_private<_AP_W2,false>& val) + { + ap_private<_AP_W,_AP_S> vval=val; + if(l_index>h_index) + { + for(int i=0, j=l_index;j>=0&&j>=h_index;j--,i++) + (vval)[i]? d_bv.set(j):d_bv.clear(j); + } else { + if (_AP_W>64 ) { + ap_private<_AP_W,_AP_S> mask(-1); + if(l_index>0) + { + ap_private<_AP_W,false> mask1(-1); + mask1>>=_AP_W-l_index; + mask1.flip(); + mask=mask1; + //vval&=mask1; + vval<<=l_index; + } + if(h_index<_AP_W-1) + { + ap_private<_AP_W,false> mask2(-1); + mask2<<=h_index+1; + mask2.flip(); + mask&=mask2; + vval&=mask2; + } + mask.flip(); + d_bv&=mask; + d_bv|=vval; + } else { + uint64_t mask = ~0ULL >> (64-_AP_W); + if(l_index>0) + { + uint64_t mask1 = mask; + mask1=mask & (mask1>>(_AP_W-l_index)); + vval =mask&( vval <> (64-_AP_W); + mask2 = mask &(mask2<<(h_index+1)); + mask&=~mask2; + vval&=~mask2; + } + d_bv&=(~mask&(~0ULL >> (64-_AP_W))); + d_bv|=vval; + } + } + } + + + INLINE ap_private<_AP_W,false> get() const + { + ap_private<_AP_W,false> val(0); + if(h_index=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } else { + val=d_bv; + val>>=l_index; + if(h_index<_AP_W-1) + { + if (_AP_W <= 64) { + const static uint64_t mask = (~0ULL>> (64>_AP_W ? (64-_AP_W):0)); + val &= (mask>> (_AP_W-(h_index-l_index+1))); + } else { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + } + } + return val; + } + + + INLINE ap_private<_AP_W,false> get() + { + ap_private<_AP_W,false> val(0); + if(h_index=0&&j>=h_index;j--,i++) + if((d_bv)[j]) val.set(i); + } else { + val=d_bv; + val>>=l_index; + if(h_index<_AP_W-1) + { + if (_AP_W <= 64 ) { + static const uint64_t mask = ~0ULL>> (64>_AP_W ? (64-_AP_W):0); + return val &= ( (mask) >> (_AP_W - (h_index-l_index+1))); + } else { + ap_private<_AP_W,false> mask(-1); + mask>>=_AP_W-(h_index-l_index+1); + val&=mask; + } + } + } + return val; + } + + + INLINE int length() const + { + return h_index>=l_index?h_index-l_index+1:l_index-h_index+1; + } + + + INLINE int to_int() const + { + ap_private<_AP_W,false> val=get(); + return val.to_int(); + } + + + INLINE unsigned int to_uint() const + { + ap_private<_AP_W,false> val=get(); + return val.to_uint(); + } + + + INLINE long to_long() const + { + ap_private<_AP_W,false> val=get(); + return val.to_long(); + } + + + INLINE unsigned long to_ulong() const + { + ap_private<_AP_W,false> val=get(); + return val.to_ulong(); + } + + + INLINE ap_slong to_int64() const + { + ap_private<_AP_W,false> val=get(); + return val.to_int64(); + } + + + INLINE ap_ulong to_uint64() const + { + ap_private<_AP_W,false> val=get(); + return val.to_uint64(); + } + + INLINE std::string to_string(uint8_t radix=2) const { + return get().to_string(radix); + } + +}; + +///Proxy class, which allows bit selection to be used as rvalue(for reading) and +//lvalue(for writing) + +///Bit reference +//-------------------------------------------------------------- +template +struct ap_bit_ref { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif + ap_private<_AP_W,_AP_S>& d_bv; + int d_index; + +public: + INLINE ap_bit_ref(const ap_bit_ref<_AP_W, _AP_S>& ref): + d_bv(ref.d_bv), d_index(ref.d_index) {} + + INLINE ap_bit_ref(ap_private<_AP_W,_AP_S>& bv, int index=0): + d_bv(bv),d_index(index) + { +#ifdef _AP_DEBUG_ + assert(d_index<_AP_W&&"index out of bound"); +#endif + } + + + INLINE operator bool () const + { + return d_bv.get_bit(d_index); + } + + + INLINE bool to_bool() const + { + return operator bool (); + } + + + INLINE ap_bit_ref& operator = (unsigned long long val) + { + if(val) + d_bv.set(d_index); + else + d_bv.clear(d_index); + return *this; + } + + +#if 0 + INLINE ap_bit_ref& operator = (bool val) + { + if(val) + d_bv.set(d_index); + else + d_bv.clear(d_index); + return *this; + } +#endif + template + INLINE ap_bit_ref& operator =(const ap_private<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(val != 0)); + } + + + template + INLINE ap_bit_ref& operator =(const ap_bit_ref<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(bool)val); + } + + INLINE ap_bit_ref& operator =(const ap_bit_ref<_AP_W,_AP_S>& val) + { + return operator =((unsigned long long)(bool)val); + } + + template + INLINE ap_bit_ref& operator =(const ap_range_ref<_AP_W2,_AP_S2>& val) + { + return operator =((unsigned long long)(bool) val); + } + + + template + INLINE ap_bit_ref& operator= (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((const ap_private<_AP_W2, false>)(val)); + } + + template + INLINE ap_bit_ref& operator= (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2>& val) { + return operator=((unsigned long long)(bool)(val)); + } + + template + INLINE ap_bit_ref& operator= (const ap_concat_ref<_AP_W2, _AP_T3, _AP_W3, _AP_T3>& val) { + return operator=((const ap_private<_AP_W2 + _AP_W3, false>)(val)); + } + + + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> > + operator , (ap_private<_AP_W2, _AP_S2>& a2) + { + return ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_range_ref<_AP_W2,_AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) + { + return + ap_concat_ref<1, ap_bit_ref, _AP_W2, ap_range_ref<_AP_W2,_AP_S2> >(*this, + const_cast &>(a2)); + } + + + template + INLINE ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref<_AP_W2,_AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) + { + return + ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref<_AP_W2,_AP_S2> >(*this, + const_cast &>(a2)); + } + + + INLINE ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref > + operator, (const ap_bit_ref &a2) + { + return + ap_concat_ref<1, ap_bit_ref, 1, ap_bit_ref >(*this, + const_cast(a2)); + } + + + template + INLINE ap_concat_ref<1, ap_bit_ref, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> > + operator, (const ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> &a2) + { + return + ap_concat_ref<1,ap_bit_ref,_AP_W2+_AP_W3, + ap_concat_ref<_AP_W2,_AP_T2,_AP_W3,_AP_T3> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<1, ap_bit_ref, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<1, ap_bit_ref, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<1, ap_bit_ref, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<1, ap_bit_ref, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, + const_cast& >(a2)); + } + + template + INLINE bool operator == (const ap_bit_ref<_AP_W2, _AP_S2>& op) { + return get() == op.get(); + } + + template + INLINE bool operator != (const ap_bit_ref<_AP_W2, _AP_S2>& op) { + return get() != op.get(); + } + + + INLINE bool get() const + { + return operator bool (); + } + + + INLINE bool get() + { + return operator bool (); + } + + + template + INLINE void set(const ap_private<_AP_W3, false>& val) + { + operator = (val); + } + + INLINE bool operator ~ () const { + bool bit = (d_bv)[d_index]; + return bit ? false : true; + } + + INLINE int length() const { return 1; } + + INLINE std::string to_string() const { + bool val = get(); + return val ? "1" : "0"; + } +}; + +/// Operators mixing Integers with AP_Int +// ---------------------------------------------------------------- +#if 1 +#define OP_BIN_MIX_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + INLINE typename ap_private<_AP_WI,_AP_SI>::template RType<_AP_W,_AP_S>::RTYPE \ + operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_WI,_AP_SI>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (ap_private<_AP_WI,_AP_SI>(i_op)); \ + } +#else +#define OP_BIN_MIX_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + INLINE typename ap_private<_AP_WI,_AP_SI>::template RType<_AP_W,_AP_S>::RTYPE \ + operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_WI,_AP_SI>(i_op).operator BIN_OP (op); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (ap_private<_AP_WI,_AP_SI>(i_op)); \ + } +#endif +#define OP_REL_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return op.operator REL_OP (ap_private<_AP_W2, _AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_private<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (op); \ + } +#define OP_ASSIGN_MIX_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return op.operator ASSIGN_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } + +#define OP_BIN_SHIFT_INT(BIN_OP, C_TYPE, _AP_WI, _AP_SI, RTYPE) \ + template \ + C_TYPE operator BIN_OP ( C_TYPE i_op, const ap_private<_AP_W,_AP_S> &op) { \ + return i_op BIN_OP (op.getVal()); \ + } \ + template \ + INLINE typename ap_private<_AP_W,_AP_S>::template RType<_AP_WI,_AP_SI>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W,_AP_S> &op, C_TYPE i_op) { \ + return op.operator BIN_OP (i_op); \ + } +#define OP_ASSIGN_RSHIFT_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + op = op.operator >> (op2); \ + return op; \ + } +#define OP_ASSIGN_LSHIFT_INT(ASSIGN_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE ap_private<_AP_W,_AP_S> &operator ASSIGN_OP ( ap_private<_AP_W,_AP_S> &op, C_TYPE op2) { \ + op = op.operator << (op2); \ + return op; \ + } + +#define OPS_MIX_INT(C_TYPE, WI, SI) \ + OP_BIN_MIX_INT(*, C_TYPE, WI, SI, mult) \ + OP_BIN_MIX_INT(+, C_TYPE, WI, SI, plus) \ + OP_BIN_MIX_INT(-, C_TYPE, WI, SI, minus) \ + OP_BIN_MIX_INT(/, C_TYPE, WI, SI, div) \ + OP_BIN_MIX_INT(%, C_TYPE, WI, SI, mod) \ + OP_BIN_MIX_INT(&, C_TYPE, WI, SI, logic) \ + OP_BIN_MIX_INT(|, C_TYPE, WI, SI, logic) \ + OP_BIN_MIX_INT(^, C_TYPE, WI, SI, logic) \ + OP_BIN_SHIFT_INT(>>, C_TYPE, WI, SI, arg1) \ + OP_BIN_SHIFT_INT(<<, C_TYPE, WI, SI, arg1) \ + \ + OP_REL_MIX_INT(==, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(!=, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(>, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(>=, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(<, C_TYPE, WI, SI) \ + OP_REL_MIX_INT(<=, C_TYPE, WI, SI) \ + \ + OP_ASSIGN_MIX_INT(+=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(-=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(*=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(/=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(%=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(&=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(|=, C_TYPE, WI, SI) \ + OP_ASSIGN_MIX_INT(^=, C_TYPE, WI, SI) \ + OP_ASSIGN_RSHIFT_INT(>>=, C_TYPE, WI, SI) \ + OP_ASSIGN_LSHIFT_INT(<<=, C_TYPE, WI, SI) + + +OPS_MIX_INT(bool, 1, false) +OPS_MIX_INT(char, 8, true) +OPS_MIX_INT(signed char, 8, true) +OPS_MIX_INT(unsigned char, 8, false) +OPS_MIX_INT(short, 16, true) +OPS_MIX_INT(unsigned short, 16, false) +OPS_MIX_INT(int, 32, true) +OPS_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +OPS_MIX_INT(long, 64, true) +OPS_MIX_INT(unsigned long, 64, false) +# else +OPS_MIX_INT(long, 32, true) +OPS_MIX_INT(unsigned long, 32, false) +# endif +OPS_MIX_INT(ap_slong, 64, true) +OPS_MIX_INT(ap_ulong, 64, false) + +#define OP_BIN_MIX_RANGE(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_range_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1, false>(op1).operator BIN_OP (op2); \ + } \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator BIN_OP (ap_private<_AP_W2, false>(op2)); \ + } + +#define OP_REL_MIX_RANGE(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_range_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1,false>(op1).operator REL_OP (op2); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator REL_OP (op2.operator ap_private<_AP_W2, false>()); \ + } + +#define OP_ASSIGN_MIX_RANGE(ASSIGN_OP) \ + template \ + INLINE ap_private<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_private<_AP_W1,_AP_S1>& op1, const ap_range_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator ASSIGN_OP (ap_private<_AP_W2, false>(op2)); \ + } \ + template \ + INLINE ap_range_ref<_AP_W1,_AP_S1>& operator ASSIGN_OP (ap_range_ref<_AP_W1,_AP_S1>& op1, ap_private<_AP_W2,_AP_S2>& op2) { \ + ap_private<_AP_W1, false> tmp(op1); \ + tmp.operator ASSIGN_OP (op2); \ + op1 = tmp; \ + return op1; \ + } + + +OP_ASSIGN_MIX_RANGE(+=) +OP_ASSIGN_MIX_RANGE(-=) +OP_ASSIGN_MIX_RANGE(*=) +OP_ASSIGN_MIX_RANGE(/=) +OP_ASSIGN_MIX_RANGE(%=) +OP_ASSIGN_MIX_RANGE(>>=) +OP_ASSIGN_MIX_RANGE(<<=) +OP_ASSIGN_MIX_RANGE(&=) +OP_ASSIGN_MIX_RANGE(|=) +OP_ASSIGN_MIX_RANGE(^=) + +OP_REL_MIX_RANGE(==) +OP_REL_MIX_RANGE(!=) +OP_REL_MIX_RANGE(>) +OP_REL_MIX_RANGE(>=) +OP_REL_MIX_RANGE(<) +OP_REL_MIX_RANGE(<=) + +OP_BIN_MIX_RANGE(+, plus) +OP_BIN_MIX_RANGE(-, minus) +OP_BIN_MIX_RANGE(*, mult) +OP_BIN_MIX_RANGE(/, div) +OP_BIN_MIX_RANGE(%, mod) +OP_BIN_MIX_RANGE(>>, arg1) +OP_BIN_MIX_RANGE(<<, arg1) +OP_BIN_MIX_RANGE(&, logic) +OP_BIN_MIX_RANGE(|, logic) +OP_BIN_MIX_RANGE(^, logic) + +#define OP_BIN_MIX_BIT(BIN_OP, RTYPE) \ + template \ + INLINE typename ap_private<1, false>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_bit_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<1, false>(op1).operator BIN_OP (op2); \ + } \ + template \ + INLINE typename ap_private<_AP_W1,_AP_S1>::template RType<1,false>::RTYPE \ + operator BIN_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator BIN_OP (ap_private<1, false>(op2)); \ + } + +#define OP_REL_MIX_BIT(REL_OP) \ + template \ + INLINE bool operator REL_OP ( const ap_bit_ref<_AP_W1,_AP_S1>& op1, const ap_private<_AP_W2,_AP_S2>& op2) { \ + return ap_private<_AP_W1,false>(op1).operator REL_OP (op2); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_private<_AP_W1,_AP_S1>& op1, const ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator REL_OP (ap_private<1, false>(op2)); \ + } + +#define OP_ASSIGN_MIX_BIT(ASSIGN_OP) \ + template \ + INLINE ap_private<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_private<_AP_W1,_AP_S1>& op1, ap_bit_ref<_AP_W2,_AP_S2>& op2) { \ + return op1.operator ASSIGN_OP (ap_private<1, false>(op2)); \ + } \ + template \ + INLINE ap_bit_ref<_AP_W1,_AP_S1>& operator ASSIGN_OP ( ap_bit_ref<_AP_W1,_AP_S1>& op1, ap_private<_AP_W2,_AP_S2>& op2) { \ + ap_private<1, false> tmp(op1); \ + tmp.operator ASSIGN_OP (op2); \ + op1 = tmp; \ + return op1; \ + } + + +OP_ASSIGN_MIX_BIT(+=) +OP_ASSIGN_MIX_BIT(-=) +OP_ASSIGN_MIX_BIT(*=) +OP_ASSIGN_MIX_BIT(/=) +OP_ASSIGN_MIX_BIT(%=) +OP_ASSIGN_MIX_BIT(>>=) +OP_ASSIGN_MIX_BIT(<<=) +OP_ASSIGN_MIX_BIT(&=) +OP_ASSIGN_MIX_BIT(|=) +OP_ASSIGN_MIX_BIT(^=) + +OP_REL_MIX_BIT(==) +OP_REL_MIX_BIT(!=) +OP_REL_MIX_BIT(>) +OP_REL_MIX_BIT(>=) +OP_REL_MIX_BIT(<) +OP_REL_MIX_BIT(<=) + +OP_BIN_MIX_BIT(+, plus) +OP_BIN_MIX_BIT(-, minus) +OP_BIN_MIX_BIT(*, mult) +OP_BIN_MIX_BIT(/, div) +OP_BIN_MIX_BIT(%, mod) +OP_BIN_MIX_BIT(>>, arg1) +OP_BIN_MIX_BIT(<<, arg1) +OP_BIN_MIX_BIT(&, logic) +OP_BIN_MIX_BIT(|, logic) +OP_BIN_MIX_BIT(^, logic) + +#define REF_REL_OP_MIX_INT(REL_OP, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE bool operator REL_OP ( const ap_range_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_range_ref<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W, false>(op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_bit_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (bool(op)) REL_OP op2; \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_bit_ref<_AP_W,_AP_S> &op) { \ + return op2 REL_OP (bool(op)); \ + } \ + template \ + INLINE bool operator REL_OP ( const ap_concat_ref<_AP_W,_AP_T, _AP_W1, _AP_T1> &op, C_TYPE op2) { \ + return (ap_private<_AP_W + _AP_W1, false>(op)).operator REL_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE bool operator REL_OP ( C_TYPE op2, const ap_concat_ref<_AP_W,_AP_T, _AP_W1, _AP_T1> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator REL_OP (ap_private<_AP_W + _AP_W1, false>(op)); \ + } + +#define REF_REL_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(>, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(<, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(>=, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(<=, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(==, C_TYPE, _AP_WI, _AP_SI) \ +REF_REL_OP_MIX_INT(!=, C_TYPE, _AP_WI, _AP_SI) + +REF_REL_MIX_INT(bool, 1, false) +REF_REL_MIX_INT(char, 8, true) +REF_REL_MIX_INT(signed char, 8, true) +REF_REL_MIX_INT(unsigned char, 8, false) +REF_REL_MIX_INT(short, 16, true) +REF_REL_MIX_INT(unsigned short, 16, false) +REF_REL_MIX_INT(int, 32, true) +REF_REL_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +REF_REL_MIX_INT(long, 64, true) +REF_REL_MIX_INT(unsigned long, 64, false) +# else +REF_REL_MIX_INT(long, 32, true) +REF_REL_MIX_INT(unsigned long, 32, false) +# endif +REF_REL_MIX_INT(ap_slong, 64, true) +REF_REL_MIX_INT(ap_ulong, 64, false) + +#define REF_BIN_OP_MIX_INT(BIN_OP, RTYPE, C_TYPE, _AP_W2, _AP_S2) \ + template \ + INLINE typename ap_private<_AP_W, false>::template RType<_AP_W2,_AP_S2>::RTYPE \ + operator BIN_OP ( const ap_range_ref<_AP_W,_AP_S> &op, C_TYPE op2) { \ + return (ap_private<_AP_W, false>(op)).operator BIN_OP (ap_private<_AP_W2,_AP_S2>(op2)); \ + } \ + template \ + INLINE typename ap_private<_AP_W2, _AP_S2>::template RType<_AP_W,false>::RTYPE \ + operator BIN_OP ( C_TYPE op2, const ap_range_ref<_AP_W,_AP_S> &op) { \ + return ap_private<_AP_W2,_AP_S2>(op2).operator BIN_OP (ap_private<_AP_W, false>(op)); \ + } + +#define REF_BIN_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(+, plus, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(-, minus, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(*, mult, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(/, div, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(%, mod, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(>>, arg1, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(<<, arg1, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(&, logic, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(|, logic, C_TYPE, _AP_WI, _AP_SI) \ +REF_BIN_OP_MIX_INT(^, logic, C_TYPE, _AP_WI, _AP_SI) + +REF_BIN_MIX_INT(bool, 1, false) +REF_BIN_MIX_INT(char, 8, true) +REF_BIN_MIX_INT(signed char, 8, true) +REF_BIN_MIX_INT(unsigned char, 8, false) +REF_BIN_MIX_INT(short, 16, true) +REF_BIN_MIX_INT(unsigned short, 16, false) +REF_BIN_MIX_INT(int, 32, true) +REF_BIN_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +REF_BIN_MIX_INT(long, 64, true) +REF_BIN_MIX_INT(unsigned long, 64, false) +#else +REF_BIN_MIX_INT(long, 32, true) +REF_BIN_MIX_INT(unsigned long, 32, false) +#endif +REF_BIN_MIX_INT(ap_slong, 64, true) +REF_BIN_MIX_INT(ap_ulong, 64, false) + +#define REF_BIN_OP(BIN_OP, RTYPE) \ +template \ +INLINE typename ap_private<_AP_W, false>::template RType<_AP_W2, false>::RTYPE \ +operator BIN_OP (const ap_range_ref<_AP_W,_AP_S> &lhs, const ap_range_ref<_AP_W2,_AP_S2> &rhs) { \ + return ap_private<_AP_W,false>(lhs).operator BIN_OP (ap_private<_AP_W2, false>(rhs)); \ +} + +REF_BIN_OP(+, plus) +REF_BIN_OP(-, minus) +REF_BIN_OP(*, mult) +REF_BIN_OP(/, div) +REF_BIN_OP(%, mod) +REF_BIN_OP(>>, arg1) +REF_BIN_OP(<<, arg1) +REF_BIN_OP(&, logic) +REF_BIN_OP(|, logic) +REF_BIN_OP(^, logic) + +#if 1 +#define CONCAT_OP_MIX_INT(C_TYPE, _AP_WI, _AP_SI) \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const ap_private<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + ret <<= _AP_WI; \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + }\ + ret |= val; \ + return ret;\ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_private<_AP_W, _AP_S>& op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + if (_AP_S) { \ + ret <<= _AP_WI; ret >>= _AP_WI; \ + } \ + ret |= val << _AP_W; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const ap_range_ref<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + ret <<= _AP_WI; \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + } \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_range_ref<_AP_W, _AP_S> &op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + int len = op2.length(); \ + val <<= len; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private<_AP_WI + 1, false > \ + operator, (const ap_bit_ref<_AP_W, _AP_S> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + 1, false> val(op2); \ + val[_AP_WI] = op1; \ + return val; \ +} \ +template \ +INLINE \ +ap_private<_AP_WI + 1, false > \ + operator, (C_TYPE op1, const ap_bit_ref<_AP_W, _AP_S> &op2) { \ + ap_private<_AP_WI + 1, false> val(op1); \ + val <<= 1; \ + val[0] = op2; \ + return val; \ +} \ +template \ +INLINE \ +ap_private<_AP_W + _AP_W2 + _AP_WI, false > \ + operator, (const ap_concat_ref<_AP_W, _AP_T, _AP_W2, _AP_T2> &op1, C_TYPE op2) {\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> val(op2);\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> ret(op1);\ + if (_AP_SI) { \ + val <<= _AP_W + _AP_W2; val >>= _AP_W + _AP_W2; \ + } \ + ret <<= _AP_WI; \ + ret |= val; \ + return ret; \ +}\ +template \ +INLINE \ +ap_private<_AP_W + _AP_W2 + _AP_WI, false > \ + operator, (C_TYPE op1, const ap_concat_ref<_AP_W, _AP_T, _AP_W2, _AP_T2> &op2) {\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> val(op1);\ + ap_private<_AP_WI + _AP_W + _AP_W2, _AP_SI> ret(op2);\ + int len = op2.length(); \ + val <<= len; \ + ret |= val;\ + return ret; \ +}\ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (const af_range_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, _AP_N> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op2); \ + ap_private<_AP_WI + _AP_W, false> ret(op1); \ + if (_AP_SI) { \ + val <<= _AP_W; val >>= _AP_W; \ + }\ + ret <<= _AP_WI; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< _AP_W + _AP_WI, false > \ + operator, (C_TYPE op1, const af_range_ref<_AP_W, _AP_I, _AP_S, \ + _AP_Q, _AP_O, _AP_N> &op2) { \ + ap_private<_AP_WI + _AP_W, false> val(op1); \ + ap_private<_AP_WI + _AP_W, false> ret(op2); \ + int len = op2.length(); \ + val <<= len; \ + ret |= val; \ + return ret; \ +} \ +template \ +INLINE \ +ap_private< 1 + _AP_WI, false> \ + operator, (const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q, _AP_O, \ + _AP_N> &op1, C_TYPE op2) { \ + ap_private<_AP_WI + 1, _AP_SI> val(op2); \ + val[_AP_WI] = op1; \ + return val; \ +} \ +template \ +INLINE \ +ap_private< 1 + _AP_WI, false> \ + operator, (C_TYPE op1, const af_bit_ref<_AP_W, _AP_I, _AP_S, _AP_Q,\ + _AP_O, _AP_N> &op2) { \ + ap_private<_AP_WI + 1, _AP_SI> val(op1); \ + val <<= 1; \ + val[0] = op2; \ + return val; \ +} + +CONCAT_OP_MIX_INT(bool, 1, false) +CONCAT_OP_MIX_INT(char, 8, true) +CONCAT_OP_MIX_INT(signed char, 8, true) +CONCAT_OP_MIX_INT(unsigned char, 8, false) +CONCAT_OP_MIX_INT(short, 16, true) +CONCAT_OP_MIX_INT(unsigned short, 16, false) +CONCAT_OP_MIX_INT(int, 32, true) +CONCAT_OP_MIX_INT(unsigned int, 32, false) +# if defined __x86_64__ +CONCAT_OP_MIX_INT(long, 64, true) +CONCAT_OP_MIX_INT(unsigned long, 64, false) +# else +CONCAT_OP_MIX_INT(long, 32, true) +CONCAT_OP_MIX_INT(unsigned long, 32, false) +# endif +CONCAT_OP_MIX_INT(ap_slong, 64, true) +CONCAT_OP_MIX_INT(ap_ulong, 64, false) +#endif + +#if 1 +#define CONCAT_SHIFT_MIX_INT(C_TYPE, op) \ +template \ +INLINE ap_uint<_AP_W+_AP_W1> operator op (const ap_concat_ref<_AP_W, _AP_T, _AP_W1, _AP_T1> lhs, C_TYPE rhs) { \ + return ((ap_uint<_AP_W+_AP_W1>)lhs.get()) op ((int)rhs); \ +} + +CONCAT_SHIFT_MIX_INT(long, <<) +CONCAT_SHIFT_MIX_INT(unsigned long, <<) +CONCAT_SHIFT_MIX_INT(unsigned int, <<) +CONCAT_SHIFT_MIX_INT(ap_ulong, <<) +CONCAT_SHIFT_MIX_INT(ap_slong, <<) +CONCAT_SHIFT_MIX_INT(long, >>) +CONCAT_SHIFT_MIX_INT(unsigned long, >>) +CONCAT_SHIFT_MIX_INT(unsigned int, >>) +CONCAT_SHIFT_MIX_INT(ap_ulong, >>) +CONCAT_SHIFT_MIX_INT(ap_slong, >>) +#endif + +#if defined(SYSTEMC_H) || defined(SYSTEMC_INCLUDED) +template +INLINE void sc_trace(sc_core::sc_trace_file *tf, const ap_private<_AP_W, _AP_S> &op, + const std::string &name) { + if (tf) + tf->trace(sc_dt::sc_lv<_AP_W>(op.to_string(2).c_str()), name); +} +#endif + +template +INLINE std::ostream& operator<<(std::ostream& out, const ap_private<_AP_W,_AP_S> &op) +{ + ap_private<_AP_W, _AP_S> v=op; + const std::ios_base::fmtflags basefield = out.flags() & std::ios_base::basefield; + unsigned radix = (basefield == std::ios_base::hex) ? 16 : + ((basefield == std::ios_base::oct) ? 8 : 10); + std::string str=v.toString(radix,_AP_S); + out< +INLINE std::istream& operator >> (std::istream& in, ap_private<_AP_W,_AP_S> &op) +{ + std::string str; + in >> str; + op = ap_private<_AP_W, _AP_S>(str.c_str()); + return in; + +} + +template +INLINE std::ostream& operator<<(std::ostream& out, const ap_range_ref<_AP_W,_AP_S> &op) +{ + return operator<<(out, ap_private<_AP_W, _AP_S>(op)); +} + +template +INLINE std::istream& operator >> (std::istream& in, ap_range_ref<_AP_W,_AP_S> &op) +{ + return operator>>(in, ap_private<_AP_W, _AP_S>(op));; +} + +template +INLINE void print(const ap_private<_AP_W,_AP_S> &op, bool fill=true ) +{ + ap_private<_AP_W, _AP_S> v=op; + uint32_t ws=v.getNumWords(); + const uint64_t *ptr=v.getRawData(); + int i=ws-1; +#if 0 + if(fill) + printf("%016llx",*(ptr+i)); + else + printf("%llx",*(ptr+i)); +#else +//match SystemC output + if(_AP_W%64 != 0) { + uint32_t offset=_AP_W%64; + uint32_t count=(offset+3)/4; + int64_t data=*(ptr+i); + if(_AP_S) + data=(data<<(64-offset))>>(64-offset); + else + count=(offset+4)/4; + while(count-->0) + printf("%llx",(data>>(count*4))&0xf); + } else { + if(_AP_S==false) + printf("0"); + printf("%016llx",*(ptr+i)); + } +#endif + for(--i;i>=0;i--) + printf("%016llx",*(ptr+i)); + printf("\n"); + +} +#endif /* #ifndef __AESL_GCC_AP_INT_H__ */ \ No newline at end of file diff --git a/hls/lines256_length128/etc/ap_private.h b/hls/lines256_length128/etc/ap_private.h new file mode 100644 index 0000000000000000000000000000000000000000..1a68a9e56e950d7108a3014149623c017023d809 --- /dev/null +++ b/hls/lines256_length128/etc/ap_private.h @@ -0,0 +1,5858 @@ +/* + * Copyright 2012 Xilinx, Inc. