Add pragma for Vivado-HLS

hls_2018/router_03/router.cpp

@@ -51,7 +51,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat
     // For each line
     // Note: Should not partition completely
     bool adjacents[MAX_LINES];              // Line has adjacent terminals?
-    ap_uint<CELL_BIT> starts[MAX_LINES];    // Start list
+    ap_uint<CELL_BIT> starts[MAX_LINES];	// Start list
     ap_uint<CELL_BIT> goals[MAX_LINES];     // Goal list
     ap_uint<BUFF_BIT> s_idx[MAX_LINES];     // Start point on line buffer
 
@@ -151,6 +151,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat
 
     // Memories for Overlap Check
     ap_uint<1> overlap_checks[MAX_CELLS];
+#pragma HLS ARRAY_PARTITION variable=overlap_checks cyclic factor=32 dim=1
     bool has_overlap = false;
 
     // ================================
@@ -188,6 +189,7 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat
         // (Step.2-1) Reset weights of target line
         WEIGHT_RESET:
         for (ap_uint<BUFF_BIT> j = s_idx[target]; j != s_idx[next_target]; j++) {
+#pragma HLS UNROLL factor=2
 #pragma HLS LOOP_TRIPCOUNT min=1 max=256
             weights[paths[j]] = 1;
         }
@@ -196,11 +198,13 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat
         ap_uint<8> current_round_weight = new_weight(round);
         WEIGHT_PATH:
         for (ap_uint<BUFF_BIT> j = s_idx[next_target]; j != pointer; j++) {
+#pragma HLS UNROLL factor=2
 #pragma HLS LOOP_TRIPCOUNT min=1 max=8192
             weights[paths[j]] = current_round_weight;
         }
         WEIGHT_TERMINAL:
         for (ap_uint<LINE_BIT> i = 0; i < (ap_uint<LINE_BIT>)(line_num); i++) {
+#pragma HLS UNROLL factor=2
 #pragma HLS LOOP_TRIPCOUNT min=2 max=999
             weights[starts[i]] = MAX_WEIGHT;
             weights[goals[i]] = MAX_WEIGHT;
@@ -230,16 +234,19 @@ bool pynqrouter(char boardstr[BOARDSTR_SIZE], ap_uint<32> seed, ap_int<32> *stat
         has_overlap = false;
         OVERLAP_RESET:
         for (ap_uint<CELL_BIT> i = 0; i < (ap_uint<CELL_BIT>)(MAX_CELLS); i++) {
+#pragma HLS UNROLL factor=32
             overlap_checks[i] = 0;
         }
         OVERLAP_CHECK_LINE:
         for (ap_uint<LINE_BIT> i = 0; i < (ap_uint<LINE_BIT>)(line_num); i++) {
+#pragma HLS UNROLL factor=2
 #pragma HLS LOOP_TRIPCOUNT min=2 max=999
             overlap_checks[starts[i]] = 1;
             overlap_checks[goals[i]] = 1;
         }
         OVERLAP_CHECK_PATH:
         for (ap_uint<BUFF_BIT> j = s_idx[next_target]; j != pointer; j++) {
+#pragma HLS UNROLL factor=2
 #pragma HLS LOOP_TRIPCOUNT min=1 max=8192
             ap_uint<CELL_BIT> cell_id = paths[j];
             if (overlap_checks[cell_id]) {
@@ -340,10 +347,12 @@ ap_uint<3> abs_uint3(ap_uint<3> a, ap_uint<3> b) {
 ap_uint<BUFF_BIT> search(ap_uint<BUFF_BIT> idx, ap_uint<CELL_BIT> paths[MAX_BUFFER], ap_uint<CELL_BIT> start, ap_uint<CELL_BIT> goal, ap_uint<8> w[MAX_CELLS]) {
 
     ap_uint<CELL_BIT> dist[MAX_CELLS];
+#pragma HLS ARRAY_PARTITION variable=dist cyclic factor=64 dim=1
     ap_uint<CELL_BIT> prev[MAX_CELLS];
 
     SEARCH_INIT_DIST:
     for (ap_uint<CELL_BIT> i = 0; i < (ap_uint<CELL_BIT>)(MAX_CELLS); i++) {
+#pragma HLS UNROLL factor=64
         dist[i] = 65535; // = (2^16 - 1)
     }
 
@@ -459,7 +468,7 @@ void pq_push(ap_uint<32> pq_nodes[MAX_PQ], ap_uint<16> priority, ap_uint<16> dat
     ap_uint<PQ_BIT> p = (*pq_len) >> 1; // parent node
     PQ_PUSH_LOOP:
     while (i > 1 && (ap_uint<16>)(pq_nodes[p] & PQ_PRIORITY_MASK) >= priority) {
-#pragma HLS LOOP_TRIPCOUNT min=0 max=15
+#pragma HLS LOOP_TRIPCOUNT min=0 max=14
 /** Set!: min=0 max=PQ_BIT **/
         pq_nodes[i] = pq_nodes[p];
         i = p;
@@ -482,7 +491,7 @@ void pq_pop(ap_uint<32> pq_nodes[MAX_PQ], ap_uint<16> *ret_priority, ap_uint<16>
 
     PQ_POP_LOOP:
     while (!(i >> (PQ_BIT-1))) { // (2018.08.24) Loop condition fixed
-#pragma HLS LOOP_TRIPCOUNT min=1 max=15
+#pragma HLS LOOP_TRIPCOUNT min=1 max=14
 /** Set!: min=0 max=PQ_BIT **/
         ap_uint<PQ_BIT> c1 = i << 1; // child node(left)
         ap_uint<PQ_BIT> c2 = c1 + 1; // child node(right)

hls_2018/router_03/router.hpp

@@ -29,11 +29,11 @@ using namespace std;
 
 #define MAX_CELLS	41472    // Max #cells (16bit)
 #define MAX_LINES	1024     // Max #lines (10bit)
-#define MAX_PQ		32768    // Queue size (15bit)
+#define MAX_PQ		16384    // Queue size (14bit)
 #define MAX_BUFFER	16384    // Line buffer size (14bit)
 #define CELL_BIT	16
 #define LINE_BIT	10
-#define PQ_BIT		15
+#define PQ_BIT		14
 #define BUFF_BIT	14
 
 #define PQ_PRIORITY_WIDTH 16