Multi-PE host code and Vivado HLS tcl script

triangle_counting_hls/run_tc_hls.tcl

@@ -0,0 +1,13 @@
+open_project -reset tc_hls
+set_top triangle_counting
+add_files triangle_counting.cc
+add_files -tb triangle_counting_tb.cc
+
+open_solution -reset "solution1"
+set_part {xc7z020clg400-1}
+create_clock -period 10
+
+csynth_design
+export_design -format ip_catalog -version "0.001"
+
+exit

triangle_counting_host/tc_host_opt.py

@@ -0,0 +1,154 @@
+# coding: utf-8
+
+import sys
+import numpy as np 
+import os
+import time
+import math
+from datetime import datetime
+from pynq import Xlnk
+from pynq import Overlay
+
+# load our design overlay
+overlay = Overlay('tc_opt.bit')
+print("tc_opt.bit loaded")
+
+myIP = overlay.triangle_counting_0
+
+t0 = time.time()
+
+neighbor_list = []
+offset_list = [0]
+edge_list = []
+
+graph_file = open("graph/soc-Epinions1_adj.tsv")
+# graph_file = open("graph/test.tsv")
+lines = graph_file.readlines()
+
+degree_count = 0
+prev_node = 0
+
+for line in lines:
+    node_a, node_b, _ = map(int, line.split())
+    if prev_node != node_b:
+        offset_list.append(degree_count)
+
+    prev_node = node_b
+    if node_a < node_b:
+        edge_list.extend([node_b, node_a])
+    else:
+        neighbor_list.append(node_a)
+        degree_count += 1
+
+offset_list.append(degree_count)
+
+print("neighbor_list size= ", len(neighbor_list))
+print("offset_list size= ", len(offset_list))
+print("edge_list size= ", len(edge_list))
+
+t1 = time.time()
+
+print("Finished reading graph file. ")
+t = t1 - t0
+print("Reading input file time: ", str(t))
+
+xlnk = Xlnk()
+
+num_edge = len(edge_list) / 2
+num_batch = 4
+num_edge_batch = int(math.floor(float(num_edge) / num_batch))
+num_edge_last_batch = num_edge - (num_batch-1)*num_edge_batch
+
+neighbor = xlnk.cma_array(shape=(len(neighbor_list),), dtype=np.int32)
+offset   = xlnk.cma_array(shape=(len(offset_list),), dtype=np.int32)
+edge1    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge2    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge3    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge4    = xlnk.cma_array(shape=(2*num_edge_last_batch,), dtype=np.int32)
+progress = xlnk.cma_array(shape=(5,), dtype=np.int32)
+
+neighbor[:] = neighbor_list
+offset[:] = offset_list
+edge1[:] = edge_list[0:2*num_edge_batch-1]
+edge2[:] = edge_list[2*num_edge_batch:4*num_edge_batch-1]
+edge3[:] = edge_list[4*num_edge_batch:6*num_edge_batch-1]
+edge4[:] = edge_list[6*num_edge_batch:-1]
+
+# neighbor[:] = [2, 4, 5, 3, 4, 5, 4, 5, 5]
+# offset[:]   = [0, 0, 3, 6, 8, 9, 9]
+# edge[:]     = [5, 4, 5, 3, 5, 2, 5, 1, 4, 3, 4, 2, 4, 1, 3, 2, 2, 1]
+
+myIP.write(0x00018, neighbor.physical_address)
+myIP.write(0x00020, offset.physical_address)
+myIP.write(0x00028, edge1.physical_address)
+myIP.write(0x00030, 2*num_edge_batch)
+myIP.write(0x00038, progress.physical_address)
+
+myIP.write(0x10018, neighbor.physical_address)
+myIP.write(0x10020, offset.physical_address)
+myIP.write(0x10028, edge2.physical_address)
+myIP.write(0x10030, 2*num_edge_batch)
+myIP.write(0x10038, progress.physical_address)
+
+myIP.write(0x20018, neighbor.physical_address)
+myIP.write(0x20020, offset.physical_address)
+myIP.write(0x20028, edge3.physical_address)
+myIP.write(0x20030, 2*num_edge_batch)
+myIP.write(0x20038, progress.physical_address)
+
+myIP.write(0x30018, neighbor.physical_address)
+myIP.write(0x30020, offset.physical_address)
+myIP.write(0x30028, edge4.physical_address)
+myIP.write(0x30030, 2*num_edge_last_batch)
+myIP.write(0x30038, progress.physical_address)
+
+# for i in range(neighbor.size):
+#     print("neighbor[%d] = %d" % (i, neighbor[i]))
+
+# for i in range(offset.size):
+#     print("offset[%d] = %d" % (i, offset[i]))
+
+# for i in range(edge.size):
+#     print("edge[%d] = %d" % (i, edge[i]))
+
+t2 = time.time()
+t = t2 - t1
+print("Preparing input data time: ", str(t))
+
+myIP.write(0x00000, 1)
+myIP.write(0x10000, 1)
+myIP.write(0x20000, 1)
+myIP.write(0x30000, 1)
+
+isready = 0;
+while( isready != 6 ):
+    isready = myIP.read(0x00000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = myIP.read(0x10000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = myIP.read(0x20000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = myIP.read(0x30000)
+
+t3 = time.time()
+t = t3 - t2
+#tbatch = tbatch + t
+#print("Computation finished")
+print("PL Time: ", str(t))
+
+result1 = myIP.read(0x00010)
+result2 = myIP.read(0x10010)
+result3 = myIP.read(0x20010)
+result4 = myIP.read(0x30010)
+
+print("Return value 1: ", result1)
+print("Return value 2: ", result2)
+print("Return value 3: ", result3)
+print("Return value 4: ", result4)
+print("Number of triangles: ", result1+result2+result3+result4)