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef LLVM_SUPPORT_MATHEXTRAS_H +#define LLVM_SUPPORT_MATHEXTRAS_H + +#ifdef _MSC_VER +#if _MSC_VER <= 1500 +typedef __int8 int8_t; +typedef unsigned __int8 uint8_t; +typedef __int16 int16_t; +typedef unsigned __int16 uint16_t; +typedef __int32 int32_t; +typedef unsigned __int32 uint32_t; +typedef __int64 int64_t; +typedef unsigned __int64 uint64_t; +#else if +#include +#endif /* #if _MSC_VER <= 1500 */ +#else +#include +#endif /* #if _MSC_VER <= 1500 */ +#undef INLINE +#if 1 +#define INLINE inline +#else +//Enable to debug ap_int/ap_fixed +#define INLINE __attribute__((weak)) +#endif +#define AP_MAX(a,b) ((a) > (b) ? (a) : (b)) +#define AP_MIN(a,b) ((a) < (b) ? (a) : (b)) +#define AP_ABS(a) ((a)>=0 ? (a):-(a)) +#ifndef AP_INT_MAX_W +#define AP_INT_MAX_W 1024 +#endif +#define BIT_WIDTH_UPPER_LIMIT (1 << 15) +#if AP_INT_MAX_W > BIT_WIDTH_UPPER_LIMIT +#error "Bitwidth exceeds 32768 (1 << 15), the maximum allowed value" +#endif +#define MAX_MODE(BITS) ((BITS + 1023) / 1024) + +// NOTE: The following support functions use the _32/_64 extensions instead of +// type overloading so that signed and unsigned integers can be used without +// ambiguity. + +/// Hi_32 - This function returns the high 32 bits of a 64 bit value. +INLINE uint32_t Hi_32(uint64_t Value) { + return static_cast(Value >> 32); +} + +/// Lo_32 - This function returns the low 32 bits of a 64 bit value. +INLINE uint32_t Lo_32(uint64_t Value) { + return static_cast(Value); +} + +/// ByteSwap_16 - This function returns a byte-swapped representation of the +/// 16-bit argument, Value. +INLINE uint16_t ByteSwap_16(uint16_t Value) { +#if defined(_MSC_VER) && !defined(_DEBUG) + // The DLL version of the runtime lacks these functions (bug!?), but in a + // release build they're replaced with BSWAP instructions anyway. + return (uint16_t)(_byteswap_ushort(Value)); +#else + uint16_t Hi = (uint16_t)((Value) << 8); + uint16_t Lo = (uint16_t)((Value) >> 8); + return Hi | Lo; +#endif +} + +/// ByteSwap_32 - This function returns a byte-swapped representation of the +/// 32-bit argument, Value. +INLINE uint32_t ByteSwap_32(uint32_t Value) { + uint32_t Byte0 = Value & 0x000000FF; + uint32_t Byte1 = Value & 0x0000FF00; + uint32_t Byte2 = Value & 0x00FF0000; + uint32_t Byte3 = Value & 0xFF000000; + return ((Byte0) << 24) | ((Byte1) << 8) | ((Byte2) >> 8) | ((Byte3) >> 24); +} + +/// ByteSwap_64 - This function returns a byte-swapped representation of the +/// 64-bit argument, Value. +INLINE uint64_t ByteSwap_64(uint64_t Value) { + uint64_t Hi = ByteSwap_32(uint32_t(Value)); + uint32_t Lo = ByteSwap_32(uint32_t(Value >> 32)); + return ((Hi) << 32) | Lo; +} + +/// CountLeadingZeros_32 - this function performs the platform optimal form of +/// counting the number of zeros from the most significant bit to the first one +/// bit. Ex. CountLeadingZeros_32(0x00F000FF) == 8. +/// Returns 32 if the word is zero. +INLINE unsigned CountLeadingZeros_32(uint32_t Value) { + unsigned Count; // result +#if __GNUC__ >= 4 + // PowerPC is defined for __builtin_clz(0) +#if !defined(__ppc__) && !defined(__ppc64__) + if (Value == 0) return 32; +#endif + Count = __builtin_clz(Value); +#else + if (Value == 0) return 32; + Count = 0; + // bisecton method for count leading zeros + for (unsigned Shift = 32 >> 1; Shift; Shift >>= 1) { + uint32_t Tmp = (Value) >> (Shift); + if (Tmp) { + Value = Tmp; + } else { + Count |= Shift; + } + } +#endif + return Count; +} + +/// CountLeadingZeros_64 - This function performs the platform optimal form +/// of counting the number of zeros from the most significant bit to the first +/// one bit (64 bit edition.) +/// Returns 64 if the word is zero. +INLINE unsigned CountLeadingZeros_64(uint64_t Value) { + unsigned Count; // result +#if __GNUC__ >= 4 + // PowerPC is defined for __builtin_clzll(0) +#if !defined(__ppc__) && !defined(__ppc64__) + if (!Value) return 64; +#endif + Count = __builtin_clzll(Value); +#else + if (sizeof(long) == sizeof(int64_t)) { + if (!Value) return 64; + Count = 0; + // bisecton method for count leading zeros + for (unsigned Shift = 64 >> 1; Shift; Shift >>= 1) { + uint64_t Tmp = (Value) >> (Shift); + if (Tmp) { + Value = Tmp; + } else { + Count |= Shift; + } + } + } else { + // get hi portion + uint32_t Hi = Hi_32(Value); + + // if some bits in hi portion + if (Hi) { + // leading zeros in hi portion plus all bits in lo portion + Count = CountLeadingZeros_32(Hi); + } else { + // get lo portion + uint32_t Lo = Lo_32(Value); + // same as 32 bit value + Count = CountLeadingZeros_32(Lo)+32; + } + } +#endif + return Count; +} + +/// CountTrailingZeros_64 - This function performs the platform optimal form +/// of counting the number of zeros from the least significant bit to the first +/// one bit (64 bit edition.) +/// Returns 64 if the word is zero. +INLINE unsigned CountTrailingZeros_64(uint64_t Value) { +#if __GNUC__ >= 4 + return (Value != 0) ? __builtin_ctzll(Value) : 64; +#else + static const unsigned Mod67Position[] = { + 64, 0, 1, 39, 2, 15, 40, 23, 3, 12, 16, 59, 41, 19, 24, 54, + 4, 64, 13, 10, 17, 62, 60, 28, 42, 30, 20, 51, 25, 44, 55, + 47, 5, 32, 65, 38, 14, 22, 11, 58, 18, 53, 63, 9, 61, 27, + 29, 50, 43, 46, 31, 37, 21, 57, 52, 8, 26, 49, 45, 36, 56, + 7, 48, 35, 6, 34, 33, 0 + }; + return Mod67Position[(uint64_t)(-(int64_t)Value & (int64_t)Value) % 67]; +#endif +} + +/// CountPopulation_64 - this function counts the number of set bits in a value, +/// (64 bit edition.) +INLINE unsigned CountPopulation_64(uint64_t Value) { +#if __GNUC__ >= 4 + return __builtin_popcountll(Value); +#else + uint64_t v = Value - (((Value) >> 1) & 0x5555555555555555ULL); + v = (v & 0x3333333333333333ULL) + (((v) >> 2) & 0x3333333333333333ULL); + v = (v + ((v) >> 4)) & 0x0F0F0F0F0F0F0F0FULL; + return unsigned((uint64_t)(v * 0x0101010101010101ULL) >> 56); +#endif +} + +#endif // LLVM_SUPPORT_MATHEXTRAS_H + + +#ifndef AP_PRIVATE_H +#define AP_PRIVATE_H + +#include +#include +#include +#include +#include +#include +#include +#include + +namespace AESL_std { + template + DataType INLINE min(DataType a, DataType b) { + // if (a >= b) return b; + // else return a; + return (a>=b) ? b : a; + } + + template + DataType INLINE max(DataType a, DataType b) { + // if (a >= b) return a; + // else return b; + return (a>=b) ? a : b; + } +} +enum ap_q_mode { + AP_RND, // rounding to plus infinity + AP_RND_ZERO,// rounding to zero + AP_RND_MIN_INF,// rounding to minus infinity + AP_RND_INF,// rounding to infinity + AP_RND_CONV, // convergent rounding + AP_TRN, // truncation + AP_TRN_ZERO // truncation to zero + +}; +enum ap_o_mode { + AP_SAT, // saturation + AP_SAT_ZERO, // saturation to zero + AP_SAT_SYM, // symmetrical saturation + AP_WRAP, // wrap-around (*) + AP_WRAP_SM // sign magnitude wrap-around (*) +}; + +template struct ap_fixed_base; +template struct af_range_ref; +template struct af_bit_ref; + +template struct ap_range_ref; +template struct ap_bit_ref; +template struct ap_concat_ref; +static bool InvalidDigit(const char* str, unsigned len, unsigned start, unsigned radix) { + unsigned i; + for (i = start; i < len; ++i) + if ((radix == 2 && (str[i] == '0' || str[i] == '1')) || + (radix == 8 && str[i] >= '0' && str[i] <= '7') || + (radix == 10 && str[i] >= '0' && str[i] <= '9') || + (radix == 16 && ((str[i] >= '0' && str[i] <= '9') || + (str[i] >= 'a' && str[i] <= 'f') || + (str[i] >= 'A' && str[i] <= 'F')))) + continue; + else + return true; + return false; +} + +static void ap_parse_sign(const char* str, uint32_t &base, bool &neg) { + if (str[0] == '+' || str[0] == '-') base = 1; + if (str[0] == '-') neg = true; + else neg = false; + return; +} + +static void ap_parse_prefix(const char* str, uint32_t &offset, uint32_t &radix) { + if (str[0] == '0') { + switch (str[1]) { + case 'b': + case 'B': offset = 2; radix = 2; break; + case 'x': + case 'X': offset = 2; radix = 16; break; + case 'd': + case 'D': offset = 2; radix = 10; break; + case 'o': + case 'O': offset = 2; radix = 8; break; + default: break; + } + } + if (offset == 0) + for (int i=0, len = strlen(str); i= 'a') || (str[i] <= 'F' && str[i] >= 'A')) { + radix = 16; + break; + } + return; +} + +/// sub_1 - This function subtracts a single "digit" (64-bit word), y, from +/// the multi-digit integer array, x[], propagating the borrowed 1 value until +/// no further borrowing is neeeded or it runs out of "digits" in x. The result +/// is 1 if "borrowing" exhausted the digits in x, or 0 if x was not exhausted. +/// In other words, if y > x then this function returns 1, otherwise 0. +/// @returns the borrow out of the subtraction +static bool sub_1(uint64_t x[], uint32_t len, uint64_t y) { + for (uint32_t i = 0; i < len; ++i) { + uint64_t __X = x[i]; + x[i] -= y; + if (y > __X) + y = 1; // We have to "borrow 1" from next "digit" + else { + y = 0; // No need to borrow + break; // Remaining digits are unchanged so exit early + } + } + return (y != 0); +} + + /// This enumeration just provides for internal constants used in this + /// translation unit. + enum { + MIN_INT_BITS = 1, ///< Minimum number of bits that can be specified + ///< Note that this must remain synchronized with IntegerType::MIN_INT_BITS + MAX_INT_BITS = (1<<23)-1 ///< Maximum number of bits that can be specified + ///< Note that this must remain synchronized with IntegerType::MAX_INT_BITS + }; + + /// A utility function for allocating memory and checking for allocation + /// failure. The content is not zeroed. + static uint64_t* getMemory(uint32_t numWords) { + return (uint64_t*) malloc(numWords*sizeof(uint64_t)); + } + + //===----------------------------------------------------------------------===// + // ap_private Class + //===----------------------------------------------------------------------===// + + /// ap_private - This class represents arbitrary precision constant integral values. + /// It is a functional replacement for common case unsigned integer type like + /// "unsigned", "unsigned long" or "uint64_t", but also allows non-byte-width + /// integer sizes and large integer value types such as 3-bits, 15-bits, or more + /// than 64-bits of precision. ap_private provides a variety of arithmetic operators + /// and methods to manipulate integer values of any bit-width. It supports both + /// the typical integer arithmetic and comparison operations as well as bitwise + /// manipulation. + /// + /// The class has several invariants worth noting: + /// * All bit, byte, and word positions are zero-based. + /// * Once the bit width is set, it doesn't change except by the Truncate, + /// SignExtend, or ZeroExtend operations. + /// * All binary operators must be on ap_private instances of the same bit width. + /// Attempting to use these operators on instances with different bit + /// widths will yield an assertion. + /// * The value is stored canonically as an unsigned value. For operations + /// where it makes a difference, there are both signed and unsigned variants + /// of the operation. For example, sdiv and udiv. However, because the bit + /// widths must be the same, operations such as Mul and Add produce the same + /// results regardless of whether the values are interpreted as signed or + /// not. + /// * In general, the class tries to follow the style of computation that LLVM + /// uses in its IR. This simplifies its use for LLVM. + /// + /// @brief Class for arbitrary precision integers. + template class ap_private; + namespace ap_private_ops{ + template + INLINE ap_private<_AP_W, _AP_S, _AP_N> lshr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt); + template + INLINE ap_private<_AP_W, _AP_S, _AP_N> shl(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt); + } + +#if defined(_MSC_VER) +# if _MSC_VER < 1400 && !defined(for) +# define for if(0);else for +# endif + typedef unsigned __int64 ap_ulong; + typedef signed __int64 ap_slong; +#else + typedef unsigned long long ap_ulong; + typedef signed long long ap_slong; +#endif + template struct retval { + }; + template<> struct retval { + typedef ap_slong Type; + }; + template<> struct retval { + typedef ap_ulong Type; + }; + + template + class ap_private { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef typename retval<_AP_S>::Type ValType; + template friend struct ap_fixed_base; + ///return type of variety of operations + //---------------------------------------------------------- + template + struct RType { + enum { + mult_w = _AP_W+_AP_W2, + mult_s = _AP_S||_AP_S2, + plus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + minus_s = true, + div_w = _AP_W+_AP_S2, + div_s = _AP_S||_AP_S2, + mod_w = AP_MIN(_AP_W,_AP_W2+(!_AP_S2&&_AP_S)), + mod_s = _AP_S, + logic_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + typedef ap_private mult; + typedef ap_private plus; + typedef ap_private minus; + typedef ap_private logic; + typedef ap_private div; + typedef ap_private mod; + typedef ap_private<_AP_W, _AP_S> arg1; + typedef bool reduce; + }; + + INLINE void report() { +#if 0 + if (_AP_W > 1024 && _AP_W <= 4096) { + fprintf(stderr, "[W] W=%d is out of bound (1<=W<=1024): for" + " synthesis: please define macro AP_INT_TYPE_EXT(N)" + " to extend the valid range.\n", _AP_W); + } else +#endif + if (_AP_W > MAX_MODE(AP_INT_MAX_W) * 1024) { + fprintf(stderr, "[E] ap_%sint<%d>: Bitwidth exceeds the " + "default max value %d. Please use macro " + "AP_INT_MAX_W to set a larger max value.\n", + _AP_S?"":"u", _AP_W, + MAX_MODE(AP_INT_MAX_W) * 1024); + exit(1); + } + } + + enum { BitWidth = _AP_W }; + /// This union is used to store the integer value. When the + /// integer bit-width <= 64, it uses VAL, otherwise it uses pVal. + + /// This enum is used to hold the constants we needed for ap_private. + uint64_t VAL; ///< Used to store the <= 64 bits integer value. + uint64_t pVal[_AP_N]; ///< Used to store the >64 bits integer value. + + /// This enum is used to hold the constants we needed for ap_private. + enum { + APINT_BITS_PER_WORD = sizeof(uint64_t) * 8, ///< Bits in a word + APINT_WORD_SIZE = sizeof(uint64_t) ///< Byte size of a word + }; + + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -(_AP_W%APINT_BITS_PER_WORD) : 0}; + static const uint64_t mask = ((uint64_t)~0ULL >> (excess_bits)); + + /// This constructor is used only internally for speed of construction of + /// temporaries. It is unsafe for general use so it is not public. + /* Constructors */ + + ap_private(const char* val) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + bool neg = false; + uint32_t radix = 16; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + ap_private ap_private_val(str.c_str(), strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + report(); + } + + ap_private(const char* val, int rd) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + uint32_t radix = rd; + bool neg = false; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + // uint32_t bitsNeeded = ap_private<_AP_W, _AP_S>::getBitsNeeded(strp, strLen, radix); + // ap_private<_AP_W, _AP_S> ap_private_val(bitsNeeded, strp , strLen, radix, base, offset); + ap_private ap_private_val(strp , strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + report(); + } + + /// Note that numWords can be smaller or larger than the corresponding bit + /// width but any extraneous bits will be dropped. + /// @param numBits the bit width of the constructed ap_private + /// @param numWords the number of words in bigVal + /// @param bigVal a sequence of words to form the initial value of the ap_private + /// @brief Construct an ap_private of numBits width, initialized as bigVal[]. + ap_private(uint32_t numWords, const uint64_t bigVal[]): VAL(0) { + assert(bigVal && "Null pointer detected!"); + { + // Get memory, cleared to 0 + memset(pVal, 0, _AP_N * sizeof(uint64_t)); + + // Calculate the number of words to copy + uint32_t words = AESL_std::min(numWords, _AP_N); + // Copy the words from bigVal to pVal + memcpy(pVal, bigVal, words * APINT_WORD_SIZE); + if (words >= _AP_W) + clearUnusedBits(); + // Make sure unused high bits are cleared + } + } + + /// This constructor interprets Val as a string in the given radix. The + /// interpretation stops when the first charater that is not suitable for the + /// radix is encountered. Acceptable radix values are 2, 8, 10 and 16. It is + /// an error for the value implied by the string to require more bits than + /// numBits. + /// @param numBits the bit width of the constructed ap_private + /// @param val the string to be interpreted + /// @param radix the radix of Val to use for the intepretation + /// @brief Construct an ap_private from a string representation. + ap_private(const std::string& val, uint8_t radix=2, int base=0, int offset=0): VAL(0) { + assert(!val.empty() && "The input string is empty."); + const char *c_str = val.c_str(); + fromString(c_str+base+offset, val.size()-base-offset, radix); + } + + /// This constructor interprets the slen characters starting at StrStart as + /// a string in the given radix. The interpretation stops when the first + /// character that is not suitable for the radix is encountered. Acceptable + /// radix values are 2, 8, 10 and 16. It is an error for the value implied by + /// the string to require more bits than numBits. + /// @param numBits the bit width of the constructed ap_private + /// @param strStart the start of the string to be interpreted + /// @param slen the maximum number of characters to interpret + /// @param radix the radix to use for the conversion + /// @brief Construct an ap_private from a string representation. + /// This method does not consider whether it is negative or not. + ap_private(const char strStart[], uint32_t slen, uint8_t radix, int base=0, int offset=0) : VAL(0) { + fromString(strStart+base+offset, slen-base-offset, radix); + } + + template + INLINE ap_private(const ap_range_ref<_AP_W2,_AP_S2>& ref) { + *this=ref.get(); + report(); + } + + template + INLINE ap_private(const ap_bit_ref<_AP_W2,_AP_S2>& ref) { + *this = ((uint64_t)(bool)ref); + report(); + } + + template + INLINE ap_private(const ap_concat_ref<_AP_W2, _AP_T2,_AP_W3, _AP_T3>& ref) { + *this=ref.get(); + report(); + } + + template + INLINE ap_private(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) { + *this = ((val.operator ap_private<_AP_W2, false> ())); + report(); + } + + template + INLINE ap_private(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) { + *this = (uint64_t)(bool)val; + report(); + } + + /// Simply makes *this a copy of that. + /// @brief Copy Constructor. + template + ap_private(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& that): VAL(0) { + operator = (const_cast& >(that)); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that): VAL(0) { + operator = (that); + } + + template + explicit ap_private(const ap_private<_AP_W1, _AP_S1, 1>& that): VAL(0) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (that.isNegative()) { + pVal[0] = that.VAL|that_sign_ext_mask; + memset(pVal+1, ~0, sizeof(uint64_t)*(_AP_N-1)); + } else { + pVal[0] = that.VAL; + memset(pVal+1, 0, sizeof(uint64_t)*(_AP_N-1)); + } + clearUnusedBits(); + } + + ap_private(const ap_private& that): VAL(0) { + memcpy(pVal, that.pVal, _AP_N * APINT_WORD_SIZE); + clearUnusedBits(); + } + + /// @brief Destructor. + virtual ~ap_private() {} + + /// Default constructor that creates an uninitialized ap_private. This is useful + /// for object deserialization (pair this with the static method Read). + ap_private(){memset(pVal, 0, sizeof(uint64_t)*(_AP_N));} + + ap_private(uint64_t* val, uint32_t bits=_AP_W) {assert(0);} + ap_private(const uint64_t *const val, uint32_t bits) {assert(0);} + + /// @name Constructors + /// @{ + /// If isSigned is true then val is treated as if it were a signed value + /// (i.e. as an int64_t) and the appropriate sign extension to the bit width + /// will be done. Otherwise, no sign extension occurs (high order bits beyond + /// the range of val are zero filled). + /// @param numBits the bit width of the constructed ap_private + /// @param val the initial value of the ap_private + /// @param isSigned how to treat signedness of val + /// @brief Create a new ap_private of numBits width, initialized as val. +#define CTOR(TYPE, SIGNED) \ + ap_private(TYPE val, bool isSigned=SIGNED) { \ + pVal[0] = val; \ + if (isSigned && int64_t(pVal[0]) < 0) { \ + memset(pVal+1, ~0, sizeof(uint64_t)*(_AP_N-1)); \ + } else { \ + memset(pVal+1, 0, sizeof(uint64_t)*(_AP_N-1)); \ + } \ + clearUnusedBits(); \ + } +#if 1 + CTOR(int, true) + CTOR(bool, false) + CTOR(signed char, true) + CTOR(unsigned char, false) + CTOR(short, true) + CTOR(unsigned short, false) + CTOR(unsigned int, false) + CTOR(long, true) + CTOR(unsigned long, false) + CTOR(unsigned long long, false) + CTOR(long long, true) + CTOR(float, false) + CTOR(double, false) +#undef CTOR +#else + CTOR(uint64_t) +#undef CTOR +#endif + + + /// @returns true if the number of bits <= 64, false otherwise. + /// @brief Determine if this ap_private just has one word to store value. + INLINE bool isSingleWord() const { + return false; + } + + /// @returns the word position for the specified bit position. + /// @brief Determine which word a bit is in. + static uint32_t whichWord(uint32_t bitPosition) { + // return bitPosition / APINT_BITS_PER_WORD; + return (bitPosition) >> 6; + } + + /// @returns the bit position in a word for the specified bit position + /// in the ap_private. + /// @brief Determine which bit in a word a bit is in. + static uint32_t whichBit(uint32_t bitPosition) { + // return bitPosition % APINT_BITS_PER_WORD; + return bitPosition & 0x3f; + } + + /// bit at a specific bit position. This is used to mask the bit in the + /// corresponding word. + /// @returns a uint64_t with only bit at "whichBit(bitPosition)" set + /// @brief Get a single bit mask. + static uint64_t maskBit(uint32_t bitPosition) { + return 1ULL << (whichBit(bitPosition)); + } + + /// @returns the corresponding word for the specified bit position. + /// @brief Get the word corresponding to a bit position + INLINE uint64_t getWord(uint32_t bitPosition) const { + return isSingleWord() ? VAL : pVal[whichWord(bitPosition)]; + } + + /// This method is used internally to clear the to "N" bits in the high order + /// word that are not used by the ap_private. This is needed after the most + /// significant word is assigned a value to ensure that those bits are + /// zero'd out. + /// @brief Clear unused high order bits + INLINE void clearUnusedBits(void) { + pVal[_AP_N-1] = _AP_S ? ((((int64_t)pVal[_AP_N-1])<<(excess_bits))>> excess_bits) : (excess_bits ? ((pVal[_AP_N-1])<<(excess_bits))>>(excess_bits) : pVal[_AP_N-1]); + } + + INLINE void clearUnusedBitsToZero(void) { + pVal[_AP_N-1] &= mask; + } + + INLINE void clearUnusedBitsToOne(void) { + pVal[_AP_N-1] |= mask; + } + + /// This is used by the constructors that take string arguments. + /// @brief Convert a char array into an ap_private + INLINE void fromString(const char *strStart, uint32_t slen, + uint8_t radix) ; + + INLINE ap_private read() volatile { + return *this; + } + + INLINE void write(const ap_private& op2) volatile { + *this = (op2); + } + + //Explicit conversions to C interger types + //----------------------------------------------------------- + operator ValType() const { + return getVal(); + } + + INLINE ValType getVal() const{ + return *pVal; + } + + INLINE int to_int() const { + return int(*this); + } + + INLINE unsigned to_uint() const { + return (unsigned) getVal(); + } + + INLINE long to_long() const { + return (long) getVal(); + } + + INLINE unsigned long to_ulong() const { + return (unsigned long) getVal(); + } + + INLINE ap_slong to_int64() const { + return (ap_slong) getVal(); + } + + INLINE ap_ulong to_uint64() const { + return (ap_ulong) getVal(); + } + + INLINE double to_double() const { + if (isNegative()) + return roundToDouble(true); + else + return roundToDouble(false); + } + + INLINE unsigned length() const { return _AP_W; } + + /*Reverse the contents of ap_private instance. I.e. LSB becomes MSB and vise versa*/ + INLINE ap_private& reverse () { + for (int i = 0; i < _AP_W/2; ++i) { + bool tmp = operator[](i); + if (operator[](_AP_W - 1 - i)) + set(i); + else + clear(i); + if (tmp) + set(_AP_W - 1 - i); + else + clear(_AP_W - 1 - i); + } + clearUnusedBits(); + return *this; + } + + /*Return true if the value of ap_private instance is zero*/ + INLINE bool iszero () const { + return isMinValue(); + } + + /* x < 0 */ + INLINE bool sign () const { + if (isNegative()) + return true; + return false; + } + + /* x[i] = !x[i] */ + INLINE void invert (int i) { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + flip(i); + } + + /* x[i] */ + INLINE bool test (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator[](i); + } + + //Set the ith bit into v + INLINE void set (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Set the ith bit into v + INLINE void set_bit (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + INLINE ap_private& set(uint32_t bitPosition) { + pVal[whichWord(bitPosition)] |= maskBit(bitPosition); + clearUnusedBits(); + return *this; + } + + INLINE void set() { + for (uint32_t i = 0; i < _AP_N; ++i) + pVal[i] = ~0ULL; + clearUnusedBits(); + } + + //Get the value of ith bit + INLINE bool get (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //Get the value of ith bit + INLINE bool get_bit (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the left + INLINE void lrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (shl(n) | lshr(_AP_W - n)); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the right + INLINE void rrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (lshr(n) | shl(_AP_W - n)); + } + + /// Set the given bit to 0 whose position is given as "bitPosition". + /// @brief Set a given bit to 0. + ap_private& clear(uint32_t bitPosition) { + pVal[whichWord(bitPosition)] &= ~maskBit(bitPosition); + clearUnusedBits(); + return *this; + } + + /// @brief Set every bit to 0. + void clear() { + memset(pVal, 0, _AP_N * APINT_WORD_SIZE); + } + + /// @brief Toggle every bit to its opposite value. + ap_private& flip() { + for (uint32_t i = 0; i < _AP_N; ++i) + pVal[i] ^= ~0ULL; + clearUnusedBits(); + return *this; + } + + /// Toggle a given bit to its opposite value whose position is given + /// as "bitPosition". + /// @brief Toggles a given bit to its opposite value. + ap_private& flip(uint32_t bitPosition) { + assert(bitPosition < BitWidth && "Out of the bit-width range!"); + if ((*this)[bitPosition]) clear(bitPosition); + else set(bitPosition); + return *this; + } + + //complements every bit + INLINE void b_not() { + flip(); + } + + ap_private getLoBits(uint32_t numBits) const { + return ap_private_ops::lshr(ap_private_ops::shl(*this, _AP_W - numBits), + _AP_W - numBits); + } + + ap_private getHiBits(uint32_t numBits) const { + return ap_private_ops::lshr(*this, _AP_W - numBits); + } + + //Binary Arithmetic + //----------------------------------------------------------- + + template + INLINE ap_private + operator & (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this & a2.get(); + } + + template + INLINE ap_private + operator | (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this | a2.get(); + } + + template + INLINE ap_private + operator ^ (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this ^ a2.get(); + } + + + ///Arithmetic assign + //------------------------------------------------------------- + +#define OP_BIN_LOGIC_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { \ + uint32_t numWords = AESL_std::min(_AP_N, _AP_N1); \ + uint32_t i; \ + for (i = 0; i < numWords; ++i) \ + pVal[i] Sym RHS.pVal[i]; \ + if (_AP_N1 < _AP_N) { \ + uint64_t ext = RHS.isNegative()?~0ULL:0; \ + for (;i<_AP_N; i++) \ + pVal[i] Sym ext; \ + } \ + clearUnusedBits(); \ + return *this; \ + } + + OP_BIN_LOGIC_ASSIGN_AP(&=); + OP_BIN_LOGIC_ASSIGN_AP(|=); + OP_BIN_LOGIC_ASSIGN_AP(^=); +#undef OP_BIN_LOGIC_ASSIGN_AP + + /// Adds the RHS APint to this ap_private. + /// @returns this, after addition of RHS. + /// @brief Addition assignment operator. + template + INLINE ap_private& operator+=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + add(pVal, pVal, RHS.pVal, _AP_N, _AP_N, _AP_N1, _AP_S, _AP_S1); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator-=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + sub(pVal, pVal, RHS.pVal, _AP_N, _AP_N, _AP_N1, _AP_S, _AP_S1); + clearUnusedBits(); + return *this; + } + + template + ap_private& operator*=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + // Get some bit facts about LHS and check for zero + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + if (!lhsWords) { + // 0 * X ===> 0 + return *this; + } + + ap_private dupRHS = RHS; + // Get some bit facts about RHS and check for zero + uint32_t rhsBits = dupRHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + if (!rhsWords) { + // X * 0 ===> 0 + clear(); + return *this; + } + + // Allocate space for the result + uint32_t destWords = rhsWords + lhsWords; + uint64_t *dest = getMemory(destWords); + + // Perform the long multiply + mul(dest, pVal, lhsWords, dupRHS.pVal, rhsWords, destWords); + + // Copy result back into *this + clear(); + uint32_t wordsToCopy = destWords >= _AP_N ? _AP_N : destWords; + + memcpy(pVal, dest, wordsToCopy* APINT_WORD_SIZE); + + uint64_t ext = (isNegative() ^ RHS.isNegative()) ? ~0ULL : 0ULL; + for (int i=wordsToCopy; i<_AP_N; i++) + pVal[i]=ext; + clearUnusedBits(); + // delete dest array and return + free(dest); + return *this; + } + +#define OP_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op); \ + return *this; \ + } \ + + OP_ASSIGN_AP(/) + OP_ASSIGN_AP(%) +#undef OP_ASSIGN_AP + +#define OP_BIN_LOGIC_AP(Sym) \ + template \ + INLINE \ + typename RType<_AP_W1, _AP_S1>::logic \ + operator Sym (const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { \ + enum { numWords = (RType<_AP_W1, _AP_S1>::logic_w +APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; \ + typename RType<_AP_W1, _AP_S1>::logic Result; \ + uint64_t *val = Result.pVal; \ + uint32_t i; \ + uint32_t min_N = std::min(_AP_N, _AP_N1); \ + uint32_t max_N = std::max(_AP_N, _AP_N1); \ + for (i = 0; i < min_N; ++i) \ + val[i] = pVal[i] Sym RHS.pVal[i]; \ + if (numWords > i) { \ + const uint64_t* tmpVal = (_AP_N>_AP_N1 ? pVal : RHS.pVal)+i; \ + uint64_t ext = ((_AP_N<_AP_N1 && isNegative() )||(_AP_N1 < _AP_N && RHS.isNegative())) ? ~0ULL : 0; \ + for (;i i) { \ + uint64_t ext2 = ((_AP_N>_AP_N1 && isNegative() )||(_AP_N1 > _AP_N && RHS.isNegative())) ? ~0ULL : 0; \ + val[i] = ext Sym ext2; \ + } \ + } \ + Result.clearUnusedBits(); \ + return Result; \ + } + + OP_BIN_LOGIC_AP(|); + OP_BIN_LOGIC_AP(&); + OP_BIN_LOGIC_AP(^); + +#undef OP_BIN_LOGIC_AP + + template + INLINE typename RType<_AP_W1,_AP_S1>::plus operator+(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + // assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); + typename RType<_AP_W1,_AP_S1>::plus Result; + bool carry = add(Result.pVal, this->pVal, RHS.pVal, (RType<_AP_W1,_AP_S1>::plus_w + 63) / 64, _AP_N, _AP_N1, _AP_S, _AP_S1); + if ((RType<_AP_W1,_AP_S1>::plus_w + 63) / 64> std::max(_AP_W, _AP_W1) ) + Result.pVal[(RType<_AP_W1,_AP_S1>::plus_w + 63)/64 - 1] = carry; + Result.clearUnusedBits(); + return Result; + } + + template + INLINE typename RType<_AP_W1,_AP_S1>::minus operator-(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + typename RType<_AP_W1,_AP_S1>::minus Result; + bool borrow = sub(Result.pVal, this->pVal, RHS.pVal, (RType<_AP_W1,_AP_S1>::minus_w + 63) / 64, _AP_N, _AP_N1, _AP_S, _AP_S1); + if ((RType<_AP_W1,_AP_S1>::minus_w + 63) / 64 > AESL_std::max(_AP_W, _AP_W1) ) { + Result.pVal[(RType<_AP_W1,_AP_S1>::minus_w+63)/64 - 1] = borrow; + } + Result.clearUnusedBits(); + return Result; + } + + template + typename RType<_AP_W1, _AP_S1>::mult operator*(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) const { + + // Get some bit facts about LHS and check for zero + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + if (!lhsWords) + // 0 * X ===> 0 + return typename RType<_AP_W1, _AP_S1>::mult(); + + // Get some bit facts about RHS and check for zero + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + if (!rhsWords) { + // X * 0 ===> 0 + return typename RType<_AP_W1, _AP_S1>::mult(); + } + + //extend size to avoid result loss + typename RType<_AP_W1, _AP_S1>::mult dupLHS = *this; + typename RType<_AP_W1, _AP_S1>::mult dupRHS = RHS; + lhsBits = dupLHS.getActiveBits(); + lhsWords = !lhsBits ? 0 : whichWord(lhsBits - 1) + 1; + rhsBits = dupRHS.getActiveBits(); + rhsWords = !rhsBits ? 0 : whichWord(rhsBits - 1) + 1; + + // Allocate space for the result + enum { destWords =(RType<_AP_W1, _AP_S1>::mult_w+APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; + int destw = destWords; + typename RType<_AP_W1, _AP_S1>::mult Result; + uint64_t *dest = Result.pVal; + uint64_t ext = (isNegative() ^ RHS.isNegative()) ? ~0ULL : 0; + + // Perform the long multiply + mul(dest, dupLHS.pVal, lhsWords, dupRHS.pVal, rhsWords, destWords); + + for (int i=lhsWords+rhsWords; i + INLINE typename RType<_AP_W2,_AP_S2>::div + operator / (const ap_private<_AP_W2,_AP_S2>& op) const { + ap_private lhs=ap_private(*this); + ap_private rhs=ap_private(op); + return typename RType<_AP_W2,_AP_S2>::div((_AP_S||_AP_S2)?lhs.sdiv(rhs):lhs.udiv(rhs)); + } + + template + INLINE typename RType<_AP_W2,_AP_S2>::mod + operator % (const ap_private<_AP_W2,_AP_S2>& op) const { + ap_private lhs=*this; + ap_private rhs= op; + typename RType<_AP_W2,_AP_S2>::mod res = typename RType<_AP_W2,_AP_S2>::mod(_AP_S?lhs.srem(rhs):lhs.urem(rhs)); + return res; + } + + template + INLINE ap_private + operator << (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh=op2.to_uint(); + return *this << sh; + } + + INLINE ap_private + operator << (uint32_t sh) const { + ap_private r(*this); + bool overflow=(sh>=length()); + if(overflow) + r.clear(); + else + r = ap_private(r.shl(sh)); + return r; + } + + template + INLINE ap_private + operator >> (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh = op2.to_uint(); + return *this >> sh; + } + + INLINE ap_private + operator >> (uint32_t sh) const { + ap_private r(*this); + bool overflow=(sh>=_AP_W); + bool neg_v=r.isNegative(); + if(_AP_S) { + if(overflow) + neg_v?r.set():r.clear(); + else + return r.ashr(sh); + } else { + if(overflow) + r.clear(); + else + return r.lshr(sh); + } + return r; + } + + ///Shift assign + //------------------------------------------------------------------ +#define OP_ASSIGN_AP(Sym) \ + template \ + INLINE ap_private& operator Sym##=(int op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } \ + INLINE ap_private& operator Sym##=(unsigned int op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } \ + template \ + INLINE ap_private& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this = operator Sym (op); \ + return *this; \ + } + OP_ASSIGN_AP(>>) + OP_ASSIGN_AP(<<) +#undef OP_ASSIGN_AP + ///Comparisons + //----------------------------------------------------------------- + bool operator==(const ap_private& RHS) const { + // Get some facts about the number of bits used in the two operands. + uint32_t n1 = getActiveBits(); + uint32_t n2 = RHS.getActiveBits(); + + // If the number of bits isn't the same, they aren't equal + if (n1 != n2) + return false; + + // If the number of bits fits in a word, we only need to compare the low word. + if (n1 <= APINT_BITS_PER_WORD) + return pVal[0] == RHS.pVal[0]; + + // Otherwise, compare everything + for (int i = whichWord(n1 - 1); i >= 0; --i) + if (pVal[i] != RHS.pVal[i]) + return false; + return true; + } + + template + INLINE bool operator == (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S|_AP_S2> lhs(*this); + ap_private<_AP_MAX_W, _AP_S|_AP_S2> rhs(op); + return lhs==rhs; + } + + bool operator==(uint64_t Val) const { + uint32_t n = getActiveBits(); + if (n <= APINT_BITS_PER_WORD) + return pVal[0] == Val; + else + return false; + } + + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2>& op) const { + return !(*this==op); + } + + template + INLINE bool operator!=(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !((*this) == RHS); + } + + INLINE bool operator!=(uint64_t Val) const { + return !((*this) == Val); + } + + + template + INLINE bool operator <= (const ap_private<_AP_W2,_AP_S2>& op) const { + return !(*this>op); + } + + INLINE bool operator <(const ap_private& op) const { + return _AP_S ? slt(op):ult(op); + } + + template + INLINE bool operator < (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.slt(rhs):lhs.ult(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + } + + template + INLINE bool operator >=(const ap_private<_AP_W2,_AP_S2>& op) const { + return !(*this + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.sgt(rhs):lhs.ugt(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ugt(rhs); + else + return lhs.sgt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ugt(rhs); + else + return lhs.sgt(rhs); + } + + ///Bit and Part Select + //-------------------------------------------------------------- + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast*>(this), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>((const_cast*> (this)), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast(this), Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (uint32_t index) { + assert(index >= 0&&"Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + template + INLINE bool operator [] (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br =operator [] (index); + return br.to_bool(); + } + + INLINE bool operator [](uint32_t bitPosition) const { + return (maskBit(bitPosition) & (pVal[whichWord(bitPosition)])) != 0; + } + + INLINE ap_bit_ref<_AP_W,_AP_S> bit (int index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> bit (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W &&"Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + INLINE bool bit (int index) const { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br(const_cast*>(this), index); + return br.to_bool(); + } + + template + INLINE bool bit (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br = bit(index); + return br.to_bool(); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(ap_private<_AP_W2,_AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(const ap_private<_AP_W2,_AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (ap_range_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (ap_bit_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE + ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + INLINE ap_private<_AP_W,false> get() const { + ap_private<_AP_W,false> ret(*this); + return ret; + } + + template + INLINE void set(const ap_private<_AP_W3, false> & val) { + operator = (ap_private<_AP_W3, _AP_S>(val)); + } + + /// @} + /// @name Value Tests + /// @{ + /// This tests the high bit of this ap_private to determine if it is set. + /// @returns true if this ap_private is negative, false otherwise + /// @brief Determine sign of this ap_private. + INLINE bool isNegative() const { + //just for get rid of warnings + enum {shift = (_AP_W-APINT_BITS_PER_WORD*(_AP_N-1)-1)}; + static const uint64_t mask = 1ULL << (shift); + return _AP_S && (pVal[_AP_N-1]&mask); + } + + /// This tests the high bit of the ap_private to determine if it is unset. + /// @brief Determine if this ap_private Value is positive (not negative). + INLINE bool isPositive() const { + return !isNegative(); + } + + /// This tests if the value of this ap_private is strictly positive (> 0). + /// @returns true if this ap_private is Positive and not zero. + /// @brief Determine if this ap_private Value is strictly positive. + INLINE bool isStrictlyPositive() const { + return isPositive() && (*this) != 0; + } + + /// This checks to see if the value has all bits of the ap_private are set or not. + /// @brief Determine if all bits are set + INLINE bool isAllOnesValue() const { + return countPopulation() == _AP_W; + } + + /// This checks to see if the value of this ap_private is the maximum unsigned + /// value for the ap_private's bit width. + /// @brief Determine if this is the largest unsigned value. + INLINE bool isMaxValue() const { + return countPopulation() == _AP_W; + } + + /// This checks to see if the value of this ap_private is the maximum signed + /// value for the ap_private's bit width. + /// @brief Determine if this is the largest signed value. + INLINE bool isMaxSignedValue() const { + return BitWidth == 1 ? VAL == 0 : + !isNegative() && countPopulation() == _AP_W - 1; + } + + /// This checks to see if the value of this ap_private is the minimum unsigned + /// value for the ap_private's bit width. + /// @brief Determine if this is the smallest unsigned value. + INLINE bool isMinValue() const { + return countPopulation() == 0; + } + + /// This checks to see if the value of this ap_private is the minimum signed + /// value for the ap_private's bit width. + /// @brief Determine if this is the smallest signed value. + INLINE bool isMinSignedValue() const { + return BitWidth == 1 ? VAL == 1 : + isNegative() && countPopulation() == 1; + } + + /// This function returns a pointer to the internal storage of the ap_private. + /// This is useful for writing out the ap_private in binary form without any + /// conversions. + INLINE const uint64_t* getRawData() const { + if (isSingleWord()) + return &VAL; + return &pVal[0]; + } + + ap_private sqrt() const; + + /// @} + /// @Assignment Operators + /// @{ + /// @returns *this after assignment of RHS. + /// @brief Copy assignment operator. + INLINE ap_private& operator=(const ap_private& RHS) { + if (this != &RHS) + memcpy(pVal, RHS.pVal, _AP_N * APINT_WORD_SIZE); + return *this; + } + INLINE ap_private& operator=(const volatile ap_private& RHS) { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + INLINE volatile ap_private& operator=(const ap_private& RHS) volatile { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + INLINE volatile ap_private& operator=(const volatile ap_private& RHS) volatile { + if (this != &RHS) + for (int i=0; i<_AP_N; ++i) + pVal[i] = RHS.pVal[i]; + return *this; + } + + template + INLINE ap_private& operator=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + if (_AP_S1) + cpSextOrTrunc(RHS); + else + cpZextOrTrunc(RHS); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + if (_AP_S1) + cpSextOrTrunc(RHS); + else + cpZextOrTrunc(RHS); + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const ap_private<_AP_W1, _AP_S1, 1>& RHS) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (RHS.isNegative()) { + pVal[0] = RHS.VAL | that_sign_ext_mask; + memset(pVal+1,~0, APINT_WORD_SIZE*(_AP_N-1)); + } else { + pVal[0] = RHS.VAL; + memset(pVal+1, 0, APINT_WORD_SIZE*(_AP_N-1)); + } + clearUnusedBits(); + return *this; + } + + template + INLINE ap_private& operator=(const volatile ap_private<_AP_W1, _AP_S1, 1>& RHS) { + static const uint64_t that_sign_ext_mask = (_AP_W1==APINT_BITS_PER_WORD)?0:~0ULL>>(_AP_W1%APINT_BITS_PER_WORD)<<(_AP_W1%APINT_BITS_PER_WORD); + if (RHS.isNegative()) { + pVal[0] = RHS.VAL | that_sign_ext_mask; + memset(pVal+1,~0, APINT_WORD_SIZE*(_AP_N-1)); + } else { + pVal[0] = RHS.VAL; + memset(pVal+1, 0, APINT_WORD_SIZE*(_AP_N-1)); + } + clearUnusedBits(); + return *this; + } + + /// @} + /// @name Unary Operators + /// @{ + /// @returns a new ap_private value representing *this incremented by one + /// @brief Postfix increment operator. + INLINE const ap_private operator++(int) { + ap_private API(*this); + ++(*this); + return API; + } + + /// @returns *this incremented by one + /// @brief Prefix increment operator. + INLINE ap_private& operator++() { + add_1(pVal, pVal, _AP_N, 1); + clearUnusedBits(); + return *this; + } + + /// @returns a new ap_private representing *this decremented by one. + /// @brief Postfix decrement operator. + INLINE const ap_private operator--(int) { + ap_private API(*this); + --(*this); + return API; + } + + /// @returns *this decremented by one. + /// @brief Prefix decrement operator. + INLINE ap_private& operator--() { + sub_1(pVal, _AP_N, 1); + clearUnusedBits(); + return *this; + } + + /// Performs a bitwise complement operation on this ap_private. + /// @returns an ap_private that is the bitwise complement of *this + /// @brief Unary bitwise complement operator. + INLINE ap_private operator~() const { + ap_private Result(*this); + Result.flip(); + return Result; + } + + /// Negates *this using two's complement logic. + /// @returns An ap_private value representing the negation of *this. + /// @brief Unary negation operator + INLINE typename RType<1,false>::minus operator-() const { + return ap_private<1,false>(0) - (*this); + } + + /// Performs logical negation operation on this ap_private. + /// @returns true if *this is zero, false otherwise. + /// @brief Logical negation operator. + INLINE bool operator !() const { + for (uint32_t i = 0; i < _AP_N; ++i) + if (pVal[i]) + return false; + return true; + } + + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> And(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator&(RHS); + } + template + INLINE ap_private Or(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator|(RHS); + } + template + ap_private Xor(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return this->operator^(RHS); + } + + ap_private Mul(const ap_private& RHS) const { + ap_private Result(*this); + Result *= RHS; + return Result; + } + + ap_private Add(const ap_private& RHS) const { + ap_private Result(0); + bool carry = add(Result.pVal, this->pVal, RHS.pVal, _AP_N, _AP_N, _AP_N, _AP_S, _AP_S); + Result.clearUnusedBits(); + return Result; + } + + ap_private Sub(const ap_private& RHS) const { + ap_private Result(0); + sub(Result.pVal, this->pVal, RHS.pVal, _AP_N, _AP_N, _AP_N, _AP_S, _AP_S); + Result.clearUnusedBits(); + return Result; + } + + /// Arithmetic right-shift this ap_private by shiftAmt. + /// @brief Arithmetic right-shift function. + ap_private ashr(uint32_t shiftAmt) const { + assert(shiftAmt <= BitWidth && "Invalid shift amount, too big"); + // Handle a degenerate case + if (shiftAmt == 0) + return *this; + + // Handle single word shifts with built-in ashr + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(/*BitWidth, 0*/); // undefined + else { + uint32_t SignBit = APINT_BITS_PER_WORD - BitWidth; + return ap_private(/*BitWidth,*/ + (((int64_t(VAL) << (SignBit)) >> (SignBit)) >> (shiftAmt))); + } + } + + // If all the bits were shifted out, the result is, technically, undefined. + // We return -1 if it was negative, 0 otherwise. We check this early to avoid + // issues in the algorithm below. + if (shiftAmt == BitWidth) { + if (isNegative()) + return ap_private(-1); + else + return ap_private(0); + } + + // Create some space for the result. + ap_private Retval(0); + uint64_t * val = Retval.pVal; + + // Compute some values needed by the following shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; // bits to shift per word + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; // word offset for shift + uint32_t breakWord = _AP_N - 1 - offset; // last word affected + uint32_t bitsInWord = whichBit(BitWidth); // how many bits in last word? + if (bitsInWord == 0) + bitsInWord = APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + // Move the words containing significant bits + for (uint32_t i = 0; i <= breakWord; ++i) + val[i] = pVal[i+offset]; // move whole word + + // Adjust the top significant word for sign bit fill, if negative + if (isNegative()) + if (bitsInWord < APINT_BITS_PER_WORD) + val[breakWord] |= ~0ULL << (bitsInWord); // set high bits + } else { + // Shift the low order words + for (uint32_t i = 0; i < breakWord; ++i) { + // This combines the shifted corresponding word with the low bits from + // the next word (shifted into this word's high bits). + val[i] = ((pVal[i+offset]) >> (wordShift)); + val[i] |= ((pVal[i+offset+1]) << (APINT_BITS_PER_WORD - wordShift)); + } + + // Shift the break word. In this case there are no bits from the next word + // to include in this word. + val[breakWord] = (pVal[breakWord+offset]) >> (wordShift); + + // Deal with sign extenstion in the break word, and possibly the word before + // it. + if (isNegative()) { + if (wordShift > bitsInWord) { + if (breakWord > 0) + val[breakWord-1] |= + ~0ULL << (APINT_BITS_PER_WORD - (wordShift - bitsInWord)); + val[breakWord] |= ~0ULL; + } else + val[breakWord] |= (~0ULL << (bitsInWord - wordShift)); + } + } + + // Remaining words are 0 or -1, just assign them. + uint64_t fillValue = (isNegative() ? ~0ULL : 0); + for (uint32_t i = breakWord+1; i < _AP_N; ++i) + val[i] = fillValue; + Retval.clearUnusedBits(); + return Retval; + } + + /// Logical right-shift this ap_private by shiftAmt. + /// @brief Logical right-shift function. + ap_private lshr(uint32_t shiftAmt) const { + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(0); + else + return ap_private((this->VAL) >> (shiftAmt)); + } + + // If all the bits were shifted out, the result is 0. This avoids issues + // with shifting by the size of the integer type, which produces undefined + // results. We define these "undefined results" to always be 0. + if (shiftAmt == BitWidth) + return ap_private(0); + + // If none of the bits are shifted out, the result is *this. This avoids + // issues with shifting byt he size of the integer type, which produces + // undefined results in the code below. This is also an optimization. + if (shiftAmt == 0) + return *this; + + // Create some space for the result. + ap_private Retval(0); + uint64_t * val = Retval.pVal; + + // If we are shifting less than a word, compute the shift with a simple carry + if (shiftAmt < APINT_BITS_PER_WORD) { + uint64_t carry = 0; + for (int i = _AP_N-1; i >= 0; --i) { + val[i] = ((pVal[i]) >> (shiftAmt)) | carry; + carry = (pVal[i]) << (APINT_BITS_PER_WORD - shiftAmt); + } + Retval.clearUnusedBits(); + return Retval; + } + + // Compute some values needed by the remaining shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + for (uint32_t i = 0; i < _AP_N - offset; ++i) + val[i] = pVal[i+offset]; + for (uint32_t i = _AP_N-offset; i < _AP_N; i++) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + // Shift the low order words + uint32_t breakWord = _AP_N - offset -1; + for (uint32_t i = 0; i < breakWord; ++i) + val[i] = ((pVal[i+offset]) >> (wordShift)) | + ((pVal[i+offset+1]) << (APINT_BITS_PER_WORD - wordShift)); + // Shift the break word. + val[breakWord] = (pVal[breakWord+offset]) >> (wordShift); + + // Remaining words are 0 + for (uint32_t i = breakWord+1; i < _AP_N; ++i) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + /// Left-shift this ap_private by shiftAmt. + /// @brief Left-shift function. + ap_private shl(uint32_t shiftAmt) const { + assert(shiftAmt <= BitWidth && "Invalid shift amount, too big"); + if (isSingleWord()) { + if (shiftAmt == BitWidth) + return ap_private(0); // avoid undefined shift results + return ap_private((VAL) << (shiftAmt)); + } + + // If all the bits were shifted out, the result is 0. This avoids issues + // with shifting by the size of the integer type, which produces undefined + // results. We define these "undefined results" to always be 0. + if (shiftAmt == BitWidth) + return ap_private(0); + + // If none of the bits are shifted out, the result is *this. This avoids a + // lshr by the words size in the loop below which can produce incorrect + // results. It also avoids the expensive computation below for a common case. + if (shiftAmt == 0) + return *this; + + // Create some space for the result. + ap_private Retval(0); + uint64_t* val = Retval.pVal; + // If we are shifting less than a word, do it the easy way + if (shiftAmt < APINT_BITS_PER_WORD) { + uint64_t carry = 0; + for (uint32_t i = 0; i < _AP_N; i++) { + val[i] = ((pVal[i]) << (shiftAmt)) | carry; + carry = (pVal[i]) >> (APINT_BITS_PER_WORD - shiftAmt); + } + Retval.clearUnusedBits(); + return Retval; + } + + // Compute some values needed by the remaining shift algorithms + uint32_t wordShift = shiftAmt % APINT_BITS_PER_WORD; + uint32_t offset = shiftAmt / APINT_BITS_PER_WORD; + + // If we are shifting whole words, just move whole words + if (wordShift == 0) { + for (uint32_t i = 0; i < offset; i++) + val[i] = 0; + for (uint32_t i = offset; i < _AP_N; i++) + val[i] = pVal[i-offset]; + Retval.clearUnusedBits(); + return Retval; + } + + // Copy whole words from this to Result. + uint32_t i = _AP_N - 1; + for (; i > offset; --i) + val[i] = (pVal[i-offset]) << (wordShift) | + (pVal[i-offset-1]) >> (APINT_BITS_PER_WORD - wordShift); + val[offset] = (pVal[0]) << (wordShift); + for (i = 0; i < offset; ++i) + val[i] = 0; + Retval.clearUnusedBits(); + return Retval; + } + + INLINE ap_private rotl(uint32_t rotateAmt) const { + if (rotateAmt == 0) + return *this; + // Don't get too fancy, just use existing shift/or facilities + ap_private hi(*this); + ap_private lo(*this); + hi.shl(rotateAmt); + lo.lshr(BitWidth - rotateAmt); + return hi | lo; + } + + INLINE ap_private rotr(uint32_t rotateAmt) const { + if (rotateAmt == 0) + return *this; + // Don't get too fancy, just use existing shift/or facilities + ap_private hi(*this); + ap_private lo(*this); + lo.lshr(rotateAmt); + hi.shl(BitWidth - rotateAmt); + return hi | lo; + } + + /// Perform an unsigned divide operation on this ap_private by RHS. Both this and + /// RHS are treated as unsigned quantities for purposes of this division. + /// @returns a new ap_private value containing the division result + /// @brief Unsigned division operation. + ap_private udiv(const ap_private& RHS) const { + assert(BitWidth == RHS.BitWidth && "Bit widths must be the same"); + + // First, deal with the easy case + if (isSingleWord()) { + assert(RHS.VAL != 0 && "Divide by zero?"); + return ap_private(VAL / RHS.VAL); + } + + // Get some facts about the LHS and RHS number of bits and words + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Divided by zero???"); + uint32_t lhsBits = this->getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + + // Deal with some degenerate cases + if (!lhsWords) + // 0 / X ===> 0 + return ap_private(0); + else if (lhsWords < rhsWords || this->ult(RHS)) { + // X / Y ===> 0, iff X < Y + return ap_private(0); + } else if (*this == RHS) { + // X / X ===> 1 + return ap_private(1); + } else if (lhsWords == 1 && rhsWords == 1) { + // All high words are zero, just use native divide + return ap_private(this->pVal[0] / RHS.pVal[0]); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Quotient(0); // to hold result. + divide(*this, lhsWords, RHS, rhsWords, &Quotient, (ap_private*)0); + return Quotient; + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + INLINE ap_private sdiv(const ap_private& RHS) const { + if (isNegative()) + if (RHS.isNegative()) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS.isNegative()) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + /// Perform an unsigned remainder operation on this ap_private with RHS being the + /// divisor. Both this and RHS are treated as unsigned quantities for purposes + /// of this operation. Note that this is a true remainder operation and not + /// a modulo operation because the sign follows the sign of the dividend + /// which is *this. + /// @returns a new ap_private value containing the remainder result + /// @brief Unsigned remainder operation. + ap_private urem(const ap_private& RHS) const { + if (isSingleWord()) { + assert(RHS.VAL != 0 && "Remainder by zero?"); + return ap_private(VAL % RHS.VAL); + } + + // Get some facts about the LHS + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + + // Get some facts about the RHS + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Performing remainder operation by zero ???"); + + // Check the degenerate cases + if (lhsWords == 0) { + // 0 % Y ===> 0 + return ap_private(0); + } else if (lhsWords < rhsWords || this->ult(RHS)) { + // X % Y ===> X, iff X < Y + return *this; + } else if (*this == RHS) { + // X % X == 0; + return ap_private(0); + } else if (lhsWords == 1) { + // All high words are zero, just use native remainder + return ap_private(pVal[0] % RHS.pVal[0]); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Remainder(0); + divide(*this, lhsWords, RHS, rhsWords, (ap_private*)(0), &Remainder); + return Remainder; + } + + ap_private urem(uint64_t RHS) const { + // Get some facts about the LHS + uint32_t lhsBits = getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (whichWord(lhsBits - 1) + 1); + // Get some facts about the RHS + uint32_t rhsBits = 64 - CountLeadingZeros_64(RHS); // RHS.getActiveBits(); + uint32_t rhsWords = 1;//!rhsBits ? 0 : (ap_private<_AP_W, _AP_S, _AP_N>::whichWord(rhsBits - 1) + 1); + assert(rhsWords && "Performing remainder operation by zero ???"); + // Check the degenerate cases + if (lhsWords == 0) { + // 0 % Y ===> 0 + return ap_private(0); + } else if (lhsWords < rhsWords || this->ult(RHS)) { + // X % Y ===> X, iff X < Y + return *this; + } else if (*this == RHS) { + // X % X == 0; + return ap_private(0); + } else if (lhsWords == 1) { + // All high words are zero, just use native remainder + return ap_private(pVal[0] % RHS); + } + + // We have to compute it the hard way. Invoke the Knuth divide algorithm. + ap_private Remainder(0); + divide(*this, lhsWords, RHS, (ap_private*)(0), &Remainder); + return Remainder; + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(const ap_private& RHS) const { + if (isNegative()) { + ap_private lhs = -(*this); + if (RHS.isNegative()) { + ap_private rhs = -RHS; + return -(lhs.urem(rhs)); + } else + return -(lhs.urem(RHS)); + } else if (RHS.isNegative()) { + ap_private rhs = -RHS; + return this->urem(rhs); + } + return this->urem(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return -((-(*this)).urem(-RHS)); + else + return -((-(*this)).urem(RHS)); + else if (RHS<0) + return this->urem(-RHS); + return this->urem(RHS); + } + + /// Sometimes it is convenient to divide two ap_private values and obtain both + /// the quotient and remainder. This function does both operations in the + /// same computation making it a little more efficient. + /// @brief Dual division/remainder interface. + static void udivrem(const ap_private& LHS, const ap_private& RHS, ap_private &Quotient, ap_private& Remainder) { + // Get some size facts about the dividend and divisor + uint32_t lhsBits = LHS.getActiveBits(); + uint32_t lhsWords = !lhsBits ? 0 : (ap_private::whichWord(lhsBits - 1) + 1); + uint32_t rhsBits = RHS.getActiveBits(); + uint32_t rhsWords = !rhsBits ? 0 : (ap_private::whichWord(rhsBits - 1) + 1); + + // Check the degenerate cases + if (lhsWords == 0) { + Quotient = 0; // 0 / Y ===> 0 + Remainder = 0; // 0 % Y ===> 0 + return; + } + + if (lhsWords < rhsWords || LHS.ult(RHS)) { + Quotient = 0; // X / Y ===> 0, iff X < Y + Remainder = LHS; // X % Y ===> X, iff X < Y + return; + } + + if (LHS == RHS) { + Quotient = 1; // X / X ===> 1 + Remainder = 0; // X % X ===> 0; + return; + } + + if (lhsWords == 1 && rhsWords == 1) { + // There is only one word to consider so use the native versions. + if (LHS.isSingleWord()) { + Quotient = ap_private(LHS.VAL / RHS.VAL); + Remainder = ap_private(LHS.VAL % RHS.VAL); + } else { + Quotient = ap_private(LHS.pVal[0] / RHS.pVal[0]); + Remainder = ap_private(LHS.pVal[0] % RHS.pVal[0]); + } + return; + } + + // Okay, lets do it the long way + divide(LHS, lhsWords, RHS, rhsWords, &Quotient, &Remainder); + } + + static void sdivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS.isNegative()) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS.isNegative()) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + /// Compares this ap_private with RHS for the validity of the equality + /// relationship. + /// @returns true if *this == Val + /// @brief Equality comparison. + template + INLINE bool eq(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return (*this) == RHS; + } + + /// Compares this ap_private with RHS for the validity of the inequality + /// relationship. + /// @returns true if *this != Val + /// @brief Inequality comparison + template + INLINE bool ne(const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) const { + return !((*this) == RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered unsigned. + /// @brief Unsigned less than comparison + template + INLINE bool ult(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + // Get active bit length of both operands + uint32_t n1 = getActiveBits(); + uint32_t n2 = RHS.getActiveBits(); + + // If magnitude of LHS is less than RHS, return true. + if (n1 < n2) + return true; + + // If magnitude of RHS is greather than LHS, return false. + if (n2 < n1) + return false; + + // If they bot fit in a word, just compare the low order word + if (n1 <= APINT_BITS_PER_WORD && n2 <= APINT_BITS_PER_WORD) + return pVal[0] < RHS.pVal[0]; + + // Otherwise, compare all words + uint32_t topWord = whichWord(AESL_std::max(n1,n2)-1); + for (int i = topWord; i >= 0; --i) { + if (pVal[i] > RHS.pVal[i]) + return false; + if (pVal[i] < RHS.pVal[i]) + return true; + } + return false; + } + + INLINE bool ult(uint64_t RHS) const { + // Get active bit length of both operands + uint32_t n1 = getActiveBits(); + uint32_t n2 = 64 - CountLeadingZeros_64(RHS); //RHS.getActiveBits(); + + // If magnitude of LHS is less than RHS, return true. + if (n1 < n2) + return true; + + // If magnitude of RHS is greather than LHS, return false. + if (n2 < n1) + return false; + + // If they bot fit in a word, just compare the low order word + if (n1 <= APINT_BITS_PER_WORD && n2 <= APINT_BITS_PER_WORD) + return pVal[0] < RHS; + assert(0); + } + + template + INLINE bool slt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + ap_private lhs(*this); + ap_private<_AP_W, _AP_S1, _AP_N> rhs(RHS); + bool lhsNeg = isNegative(); + bool rhsNeg = rhs.isNegative(); + if (lhsNeg) { + // Sign bit is set so perform two's complement to make it positive + lhs.flip(); + lhs++; + } + if (rhsNeg) { + // Sign bit is set so perform two's complement to make it positive + rhs.flip(); + rhs++; + } + + // Now we have unsigned values to compare so do the comparison if necessary + // based on the negativeness of the values. + if (lhsNeg) + if (rhsNeg) + return lhs.ugt(rhs); + else + return true; + else if (rhsNeg) + return false; + else + return lhs.ult(rhs); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered unsigned. + /// @brief Unsigned less or equal comparison + template + INLINE bool ule(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return ult(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered signed. + /// @brief Signed less or equal comparison + template + INLINE bool sle(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return slt(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered unsigned. + /// @brief Unsigned greather than comparison + template + INLINE bool ugt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !ult(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered signed. + /// @brief Signed greather than comparison + template + INLINE bool sgt(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !slt(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered unsigned. + /// @brief Unsigned greater or equal comparison + template + INLINE bool uge(const ap_private<_AP_W, _AP_S, _AP_N>& RHS) const { + return !ult(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered signed. + /// @brief Signed greather or equal comparison + template + INLINE bool sge(const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) const { + return !slt(RHS); + } + + template + void cpTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 > BitWidth && "Invalid ap_private Truncate request"); + assert(_AP_W1 >= MIN_INT_BITS && "Can't truncate to 0 bits"); + memcpy(pVal, that.pVal, _AP_N*APINT_WORD_SIZE); + } + + // Sign extend to a new width. + template + void cpSext(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 < BitWidth && "Invalid ap_private SignExtend request"); + assert(_AP_W1 <= MAX_INT_BITS && "Too many bits"); + // If the sign bit isn't set, this is the same as zext. + if (!that.isNegative()) { + cpZext(that); + return; + } + + // The sign bit is set. First, get some facts + enum { wordBits = _AP_W1 % APINT_BITS_PER_WORD}; + + // Mask the high order word appropriately + if (_AP_N1 == _AP_N) { + enum { newWordBits = _AP_W % APINT_BITS_PER_WORD}; + // The extension is contained to the wordsBefore-1th word. + static const uint64_t mask = wordBits?(~0ULL<<(wordBits)):0ULL; + if (_AP_N1 == 1) { + assert(0); + } else { + for (uint32_t i = 0; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + pVal[_AP_N1-1] |= mask; + return; + } + } + + if (_AP_N1 == 1) { + assert(0);// newVal[0] = VAL | mask; + } else { + enum { newWordBits = _AP_W % APINT_BITS_PER_WORD}; + // The extension is contained to the wordsBefore-1th word. + static const uint64_t mask = wordBits?(~0ULL<<(wordBits)):0ULL; + for (uint32_t i = 0; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + pVal[_AP_N1-1] |= mask; + } + for (uint32_t i=_AP_N1; i < _AP_N-1; i++) + pVal[i] = ~0ULL; + pVal[_AP_N-1] = ~0ULL; + clearUnusedBits(); + return; + } + + // Zero extend to a new width. + template + void cpZext(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + assert(_AP_W1 < BitWidth && "Invalid ap_private ZeroExtend request"); + assert(_AP_W1 <= MAX_INT_BITS && "Too many bits"); + uint32_t wordsAfter = _AP_N; + if (wordsAfter==1) { + assert(0); // return ap_private<_AP_W1, _AP_S, _AP_N1> (_AP_W1, VAL, _AP_S); + } else { + if (_AP_N1 == 1) { + assert(0); + // newVal[0] = VAL; + } else { + uint32_t i = 0; + for (; i < _AP_N1; ++i) + pVal[i] = that.pVal[i]; + for (; i < _AP_N; ++i) + pVal[i] = 0; + } + } + clearUnusedBits(); + } + + template + void cpZextOrTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + if (BitWidth > _AP_W1) + cpZext(that); + else if (BitWidth < _AP_W1) + cpTrunc(that); + else { + for (int i=0; i<_AP_N1; ++i) + pVal[i]=that.pVal[i]; + clearUnusedBits(); + } + } + + template + void cpSextOrTrunc(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) { + if (BitWidth > _AP_W1) + cpSext(that); + else if (BitWidth < _AP_W1) + cpTrunc(that); + else { + for (int i=0; i<_AP_N1; ++i) + pVal[i] = that.pVal[i]; + clearUnusedBits(); + } + } + + /// @name Value Characterization Functions + /// @{ + + /// @returns the total number of bits. + INLINE uint32_t getBitWidth() const { + return BitWidth; + } + + /// Here one word's bitwidth equals to that of uint64_t. + /// @returns the number of words to hold the integer value of this ap_private. + /// @brief Get the number of words. + INLINE uint32_t getNumWords() const { + return (BitWidth + APINT_BITS_PER_WORD - 1) / APINT_BITS_PER_WORD; + } + + /// This function returns the number of active bits which is defined as the + /// bit width minus the number of leading zeros. This is used in several + /// computations to see how "wide" the value is. + /// @brief Compute the number of active bits in the value + INLINE uint32_t getActiveBits() const { + uint32_t bits=BitWidth - countLeadingZeros(); + return bits?bits:1; + } + + + /// This method attempts to return the value of this ap_private as a zero extended + /// uint64_t. The bitwidth must be <= 64 or the value must fit within a + /// uint64_t. Otherwise an assertion will result. + /// @brief Get zero extended value + INLINE uint64_t getZExtValue() const { + assert(getActiveBits() <= 64 && "Too many bits for uint64_t"); + return *pVal; + } + + /// This method attempts to return the value of this ap_private as a sign extended + /// int64_t. The bit width must be <= 64 or the value must fit within an + /// int64_t. Otherwise an assertion will result. + /// @brief Get sign extended value + INLINE int64_t getSExtValue() const { + assert(getActiveBits() <= 64 && "Too many bits for int64_t"); + return int64_t(pVal[0]); + } + + /// This method determines how many bits are required to hold the ap_private + /// equivalent of the string given by \p str of length \p slen. + /// @brief Get bits required for string value. + static uint32_t getBitsNeeded(const char* str, uint32_t slen, uint8_t radix); + + /// countLeadingZeros - This function is an ap_private version of the + /// countLeadingZeros_{32,64} functions in MathExtras.h. It counts the number + /// of zeros from the most significant bit to the first one bit. + /// @returns BitWidth if the value is zero. + /// @returns the number of zeros from the most significant bit to the first + /// one bits. + INLINE uint32_t countLeadingZeros() const ; + + /// countLeadingOnes - This function counts the number of contiguous 1 bits + /// in the high order bits. The count stops when the first 0 bit is reached. + /// @returns 0 if the high order bit is not set + /// @returns the number of 1 bits from the most significant to the least + /// @brief Count the number of leading one bits. + INLINE uint32_t countLeadingOnes() const ; + + /// countTrailingZeros - This function is an ap_private version of the + /// countTrailingZoers_{32,64} functions in MathExtras.h. It counts + /// the number of zeros from the least significant bit to the first set bit. + /// @returns BitWidth if the value is zero. + /// @returns the number of zeros from the least significant bit to the first + /// one bit. + /// @brief Count the number of trailing zero bits. + INLINE uint32_t countTrailingZeros() const ; + + /// countPopulation - This function is an ap_private version of the + /// countPopulation_{32,64} functions in MathExtras.h. It counts the number + /// of 1 bits in the ap_private value. + /// @returns 0 if the value is zero. + /// @returns the number of set bits. + /// @brief Count the number of bits set. + INLINE uint32_t countPopulation() const { + uint32_t Count = 0; + for (uint32_t i = 0; i<_AP_N-1 ; ++i) + Count += CountPopulation_64(pVal[i]); + Count += CountPopulation_64(pVal[_AP_N-1]&mask); + return Count; + } + + /// @} + /// @name Conversion Functions + /// @{ + + /// This is used internally to convert an ap_private to a string. + /// @brief Converts an ap_private to a std::string + INLINE std::string toString(uint8_t radix, bool wantSigned) const + ; + + /// Considers the ap_private to be unsigned and converts it into a string in the + /// radix given. The radix can be 2, 8, 10 or 16. + /// @returns a character interpretation of the ap_private + /// @brief Convert unsigned ap_private to string representation. + INLINE std::string toStringUnsigned(uint8_t radix = 10) const { + return toString(radix, false); + } + + /// Considers the ap_private to be unsigned and converts it into a string in the + /// radix given. The radix can be 2, 8, 10 or 16. + /// @returns a character interpretation of the ap_private + /// @brief Convert unsigned ap_private to string representation. + INLINE std::string toStringSigned(uint8_t radix = 10) const { + return toString(radix, true); + } + + /// @returns a byte-swapped representation of this ap_private Value. + INLINE ap_private byteSwap() const ; + + /// @brief Converts this ap_private to a double value. + INLINE double roundToDouble(bool isSigned) const ; + + /// @brief Converts this unsigned ap_private to a double value. + INLINE double roundToDouble() const { + return roundToDouble(false); + } + + /// @brief Converts this signed ap_private to a double value. + INLINE double signedRoundToDouble() const { + return roundToDouble(true); + } + + /// The conversion does not do a translation from integer to double, it just + /// re-interprets the bits as a double. Note that it is valid to do this on + /// any bit width. Exactly 64 bits will be translated. + /// @brief Converts ap_private bits to a double + INLINE double bitsToDouble() const { + union { + uint64_t __I; + double __D; + } __T; + __T.__I = pVal[0]; + return __T.__D; + } + + /// The conversion does not do a translation from integer to float, it just + /// re-interprets the bits as a float. Note that it is valid to do this on + /// any bit width. Exactly 32 bits will be translated. + /// @brief Converts ap_private bits to a double + INLINE float bitsToFloat() const { + union { + uint32_t __I; + float __F; + } __T; + __T.__I = uint32_t(pVal[0]); + return __T.__F; + } + + /// The conversion does not do a translation from double to integer, it just + /// re-interprets the bits of the double. Note that it is valid to do this on + /// any bit width but bits from V may get truncated. + /// @brief Converts a double to ap_private bits. + INLINE ap_private& doubleToBits(double __V) { + union { + uint64_t __I; + double __D; + } __T; + __T.__D = __V; + pVal[0] = __T.__I; + return *this; + } + + /// The conversion does not do a translation from float to integer, it just + /// re-interprets the bits of the float. Note that it is valid to do this on + /// any bit width but bits from V may get truncated. + /// @brief Converts a float to ap_private bits. + INLINE ap_private& floatToBits(float __V) { + union { + uint32_t __I; + float __F; + } __T; + __T.__F = __V; + pVal[0] = __T.__I; + } + + //Reduce operation + //----------------------------------------------------------- + INLINE bool and_reduce() const { + return isMaxValue(); + } + + INLINE bool nand_reduce() const { + return isMinValue(); + } + + INLINE bool or_reduce() const { + return (bool)countPopulation(); + } + + INLINE bool nor_reduce() const { + return countPopulation()==0; + } + + INLINE bool xor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?true:false; + } + + INLINE bool xnor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?false:true; + } + INLINE std::string to_string(uint8_t radix=16, bool sign=false) const { + return toString(radix, radix==10?_AP_S:sign); + } +}; + +template +INLINE bool operator==(uint64_t V1, const ap_private<_AP_W, _AP_S, _AP_N>& V2) { + return V2 == V1; +} + +template +INLINE bool operator!=(uint64_t V1, const ap_private<_AP_W, _AP_S, _AP_N>& V2) { + return V2 != V1; +} + +namespace ap_private_ops { + enum {APINT_BITS_PER_WORD=64}; + /// @brief Determine the smaller of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> smin(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.slt(RHS) ? LHS : RHS; + } + + /// @brief Determine the larger of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> smax(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.sgt(RHS) ? LHS : RHS; + } + + /// @brief Determine the smaller of two ap_privates considered to be signed. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> umin(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.ult(RHS) ? LHS : RHS; + } + + /// @brief Determine the larger of two ap_privates considered to be unsigned. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> umax(const ap_private<_AP_W, _AP_S, _AP_N> &LHS, const ap_private<_AP_W, _AP_S1, _AP_N> &RHS) { + return LHS.ugt(RHS) ? LHS : RHS; + } + + /// @brief Check if the specified ap_private has a N-bits integer value. + template + INLINE bool isIntN(uint32_t __N, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.isIntN(__N); + } + + /// @returns true if the argument ap_private value is a sequence of ones + /// starting at the least significant bit with the remainder zero. + template + INLINE bool isMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.getBoolValue() && ((APIVal + ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) & APIVal) == 0; + } + + /// @returns true if the argument ap_private value contains a sequence of ones + /// with the remainder zero. + template + INLINE bool isShiftedMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return isMask(numBits, (APIVal - ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) | APIVal); + } + + /// @returns a byte-swapped representation of the specified ap_private Value. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> byteSwap(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.byteSwap(); + } + + /// @returns the floor log base 2 of the specified ap_private value. + template INLINE uint32_t logBase2(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.logBase2(); + } + + /// GreatestCommonDivisor - This function returns the greatest common + /// divisor of the two ap_private values using Enclid's algorithm. + /// @returns the greatest common divisor of Val1 and Val2 + /// @brief Compute GCD of two ap_private values. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> GreatestCommonDivisor(const ap_private<_AP_W, _AP_S, _AP_N>& Val1, const ap_private<_AP_W, _AP_S, _AP_N>& Val2) + ; + + /// Treats the ap_private as an unsigned value for conversion purposes. + /// @brief Converts the given ap_private to a double value. + template INLINE double Roundap_privateToDouble(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.roundToDouble(); + } + + /// Treats the ap_private as a signed value for conversion purposes. + /// @brief Converts the given ap_private to a double value. + template INLINE double RoundSignedap_privateToDouble(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.signedRoundToDouble(); + } + + /// @brief Converts the given ap_private to a float vlalue. + template INLINE float Roundap_privateToFloat(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return float(Roundap_privateToDouble(APIVal)); + } + + /// Treast the ap_private as a signed value for conversion purposes. + /// @brief Converts the given ap_private to a float value. + template INLINE float RoundSignedap_privateToFloat(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return float(APIVal.signedRoundToDouble()); + } + + /// RoundDoubleToap_private - This function convert a double value to an ap_private value. + /// @brief Converts the given double value into a ap_private. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> RoundDoubleToap_private(double Double, uint32_t width) ; + + /// RoundFloatToap_private - Converts a float value into an ap_private value. + /// @brief Converts a float value into a ap_private. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> RoundFloatToap_private(float Float, uint32_t width) { + return RoundDoubleToap_private<_AP_W, _AP_S, _AP_N>(double(Float), width); + } + + /// Arithmetic right-shift the ap_private by shiftAmt. + /// @brief Arithmetic right-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> ashr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.ashr(shiftAmt); + } + + /// Logical right-shift the ap_private by shiftAmt. + /// @brief Logical right-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> lshr(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.lshr(shiftAmt); + } + + /// Left-shift the ap_private by shiftAmt. + /// @brief Left-shift function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> shl(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t shiftAmt) { + return LHS.shl(shiftAmt); + } + + /// Signed divide ap_private LHS by ap_private RHS. + /// @brief Signed division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> sdiv(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.sdiv(RHS); + } + + /// Unsigned divide ap_private LHS by ap_private RHS. + /// @brief Unsigned division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> udiv(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.udiv(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> srem(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.srem(RHS); + } + + /// Unsigned remainder operation on ap_private. + /// @brief Function for unsigned remainder operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> urem(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS.urem(RHS); + } + + /// Performs multiplication on ap_private values. + /// @brief Function for multiplication operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> mul(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS * RHS; + } + + /// Performs addition on ap_private values. + /// @brief Function for addition operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> add(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS + RHS; + } + + /// Performs subtraction on ap_private values. + /// @brief Function for subtraction operation. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> sub(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS - RHS; + } + + /// Performs bitwise AND operation on ap_private LHS and + /// ap_private RHS. + /// @brief Bitwise AND function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> And(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS & RHS; + } + + /// Performs bitwise OR operation on ap_private LHS and ap_private RHS. + /// @brief Bitwise OR function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> Or(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS | RHS; + } + + /// Performs bitwise XOR operation on ap_private. + /// @brief Bitwise XOR function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1, _AP_N> Xor(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, const ap_private<_AP_W, _AP_S1, _AP_N>& RHS) { + return LHS ^ RHS; + } + + /// Performs a bitwise complement operation on ap_private. + /// @brief Bitwise complement function. + template INLINE ap_private<_AP_W, _AP_S, _AP_N> Not(const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return ~APIVal; + } + + template void clearUnusedBits(uint64_t& msw) { + // Compute how many bits are used in the final word + // uint32_t wordBits = APIVal.getBitWidth() & 0x3f; + if (wordBits == 0) + // If all bits are used, we want to leave the value alone. This also + // avoids the undefined behavior of >> when the shfit is the same size as + // the word size (64). + return; + + // Mask out the hight bits. + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- wordBits); + msw &= mask; + } + template <> INLINE void clearUnusedBits<1>(uint64_t& msw) { + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- 1); + msw &= mask; + } + template void clearUnusedBits(int64_t& msw) { + // Compute how many bits are used in the final word + // uint32_t wordBits = APIVal.getBitWidth() & 0x3f; + if (wordBits == 0) + // If all bits are used, we want to leave the value alone. This also + // avoids the undefined behavior of >> when the shfit is the same size as + // the word size (64). + return; + + // Mask out the hight bits. + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- wordBits); + msw &= mask; + } + template <> INLINE void clearUnusedBits<1>(int64_t& msw) { + uint64_t mask = ~uint64_t(0ULL) >> (64 /*ap_private::APINT_BITS_PER_WORD */- 1); + msw &= mask; + } + // template + template + INLINE ap_private<_AP_W, _AP_S> ashr(const ap_private<_AP_W, _AP_S>& a) { + return ashr(a, shiftAmt); + } + + template + INLINE ap_private<_AP_W, _AP_S> lshr(const ap_private<_AP_W, _AP_S>& a) { + return lshr(a, shiftAmt); + } + + template + INLINE ap_private<_AP_W, true> ashr(const ap_private<_AP_W, true, 1>& a) { + enum {APINT_BITS_PER_WORD=64, excess_bits=APINT_BITS_PER_WORD-_AP_W}; + static const uint64_t sign_bit = (1ULL<<(_AP_W-1)); + static const uint64_t sign_ext_mask = (_AP_W-shiftAmt>0)?~0ULL<<(APINT_BITS_PER_WORD-_AP_W+shiftAmt):~0ULL; + return ap_private<_AP_W, true>((((int64_t)a.VAL) >> (shiftAmt)) | (a.VAL & sign_bit? sign_ext_mask : 0ULL)); + } + + template + INLINE ap_private<_AP_W, false> ashr(const ap_private<_AP_W, false, 1>& a) { + return ap_private<_AP_W, false>((a.VAL) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W, _AP_S> lshr(const ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask = ~0ULL<<_AP_W; + return ap_private<_AP_W, _AP_S>((a.VAL&mask) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W-shiftAmt, _AP_S> shr(const ap_private<_AP_W, _AP_S>& a) { + return ap_private<_AP_W-shiftAmt, _AP_S>((a.VAL) >> (shiftAmt)); + } + + template + INLINE ap_private<_AP_W+shiftAmt, _AP_S> shl(const ap_private<_AP_W, _AP_S>& a) { + return ap_private<_AP_W+shiftAmt, _AP_S>((a.VAL) << (shiftAmt)); + } + + template + INLINE bool get(const ap_private<_AP_W, _AP_S, 1>& a) { + unsigned shift = (index%APINT_BITS_PER_WORD); + static const uint64_t mask=1ULL << (shift); + return ((mask & a.VAL) != 0); + } + + template + INLINE bool get(const ap_private<_AP_W, _AP_S>& a) { + static const uint64_t mask=1ULL << (index&0x3f); + return ((mask & a.pVal[(index)>>6]) != 0); + } + + template + INLINE void set(ap_private<_AP_W, _AP_S, 1>& a) { + const uint64_t mask = ~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.VAL |= mask; + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask = ~(~0ULL >> (lsb) <<(APINT_BITS_PER_WORD-msb+lsb-1) >> (APINT_BITS_PER_WORD-msb-1)); + a.VAL &= mask; + } + + template + INLINE void set(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, + lsb_word = lsb_index /APINT_BITS_PER_WORD, + msb_word = msb_index / APINT_BITS_PER_WORD, + msb = msb_index % APINT_BITS_PER_WORD, + lsb=lsb_index % APINT_BITS_PER_WORD}; + if (msb_word==lsb_word) { + const uint64_t mask = ~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.pVal[msb_word] |= mask; + } else { + const uint64_t lsb_mask = ~0ULL >> (lsb) << (lsb); + const uint64_t msb_mask = ~0ULL << (APINT_BITS_PER_WORD-msb-1)>>(APINT_BITS_PER_WORD-msb-1); + a.pVal[lsb_word] |=lsb_mask; + for (int i=lsb_word+1; i + INLINE void clear(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, + lsb_word = lsb_index /APINT_BITS_PER_WORD, + msb_word = msb_index / APINT_BITS_PER_WORD, + msb = msb_index % APINT_BITS_PER_WORD, + lsb=lsb_index % APINT_BITS_PER_WORD}; + if (msb_word == lsb_word) { + const uint64_t mask = ~(~0ULL >> (lsb) << (APINT_BITS_PER_WORD-msb+lsb-1)>>(APINT_BITS_PER_WORD-msb-1)); + a.pVal[msb_word] &= mask; + } else { + const uint64_t lsb_mask = ~(~0ULL >> (lsb) << (lsb)); + const uint64_t msb_mask = ~(~0ULL << (APINT_BITS_PER_WORD-msb-1)>>(APINT_BITS_PER_WORD-msb-1)); + a.pVal[lsb_word] &=lsb_mask; + for (int i=lsb_word+1; i + INLINE void set(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask=1ULL << (index); + a.VAL |= mask; + a.clearUnusedBits(); + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S, 1>& a) { + static const uint64_t mask=~(1ULL << (index)); + a.VAL &= mask; + a.clearUnusedBits(); + } + + template + INLINE void set(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, word = index/APINT_BITS_PER_WORD}; + static const uint64_t mask=1ULL << (index%APINT_BITS_PER_WORD); + a.pVal[word] |= mask; + a.clearUnusedBits(); + } + + template + INLINE void clear(ap_private<_AP_W, _AP_S>& a) { + enum { APINT_BITS_PER_WORD=64, word = index/APINT_BITS_PER_WORD}; + static const uint64_t mask=~(1ULL << (index%APINT_BITS_PER_WORD)); + a.pVal[word] &= mask; + a.clearUnusedBits(); + } + + template + INLINE bool isNegative(const ap_private<_AP_W, false>& a) { + return false; + } + + template + INLINE bool isNegative(const ap_private<_AP_W, true, 1>& a) { + static const uint64_t sign_mask = (1ULL << (_AP_W-1)); + return ((sign_mask & a.VAL) != 0); + } + + template + INLINE bool isNegative(const ap_private<_AP_W, true>& a) { + enum {APINT_BITS_PER_WORD=64,_AP_N=(_AP_W+APINT_BITS_PER_WORD-1)/APINT_BITS_PER_WORD}; + static const uint64_t sign_mask = (1ULL << (_AP_W%APINT_BITS_PER_WORD-1)); + return sign_mask & a.pVal[_AP_N-1]; + } +} // End of ap_private_ops namespace + +/// @brief Check if the specified ap_private has a N-bits integer value. +template +INLINE bool isIntN(uint32_t __N, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.isIntN(__N); +} + +/// @returns true if the argument ap_private value is a sequence of ones +/// starting at the least significant bit with the remainder zero. +template +INLINE bool isMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return APIVal.getBoolValue() && ((APIVal + ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) & APIVal) == 0; +} + +/// @returns true if the argument ap_private value contains a sequence of ones +/// with the remainder zero. +template +INLINE bool isShiftedMask(uint32_t numBits, const ap_private<_AP_W, _AP_S, _AP_N>& APIVal) { + return isMask(numBits, (APIVal - ap_private<_AP_W, _AP_S, _AP_N>(numBits,1)) | APIVal); +} + +#if 0 +/// add_1 - This function adds a single "digit" integer, y, to the multiple +/// "digit" integer array, x[]. x[] is modified to reflect the addition and +/// 1 is returned if there is a carry out, otherwise 0 is returned. +/// @returns the carry of the addition. +static bool add_1(uint64_t dest[], uint64_t x[], uint32_t len, uint64_t y) { + for (uint32_t i = 0; i < len; ++i) { + dest[i] = y + x[i]; + if (dest[i] < y) + y = 1; // Carry one to next digit. + else { + y = 0; // No need to carry so exit early + break; + } + } + return (y != 0); +} +#endif + +#if 0 +/// add - This function adds the integer array x to the integer array Y and +/// places the result in dest. +/// @returns the carry out from the addition +/// @brief General addition of 64-bit integer arrays +static bool add(uint64_t *dest, const uint64_t *x, const uint64_t *y, + uint32_t destlen, uint32_t xlen, uint32_t ylen, bool xsigned, bool ysigned) { + bool carry = false; + uint32_t len = AESL_std::min(xlen, ylen); + uint32_t i; + for (i = 0; i< len && i < destlen; ++i) { + uint64_t limit = AESL_std::min(x[i],y[i]); // must come first in case dest == x + dest[i] = x[i] + y[i] + carry; + carry = dest[i] < limit || (carry && dest[i] == limit); + } + if (xlen > ylen) { + const uint64_t yext = xsigned && int64_t(y[ylen-1])<0 ? -1 : 0; + for (i=ylen; i< xlen && i < destlen; i++) { + uint64_t limit = AESL_std::min(x[i], yext); + dest[i] = x[i] + yext + carry; + carry = (dest[i] < limit)||(carry && dest[i] == x[i]); + } + } else if (ylen> xlen) { + const uint64_t xext = ysigned && int64_t(x[xlen-1])<0 ? -1 : 0; + for (i=xlen; i< ylen && i < destlen; i++) { + uint64_t limit = AESL_std::min(xext, y[i]); + dest[i] = xext + y[i] + carry; + carry = (dest[i] < limit)||(carry && dest[i] == y[i]); + } + } + return carry; +} +#endif + +#if 0 +/// @returns returns the borrow out. +/// @brief Generalized subtraction of 64-bit integer arrays. +static bool sub(uint64_t *dest, const uint64_t *x, const uint64_t *y, + uint32_t destlen, uint32_t xlen, uint32_t ylen, bool xsigned, bool ysigned) { + bool borrow = false; + uint32_t i; + uint32_t len = AESL_std::min(xlen, ylen); + for (i = 0; i < len && i < destlen; ++i) { + uint64_t x_tmp = borrow ? x[i] - 1 : x[i]; + borrow = y[i] > x_tmp || (borrow && x[i] == 0); + dest[i] = x_tmp - y[i]; + } + if (xlen > ylen) { + const uint64_t yext = ysigned && int64_t(y[ylen-1])<0 ? -1 : 0; + for (i=ylen; i< xlen && i < destlen; i++) { + uint64_t x_tmp = borrow ? x[i] - 1 : x[i]; + borrow = yext > x_tmp || (borrow && x[i] == 0); + dest[i] = x_tmp - yext; + } + } else if (ylen> xlen) { + const uint64_t xext = xsigned && int64_t(x[xlen-1])<0 ? -1 : 0; + for (i=xlen; i< ylen && i < destlen; i++) { + uint64_t x_tmp = borrow ? xext - 1 : xext; + borrow = y[i] > x_tmp || (borrow && xext==0); + dest[i] = x_tmp - y[i]; + } + } + return borrow; +} +#endif + +/// Subtracts the RHS ap_private from this ap_private +/// @returns this, after subtraction +/// @brief Subtraction assignment operator. + +#if 0 +/// Multiplies an integer array, x by a a uint64_t integer and places the result +/// into dest. +/// @returns the carry out of the multiplication. +/// @brief Multiply a multi-digit ap_private by a single digit (64-bit) integer. +static uint64_t mul_1(uint64_t dest[], const uint64_t x[], uint32_t len, uint64_t y) { + // Split y into high 32-bit part (hy) and low 32-bit part (ly) + uint64_t ly = y & 0xffffffffULL, hy = (y) >> 32; + uint64_t carry = 0; + static const uint64_t two_power_32 = 1ULL << 32; + // For each digit of x. + for (uint32_t i = 0; i < len; ++i) { + // Split x into high and low words + uint64_t lx = x[i] & 0xffffffffULL; + uint64_t hx = (x[i]) >> 32; + // hasCarry - A flag to indicate if there is a carry to the next digit. + // hasCarry == 0, no carry + // hasCarry == 1, has carry + // hasCarry == 2, no carry and the calculation result == 0. + uint8_t hasCarry = 0; + dest[i] = carry + lx * ly; + // Determine if the add above introduces carry. + hasCarry = (dest[i] < carry) ? 1 : 0; + carry = hx * ly + ((dest[i]) >> 32) + (hasCarry ? two_power_32 : 0); + // The upper limit of carry can be (2^32 - 1)(2^32 - 1) + + // (2^32 - 1) + 2^32 = 2^64. + hasCarry = (!carry && hasCarry) ? 1 : (!carry ? 2 : 0); + + carry += (lx * hy) & 0xffffffffULL; + dest[i] = ((carry) << 32) | (dest[i] & 0xffffffffULL); + carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? two_power_32 : 0) + + ((carry) >> 32) + ((lx * hy) >> 32) + hx * hy; + } + return carry; +} +#endif + +#if 0 +/// Multiplies integer array x by integer array y and stores the result into +/// the integer array dest. Note that dest's size must be >= xlen + ylen. +/// @brief Generalized multiplicate of integer arrays. +static void mul(uint64_t dest[], const uint64_t x[], uint32_t xlen, const uint64_t y[], + uint32_t ylen, uint32_t destlen) { + dest[xlen] = mul_1(dest, x, xlen, y[0]); + for (uint32_t i = 1; i < ylen; ++i) { + uint64_t ly = y[i] & 0xffffffffULL, hy = (y[i]) >> 32; + uint64_t carry = 0, lx = 0, hx = 0; + for (uint32_t j = 0; j < xlen; ++j) { + lx = x[j] & 0xffffffffULL; + hx = (x[j]) >> 32; + // hasCarry - A flag to indicate if has carry. + // hasCarry == 0, no carry + // hasCarry == 1, has carry + // hasCarry == 2, no carry and the calculation result == 0. + uint8_t hasCarry = 0; + uint64_t resul = carry + lx * ly; + hasCarry = (resul < carry) ? 1 : 0; + carry = (hasCarry ? (1ULL << 32) : 0) + hx * ly + ((resul) >> 32); + hasCarry = (!carry && hasCarry) ? 1 : (!carry ? 2 : 0); + carry += (lx * hy) & 0xffffffffULL; + resul = ((carry) << 32) | (resul & 0xffffffffULL); + dest[i+j] += resul; + carry = (((!carry && hasCarry != 2) || hasCarry == 1) ? (1ULL << 32) : 0)+ + ((carry) >> 32) + (dest[i+j] < resul ? 1 : 0) + + ((lx * hy) >> 32) + hx * hy; + } + dest[i+xlen] = carry; + } +} +#endif + + + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { + assert(str != 0 && "Invalid value string"); + assert(slen > 0 && "Invalid string length"); + + // Each computation below needs to know if its negative + uint32_t isNegative = str[0] == '-'; + if (isNegative) { + slen--; + str++; + } + // For radixes of power-of-two values, the bits required is accurately and + // easily computed + if (radix == 2) + return slen + isNegative; + if (radix == 8) + return slen * 3 + isNegative; + if (radix == 16) + return slen * 4 + isNegative; + + // Otherwise it must be radix == 10, the hard case + assert(radix == 10 && "Invalid radix"); + + // Convert to the actual binary value. + //ap_private<_AP_W, _AP_S, _AP_N> tmp(sufficient, str, slen, radix); + + // Compute how many bits are required. + //return isNegative + tmp.logBase2() + 1; + return isNegative + slen * 4; +} + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countLeadingZeros() const { + enum { msw_bits = (BitWidth % APINT_BITS_PER_WORD)?(BitWidth % APINT_BITS_PER_WORD):APINT_BITS_PER_WORD, + excessBits = APINT_BITS_PER_WORD - msw_bits }; + uint32_t Count = CountLeadingZeros_64(pVal[_AP_N-1]); + if (Count>=excessBits) + Count -= excessBits; + if (!pVal[_AP_N-1]) { + for (uint32_t i = _AP_N-1 ; i ; --i) { + if (!pVal[i-1]) + Count += APINT_BITS_PER_WORD; + else { + Count += CountLeadingZeros_64(pVal[i-1]); + break; + } + } + } + return Count; +} + +static uint32_t countLeadingOnes_64(uint64_t __V, uint32_t skip) { + uint32_t Count = 0; + if (skip) + (__V) <<= (skip); + while (__V && (__V & (1ULL << 63))) { + Count++; + (__V) <<= 1; + } + return Count; +} + +template +uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countLeadingOnes() const { + if (isSingleWord()) + return countLeadingOnes_64(VAL, APINT_BITS_PER_WORD - BitWidth); + + uint32_t highWordBits = BitWidth % APINT_BITS_PER_WORD; + uint32_t shift = (highWordBits == 0 ? 0 : APINT_BITS_PER_WORD - highWordBits); + int i = _AP_N - 1; + uint32_t Count = countLeadingOnes_64(pVal[i], shift); + if (Count == highWordBits) { + for (i--; i >= 0; --i) { + if (pVal[i] == ~0ULL) + Count += APINT_BITS_PER_WORD; + else { + Count += countLeadingOnes_64(pVal[i], 0); + break; + } + } + } + return Count; +} + +template +INLINE uint32_t ap_private<_AP_W, _AP_S, _AP_N>::countTrailingZeros() const { + if (isSingleWord()) + return AESL_std::min(uint32_t(CountTrailingZeros_64(VAL)), BitWidth); + uint32_t Count = 0; + uint32_t i = 0; + for (; i < _AP_N && pVal[i] == 0; ++i) + Count += APINT_BITS_PER_WORD; + if (i < _AP_N) + Count += CountTrailingZeros_64(pVal[i]); + return AESL_std::min(Count, BitWidth); +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private<_AP_W, _AP_S, _AP_N>::byteSwap() const { + assert(BitWidth >= 16 && BitWidth % 16 == 0 && "Cannot byteswap!"); + if (BitWidth == 16) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_16(uint16_t(VAL))); + else if (BitWidth == 32) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_32(uint32_t(VAL))); + else if (BitWidth == 48) { + uint32_t Tmp1 = uint32_t((VAL) >> 16); + Tmp1 = ByteSwap_32(Tmp1); + uint16_t Tmp2 = uint16_t(VAL); + Tmp2 = ByteSwap_16(Tmp2); + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ((uint64_t(Tmp2)) << 32) | Tmp1); + } else if (BitWidth == 64) + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ ByteSwap_64(VAL)); + else { + ap_private<_AP_W, _AP_S, _AP_N> Result(0); + char *pByte = (char*)Result.pVal; + for (uint32_t i = 0; i < BitWidth / APINT_WORD_SIZE / 2; ++i) { + char Tmp = pByte[i]; + pByte[i] = pByte[BitWidth / APINT_WORD_SIZE - 1 - i]; + pByte[BitWidth / APINT_WORD_SIZE - i - 1] = Tmp; + } + return Result; + } +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private_ops::GreatestCommonDivisor(const ap_private<_AP_W, _AP_S, _AP_N>& API1, const ap_private<_AP_W, _AP_S, _AP_N>& API2) { + ap_private<_AP_W, _AP_S, _AP_N> __A = API1, __B = API2; + while (!!__B) { + ap_private<_AP_W, _AP_S, _AP_N> __T = __B; + __B = ap_private_ops::urem(__A, __B); + __A = __T; + } + return __A; +} + +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private_ops::RoundDoubleToap_private(double Double, uint32_t width) { + union { + double __D; + uint64_t __I; + } __T; + __T.__D = Double; + + // Get the sign bit from the highest order bit + bool isNeg = (__T.__I) >> 63; + + // Get the 11-bit exponent and adjust for the 1023 bit bias + int64_t exp = (((__T.__I) >> 52) & 0x7ffULL) - 1023; + + // If the exponent is negative, the value is < 0 so just return 0. + if (exp < 0) + return ap_private<_AP_W, _AP_S, _AP_N>(width, 0u); + + // Extract the mantissa by clearing the top 12 bits (sign + exponent). + uint64_t mantissa = (__T.__I & (~0ULL >> 12)) | 1ULL << 52; + + // If the exponent doesn't shift all bits out of the mantissa + if (exp < 52) + return isNeg ? -ap_private<_AP_W, _AP_S, _AP_N>(width, (mantissa) >> (52 - exp)) : + ap_private<_AP_W, _AP_S, _AP_N>((mantissa) >> (52 - exp)); + + // If the client didn't provide enough bits for us to shift the mantissa into + // then the result is undefined, just return 0 + if (width <= exp - 52) + return ap_private<_AP_W, _AP_S, _AP_N>(width, 0); + + // Otherwise, we have to shift the mantissa bits up to the right location + ap_private<_AP_W, _AP_S, _AP_N> Tmp(width, mantissa); + Tmp = Tmp.shl(exp - 52); + return isNeg ? -Tmp : Tmp; +} + +/// RoundToDouble - This function convert this ap_private to a double. +/// The layout for double is as following (IEEE Standard 754): +/// -------------------------------------- +/// | Sign Exponent Fraction Bias | +/// |-------------------------------------- | +/// | 1[63] 11[62-52] 52[51-00] 1023 | +/// -------------------------------------- +template +double ap_private<_AP_W, _AP_S, _AP_N>::roundToDouble(bool isSigned) const { + + // Handle the simple case where the value is contained in one uint64_t. + if (isSingleWord() || getActiveBits() <= APINT_BITS_PER_WORD) { + uint64_t val; + if (isSingleWord()) val = VAL; + else val = pVal[0]; + if (isSigned) { + int64_t sext = ((int64_t(val)) << (64-BitWidth)) >> (64-BitWidth); + return double(sext); + } else + return double(val); + } + + // Determine if the value is negative. + bool isNeg = isSigned ? (*this)[BitWidth-1] : false; + + // Construct the absolute value if we're negative. + ap_private<_AP_W, _AP_S, _AP_N> Tmp(isNeg ? -(*this) : (*this)); + + // Figure out how many bits we're using. + uint32_t n = Tmp.getActiveBits(); + + // The exponent (without bias normalization) is just the number of bits + // we are using. Note that the sign bit is gone since we constructed the + // absolute value. + uint64_t exp = n; + + // Return infinity for exponent overflow + if (exp > 1023) { + if (!isSigned || !isNeg) + return std::numeric_limits::infinity(); + else + return -std::numeric_limits::infinity(); + } + exp += 1023; // Increment for 1023 bias + + // Number of bits in mantissa is 52. To obtain the mantissa value, we must + // extract the high 52 bits from the correct words in pVal. + uint64_t mantissa; + unsigned hiWord = whichWord(n-1); + if (hiWord == 0) { + mantissa = Tmp.pVal[0]; + if (n > 52) + (mantissa) >>= (n - 52); // shift down, we want the top 52 bits. + } else { + assert(hiWord > 0 && "High word is negative?"); + uint64_t hibits = (Tmp.pVal[hiWord]) << (52 - n % APINT_BITS_PER_WORD); + uint64_t lobits = (Tmp.pVal[hiWord-1]) >> (11 + n % APINT_BITS_PER_WORD); + mantissa = hibits | lobits; + } + + // The leading bit of mantissa is implicit, so get rid of it. + uint64_t sign = isNeg ? (1ULL << (APINT_BITS_PER_WORD - 1)) : 0; + union { + double __D; + uint64_t __I; + } __T; + __T.__I = sign | ((exp) << 52) | mantissa; + return __T.__D; +} + +// Square Root - this method computes and returns the square root of "this". +// Three mechanisms are used for computation. For small values (<= 5 bits), +// a table lookup is done. This gets some performance for common cases. For +// values using less than 52 bits, the value is converted to double and then +// the libc sqrt function is called. The result is rounded and then converted +// back to a uint64_t which is then used to construct the result. Finally, +// the Babylonian method for computing square roots is used. +template +ap_private<_AP_W, _AP_S, _AP_N> ap_private<_AP_W, _AP_S, _AP_N>::sqrt() const { + + // Determine the magnitude of the value. + uint32_t magnitude = getActiveBits(); + + // Use a fast table for some small values. This also gets rid of some + // rounding errors in libc sqrt for small values. + if (magnitude <= 5) { + static const uint8_t results[32] = { + /* 0 */ 0, + /* 1- 2 */ 1, 1, + /* 3- 6 */ 2, 2, 2, 2, + /* 7-12 */ 3, 3, 3, 3, 3, 3, + /* 13-20 */ 4, 4, 4, 4, 4, 4, 4, 4, + /* 21-30 */ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + /* 31 */ 6 + }; + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ results[ (isSingleWord() ? VAL : pVal[0]) ]); + } + + // If the magnitude of the value fits in less than 52 bits (the precision of + // an IEEE double precision floating point value), then we can use the + // libc sqrt function which will probably use a hardware sqrt computation. + // This should be faster than the algorithm below. + if (magnitude < 52) { +#ifdef _MSC_VER + // Amazingly, VC++ doesn't have round(). + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ + uint64_t(::sqrt(double(isSingleWord()?VAL:pVal[0]))) + 0.5); +#else + return ap_private<_AP_W, _AP_S, _AP_N>(/*BitWidth,*/ + uint64_t(::round(::sqrt(double(isSingleWord()?VAL:pVal[0]))))); +#endif + } + + // Okay, all the short cuts are exhausted. We must compute it. The following + // is a classical Babylonian method for computing the square root. This code + // was adapted to APINt from a wikipedia article on such computations. + // See http://www.wikipedia.org/ and go to the page named + // Calculate_an_integer_square_root. + uint32_t nbits = BitWidth, i = 4; + ap_private<_AP_W, _AP_S, _AP_N> testy(16); + ap_private<_AP_W, _AP_S, _AP_N> x_old(/*BitWidth,*/ 1); + ap_private<_AP_W, _AP_S, _AP_N> x_new(0); + ap_private<_AP_W, _AP_S, _AP_N> two(/*BitWidth,*/ 2); + + // Select a good starting value using binary logarithms. + for (;; i += 2, testy = testy.shl(2)) + if (i >= nbits || this->ule(testy)) { + x_old = x_old.shl(i / 2); + break; + } + + // Use the Babylonian method to arrive at the integer square root: + for (;;) { + x_new = (this->udiv(x_old) + x_old).udiv(two); + if (x_old.ule(x_new)) + break; + x_old = x_new; + } + + // Make sure we return the closest approximation + // NOTE: The rounding calculation below is correct. It will produce an + // off-by-one discrepancy with results from pari/gp. That discrepancy has been + // determined to be a rounding issue with pari/gp as it begins to use a + // floating point representation after 192 bits. There are no discrepancies + // between this algorithm and pari/gp for bit widths < 192 bits. + ap_private<_AP_W, _AP_S, _AP_N> square(x_old * x_old); + ap_private<_AP_W, _AP_S, _AP_N> nextSquare((x_old + 1) * (x_old +1)); + if (this->ult(square)) + return x_old; + else if (this->ule(nextSquare)) { + ap_private<_AP_W, _AP_S, _AP_N> midpoint((nextSquare - square).udiv(two)); + ap_private<_AP_W, _AP_S, _AP_N> offset(*this - square); + if (offset.ult(midpoint)) + return x_old; + else + return x_old + 1; + } else + assert(0 && "Error in ap_private<_AP_W, _AP_S, _AP_N>::sqrt computation"); + return x_old + 1; +} + +/// Implementation of Knuth's Algorithm D (Division of nonnegative integers) +/// from "Art of Computer Programming, Volume 2", section 4.3.1, p. 272. The +/// variables here have the same names as in the algorithm. Comments explain +/// the algorithm and any deviation from it. +static void KnuthDiv(uint32_t *u, uint32_t *v, uint32_t *q, uint32_t* r, + uint32_t m, uint32_t n) { + assert(u && "Must provide dividend"); + assert(v && "Must provide divisor"); + assert(q && "Must provide quotient"); + assert(u != v && u != q && v != q && "Must us different memory"); + assert(n>1 && "n must be > 1"); + + // Knuth uses the value b as the base of the number system. In our case b + // is 2^31 so we just set it to -1u. + uint64_t b = uint64_t(1) << 32; + + //DEBUG(cerr << "KnuthDiv: m=" << m << " n=" << n << '\n'); + //DEBUG(cerr << "KnuthDiv: original:"); + //DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); + //DEBUG(cerr << " by"); + //DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); + //DEBUG(cerr << '\n'); + // D1. [Normalize.] Set d = b / (v[n-1] + 1) and multiply all the digits of + // u and v by d. Note that we have taken Knuth's advice here to use a power + // of 2 value for d such that d * v[n-1] >= b/2 (b is the base). A power of + // 2 allows us to shift instead of multiply and it is easy to determine the + // shift amount from the leading zeros. We are basically normalizing the u + // and v so that its high bits are shifted to the top of v's range without + // overflow. Note that this can require an extra word in u so that u must + // be of length m+n+1. + uint32_t shift = CountLeadingZeros_32(v[n-1]); + uint32_t v_carry = 0; + uint32_t u_carry = 0; + if (shift) { + for (uint32_t i = 0; i < m+n; ++i) { + uint32_t u_tmp = (u[i]) >> (32 - shift); + u[i] = ((u[i]) << (shift)) | u_carry; + u_carry = u_tmp; + } + for (uint32_t i = 0; i < n; ++i) { + uint32_t v_tmp = (v[i]) >> (32 - shift); + v[i] = ((v[i]) << (shift)) | v_carry; + v_carry = v_tmp; + } + } + u[m+n] = u_carry; + //DEBUG(cerr << "KnuthDiv: normal:"); + //DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << std::setbase(16) << u[i]); + //DEBUG(cerr << " by"); + //DEBUG(for (int i = n; i >0; i--) cerr << " " << std::setbase(16) << v[i-1]); + //DEBUG(cerr << '\n'); + + // D2. [Initialize j.] Set j to m. This is the loop counter over the places. + int j = m; + do { + //DEBUG(cerr << "KnuthDiv: quotient digit #" << j << '\n'); + // D3. [Calculate q'.]. + // Set qp = (u[j+n]*b + u[j+n-1]) / v[n-1]. (qp=qprime=q') + // Set rp = (u[j+n]*b + u[j+n-1]) % v[n-1]. (rp=rprime=r') + // Now test if qp == b or qp*v[n-2] > b*rp + u[j+n-2]; if so, decrease + // qp by 1, inrease rp by v[n-1], and repeat this test if rp < b. The test + // on v[n-2] determines at high speed most of the cases in which the trial + // value qp is one too large, and it eliminates all cases where qp is two + // too large. + uint64_t dividend = ((uint64_t(u[j+n]) << 32) + u[j+n-1]); + //DEBUG(cerr << "KnuthDiv: dividend == " << dividend << '\n'); + uint64_t qp = dividend / v[n-1]; + uint64_t rp = dividend % v[n-1]; + if (qp == b || qp*v[n-2] > b*rp + u[j+n-2]) { + qp--; + rp += v[n-1]; + if (rp < b && (qp == b || qp*v[n-2] > b*rp + u[j+n-2])) + qp--; + } + //DEBUG(cerr << "KnuthDiv: qp == " << qp << ", rp == " << rp << '\n'); + + // D4. [Multiply and subtract.] Replace (u[j+n]u[j+n-1]...u[j]) with + // (u[j+n]u[j+n-1]..u[j]) - qp * (v[n-1]...v[1]v[0]). This computation + // consists of a simple multiplication by a one-place number, combined with + // a subtraction. + bool isNeg = false; + for (uint32_t i = 0; i < n; ++i) { + uint64_t u_tmp = uint64_t(u[j+i]) | ((uint64_t(u[j+i+1])) << 32); + uint64_t subtrahend = uint64_t(qp) * uint64_t(v[i]); + bool borrow = subtrahend > u_tmp; + /*DEBUG(cerr << "KnuthDiv: u_tmp == " << u_tmp + << ", subtrahend == " << subtrahend + << ", borrow = " << borrow << '\n');*/ + + uint64_t result = u_tmp - subtrahend; + uint32_t k = j + i; + u[k++] = (uint32_t)(result & (b-1)); // subtract low word + u[k++] = (uint32_t)((result) >> 32); // subtract high word + while (borrow && k <= m+n) { // deal with borrow to the left + borrow = u[k] == 0; + u[k]--; + k++; + } + isNeg |= borrow; + /*DEBUG(cerr << "KnuthDiv: u[j+i] == " << u[j+i] << ", u[j+i+1] == " << + u[j+i+1] << '\n');*/ + } + /*DEBUG(cerr << "KnuthDiv: after subtraction:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); + DEBUG(cerr << '\n');*/ + // The digits (u[j+n]...u[j]) should be kept positive; if the result of + // this step is actually negative, (u[j+n]...u[j]) should be left as the + // true value plus b**(n+1), namely as the b's complement of + // the true value, and a "borrow" to the left should be remembered. + // + if (isNeg) { + bool carry = true; // true because b's complement is "complement + 1" + for (uint32_t i = 0; i <= m+n; ++i) { + u[i] = ~u[i] + carry; // b's complement + carry = carry && u[i] == 0; + } + } + /*DEBUG(cerr << "KnuthDiv: after complement:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr << " " << u[i]); + DEBUG(cerr << '\n');*/ + + // D5. [Test remainder.] Set q[j] = qp. If the result of step D4 was + // negative, go to step D6; otherwise go on to step D7. + q[j] = (uint32_t)qp; + if (isNeg) { + // D6. [Add back]. The probability that this step is necessary is very + // small, on the order of only 2/b. Make sure that test data accounts for + // this possibility. Decrease q[j] by 1 + q[j]--; + // and add (0v[n-1]...v[1]v[0]) to (u[j+n]u[j+n-1]...u[j+1]u[j]). + // A carry will occur to the left of u[j+n], and it should be ignored + // since it cancels with the borrow that occurred in D4. + bool carry = false; + for (uint32_t i = 0; i < n; i++) { + uint32_t limit = AESL_std::min(u[j+i],v[i]); + u[j+i] += v[i] + carry; + carry = u[j+i] < limit || (carry && u[j+i] == limit); + } + u[j+n] += carry; + } + /*DEBUG(cerr << "KnuthDiv: after correction:"); + DEBUG(for (int i = m+n; i >=0; i--) cerr <<" " << u[i]); + DEBUG(cerr << "\nKnuthDiv: digit result = " << q[j] << '\n');*/ + + // D7. [Loop on j.] Decrease j by one. Now if j >= 0, go back to D3. + } while (--j >= 0); + + /*DEBUG(cerr << "KnuthDiv: quotient:"); + DEBUG(for (int i = m; i >=0; i--) cerr <<" " << q[i]); + DEBUG(cerr << '\n');*/ + + // D8. [Unnormalize]. Now q[...] is the desired quotient, and the desired + // remainder may be obtained by dividing u[...] by d. If r is non-null we + // compute the remainder (urem uses this). + if (r) { + // The value d is expressed by the "shift" value above since we avoided + // multiplication by d by using a shift left. So, all we have to do is + // shift right here. In order to mak + if (shift) { + uint32_t carry = 0; + //DEBUG(cerr << "KnuthDiv: remainder:"); + for (int i = n-1; i >= 0; i--) { + r[i] = ((u[i]) >> (shift)) | carry; + carry = (u[i]) << (32 - shift); + //DEBUG(cerr << " " << r[i]); + } + } else { + for (int i = n-1; i >= 0; i--) { + r[i] = u[i]; + //DEBUG(cerr << " " << r[i]); + } + } + //DEBUG(cerr << '\n'); + } + //DEBUG(cerr << std::setbase(10) << '\n'); +} + +template +void divide(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t lhsWords, + const ap_private<_AP_W, _AP_S, _AP_N>& RHS, uint32_t rhsWords, + ap_private<_AP_W, _AP_S, _AP_N> *Quotient, ap_private<_AP_W, _AP_S, _AP_N> *Remainder) { + assert(lhsWords >= rhsWords && "Fractional result"); + enum {APINT_BITS_PER_WORD=64}; + // First, compose the values into an array of 32-bit words instead of + // 64-bit words. This is a necessity of both the "short division" algorithm + // and the the Knuth "classical algorithm" which requires there to be native + // operations for +, -, and * on an m bit value with an m*2 bit result. We + // can't use 64-bit operands here because we don't have native results of + // 128-bits. Furthremore, casting the 64-bit values to 32-bit values won't + // work on large-endian machines. + uint64_t mask = ~0ull >> (sizeof(uint32_t)*8); + uint32_t n = rhsWords * 2; + uint32_t m = (lhsWords * 2) - n; + + // Allocate space for the temporary values we need either on the stack, if + // it will fit, or on the heap if it won't. + uint32_t SPACE[128]; + uint32_t *__U = 0; + uint32_t *__V = 0; + uint32_t *__Q = 0; + uint32_t *__R = 0; + if ((Remainder?4:3)*n+2*m+1 <= 128) { + __U = &SPACE[0]; + __V = &SPACE[m+n+1]; + __Q = &SPACE[(m+n+1) + n]; + if (Remainder) + __R = &SPACE[(m+n+1) + n + (m+n)]; + } else { + __U = new uint32_t[m + n + 1]; + __V = new uint32_t[n]; + __Q = new uint32_t[m+n]; + if (Remainder) + __R = new uint32_t[n]; + } + + // Initialize the dividend + memset(__U, 0, (m+n+1)*sizeof(uint32_t)); + for (unsigned i = 0; i < lhsWords; ++i) { + uint64_t tmp = (LHS.getNumWords() == 1 ? LHS.VAL : LHS.pVal[i]); + __U[i * 2] = (uint32_t)(tmp & mask); + __U[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + __U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm. + + // Initialize the divisor + memset(__V, 0, (n)*sizeof(uint32_t)); + for (unsigned i = 0; i < rhsWords; ++i) { + uint64_t tmp = (RHS.getNumWords() == 1 ? RHS.VAL : RHS.pVal[i]); + __V[i * 2] = (uint32_t)(tmp & mask); + __V[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + + // initialize the quotient and remainder + memset(__Q, 0, (m+n) * sizeof(uint32_t)); + if (Remainder) + memset(__R, 0, n * sizeof(uint32_t)); + + // Now, adjust m and n for the Knuth division. n is the number of words in + // the divisor. m is the number of words by which the dividend exceeds the + // divisor (i.e. m+n is the length of the dividend). These sizes must not + // contain any zero words or the Knuth algorithm fails. + for (unsigned i = n; i > 0 && __V[i-1] == 0; i--) { + n--; + m++; + } + for (unsigned i = m+n; i > 0 && __U[i-1] == 0; i--) + m--; + + // If we're left with only a single word for the divisor, Knuth doesn't work + // so we implement the short division algorithm here. This is much simpler + // and faster because we are certain that we can divide a 64-bit quantity + // by a 32-bit quantity at hardware speed and short division is simply a + // series of such operations. This is just like doing short division but we + // are using base 2^32 instead of base 10. + assert(n != 0 && "Divide by zero?"); + if (n == 1) { + uint32_t divisor = __V[0]; + uint32_t remainder = 0; + for (int i = m+n-1; i >= 0; i--) { + uint64_t partial_dividend = (uint64_t(remainder)) << 32 | __U[i]; + if (partial_dividend == 0) { + __Q[i] = 0; + remainder = 0; + } else if (partial_dividend < divisor) { + __Q[i] = 0; + remainder = (uint32_t)partial_dividend; + } else if (partial_dividend == divisor) { + __Q[i] = 1; + remainder = 0; + } else { + __Q[i] = (uint32_t)(partial_dividend / divisor); + remainder = (uint32_t)(partial_dividend - (__Q[i] * divisor)); + } + } + if (__R) + __R[0] = remainder; + } else { + // Now we're ready to invoke the Knuth classical divide algorithm. In this + // case n > 1. + KnuthDiv(__U, __V, __Q, __R, m, n); + } + + // If the caller wants the quotient + if (Quotient) { + // Set up the Quotient value's memory. + if (Quotient->BitWidth != LHS.BitWidth) { + if (Quotient->isSingleWord()) + Quotient->VAL = 0; + } else + Quotient->clear(); + + // The quotient is in Q. Reconstitute the quotient into Quotient's low + // order words. + if (lhsWords == 1) { + uint64_t tmp = + uint64_t(__Q[0]) | ((uint64_t(__Q[1])) << (APINT_BITS_PER_WORD / 2)); + if (Quotient->isSingleWord()) + Quotient->VAL = tmp; + else + Quotient->pVal[0] = tmp; + } else { + assert(!Quotient->isSingleWord() && "Quotient ap_private not large enough"); + for (unsigned i = 0; i < lhsWords; ++i) + Quotient->pVal[i] = + uint64_t(__Q[i*2]) | ((uint64_t(__Q[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Quotient->clearUnusedBits(); + } + + // If the caller wants the remainder + if (Remainder) { + // Set up the Remainder value's memory. + if (Remainder->BitWidth != RHS.BitWidth) { + if (Remainder->isSingleWord()) + Remainder->VAL = 0; + } else + Remainder->clear(); + + // The remainder is in R. Reconstitute the remainder into Remainder's low + // order words. + if (rhsWords == 1) { + uint64_t tmp = + uint64_t(__R[0]) | ((uint64_t(__R[1])) << (APINT_BITS_PER_WORD / 2)); + if (Remainder->isSingleWord()) + Remainder->VAL = tmp; + else + Remainder->pVal[0] = tmp; + } else { + assert(!Remainder->isSingleWord() && "Remainder ap_private not large enough"); + for (unsigned i = 0; i < rhsWords; ++i) + Remainder->pVal[i] = + uint64_t(__R[i*2]) | ((uint64_t(__R[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Remainder->clearUnusedBits(); + } + + // Clean up the memory we allocated. + if (__U != &SPACE[0]) { + delete [] __U; + delete [] __V; + delete [] __Q; + delete [] __R; + } +} + + +template +void ap_private<_AP_W, _AP_S, _AP_N>::fromString(const char *str, uint32_t slen, uint8_t radix) { + enum { numbits=_AP_W}; + // Check our assumptions here + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + assert(str && "String is null?"); + bool isNeg = str[0] == '-'; + if (isNeg) + str++, slen--; + + //skip any leading zero + while(*str == '0' && *(str+1) != '\0') {str++; slen--;} + assert((slen <= numbits || radix != 2) && "Insufficient bit width"); + assert(((slen - 1)*3 <= numbits || radix != 8) && "Insufficient bit width"); + assert(((slen - 1)*4 <= numbits || radix != 16) && "Insufficient bit width"); + assert((((slen -1)*64)/22 <= numbits || radix != 10) && "Insufficient bit width"); + + memset(pVal, 0, _AP_N * sizeof(uint64_t)); + + // Figure out if we can shift instead of multiply + uint32_t shift = (radix == 16 ? 4 : radix == 8 ? 3 : radix == 2 ? 1 : 0); + + // Set up an ap_private for the digit to add outside the loop so we don't + // constantly construct/destruct it. + uint64_t bigVal[_AP_N]; + memset(bigVal, 0, _AP_N * sizeof(uint64_t)); + ap_private<_AP_W, _AP_S, _AP_N> apdigit(getBitWidth(), bigVal); + ap_private<_AP_W, _AP_S, _AP_N> apradix(radix); + + // Enter digit traversal loop + for (unsigned i = 0; i < slen; i++) { + // Get a digit + uint32_t digit = 0; + char cdigit = str[i]; + if (radix == 16) { +#define isxdigit(c) (((c) >= '0' && (c) <= '9') || ((c) >= 'a' && (c) <= 'f') || ((c) >= 'A' && (c) <= 'F')) +#define isdigit(c) ((c) >= '0' && (c) <= '9') + if (!isxdigit(cdigit)) + assert(0 && "Invalid hex digit in string"); + if (isdigit(cdigit)) + digit = cdigit - '0'; + else if (cdigit >= 'a') + digit = cdigit - 'a' + 10; + else if (cdigit >= 'A') + digit = cdigit - 'A' + 10; + else + assert(0 && "huh? we shouldn't get here"); + } else if (isdigit(cdigit)) { + digit = cdigit - '0'; + } else { + assert(0 && "Invalid character in digit string"); + } +#undef isxdigit +#undef isdigit + // Shift or multiply the value by the radix + if (shift) + *this <<= shift; + else + *this *= apradix; + + // Add in the digit we just interpreted + if (apdigit.isSingleWord()) + apdigit.VAL = digit; + else + apdigit.pVal[0] = digit; + *this += apdigit; + } + // If its negative, put it in two's complement form + if (isNeg) { + (*this)--; + this->flip(); + } + clearUnusedBits(); +} + +template +std::string ap_private<_AP_W, _AP_S, _AP_N>::toString(uint8_t radix, bool wantSigned) const { + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + static const char *digits[] = { + "0","1","2","3","4","5","6","7","8","9","A","B","C","D","E","F" + }; + std::string result; + uint32_t bits_used = getActiveBits(); + + if (radix != 10) { + // For the 2, 8 and 16 bit cases, we can just shift instead of divide + // because the number of bits per digit (1,3 and 4 respectively) divides + // equaly. We just shift until there value is zero. + + // First, check for a zero value and just short circuit the logic below. + if (*this == (uint64_t)(0)) + result = "0"; + else { + ap_private<_AP_W, false, _AP_N> tmp(*this); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + tmp.clearUnusedBitsToZero(); + result = "-"; + insert_at = 1; + } + // Just shift tmp right for each digit width until it becomes zero + uint32_t shift = (radix == 16 ? 4 : (radix == 8 ? 3 : 1)); + uint64_t mask = radix - 1; + ap_private<_AP_W, false, _AP_N> zero(0); + while (tmp.ne(zero)) { + unsigned digit = (tmp.isSingleWord() ? tmp.VAL : tmp.pVal[0]) & mask; + result.insert(insert_at, digits[digit]); + tmp = tmp.lshr(shift); + } + } + return result; + } + + ap_private<_AP_W, false, _AP_N> tmp(*this); + ap_private<_AP_W, false, _AP_N> divisor(radix); + ap_private<_AP_W, false, _AP_N> zero(0); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + tmp.clearUnusedBitsToZero(); + result = "-"; + insert_at = 1; + } + if (tmp == ap_private<_AP_W, false, _AP_N>(0)) + result = "0"; + else while (tmp.ne(zero)) { + ap_private<_AP_W, false, _AP_N> APdigit(0); + ap_private<_AP_W, false, _AP_N> tmp2(0); + divide(tmp, tmp.getNumWords(), divisor, divisor.getNumWords(), &tmp2, + &APdigit); + uint32_t digit = APdigit.getZExtValue(); + assert(digit < radix && "divide failed"); + result.insert(insert_at,digits[digit]); + tmp = tmp2; + } + + return result; +} + +// This implements a variety of operations on a representation of +// arbitrary precision, two's-complement, bignum integer values. + +/* Assumed by lowHalf, highHalf, partMSB and partLSB. A fairly safe + and unrestricting assumption. */ + +/* Some handy functions local to this file. */ + +template +void divide(const ap_private<_AP_W, _AP_S, _AP_N>& LHS, uint32_t lhsWords, + uint64_t RHS, + ap_private<_AP_W, _AP_S, _AP_N> *Quotient, ap_private<_AP_W, _AP_S, _AP_N> *Remainder) { + uint32_t rhsWords=1; + assert(lhsWords >= rhsWords && "Fractional result"); + enum {APINT_BITS_PER_WORD=64}; + // First, compose the values into an array of 32-bit words instead of + // 64-bit words. This is a necessity of both the "short division" algorithm + // and the the Knuth "classical algorithm" which requires there to be native + // operations for +, -, and * on an m bit value with an m*2 bit result. We + // can't use 64-bit operands here because we don't have native results of + // 128-bits. Furthremore, casting the 64-bit values to 32-bit values won't + // work on large-endian machines. + uint64_t mask = ~0ull >> (sizeof(uint32_t)*8); + uint32_t n = 2; + uint32_t m = (lhsWords * 2) - n; + + // Allocate space for the temporary values we need either on the stack, if + // it will fit, or on the heap if it won't. + uint32_t SPACE[128]; + uint32_t *__U = 0; + uint32_t *__V = 0; + uint32_t *__Q = 0; + uint32_t *__R = 0; + if ((Remainder?4:3)*n+2*m+1 <= 128) { + __U = &SPACE[0]; + __V = &SPACE[m+n+1]; + __Q = &SPACE[(m+n+1) + n]; + if (Remainder) + __R = &SPACE[(m+n+1) + n + (m+n)]; + } else { + __U = new uint32_t[m + n + 1]; + __V = new uint32_t[n]; + __Q = new uint32_t[m+n]; + if (Remainder) + __R = new uint32_t[n]; + } + + // Initialize the dividend + memset(__U, 0, (m+n+1)*sizeof(uint32_t)); + for (unsigned i = 0; i < lhsWords; ++i) { + uint64_t tmp = (LHS.getNumWords() == 1 ? LHS.VAL : LHS.pVal[i]); + __U[i * 2] = tmp & mask; + __U[i * 2 + 1] = (tmp) >> (sizeof(uint32_t)*8); + } + __U[m+n] = 0; // this extra word is for "spill" in the Knuth algorithm. + + // Initialize the divisor + memset(__V, 0, (n)*sizeof(uint32_t)); + __V[0] = RHS & mask; + __V[1] = (RHS) >> (sizeof(uint32_t)*8); + + // initialize the quotient and remainder + memset(__Q, 0, (m+n) * sizeof(uint32_t)); + if (Remainder) + memset(__R, 0, n * sizeof(uint32_t)); + + // Now, adjust m and n for the Knuth division. n is the number of words in + // the divisor. m is the number of words by which the dividend exceeds the + // divisor (i.e. m+n is the length of the dividend). These sizes must not + // contain any zero words or the Knuth algorithm fails. + for (unsigned i = n; i > 0 && __V[i-1] == 0; i--) { + n--; + m++; + } + for (unsigned i = m+n; i > 0 && __U[i-1] == 0; i--) + m--; + + // If we're left with only a single word for the divisor, Knuth doesn't work + // so we implement the short division algorithm here. This is much simpler + // and faster because we are certain that we can divide a 64-bit quantity + // by a 32-bit quantity at hardware speed and short division is simply a + // series of such operations. This is just like doing short division but we + // are using base 2^32 instead of base 10. + assert(n != 0 && "Divide by zero?"); + if (n == 1) { + uint32_t divisor = __V[0]; + uint32_t remainder = 0; + for (int i = m+n-1; i >= 0; i--) { + uint64_t partial_dividend = (uint64_t(remainder)) << 32 | __U[i]; + if (partial_dividend == 0) { + __Q[i] = 0; + remainder = 0; + } else if (partial_dividend < divisor) { + __Q[i] = 0; + remainder = partial_dividend; + } else if (partial_dividend == divisor) { + __Q[i] = 1; + remainder = 0; + } else { + __Q[i] = partial_dividend / divisor; + remainder = partial_dividend - (__Q[i] * divisor); + } + } + if (__R) + __R[0] = remainder; + } else { + // Now we're ready to invoke the Knuth classical divide algorithm. In this + // case n > 1. + KnuthDiv(__U, __V, __Q, __R, m, n); + } + + // If the caller wants the quotient + if (Quotient) { + // Set up the Quotient value's memory. + if (Quotient->BitWidth != LHS.BitWidth) { + if (Quotient->isSingleWord()) + Quotient->VAL = 0; + else + delete [] Quotient->pVal; + } else + Quotient->clear(); + + // The quotient is in Q. Reconstitute the quotient into Quotient's low + // order words. + if (lhsWords == 1) { + uint64_t tmp = + uint64_t(__Q[0]) | ((uint64_t(__Q[1])) << (APINT_BITS_PER_WORD / 2)); + if (Quotient->isSingleWord()) + Quotient->VAL = tmp; + else + Quotient->pVal[0] = tmp; + } else { + assert(!Quotient->isSingleWord() && "Quotient ap_private not large enough"); + for (unsigned i = 0; i < lhsWords; ++i) + Quotient->pVal[i] = + uint64_t(__Q[i*2]) | ((uint64_t(__Q[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Quotient->clearUnusedBits(); + } + + // If the caller wants the remainder + if (Remainder) { + // Set up the Remainder value's memory. + if (Remainder->BitWidth != 64 /* RHS.BitWidth */) { + if (Remainder->isSingleWord()) + Remainder->VAL = 0; + } else + Remainder->clear(); + + // The remainder is in __R. Reconstitute the remainder into Remainder's low + // order words. + if (rhsWords == 1) { + uint64_t tmp = + uint64_t(__R[0]) | ((uint64_t(__R[1])) << (APINT_BITS_PER_WORD / 2)); + if (Remainder->isSingleWord()) + Remainder->VAL = tmp; + else + Remainder->pVal[0] = tmp; + } else { + assert(!Remainder->isSingleWord() && "Remainder ap_private not large enough"); + for (unsigned i = 0; i < rhsWords; ++i) + Remainder->pVal[i] = + uint64_t(__R[i*2]) | ((uint64_t(__R[i*2+1])) << (APINT_BITS_PER_WORD / 2)); + } + Remainder->clearUnusedBits(); + } + + // Clean up the memory we allocated. + if (__U != &SPACE[0]) { + delete [] __U; + delete [] __V; + delete [] __Q; + delete [] __R; + } +} + +//When bitwidth < 64 +template class ap_private <_AP_W, _AP_S, 1> { +#ifdef _MSC_VER +#pragma warning( disable : 4521 4522 ) +#endif +public: + typedef typename retval<_AP_S>::Type ValType; + template + struct RType { + enum { + _AP_N =1, + mult_w = _AP_W+_AP_W2, + mult_s = _AP_S||_AP_S2, //?? why + plus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, //shouldn't it be AP_MAX(_AP_W,_AP_W2)+!(_AP_S^_AP_S2)+1 ???? + plus_s = _AP_S||_AP_S2, + minus_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2))+1, + minus_s = true, + div_w = _AP_W+_AP_S2, + div_s = _AP_S||_AP_S2, + mod_w = AP_MIN(_AP_W,_AP_W2+(!_AP_S2&&_AP_S)), + mod_s = _AP_S, + logic_w = AP_MAX(_AP_W+(_AP_S2&&!_AP_S),_AP_W2+(_AP_S&&!_AP_S2)), + logic_s = _AP_S||_AP_S2 + }; + typedef ap_private mult; + typedef ap_private plus; + typedef ap_private minus; + typedef ap_private logic; + typedef ap_private div; + typedef ap_private mod; + typedef ap_private<_AP_W, _AP_S> arg1; + typedef bool reduce; + }; + enum { APINT_BITS_PER_WORD = 64}; + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -(_AP_W%APINT_BITS_PER_WORD) : 0}; + static const uint64_t mask = ((uint64_t)~0ULL >> (excess_bits)); + static const uint64_t not_mask = ~mask; + static const uint64_t sign_bit_mask = 1ULL << (APINT_BITS_PER_WORD-1); + template struct sign_ext_mask { static const uint64_t mask=~0ULL<<_AP_W1;}; + + enum { BitWidth=_AP_W}; + uint64_t VAL; ///< Used to store the <= 64 bits integer value. + const uint64_t *const pVal; + + INLINE uint32_t getBitWidth() const { + return BitWidth; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + VAL = RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& RHS) { + VAL = RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const ap_private<_AP_W1, _AP_S1, 1>& RHS) { + VAL = RHS.VAL; + clearUnusedBits(); + return *this; + } + + template + ap_private<_AP_W, _AP_S, 1>& operator=(const volatile ap_private<_AP_W1, _AP_S1, 1>& RHS) { + VAL = RHS.VAL; + clearUnusedBits(); + return *this; + } + + volatile ap_private& operator=(const ap_private& RHS) volatile { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + ap_private& operator=(const ap_private& RHS) { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + + volatile ap_private& operator=(const volatile ap_private& RHS) volatile { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + ap_private& operator=(const volatile ap_private& RHS) { + // Don't do anything for X = X + VAL = RHS.VAL; // No need to check because no harm done by copying. + return *this; + } + + template + INLINE ap_private& operator = (const ap_range_ref<_AP_W2, _AP_S2>& op2) { + *this = ap_private<_AP_W2, false>(op2); + return *this; + } + + explicit INLINE ap_private(uint64_t* val) : VAL(val[0]), pVal(&VAL){ + clearUnusedBits(); + } + + INLINE bool isSingleWord() const { return true; } + + INLINE void fromString(const char *strStart, uint32_t slen, + uint8_t radix, int offset=0) { + // Check our assumptions here + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + assert(strStart && "String is null?"); + strStart+=offset; + switch(radix) { + case 2: + // sscanf(strStart,"%b",&VAL); + VAL = *strStart =='1' ? ~0ULL : 0; + for (;*strStart; ++strStart) { + assert((*strStart=='0'|| *strStart=='1')&&("Wrong binary number") ); + VAL <<=1; + VAL |= (*strStart-'0'); + } + break; + case 8: +#if __WIN32__ + sscanf(strStart,"%I64o",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lo",&VAL); +#else + sscanf(strStart,"%llo",&VAL); +#endif + +#endif + break; + case 10: +#if __WIN32__ + sscanf(strStart,"%I64u",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lu",&VAL); +#else + sscanf(strStart,"%llu",&VAL); +#endif + +#endif + break; + case 16: +#if __WIN32__ + sscanf(strStart,"%I64x",&VAL); +#else + +#if defined __x86_64__ + sscanf(strStart,"%lx",&VAL); +#else + sscanf(strStart,"%llx",&VAL); +#endif + +#endif + break; + default: + assert(true && "Unknown radix"); + // error + } + clearUnusedBits(); + } + + INLINE ap_private() : pVal(&VAL){VAL = 0ULL;} + +#define CTOR(TYPE) \ + INLINE ap_private(TYPE v) : VAL((uint64_t)v), pVal(&VAL) { \ + clearUnusedBits(); \ + } + CTOR(int) + CTOR(bool) + CTOR(signed char) + CTOR(unsigned char) + CTOR(short) + CTOR(unsigned short) + CTOR(unsigned int) + CTOR(long) + CTOR(unsigned long) + CTOR(unsigned long long) + CTOR(long long) + CTOR(float) + CTOR(double) +#undef CTOR + ap_private(uint32_t numWords, const uint64_t bigVal[]): VAL(bigVal[0]), pVal(&VAL) {clearUnusedBits();} + + ap_private(const std::string& val, uint8_t radix=2, int base=0, int offset=0): VAL(0), pVal(&VAL) { + assert(!val.empty() && "String empty?"); + fromString(val.c_str()+base, val.size()-base, radix); + } + + ap_private(const char strStart[], uint32_t slen, uint8_t radix, int base=0, int offset=0) : VAL(0), pVal(&VAL) { + fromString(strStart+base, slen-base, radix, offset); + } + + ap_private(const ap_private& that) : VAL(that.VAL), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, 1>& that) : VAL(that.VAL), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const ap_private<_AP_W1, _AP_S1, _AP_N1>& that) : VAL(that.pVal[0]), pVal(&VAL) { + clearUnusedBits(); + } + + template + ap_private(const volatile ap_private<_AP_W1, _AP_S1, _AP_N1>& that) : VAL(that.pVal[0]), pVal(&VAL) { + clearUnusedBits(); + } + +#if 0 +template + explicit ap_private(const ap_private<_AP_W1, true, 1+_AP_W1/64>& that) + : VAL((_AP_W1>_AP_W) ? that.VAL & mask : ((1ULL<<(_AP_W1-1)&that.pVal[0]) ? sign_ext_mask<_AP_W1>::mask | that.VAL : that.pVal[0])), pVal(&VAL) {} + +template + explicit ap_private(const ap_private<_AP_W1, false, (_AP_W1+63)/64>& that) + : VAL(that.VAL & mask), pVal(&VAL) {} +#endif + + explicit ap_private(const char* val) : pVal(&VAL) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + bool neg = false; + uint32_t radix = 10; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + ap_private<_AP_W, _AP_S> ap_private_val(str.c_str(), strLen, radix, base, offset); + if (neg) + ap_private_val = -ap_private_val; + operator = (ap_private_val); + } + + ap_private(const char* val, signed char rd): pVal(&VAL) { + std::string str(val); + uint32_t strLen = str.length(); + const char *strp = str.c_str(); + uint32_t offset = 0; + uint32_t base = 0; + uint32_t radix = rd; + bool neg = false; + ap_parse_sign(strp, base, neg); + ap_parse_prefix(strp + base, offset, radix); + + if ((radix != 10 && neg) || + (strLen - base - offset <= 0) || + InvalidDigit(strp, strLen, base + offset, radix)) { + fprintf(stderr, "invalid character string %s !\n", val); + assert(0); + } + + uint32_t bitsNeeded = ap_private<_AP_W, _AP_S>::getBitsNeeded(strp, strLen, radix); + ap_private<_AP_W, _AP_S> ap_private_val(strp , strLen, radix, base, offset); + //ap_private<_AP_W, _AP_S> ap_private_val(bitsNeeded, strp , strLen, radix, base, offset); + if (strp[0] == '-') + ap_private_val = -ap_private_val; + operator = (ap_private_val); + } + + INLINE bool isNegative() const { + static const uint64_t sign_mask = 1ULL << (_AP_W-1); + return _AP_S && (sign_mask & VAL); + } + + INLINE bool isPositive() const { + return !isNegative(); + } + + INLINE bool isStrictlyPositive() const { + return !isNegative() && VAL!=0; + } + + INLINE bool isAllOnesValue() const { + return (mask & VAL) == mask; + } + + template + INLINE bool operator==(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + return (VAL == RHS.VAL); + } + + INLINE bool operator==(const ap_private<_AP_W, _AP_S>& RHS) const { return VAL == RHS.VAL; } + INLINE bool operator==(const ap_private<_AP_W, !_AP_S>& RHS) const { return getVal() == RHS.getVal(); } + INLINE bool operator==(uint64_t Val) const { return (VAL == Val); } + INLINE bool operator!=(uint64_t Val) const { return (VAL != Val); } + INLINE bool operator!=(const ap_private<_AP_W, _AP_S>& RHS) const { return VAL != RHS.VAL; } + INLINE bool operator!=(const ap_private<_AP_W, !_AP_S>& RHS) const { return getVal() != RHS.getVal(); } + const ap_private operator++() { ++VAL; clearUnusedBits(); return *this; } + const ap_private operator--(int) { + ap_private orig(*this); + --VAL; clearUnusedBits(); + return orig; + } + const ap_private operator--() { --VAL; clearUnusedBits(); return *this;} + INLINE bool operator !() const { return !VAL;} + + const ap_private operator++(int) { + ap_private orig(*this); + VAL++; clearUnusedBits(); + return orig; + } + + const ap_private operator~() {return ap_private(~VAL);} + INLINE typename RType<1,false>::minus operator-() const { + return ap_private<1,false>(0) - (*this); + } + + INLINE std::string toString(uint8_t radix, bool wantSigned) const ; + INLINE std::string toStringUnsigned(uint8_t radix = 10) const { + return toString(radix, false); + } + INLINE std::string toStringSigned(uint8_t radix = 10) const { + return toString(radix, true); + } + INLINE void clear() { + VAL=0; + } + INLINE ap_private& clear(uint32_t bitPosition) { VAL &= ~(1ULL<<(bitPosition)); clearUnusedBits(); return *this;} + + INLINE ap_private ashr(uint32_t shiftAmt) const { + enum {excess_bits = APINT_BITS_PER_WORD - BitWidth}; + if (_AP_S) + return ap_private((shiftAmt == BitWidth) ? 0 : ((int64_t)VAL) >> (shiftAmt)); + else + return ap_private((shiftAmt == BitWidth) ? 0 : (VAL) >> (shiftAmt)); + } + + INLINE ap_private lshr(uint32_t shiftAmt) const { + return ap_private((shiftAmt == BitWidth) ? ap_private(0) : ap_private((VAL&mask) >> (shiftAmt))); + } + + INLINE ap_private shl(uint32_t shiftAmt) const { + if (shiftAmt > BitWidth) { + if (!isNegative()) + return ap_private(0); + else return ap_private(-1); + } + if (shiftAmt == BitWidth) return ap_private(0); + else return ap_private((VAL) << (shiftAmt)); + //return ap_private((shiftAmt == BitWidth) ? ap_private(0ULL) : ap_private(VAL << shiftAmt)); + } + + INLINE int64_t getSExtValue() const { + return VAL; + } + + INLINE uint64_t getZExtValue() const { + return VAL & mask; + } + + template + INLINE ap_private(const ap_range_ref<_AP_W2,_AP_S2>& ref) : pVal(&VAL) { + *this=ref.get(); + } + + template + INLINE ap_private(const ap_bit_ref<_AP_W2,_AP_S2>& ref) : pVal(&VAL) { + *this = ((uint64_t)(bool)ref); + } + + template + INLINE ap_private(const ap_concat_ref<_AP_W2, _AP_T2,_AP_W3, _AP_T3>& ref) : pVal(&VAL) { + *this=ref.get(); + } + + template + INLINE ap_private(const af_range_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) : pVal(&VAL) { + *this = ((val.operator ap_private<_AP_W2, false> ())); + } + + template + INLINE ap_private(const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, + _AP_Q2, _AP_O2, _AP_N2> &val) : pVal(&VAL) { + *this = (uint64_t)(bool)val; + } + + INLINE void write(const ap_private<_AP_W, _AP_S>& op2) volatile { + *this = (op2); + } + + //Explicit conversions to C interger types + //----------------------------------------------------------- + ValType getVal() const { + return VAL; + } + operator ValType () const { + return getVal(); + } + INLINE int to_int() const { + // ap_private<64 /* _AP_W */, _AP_S> res(V); + return (int) getVal(); + } + + INLINE unsigned to_uint() const { + return (unsigned) getVal(); + } + + INLINE long to_long() const { + return (long) getVal(); + } + + INLINE unsigned long to_ulong() const { + return (unsigned long) getVal(); + } + + INLINE ap_slong to_int64() const { + return (ap_slong) getVal(); + } + + INLINE ap_ulong to_uint64() const { + return (ap_ulong) getVal(); + } + + INLINE double to_double() const { + if (isNegative()) + return roundToDouble(true); + else + return roundToDouble(false); + } + + INLINE bool isMinValue() const { return VAL == 0;} + template INLINE ap_private& operator&=(const ap_private<_AP_W1, _AP_S1>& RHS) { + VAL = VAL&RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator|=(const ap_private<_AP_W1, _AP_S1>& RHS) { + VAL = VAL|RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator^=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL^RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator*=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL*RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator+=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL+RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + + template INLINE ap_private& operator-=(const ap_private<_AP_W1, _AP_S1>& RHS){ + VAL = VAL-RHS.pVal[0]; + clearUnusedBits(); + return *this; + } + INLINE const ap_private& operator<<=(uint32_t shiftAmt) { VAL<<=shiftAmt; clearUnusedBits(); return *this; } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator&(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL & RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret & RHS; + } + } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator^(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL ^ RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret ^ RHS; + } + } + + template INLINE typename RType<_AP_W1, _AP_S1>::logic operator|(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::logic_w <= 64) { + typename RType<_AP_W1, _AP_S1>::logic Ret(VAL | RHS.VAL); + return Ret; + } else { + typename RType<_AP_W1, _AP_S1>::logic Ret = *this; + return Ret | RHS; + } + } + + INLINE ap_private<_AP_W, _AP_S> And(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL & RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Or(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL | RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Xor(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL ^ RHS.VAL); + } +#if 1 + template + INLINE typename RType<_AP_W1, _AP_S1>::mult operator*(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1, _AP_S1>::mult_w <= 64) { + typename RType<_AP_W1, _AP_S1>::mult Result(VAL * RHS.VAL); + return Result; + } else { + typename RType<_AP_W1, _AP_S1>::mult Result = typename RType<_AP_W1, _AP_S1>::mult(*this); + Result *= RHS; + return Result; + } + } +#endif + INLINE ap_private<_AP_W, _AP_S> Mul(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL * RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Add(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL + RHS.VAL); + } + + INLINE ap_private<_AP_W, _AP_S> Sub(const ap_private<_AP_W, _AP_S>& RHS) const { + return ap_private<_AP_W, _AP_S>(VAL - RHS.VAL); + } + +#if 1 + INLINE ap_private& operator&=(uint64_t RHS) { VAL &= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator|=(uint64_t RHS) { VAL |= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator^=(uint64_t RHS){ VAL ^= RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator*=(uint64_t RHS){ VAL *= RHS; clearUnusedBits(); return *this; } + INLINE ap_private& operator+=(uint64_t RHS){ VAL += RHS; clearUnusedBits(); return *this;} + INLINE ap_private& operator-=(uint64_t RHS){ VAL -= RHS; clearUnusedBits(); return *this; } + INLINE ap_private operator&(uint64_t RHS) const { return ap_private(VAL & RHS); } + INLINE ap_private operator|(uint64_t RHS) const { return ap_private(VAL | RHS); } + INLINE ap_private operator^(uint64_t RHS) const { return ap_private(VAL ^ RHS); } + INLINE ap_private operator*(uint64_t RHS) const { return ap_private(VAL * RHS); } + INLINE ap_private operator/(uint64_t RHS) const { return ap_private(VAL / RHS); } + INLINE ap_private operator+(uint64_t RHS) const { return ap_private(VAL + RHS); } + INLINE ap_private operator-(uint64_t RHS) const { return ap_private(VAL - RHS); } +#endif + INLINE bool isMinSignedValue() const { + static const uint64_t min_mask = ~(~0ULL << (_AP_W-1)); + return BitWidth == 1 ? VAL == 1 : + (ap_private_ops::isNegative<_AP_W>(*this) && ((min_mask & VAL)==0)); + } + +#if 1 + + template INLINE + typename RType<_AP_W1,_AP_S1>::plus operator+(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1,_AP_S1>::plus_w <=64) + return typename RType<_AP_W1,_AP_S1>::plus(RType<_AP_W1,_AP_S1>::plus_s ? int64_t(VAL+RHS.VAL):uint64_t(VAL+RHS.VAL)); + typename RType<_AP_W1,_AP_S1>::plus Result=RHS; + Result += VAL; + return Result; + } + + template INLINE + typename RType<_AP_W1,_AP_S1>::minus operator-(const ap_private<_AP_W1, _AP_S1>& RHS) const { + if (RType<_AP_W1,_AP_S1>::minus_w <=64) + return typename RType<_AP_W1,_AP_S1>::minus(int64_t(VAL-RHS.VAL)); + typename RType<_AP_W1,_AP_S1>::minus Result=*this; + Result -= RHS; + return Result; + } +#endif // #if 1 + + INLINE ap_private& flip() { + VAL = (~0ULL^VAL)&mask; + clearUnusedBits(); + return *this; + } + + uint32_t countPopulation() const { return CountPopulation_64(VAL);} + uint32_t countLeadingZeros() const { + int remainder = BitWidth % APINT_BITS_PER_WORD; + int excessBits = (APINT_BITS_PER_WORD - remainder) % APINT_BITS_PER_WORD; + //enum { remainder = BitWidth % APINT_BITS_PER_WORD, excessBits = APINT_BITS_PER_WORD - remainder}; + uint32_t Count = CountLeadingZeros_64(VAL); + if (Count) + Count-=excessBits; + return AESL_std::min(Count, (uint32_t)_AP_W); + } + + /// HiBits - This function returns the high "numBits" bits of this ap_private. + ap_private<_AP_W, _AP_S, 1> getHiBits(uint32_t numBits) const { + ap_private<_AP_W, _AP_S, 1> ret(*this); + ret = (ret)>>(BitWidth - numBits); + return ret; + } + + /// LoBits - This function returns the low "numBits" bits of this ap_private. + ap_private<_AP_W, _AP_S, 1> getLoBits(uint32_t numBits) const { + ap_private<_AP_W, _AP_S, 1> ret((VAL) << (BitWidth - numBits)); + ret = (ret)>>(BitWidth - numBits); + return ret; + //return ap_private(numBits, (VAL << (BitWidth - numBits))>> (BitWidth - numBits)); + } + + ap_private<_AP_W, _AP_S,1>& set(uint32_t bitPosition) { + VAL |= (1ULL << (bitPosition)); + clearUnusedBits(); + return *this; // clearUnusedBits(); + } + + void set() { + VAL = ~0ULL; + clearUnusedBits(); + } + + template + INLINE void set(const ap_private<_AP_W3, false> & val) { + operator = (ap_private<_AP_W3, _AP_S>(val)); + } + + INLINE void set(const ap_private & val) { + operator = (val); + } + + bool operator[](uint32_t bitPosition) const { + return (((1ULL << (bitPosition)) & VAL) != 0); + } + + INLINE void clearUnusedBits(void) { + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -_AP_W%APINT_BITS_PER_WORD : 0}; + VAL = _AP_S ? ((((int64_t)VAL)<<(excess_bits))>> (excess_bits)) : (excess_bits ? ((VAL)<<(excess_bits))>>(excess_bits) : VAL); + } + + INLINE void clearUnusedBitsToZero(void) { + enum { excess_bits = (_AP_W%APINT_BITS_PER_WORD) ? APINT_BITS_PER_WORD -_AP_W%APINT_BITS_PER_WORD : 0}; + static uint64_t mask = ~0ULL >> (excess_bits); + VAL &= mask; + } + + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1> udiv(const ap_private<_AP_W, _AP_S1>& RHS) const { + return ap_private<_AP_W, _AP_S||_AP_S1>(VAL / RHS.VAL); + } + + INLINE ap_private udiv(uint64_t RHS) const { + return ap_private(VAL / RHS); + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + template + INLINE ap_private<_AP_W, _AP_S||_AP_S1> sdiv(const ap_private<_AP_W, _AP_S1> & RHS) const { + if (isNegative()) + if (RHS.isNegative()) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS.isNegative()) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + /// Signed divide this ap_private by ap_private RHS. + /// @brief Signed division function for ap_private. + INLINE ap_private sdiv(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return (-(*this)).udiv(-RHS); + else + return -((-(*this)).udiv(RHS)); + else if (RHS<0) + return -(this->udiv(-RHS)); + return this->udiv(RHS); + } + + template + INLINE ap_private urem(const ap_private<_AP_W, _AP_S2>& RHS) const { + assert(RHS.VAL != 0 && "Divide by 0"); + return ap_private(VAL%RHS.VAL); + } + + INLINE ap_private urem(uint64_t RHS) const { + assert(RHS != 0 && "Divide by 0"); + return ap_private(VAL%RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + template + INLINE ap_private srem(const ap_private<_AP_W, _AP_S2>& RHS) const { + if (isNegative()) { + ap_private lhs = -(*this); + if (RHS.isNegative()) { + ap_private rhs = -RHS; + return -(lhs.urem(rhs)); + } else + return -(lhs.urem(RHS)); + } else if (RHS.isNegative()) { + ap_private rhs = -RHS; + return this->urem(rhs); + } + return this->urem(RHS); + } + + /// Signed remainder operation on ap_private. + /// @brief Function for signed remainder operation. + INLINE ap_private srem(int64_t RHS) const { + if (isNegative()) + if (RHS<0) + return -((-(*this)).urem(-RHS)); + else + return -((-(*this)).urem(RHS)); + else if (RHS<0) + return this->urem(-RHS); + return this->urem(RHS); + } + + INLINE static void udivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder){ + assert(RHS!=0 && "Divide by 0"); + Quotient = LHS.VAl/RHS.VAl; + Remainder = LHS.VAL % RHS.VAL; + } + + INLINE static void udivrem(const ap_private &LHS, uint64_t RHS, + ap_private &Quotient, ap_private &Remainder){ + assert(RHS!=0 && "Divide by 0"); + Quotient = LHS.VAl/RHS; + Remainder = LHS.VAL % RHS; + } + + INLINE static void sdivrem(const ap_private &LHS, const ap_private &RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS.isNegative()) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS.isNegative()) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + INLINE static void sdivrem(const ap_private &LHS, int64_t RHS, + ap_private &Quotient, ap_private &Remainder) { + if (LHS.isNegative()) { + if (RHS<0) + ap_private::udivrem(-LHS, -RHS, Quotient, Remainder); + else + ap_private::udivrem(-LHS, RHS, Quotient, Remainder); + Quotient = -Quotient; + Remainder = -Remainder; + } else if (RHS<0) { + ap_private::udivrem(LHS, -RHS, Quotient, Remainder); + Quotient = -Quotient; + } else { + ap_private::udivrem(LHS, RHS, Quotient, Remainder); + } + } + + template INLINE bool eq(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return (*this) == RHS; + } + + template INLINE bool ne(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !((*this) == RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered unsigned. + /// @brief Unsigned less than comparison + template INLINE bool ult(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + uint64_t lhsZext = ((uint64_t(VAL)) << (64-_AP_W)) >> (64-_AP_W); + uint64_t rhsZext = ((uint64_t(RHS.VAL)) << (64-_AP_W1)) >> (64-_AP_W1); + return lhsZext < rhsZext; + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-than relationship. + /// @returns true if *this < RHS when both are considered signed. + /// @brief Signed less than comparison + template INLINE bool slt(const ap_private<_AP_W1, _AP_S1, 1>& RHS) const { + int64_t lhsSext = ((int64_t(VAL)) << (64-_AP_W)) >> (64-_AP_W); + int64_t rhsSext = ((int64_t(RHS.VAL)) << (64-_AP_W1)) >> (64-_AP_W1); + return lhsSext < rhsSext; + } + + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered unsigned. + /// @brief Unsigned less or equal comparison + template INLINE bool ule(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return ult(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the less-or-equal relationship. + /// @returns true if *this <= RHS when both are considered signed. + /// @brief Signed less or equal comparison + template INLINE bool sle(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return slt(RHS) || eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered unsigned. + /// @brief Unsigned greather than comparison + template INLINE bool ugt(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !ult(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// the validity of the greater-than relationship. + /// @returns true if *this > RHS when both are considered signed. + /// @brief Signed greather than comparison + template INLINE bool sgt(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !slt(RHS) && !eq(RHS); + } + + /// Regards both *this and RHS as unsigned quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered unsigned. + /// @brief Unsigned greater or equal comparison + template INLINE bool uge(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !ult(RHS); + } + + /// Regards both *this and RHS as signed quantities and compares them for + /// validity of the greater-or-equal relationship. + /// @returns true if *this >= RHS when both are considered signed. + /// @brief Signed greather or equal comparison + template INLINE bool sge(const ap_private<_AP_W1, _AP_S1>& RHS) const { + return !slt(RHS); + } + + INLINE ap_private abs() const { + if (isNegative()) + return -(*this); + return *this; + } + + ap_private<_AP_W, false> get() const { + ap_private<_AP_W,false> ret(*this); + return ret; + } + + INLINE static uint32_t getBitsNeeded(const char* str, uint32_t slen, uint8_t radix) { + return _AP_W; + } + + INLINE uint32_t getActiveBits() const { + uint32_t bits=_AP_W - countLeadingZeros(); + return bits?bits:1; + } + + INLINE double roundToDouble(bool isSigned=false) const { + const static uint64_t mask = ~0ULL << (APINT_BITS_PER_WORD - _AP_W); + return double(VAL); + } + + INLINE unsigned length() const { return _AP_W; } + + /*Reverse the contents of ap_private instance. I.e. LSB becomes MSB and vise versa*/ + INLINE ap_private& reverse () { + for (int i = 0; i < _AP_W/2; ++i) { + bool tmp = operator[](i); + if (operator[](_AP_W - 1 - i)) + set(i); + else + clear(i); + if (tmp) + set(_AP_W - 1 - i); + else + clear(_AP_W - 1 - i); + } + clearUnusedBits(); + return *this; + } + + /*Return true if the value of ap_private instance is zero*/ + INLINE bool iszero () const { + return isMinValue(); + } + + /* x < 0 */ + INLINE bool sign () const { + if (isNegative()) + return true; + return false; + } + + /* x[i] = !x[i] */ + INLINE void invert (int i) { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + flip(i); + } + + /* x[i] */ + INLINE bool test (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator[](i); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the left + INLINE void lrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (shl(n) | lshr(_AP_W - n)); + } + + //This is used for sc_lv and sc_bv, which is implemented by sc_uint + //Rotate an ap_private object n places to the right + INLINE void rrotate(int n) { + assert( n >= 0 && "Attempting to shift negative index"); + assert( n < _AP_W && "Shift value larger than bit width"); + operator = (lshr(n) | shl(_AP_W - n)); + } + + //Set the ith bit into v + INLINE void set (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Set the ith bit into v + INLINE void set_bit (int i, bool v) { + assert( i >= 0 && "Attempting to write bit with negative index"); + assert( i < _AP_W && "Attempting to write bit beyond MSB"); + v ? set(i) : clear(i); + } + + //Get the value of ith bit + INLINE bool get_bit (int i) const { + assert( i >= 0 && "Attempting to read bit with negative index"); + assert( i < _AP_W && "Attempting to read bit beyond MSB"); + return operator [](i); + } + + //complements every bit + INLINE void b_not() { + flip(); + } + + //Binary Arithmetic + //----------------------------------------------------------- +#define OP_BIN_AP(Sym,Rty, Fun) \ + template \ + INLINE \ + typename RType<_AP_W2,_AP_S2>::Rty \ + operator Sym (const ap_private<_AP_W2,_AP_S2>& op) const { \ + typename RType<_AP_W2,_AP_S2>::Rty lhs(*this); \ + typename RType<_AP_W2,_AP_S2>::Rty rhs(op); \ + return lhs.Fun(rhs); \ + } \ + + ///Bitwise and, or, xor + //OP_BIN_AP(&,logic, And) + //OP_BIN_AP(|,logic, Or) + //OP_BIN_AP(^,logic, Xor) + +#undef OP_BIN_AP + template + INLINE typename RType<_AP_W2,_AP_S2>::div + operator / (const ap_private<_AP_W2,_AP_S2>&op) const { + ap_private lhs=ap_private(*this); + ap_private rhs=ap_private(op); + return typename RType<_AP_W2,_AP_S2>::div((_AP_S||_AP_S2)?lhs.sdiv(rhs):lhs.udiv(rhs)); + } + + + template + INLINE typename RType<_AP_W2,_AP_S2>::mod + operator % (const ap_private<_AP_W2,_AP_S2>&op) const { + ap_private lhs=*this; + ap_private rhs=op; + typename RType<_AP_W2,_AP_S2>::mod res = typename RType<_AP_W2,_AP_S2>::mod (_AP_S?lhs.srem(rhs):lhs.urem(rhs)); + return res; + } + + +#define OP_ASSIGN_AP_2(Sym) \ + template \ + INLINE ap_private<_AP_W, _AP_S>& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op); \ + return *this; \ + } \ + + OP_ASSIGN_AP_2(/) + OP_ASSIGN_AP_2(%) +#undef OP_ASSIGN_AP_2 + + ///Bitwise assign: and, or, xor + //------------------------------------------------------------- + // OP_ASSIGN_AP(&) + // OP_ASSIGN_AP(^) + // OP_ASSIGN_AP(|) +#undef OP_ASSIGN_AP +#if 1 + + template + INLINE ap_private<_AP_W, _AP_S> + operator << (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh=op2.to_uint(); + return *this << sh; + } + + INLINE ap_private<_AP_W, _AP_S> + operator << (uint32_t sh) const { + return shl(sh); + } + +#endif + + template + INLINE ap_private<_AP_W, _AP_S> + operator >> (const ap_private<_AP_W2, _AP_S2>& op2) const { + uint32_t sh = op2.to_uint(); + return *this >> sh; + } + + INLINE ap_private<_AP_W, _AP_S> + operator >>(uint32_t sh) const { + ap_private<_AP_W, _AP_S> r(*this); + bool overflow=(sh>=_AP_W); + bool neg_v=r.isNegative(); + if(_AP_S) { + if(overflow) + neg_v?r.set():r.clear(); + else + return r.ashr(sh); + } else { + if(overflow) + r.clear(); + else + return r.lshr(sh); + } + return r; + } + + ///Shift assign + //------------------------------------------------------------------ +#define OP_ASSIGN_AP_3_SINGLE(Sym) \ + template \ + INLINE ap_private<_AP_W, _AP_S>& operator Sym##=(const ap_private<_AP_W2,_AP_S2>& op) \ + { \ + *this=operator Sym (op.getVal()); \ + return *this; \ + } + OP_ASSIGN_AP_3_SINGLE(>>) +#undef OP_ASSIGN_AP_3_SINGLE + + ///Comparisons + //----------------------------------------------------------------- + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this==op); + } + + template + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return op < *this; + } + + template + INLINE bool operator <= (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this>op); + } + + template + INLINE bool operator < (const ap_private<_AP_W2, _AP_S2, 1>& op) const { + enum { _AP_MAX_W = AP_MAX(_AP_W+(_AP_S||_AP_S2),_AP_W2+(_AP_S||_AP_S2))}; + ap_private<_AP_MAX_W, _AP_S> lhs(*this); + ap_private<_AP_MAX_W, _AP_S2> rhs(op); + if (_AP_S == _AP_S2) + return _AP_S?lhs.slt(rhs):lhs.ult(rhs); + else if (_AP_W < 32 && _AP_W2 < 32) + return lhs.slt(rhs); + else + if (_AP_S) + if (_AP_W2 >= _AP_W) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + else + if (_AP_W >= _AP_W2) + return lhs.ult(rhs); + else + return lhs.slt(rhs); + } + + template + INLINE bool operator >=(const ap_private<_AP_W2, _AP_S2, 1>& op) const { + return !(*this + INLINE bool operator == (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op == *this; + } + + template + INLINE bool operator != (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return !(op==*this); + } + + template + INLINE bool operator > (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op < (*this); + } + + template + INLINE bool operator <= (const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op >= *this; + } + + template + INLINE bool operator <(const ap_private<_AP_W2, _AP_S2, _AP_N2>& op) const { + return op > *this; + } + + template + INLINE bool operator >=(const ap_private<_AP_W2,_AP_S2,_AP_N2>& op) const { + return op <= *this; + } + ///Bit and Part Select + //-------------------------------------------------------------- + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast*>(this), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) const { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>((const_cast*> (this)), Hi, Lo); + } + + INLINE ap_range_ref<_AP_W,_AP_S> + range (int Hi, int Lo) { + assert((Hi < _AP_W) && (Lo < _AP_W)&&"Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(this, Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + range (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + assert((Hi < _AP_W) && (Lo < _AP_W) && "Out of bounds in range()"); + return ap_range_ref<_AP_W,_AP_S>(const_cast(this), Hi, Lo); + } + + template + INLINE ap_range_ref<_AP_W,_AP_S> + operator () (const ap_private<_AP_W2, _AP_S2> &HiIdx, + const ap_private<_AP_W3, _AP_S3> &LoIdx) const { + int Hi = HiIdx.to_int(); + int Lo = LoIdx.to_int(); + return this->range(Hi, Lo); + } + + + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (uint32_t index) { + assert(index >= 0&&"Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S> (*this, (int)index); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> operator [] (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + template + INLINE bool operator [] (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br =operator [] (index); + return br.to_bool(); + } + + INLINE ap_bit_ref<_AP_W,_AP_S> bit (int index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index ); + } + + template + INLINE ap_bit_ref<_AP_W,_AP_S> bit (const ap_private<_AP_W2,_AP_S2> &index) { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W &&"Attempting to read bit beyond MSB"); + return ap_bit_ref<_AP_W,_AP_S>( *this, index.to_int() ); + } + + INLINE bool bit (int index) const { + assert(index >= 0 && "Attempting to read bit with negative index"); + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br(const_cast*>(this), index); + return br.to_bool(); + } + + template + INLINE bool bit (const ap_private<_AP_W2,_AP_S2>& index) const { + assert(index < _AP_W && "Attempting to read bit beyond MSB"); + ap_bit_ref<_AP_W,_AP_S> br = bit(index); + return br.to_bool(); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(const ap_private<_AP_W2,_AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W,ap_private<_AP_W, _AP_S>,_AP_W2,ap_private<_AP_W2,_AP_S2> > concat(ap_private<_AP_W2,_AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_private<_AP_W2,_AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (const ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(const_cast& >(*this), a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, _AP_S2> > + operator, (ap_private<_AP_W2, _AP_S2>& a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, ap_private<_AP_W2, + _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (const ap_range_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, ap_range_ref<_AP_W2, _AP_S2> > + operator, (ap_range_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2, + ap_range_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (const ap_bit_ref<_AP_W2, _AP_S2> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, ap_bit_ref<_AP_W2, _AP_S2> > + operator, (ap_bit_ref<_AP_W2, _AP_S2> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, 1, + ap_bit_ref<_AP_W2, _AP_S2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) const { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> > + operator, (ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> &a2) { + return ap_concat_ref<_AP_W, ap_private<_AP_W, _AP_S>, _AP_W2+_AP_W3, + ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_range_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, _AP_W2, af_range_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (const af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) const { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(const_cast& >(*this), + const_cast& >(a2)); + } + + template + INLINE ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> > + operator, (af_bit_ref<_AP_W2, _AP_I2, _AP_S2, _AP_Q2, + _AP_O2, _AP_N2> &a2) { + return ap_concat_ref<_AP_W, ap_private, 1, af_bit_ref<_AP_W2, + _AP_I2, _AP_S2, _AP_Q2, _AP_O2, _AP_N2> >(*this, a2); + } + + template + INLINE ap_private + operator & (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this & a2.get(); + } + + template + INLINE ap_private + operator | (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this | a2.get(); + } + + template + INLINE ap_private + operator ^ (const ap_concat_ref<_AP_W2, _AP_T2, _AP_W3, _AP_T3>& a2) { + return *this ^ a2.get(); + } + + + //Reduce operation + //----------------------------------------------------------- + INLINE bool and_reduce() const { + return (VAL & mask) == mask; + } + + INLINE bool nand_reduce() const { + return (VAL & mask) != mask; + } + + INLINE bool or_reduce() const { + return (bool)VAL; + } + + INLINE bool nor_reduce() const { + return VAL==0; + } + + INLINE bool xor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?true:false; + } + + INLINE bool xnor_reduce() const { + unsigned int i=countPopulation(); + return (i%2)?false:true; + } + + INLINE std::string to_string(uint8_t radix=2, bool sign=false) const { + return toString(radix, radix==10?_AP_S:sign); + } +}; +template +std::string ap_private<_AP_W, _AP_S, 1>::toString(uint8_t radix, bool wantSigned) const { + assert((radix == 10 || radix == 8 || radix == 16 || radix == 2) && + "Radix should be 2, 8, 10, or 16!"); + static const char *digits[] = { + "0","1","2","3","4","5","6","7","8","9","a","b","c","d","e","f" + }; + std::string result; + if (radix != 10) { + // For the 2, 8 and 16 bit cases, we can just shift instead of divide + // because the number of bits per digit (1,3 and 4 respectively) divides + // equaly. We just shift until there value is zero. + + // First, check for a zero value and just short circuit the logic below. + if (*this == (uint64_t)(0)) + result = "0"; + else { + ap_private<_AP_W, false, 1> tmp(*this); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + result = "-"; + insert_at = 1; + } + // Just shift tmp right for each digit width until it becomes zero + uint32_t shift = (radix == 16 ? 4 : (radix == 8 ? 3 : 1)); + uint64_t mask = radix - 1; + ap_private<_AP_W, false, 1> zero(0); + while (tmp.ne(zero)) { + unsigned digit = (unsigned)(tmp.VAL & mask); + result.insert(insert_at, digits[digit]); + tmp = tmp.lshr(shift); + } + } + return result; + } + + ap_private<_AP_W, false, 1> tmp(*this); + ap_private<6, false, 1> divisor(radix); + ap_private<_AP_W, _AP_S, 1> zero(0); + size_t insert_at = 0; + if (wantSigned && isNegative()) { + // They want to print the signed version and it is a negative value + // Flip the bits and add one to turn it into the equivalent positive + // value and put a '-' in the result. + tmp.flip(); + tmp++; + result = "-"; + insert_at = 1; + } + if (tmp == ap_private<_AP_W, false, 1>(0ULL)) + result = "0"; + else while (tmp.ne(zero)) { + ap_private<_AP_W, false, 1> APdigit = tmp%divisor; + ap_private<_AP_W, false, 1> tmp2 = tmp/divisor; + uint32_t digit = (uint32_t)(APdigit.getZExtValue()); + assert(digit < radix && "divide failed"); + result.insert(insert_at,digits[digit]); + tmp = tmp2; + } + return result; + +} + +#endif /* #ifndef LLVM_SUPPORT_MATHEXTRAS_H */ \ No newline at end of file diff --git a/hls/lines256_length128/main.cpp b/hls/lines256_length128/main.cpp new file mode 100644 index 0000000000000000000000000000000000000000..38033932d4339090fd942d68ec2af89117b2b48c --- /dev/null +++ b/hls/lines256_length128/main.cpp @@ -0,0 +1,91 @@ +/** + * main.cpp + * + * for Vivado HLS + */ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +#ifdef CALCTIME +#include +#include +#endif + +#include "router.hpp" + + +int main(int argc, char *argv[]) { + using namespace std; + + // eXgf[^ (`) + // NL_Q00.txt + //char boardstr[BOARDSTR_SIZE] = "X10Y05Z3L0000107041L0004107002L0102102021L0900100003"; + // NL_Q06.txt + char boardstr[BOARDSTR_SIZE] = "X10Y18Z2L0900109002L0901105012L0902103052L0903103062L0904100102L0905106012L0906109022L0717109102L0808109112L0017209172L0401200072L0912208152L0009201092L0709209092L0901206052L0309204092L0701209072L0101201022L0011202152L0016202162"; + // NL_Q08.txt + //char boardstr[BOARDSTR_SIZE] = "X17Y20Z2L0000103022L1603115052L0916107032L0302108012L1104111042L1002100002L0919116162L1616113182L1001115012L0500201182L1603213152L0600210022"; + + // w肳Ă΃R}hC蕶ǂݍ + if (1 < argc) { + //擪Xł͂ȂȂΕW͂ǂݍ + if(argv[1][0]!='X') + { + char* c_p=fgets(boardstr, BOARDSTR_SIZE, stdin); + int length=strlen(c_p); + boardstr[length-1]=0; + } + else + { + strcpy(boardstr, argv[1]); + } + } + + // w肳Ă΃V[hlǂݍ + int seed = 12345; + if (2 < argc) { + seed = atoi(argv[2]); + } + + int size_x = (boardstr[1] - '0') * 10 + (boardstr[2] - '0'); + int size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); + int size_z = (boardstr[7] - '0'); + + // \os + ap_int<32> status; + clock_t clock_start, clock_done; + clock_start = clock(); + bool result = pynqrouter_256x128(boardstr, seed, &status); + clock_done = clock(); + if (result) { + cout << endl << "Test Passed!" << endl; + } else { + cout << endl << "Test Failed!" << endl; + } + cout << "status = " << (int)status << endl; + cout << "elapsed = " << ((double)(clock_done - clock_start) / CLOCKS_PER_SEC) << endl << endl; + + // \ + cout << "SOLUTION" << endl; + cout << "========" << endl; + cout << "SIZE " << size_x << "X" << size_y << "X" << size_z << endl; + for (int z = 0; z < size_z; z++) { + cout << "LAYER " << (z + 1) << endl; + for (int y = 0; y < size_y; y++) { + for (int x = 0; x < size_x; x++) { + if (x != 0) { + cout << ","; + } + int i = ((x * MAX_WIDTH + y) << BITWIDTH_Z) | z; + //cout << setfill('0') << setw(2) << right << (unsigned int)(unsigned char)(boardstr[i]); + cout << (unsigned int)(unsigned char)(boardstr[i]); + } + cout << endl; + } + } + + return 0; +} diff --git a/hls/lines256_length128/router.cpp b/hls/lines256_length128/router.cpp new file mode 100644 index 0000000000000000000000000000000000000000..08eb1f5a56f3e112d12223540e246cf532c26dcc --- /dev/null +++ b/hls/lines256_length128/router.cpp @@ -0,0 +1,745 @@ +/** + * router.cpp + * + * for Vivado HLS + */ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +#include "./router.hpp" + + +// ================================ // +// LFST +// ================================ // + +// 参考 https://highlevel-synthesis.com/2017/02/10/lfsr-in-hls/ +static ap_uint<32> lfsr; + +void lfsr_random_init(ap_uint<32> seed) { +#pragma HLS INLINE + lfsr = seed; +} + +ap_uint<32> lfsr_random() { +#pragma HLS INLINE + bool b_32 = lfsr.get_bit(32-32); + bool b_22 = lfsr.get_bit(32-22); + bool b_2 = lfsr.get_bit(32-2); + bool b_1 = lfsr.get_bit(32-1); + bool new_bit = b_32 ^ b_22 ^ b_2 ^ b_1; + lfsr = lfsr >> 1; + lfsr.set_bit(31, new_bit); + + return lfsr.to_uint(); +} + +// AからBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +/*ap_uint<32> lfsr_random_uint32(ap_uint<32> a, ap_uint<32> b) { +#pragma HLS INLINE + return lfsr_random() % (b - a + 1) + a; +}*/ + +// 0からBの範囲 (AとBを含む) の整数の乱数が欲しいとき +// 参考 http://www.sat.t.u-tokyo.ac.jp/~omi/random_variables_generation.html +/*ap_uint<32> lfsr_random_uint32_0(ap_uint<32> b) { +#pragma HLS INLINE + return lfsr_random() % (b + 1); +}*/ + + +// ================================ // +// メインモジュール +// ================================ // + +// 重みの更新 +// TODO 調整 +// min_uint8(r, MAX_WEIGHT) と同じ +ap_uint<8> new_weight(ap_uint<16> x) { +#pragma HLS INLINE +#if 1 + // 下位8ビット (最大 255) を抜き出して、1/8 をかけて最大 31 (32) にする + ap_uint<8> y = x & 255; + return (ap_uint<8>)(y / 8 + 1); +#endif +#if 0 + // 下位10ビット (最大 1023) を抜き出して、1/32 をかけて最大 31 (32) にする + ap_uint<10> y = x & 1023; + return (ap_uint<8>)(y / 32 + 1); +#endif +#if 0 + ap_uint<8> y = x / 8; + if (y < (ap_uint<16>)MAX_WEIGHT) { return y; } + else { return MAX_WEIGHT; } +#endif +} + +// ボードに関する変数 +static ap_uint<7> size_x; // ボードの X サイズ +static ap_uint<7> size_y; // ボードの Y サイズ +static ap_uint<4> size_z; // ボードの Z サイズ + +static ap_uint<8> line_num = 0; // ラインの総数 + +// グローバル変数で定義する +#ifdef GLOBALVARS + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト + + ap_uint<8> weights[MAX_CELLS]; // セルの重み + + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン +#endif + +bool pynqrouter_256x128(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *status) { +#pragma HLS INTERFACE s_axilite port=boardstr bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=seed bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=status bundle=AXI4LS +#pragma HLS INTERFACE s_axilite port=return bundle=AXI4LS + + *status = -1; + +// グローバル変数では定義しない +#ifndef GLOBALVARS + ap_uint<16> starts[MAX_LINES]; // ラインのスタートリスト +#pragma HLS ARRAY_PARTITION variable=starts complete dim=1 + ap_uint<16> goals[MAX_LINES]; // ラインのゴールリスト +#pragma HLS ARRAY_PARTITION variable=goals complete dim=1 + + ap_uint<8> weights[MAX_CELLS]; // セルの重み +//#pragma HLS ARRAY_PARTITION variable=weights cyclic factor=8 dim=1 partition +// Note: weights は様々な順番でアクセスされるからパーティションしても全然効果ない + + ap_uint<8> paths_size[MAX_LINES]; // ラインが対応するセルIDのサイズ +//#pragma HLS ARRAY_PARTITION variable=paths_size complete dim=1 + ap_uint<16> paths[MAX_LINES][MAX_PATH]; // ラインが対応するセルIDの集合 (スタートとゴールは除く) +//#pragma HLS ARRAY_PARTITION variable=paths cyclic factor=16 dim=2 partition + bool adjacents[MAX_LINES]; // スタートとゴールが隣接しているライン +//#pragma HLS ARRAY_PARTITION variable=adjacents complete dim=1 +#endif + + // ================================ + // 初期化 BEGIN + // ================================ + + // ループカウンタは1ビット余分に用意しないと終了判定できない + INIT_ADJACENTS: + for (ap_uint<9> i = 0; i < (ap_uint<9>)(MAX_LINES); i++) { + adjacents[i] = false; + } + + INIT_WEIGHTS: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=8 + weights[i] = 1; + } + + // ボードストリングの解釈 + + size_x = (boardstr[1] - '0') * 10 + (boardstr[2] - '0'); + size_y = (boardstr[4] - '0') * 10 + (boardstr[5] - '0'); + size_z = (boardstr[7] - '0'); + + INIT_BOARDS: + for (ap_uint<16> idx = 8; idx < BOARDSTR_SIZE; idx+=11) { +//#pragma HLS LOOP_TRIPCOUNT min=100 max=32768 avg=1000 + + // 終端 (null) 文字 + if (boardstr[idx] == 0) { + break; + } + + ap_uint<7> s_x = (boardstr[idx+1] - '0') * 10 + (boardstr[idx+2] - '0'); + ap_uint<7> s_y = (boardstr[idx+3] - '0') * 10 + (boardstr[idx+4] - '0'); + ap_uint<3> s_z = (boardstr[idx+5] - '0') - 1; + ap_uint<7> g_x = (boardstr[idx+6] - '0') * 10 + (boardstr[idx+7] - '0'); + ap_uint<7> g_y = (boardstr[idx+8] - '0') * 10 + (boardstr[idx+9] - '0'); + ap_uint<3> g_z = (boardstr[idx+10] - '0') - 1; + //cout << "L " << line_num << ": (" << s_x << ", " << s_y << ", " << s_z << ") " + // "(" << g_x << ", " << g_y << ", " << g_z << ")" << endl; + + // スタートとゴール + ap_uint<16> start_id = (((ap_uint<16>)s_x * MAX_WIDTH + (ap_uint<16>)s_y) << BITWIDTH_Z) | (ap_uint<16>)s_z; + ap_uint<16> goal_id = (((ap_uint<16>)g_x * MAX_WIDTH + (ap_uint<16>)g_y) << BITWIDTH_Z) | (ap_uint<16>)g_z; + starts[line_num] = start_id; + goals[line_num] = goal_id; + + // 初期状態で数字が隣接しているか判断 + ap_int<8> dx = (ap_int<8>)g_x - (ap_int<8>)s_x; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<8> dy = (ap_int<8>)g_y - (ap_int<8>)s_y; // 最小-71 最大71 (-> 符号付き8ビット) + ap_int<4> dz = (ap_int<4>)g_z - (ap_int<4>)s_z; // 最小-7 最大7 (-> 符号付き4ビット) + if ((dx == 0 && dy == 0 && (dz == 1 || dz == -1)) || (dx == 0 && (dy == 1 || dy == -1) && dz == 0) || ((dx == 1 || dx == -1) && dy == 0 && dz == 0)) { + adjacents[line_num] = true; + paths_size[line_num] = 0; + } else { + adjacents[line_num] = false; + } + + paths_size[line_num] = 0; + weights[start_id] = MAX_WEIGHT; + weights[goal_id] = MAX_WEIGHT; + + line_num++; + } + //cout << size_x << " " << size_y << " " << size_z << endl; + + // 乱数の初期化 + lfsr_random_init(seed); + + // TODO + // すべてのラインが隣接してたらソルバを終わりにする + + // ================================ + // 初期化 END + // ================================ + + // ================================ + // ルーティング BEGIN + // ================================ + + // [Step 1] 初期ルーティング + cout << "Initial Routing" << endl; + FIRST_ROUTING: + for (ap_uint<9> i = 0; i < (ap_uint<9>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=255 avg=50 +//#pragma HLS LOOP_TRIPCOUNT min=2 max=127 avg=50 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[i] == false) { + + // 経路探索 +#ifdef DEBUG_PRINT + cout << "LINE #" << (int)(i + 1) << endl; +#endif + search(&(paths_size[i]), paths[i], starts[i], goals[i], weights); + + } + } + + ap_uint<1> overlap_checks[MAX_CELLS]; +#pragma HLS ARRAY_PARTITION variable=overlap_checks cyclic factor=16 dim=1 partition + bool has_overlap = false; + +#ifndef USE_MOD_CALC + // line_num_2: line_num 以上で最小の2のべき乗数 + ap_uint<9> line_num_2; + CALC_LINE_NUM_2: + for (line_num_2 = 1; line_num_2 < (ap_uint<9>)line_num; line_num_2 *= 2) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 + ; + } + //cout << "line_num: " << line_num << endl; + //cout << "line_num_2: " << line_num_2 << endl; +#endif + + ap_uint<8> last_target = 255; + + // [Step 2] Rip-up 再ルーティング + ROUTING: + for (ap_uint<16> round = 1; round <= 32768 /* = (2048 * 16) */; round++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=4000 avg=50 + +//#ifdef DEBUG_PRINT + //cout << "ITERATION " << round; +//#endif + + // 対象ラインを選択 +#ifdef USE_MOD_CALC + // (1) 剰余演算を用いる方法 + ap_uint<8> target = lfsr_random() % line_num; // i.e., lfsr_random_uint32(0, line_num - 1); +#else + // (2) 剰余演算を用いない方法 + ap_uint<8> target = lfsr_random() & (line_num_2 - 1); + if (line_num <= target) { + //cout << endl; + continue; + } +#endif + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[target] == true) { + //cout << endl; + continue; + } + + // 直前のイテレーション (ラウンド) と同じ対象ラインだったらルーティングスキップする + if (target == last_target) { + //cout << endl; + continue; + } + last_target = target; + + // (1) 引きはがすラインの重みをリセット + ROUTING_RESET: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[target]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + weights[paths[target][j]] = 1; + } + // 対象ラインのスタートの重みも一旦リセット あとで (*) で戻す + weights[starts[target]] = 1; + + // (2) 重みを更新 + ap_uint<8> current_round_weight = new_weight(round); + //cout << " weight " << current_round_weight << endl; + ROUTING_UPDATE: + for (ap_uint<9> i = 0; i < (ap_uint<9>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=255 avg=50 + + // 数字が隣接する場合スキップ、そうでない場合は実行 + if (adjacents[i] == false && i != target) { + ROUTING_UPDATE_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + weights[paths[i][j]] = current_round_weight; + } + } + } + + // 経路探索 +#ifdef DEBUG_PRINT + cout << "LINE #" << (int)(target + 1) << endl; +#endif + search(&(paths_size[target]), paths[target], starts[target], goals[target], weights); + + // (*) 対象ラインのスタートの重みを戻す + weights[starts[target]] = MAX_WEIGHT; + + // ルーティング後 + // オーバーラップのチェック + has_overlap = false; + OVERLAP_RESET: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { +#pragma HLS UNROLL factor=16 + overlap_checks[i] = 0; + } + OVERLAP_CHECK: + for (ap_uint<9> i = 0; i < (ap_uint<9>)(line_num); i++) { +#pragma HLS LOOP_FLATTEN off +#pragma HLS LOOP_TRIPCOUNT min=2 max=255 avg=50 + overlap_checks[starts[i]] = 1; + overlap_checks[goals[i]] = 1; + + // 数字が隣接する場合スキップ、そうでない場合は実行 + //if (adjacents[i] == false) { + + OVERLAP_CHECK_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 +//#pragma HLS PIPELINE rewind II=33 +#pragma HLS PIPELINE II=17 +#pragma HLS UNROLL factor=8 + ap_uint<16> cell_id = paths[i][j]; + if (overlap_checks[cell_id] == 1) { + has_overlap = true; + break; + } + overlap_checks[cell_id] = 1; + } + //} + } + // オーバーラップなければ探索終了 + if (has_overlap == false) { + break; + } + + } + + // 解導出できなかった場合 + if (has_overlap == true) { + *status = 1; + return false; + } + + // ================================ + // ルーティング END + // ================================ + + // ================================ + // 解生成 BEGIN + // ================================ + + // 空白 + OUTPUT_INIT: + for (ap_uint<16> i = 0; i < (ap_uint<16>)(MAX_CELLS); i++) { + boardstr[i] = 0; + } + // ライン + // このソルバでのラインIDを+1して表示する + // なぜなら空白を 0 で表すことにするからラインIDは 1 以上にしたい + OUTPUT_LINE: + for (ap_uint<9> i = 0; i < (ap_uint<9>)(line_num); i++) { +#pragma HLS LOOP_TRIPCOUNT min=2 max=255 avg=50 + boardstr[starts[i]] = (i + 1); + boardstr[goals[i]] = (i + 1); + OUTPUT_LINE_PATH: + for (ap_uint<9> j = 0; j < (ap_uint<9>)(paths_size[i]); j++) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 + boardstr[paths[i][j]] = (i + 1); + } + } + + // ================================ + // 解生成 END + // ================================ + + *status = 0; + return true; +} + + +// ================================ // +// 探索 +// ================================ // + +#ifdef USE_ASTAR +// A* ヒューリスティック用 +// 最大71 最小0 +ap_uint<7> abs_uint7(ap_uint<7> a, ap_uint<7> b) { +#pragma HLS INLINE + if (a < b) { return b - a; } + else { return a - b; } +} +// 最大7 最小0 +ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b) { +#pragma HLS INLINE + if (a < b) { return b - a; } + else { return a - b; } +} +#endif + +// * Pythonでダイクストラアルゴリズムを実装した - フツーって言うなぁ！ +// http://lethe2211.hatenablog.com/entry/2014/12/30/011030 +// * Implementation of A* +// http://www.redblobgames.com/pathfinding/a-star/implementation.html +// をベース +void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start, ap_uint<16> goal, ap_uint<8> w[MAX_CELLS]) { +//#pragma HLS INLINE // search関数はインラインすると遅くなるしBRAM足りなくなる +//#pragma HLS FUNCTION_INSTANTIATE variable=start +//#pragma HLS FUNCTION_INSTANTIATE variable=goal + + ap_uint<16> dist[MAX_CELLS]; // 始点から各頂点までの最短距離を格納する +#pragma HLS ARRAY_PARTITION variable=dist cyclic factor=64 dim=1 partition +// Note: dist のパーティションの factor は 128 にするとBRAMが足りなくなる + ap_uint<16> prev[MAX_CELLS]; // 最短経路における，その頂点の前の頂点のIDを格納する + + SEARCH_INIT_DIST: + for (ap_uint<16> i = 0; i < MAX_CELLS; i++) { +#pragma HLS UNROLL factor=64 + dist[i] = 65535; // = (2^16 - 1) + } + + // プライオリティ・キュー + ap_uint<13> pq_len = 0; + ap_uint<32> pq_nodes[MAX_PQ]; +//#pragma HLS ARRAY_PARTITION variable=pq_nodes complete dim=1 +//#pragma HLS ARRAY_PARTITION variable=pq_nodes cyclic factor=2 dim=1 partition + +#ifdef DEBUG_PRINT + // キューの最大長さチェック用 + ap_uint<13> max_pq_len = 0; +#endif + +#ifdef USE_ASTAR + // ゴールの座標 + ap_uint<13> goal_xy = (ap_uint<13>)(goal >> BITWIDTH_Z); + ap_uint<7> goal_x = (ap_uint<7>)(goal_xy / MAX_WIDTH); + ap_uint<7> goal_y = (ap_uint<7>)(goal_xy - goal_x * MAX_WIDTH); + ap_uint<3> goal_z = (ap_uint<3>)(goal & BITMASK_Z); +#endif + + dist[start] = 0; + pq_push(0, start, &pq_len, pq_nodes); // 始点をpush +#ifdef DEBUG_PRINT + if (max_pq_len < pq_len) { max_pq_len = pq_len; } +#endif + + SEARCH_PQ: + while (0 < pq_len) { +#pragma HLS LOOP_FLATTEN off +#pragma HLS LOOP_TRIPCOUNT min=1 max=1000 avg=100 + + ap_uint<16> prov_cost; + ap_uint<16> src; + pq_pop(&prov_cost, &src, &pq_len, pq_nodes); +#ifdef DEBUG_PRINT +//ap_uint<13> _src_xy = (ap_uint<13>)(src >> BITWIDTH_Z); +//ap_uint<7> _src_x = (ap_uint<7>)(_src_xy / MAX_WIDTH); +//ap_uint<7> _src_y = (ap_uint<7>)(_src_xy % MAX_WIDTH); +//ap_uint<3> _src_z = (ap_uint<3>)(src & BITMASK_Z); +//cout << "Picked up " << (int)src << " (" << (int)_src_x << "," << (int)_src_y << "," << (int)_src_z << ") prov_cost=" << (int)prov_cost << " this_cost=" << w[src] << endl; +#endif + ap_uint<16> dist_src = dist[src]; + +#ifndef USE_ASTAR + // プライオリティキューに格納されている最短距離が，現在計算できている最短距離より大きければ，distの更新をする必要はない + if (dist_src < prov_cost) { + continue; + } +#endif + + // PQの先頭がゴールの場合にPQがまだ空じゃなくても探索終わらせしまう + if (src == goal) { + break; + } + + // 隣接する他の頂点の探索 + // (0) コスト + ap_uint<8> cost = w[src]; + // (1) ノードIDから3次元座標をマスクして抜き出す + ap_uint<13> src_xy = (ap_uint<13>)(src >> BITWIDTH_Z); + ap_uint<7> src_x = (ap_uint<7>)(src_xy / MAX_WIDTH); + ap_uint<7> src_y = (ap_uint<7>)(src_xy - src_x * MAX_WIDTH); + ap_uint<3> src_z = (ap_uint<3>)(src & BITMASK_Z); + //cout << src << " " << src_x << " " << src_y << " " << src_z << endl; + // (2) 3次元座標で隣接するノード (6個) を調べる // 手動ループ展開 +/*********************************************************** + if (src_x > 0) { // x-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x-1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x-1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_x < (size_x-1)) { // x+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x+1) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x+1, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_y > 0) { // y-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y-1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y-1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_y < (size_y-1)) { // y+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y+1)) << BITWIDTH_Z) | (ap_uint<16>)(src_z); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y+1, goal_y) + abs_uint3(src_z, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_z > 0) { // z-を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z-1); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z-1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } + if (src_z < (size_z-1)) { // y+を調査 + ap_uint<16> dest = (((ap_uint<16>)(src_x) * MAX_WIDTH + (ap_uint<16>)(src_y)) << BITWIDTH_Z) | (ap_uint<16>)(src_z+1); + ap_uint<16> dist_new = dist_src + cost; + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 + dist_new += abs_uint7(src_x, goal_x) + abs_uint7(src_y, goal_y) + abs_uint3(src_z+1, goal_z); // A* ヒューリスティック + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush + } + } +***********************************************************/ + + SEARCH_ADJACENTS: + for (ap_uint<3> a = 0; a < 6; a++) { +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + ap_int<8> dest_x = (ap_int<8>)src_x; // 最小-1 最大72 (->符号付き8ビット) + ap_int<8> dest_y = (ap_int<8>)src_y; // 最小-1 最大72 (->符号付き8ビット) + ap_int<5> dest_z = (ap_int<5>)src_z; // 最小-1 最大8 (->符号付き5ビット) + if (a == 0) { dest_x -= 1; } + if (a == 1) { dest_x += 1; } + if (a == 2) { dest_y -= 1; } + if (a == 3) { dest_y += 1; } + if (a == 4) { dest_z -= 1; } + if (a == 5) { dest_z += 1; } + + if (0 <= dest_x && dest_x < (ap_int<8>)size_x && 0 <= dest_y && dest_y < (ap_int<8>)size_y && 0 <= dest_z && dest_z < (ap_int<5>)size_z) { + ap_uint<16> dest = (((ap_uint<16>)dest_x * MAX_WIDTH + (ap_uint<16>)dest_y) << BITWIDTH_Z) | (ap_uint<16>)dest_z; + ap_uint<16> dist_new = dist_src + cost; +#ifdef DEBUG_PRINT +//cout << " adjacent " << (int)dest << " (" << (int)dest_x << "," << (int)dest_y << "," << (int)dest_z << ") dist_new=" << (int)dist_new; +#endif + if (dist[dest] > dist_new) { + dist[dest] = dist_new; // distの更新 + prev[dest] = src; // 前の頂点を記録 +#ifdef USE_ASTAR + dist_new += abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z); // A* ヒューリスティック +#endif + pq_push(dist_new, dest, &pq_len, pq_nodes); // キューに新たな仮の距離の情報をpush +#ifdef DEBUG_PRINT +//cout << " h=" << (int)(abs_uint7(dest_x, goal_x) + abs_uint7(dest_y, goal_y) + abs_uint3(dest_z, goal_z)) << endl; +//cout << (int)dest_x << " " << (int)goal_x << " " << (int)abs_uint7(dest_x, goal_x) << endl; +//cout << (int)dest_y << " " << (int)goal_y << " " << (int)abs_uint7(dest_y, goal_y) << endl; +//cout << (int)dest_z << " " << (int)goal_z << " " << (int)abs_uint7(dest_z, goal_z) << endl; + if (max_pq_len < pq_len) { max_pq_len = pq_len; } +#endif + } +#ifdef DEBUG_PRINT +//else { cout << " -> skip pushing" << endl; } +#endif + } + } + } + + // 経路を出力 + // ゴールからスタートへの順番で表示される (ゴールとスタートは含まれない) + ap_uint<16> t = prev[goal]; + +#ifdef DEBUG_PRINT + int dbg_start_xy = start >> BITWIDTH_Z; + int dbg_start_x = dbg_start_xy / MAX_WIDTH; + int dbg_start_y = dbg_start_xy % MAX_WIDTH; + int dbg_start_z = start & BITMASK_Z; + + int dbg_goal_xy = goal >> BITWIDTH_Z; + int dbg_goal_x = dbg_goal_xy / MAX_WIDTH; + int dbg_goal_y = dbg_goal_xy % MAX_WIDTH; + int dbg_goal_z = goal & BITMASK_Z; + + cout << "(" << dbg_start_x << ", " << dbg_start_y << ", " << dbg_start_z << ") #" << start << " -> " + << "(" << dbg_goal_x << ", " << dbg_goal_y << ", " << dbg_goal_z << ") #" << goal << endl; +#endif + + // バックトラック + ap_uint<8> p = 0; + SEARCH_BACKTRACK: + while (t != start) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=255 avg=50 +#pragma HLS PIPELINE II=2 + +#ifdef DEBUG_PRINT + int t_xy = prev[t] >> BITWIDTH_Z; + int t_x = t_xy / MAX_WIDTH; + int t_y = t_xy % MAX_WIDTH; + int t_z = prev[t] & BITMASK_Z; + cout << " via " << "(" << t_x << ", " << t_y << ", " << t_z << ") #" << prev[t] << " dist=" << dist[t] << endl; +#endif + + path[p] = t; // 記録 + p++; + + t = prev[t]; // 次に移動 + } + *path_size = p; + +#ifdef DEBUG_PRINT + cout << "max_path_len = " << p << endl; + cout << "max_pq_len = " << max_pq_len << endl; +#endif + +} + +// プライオリティ・キュー (ヒープで実装) +// 優先度の最小値がヒープのルートに来る +// 参考 +// * ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E3%83%92%E3%83%BC%E3%83%97 +// * 二分ヒープ - Wikipedia https://ja.wikipedia.org/wiki/%E4%BA%8C%E5%88%86%E3%83%92%E3%83%BC%E3%83%97 +// * ヒープの正体 - http://www.maroontress.com/Heap/ +// * Priority queue - Rosetta Code https://rosettacode.org/wiki/Priority_queue#C +// Note +// インデックスが0から始まるとき (0-origin index) +// --> 親: (n-1)/2, 左の子: 2n+1, 右の子: 2n+2 +// インデックスが1から始まるとき (1-origin index) +// --> 親: n/2, 左の子: 2n, 右の子: 2n+1 +// FPGA的にはどちらも遅延は同じだけど 1-origin の方がLUTリソース少なくて済む (ただし配列の0要素が無駄になる) + +// ノードの挿入は，末尾に追加してから優先度が正しい高さの位置までノードを上げていく +// 探索の都合上，同じ優先度では後から入れた方を先に出したいから， +// ループの終了条件は挿入ノードの優先度が比較対象の優先度よりも小さくなったとき +void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<13> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { +#pragma HLS INLINE + + (*pq_len)++; + ap_uint<13> i = (*pq_len); // target + ap_uint<13> p = (*pq_len) >> 1; // i.e., (*pq_len) / 2; // 親 + PQ_PUSH_LOOP: + while (i > 1 && (ap_uint<16>)(pq_nodes[p] & PQ_PRIORITY_MASK) >= priority) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + pq_nodes[i] = pq_nodes[p]; + i = p; + p = p >> 1; // i.e., p / 2; // 親 + } + pq_nodes[i] = ((ap_uint<32>)data << 16) | (ap_uint<32>)priority; +} + +// ノードの取り出しは，ルートを取ってくるだけ +// 次に最小の優先度をもつノードをルートに移動させるために， +// まず，末尾のノードをルートに移動する +// 両方の子で優先度が小さい方を上にもっていく (ルートを適切な高さまで下げる) +// これを再帰的に繰り返す +void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<13> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]) { +#pragma HLS INLINE + + *ret_priority = (ap_uint<16>)(pq_nodes[1] & PQ_PRIORITY_MASK); + *ret_data = (ap_uint<16>)(pq_nodes[1] >> PQ_PRIORITY_WIDTH); + + //pq_nodes[1] = pq_nodes[*pq_len]; + ap_uint<13> i = 1; // 親ノード + //ap_uint<13> t = 1; // 交換対象ノード + + ap_uint<16> last_priority = (ap_uint<16>)(pq_nodes[*pq_len] & PQ_PRIORITY_MASK); // 末尾ノードの優先度 + + PQ_POP_LOOP: + while (1) { +#pragma HLS LOOP_TRIPCOUNT min=1 max=8 avg=4 +//#pragma HLS PIPELINE +//#pragma HLS UNROLL factor=2 + ap_uint<13> c1 = i << 1; // i.e., 2 * i; // 左の子 + ap_uint<13> c2 = c1 + 1; // i.e., 2 * i + 1; // 右の子 + if (c1 < *pq_len && (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK) <= last_priority) { + if (c2 < *pq_len && (ap_uint<16>)(pq_nodes[c2] & PQ_PRIORITY_MASK) <= (ap_uint<16>)(pq_nodes[c1] & PQ_PRIORITY_MASK)) { + pq_nodes[i] = pq_nodes[c2]; + i = c2; + } + else { + pq_nodes[i] = pq_nodes[c1]; + i = c1; + } + } + else { + if (c2 < *pq_len && (ap_uint<16>)(pq_nodes[c2] & PQ_PRIORITY_MASK) <= last_priority) { + pq_nodes[i] = pq_nodes[c2]; + i = c2; + } + else { + break; + } + } + } + pq_nodes[i] = pq_nodes[*pq_len]; + (*pq_len)--; + +// For verification + //for (int k = 1;k<(*pq_len);k++){ + // cout << (ap_uint<16>)(pq_nodes[k] & PQ_PRIORITY_MASK) << " "; + //} + //cout << endl; +} diff --git a/hls/lines256_length128/router.hpp b/hls/lines256_length128/router.hpp new file mode 100644 index 0000000000000000000000000000000000000000..b0d9522b25bbe88d35a0195c1bae249bb75a73fe --- /dev/null +++ b/hls/lines256_length128/router.hpp @@ -0,0 +1,62 @@ +/** + * router.hpp + * + * for Vivado HLS + */ + +#ifndef __ROUTER_HPP__ +#define __ROUTER_HPP__ + +#ifdef SOFTWARE +#include "ap_int.h" +#else +#include +#endif + +//#define DEBUG_PRINT // いろいろ表示する +#define USE_ASTAR // A* 探索を使う +#define USE_MOD_CALC // ターゲットラインの選択に剰余演算を用いる + +using namespace std; + + +// 各種設定値 +#define MAX_WIDTH 72 // X, Y の最大値 (7ビットで収まる) +#define BITWIDTH_XY 13 +#define BITMASK_XY 65528 // 1111 1111 1111 1000 +#define MAX_LAYER 8 // Z の最大値 (3ビット) +#define BITWIDTH_Z 3 +#define BITMASK_Z 7 // 0000 0000 0000 0111 + +#define MAX_CELLS 41472 // セルの総数 =72*72*8 (16ビットで収まる) +#define MAX_LINES 256 // ライン数の最大値 (7ビット) +#define MAX_PATH 128 // 1つのラインが対応するセル数の最大値 (8ビット) +#define MAX_PQ 8192 // 探索時のプライオリティ・キューの最大サイズ (12ビット) 足りないかも？ + +#define PQ_PRIORITY_WIDTH 16 +#define PQ_PRIORITY_MASK 65535 // 0000 0000 0000 0000 1111 1111 1111 1111 +#define PQ_DATA_WIDTH 16 +#define PQ_DATA_MASK 4294901760 // 1111 1111 1111 1111 0000 0000 0000 0000 + +#define MAX_WEIGHT 255 // 重みの最大値 (8ビットで収まる) +#define BOARDSTR_SIZE 41472 // ボードストリングの最大文字数 (セル数 72*72*8 あれば良い) + + +void lfsr_random_init(ap_uint<32> seed); +ap_uint<32> lfsr_random(); +//ap_uint<32> lfsr_random_uint32(ap_uint<32> a, ap_uint<32> b); +//ap_uint<32> lfsr_random_uint32_0(ap_uint<32> b); + +ap_uint<8> new_weight(ap_uint<16> x); +bool pynqrouter_256x128(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *status); + +#ifdef USE_ASTAR +ap_uint<7> abs_uint7(ap_uint<7> a, ap_uint<7> b); +ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b); +#endif + +void search(ap_uint<8> *path_size, ap_uint<16> path[MAX_PATH], ap_uint<16> start, ap_uint<16> goal, ap_uint<8> w[MAX_WEIGHT]); +void pq_push(ap_uint<16> priority, ap_uint<16> data, ap_uint<13> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]); +void pq_pop(ap_uint<16> *ret_priority, ap_uint<16> *ret_data, ap_uint<13> *pq_len, ap_uint<32> pq_nodes[MAX_PQ]); + +#endif /* __ROUTER_HPP__ */