triangle_counting_host/tc_host_opt_4.py

@@ -0,0 +1,162 @@
+# coding: utf-8
+
+import sys
+import numpy as np 
+import os
+import time
+import math
+from datetime import datetime
+from pynq import Xlnk
+from pynq import Overlay
+
+# load our design overlay
+overlay = Overlay('tc_opt_4.bit')
+print("tc_opt_4.bit loaded")
+
+acc0 = overlay.triangle_counting_0
+acc1 = overlay.triangle_counting_1
+acc2 = overlay.triangle_counting_2
+acc3 = overlay.triangle_counting_3
+
+t0 = time.time()
+
+neighbor_list = []
+offset_list = [0]
+edge_list = []
+
+graph_file = open("../../graph/soc-Epinions1_adj.tsv")
+# graph_file = open("graph/test.tsv")
+lines = graph_file.readlines()
+
+degree_count = 0
+prev_node = 0
+
+for line in lines:
+    node_a, node_b, _ = map(int, line.split())
+    if prev_node != node_b:
+        offset_list.append(degree_count)
+
+    prev_node = node_b
+    if node_a < node_b:
+        edge_list.extend([node_b, node_a])
+    else:
+        neighbor_list.append(node_a)
+        degree_count += 1
+
+offset_list.append(degree_count)
+
+print("neighbor_list size= ", len(neighbor_list))
+print("offset_list size= ", len(offset_list))
+print("edge_list size= ", len(edge_list))
+
+t1 = time.time()
+
+print("Finished reading graph file. ")
+t = t1 - t0
+print("Reading input file time: ", str(t))
+
+xlnk = Xlnk()
+
+num_edge = int(len(edge_list) / 2)
+num_batch = 4
+num_edge_batch = int(math.floor(float(num_edge) / num_batch))
+num_edge_last_batch = num_edge - (num_batch-1)*num_edge_batch
+
+print(num_edge)
+print(num_batch)
+print(num_edge_batch)
+print(num_edge_last_batch)
+
+neighbor = xlnk.cma_array(shape=(len(neighbor_list),), dtype=np.int32)
+offset   = xlnk.cma_array(shape=(len(offset_list),), dtype=np.int32)
+edge1    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge2    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge3    = xlnk.cma_array(shape=(2*num_edge_batch,), dtype=np.int32)
+edge4    = xlnk.cma_array(shape=(2*num_edge_last_batch,), dtype=np.int32)
+progress = xlnk.cma_array(shape=(5,), dtype=np.int32)
+
+neighbor[:] = neighbor_list
+offset[:] = offset_list
+edge1[:] = edge_list[0:2*num_edge_batch]
+edge2[:] = edge_list[2*num_edge_batch:4*num_edge_batch]
+edge3[:] = edge_list[4*num_edge_batch:6*num_edge_batch]
+edge4[:] = edge_list[6*num_edge_batch:]
+
+# neighbor[:] = [2, 4, 5, 3, 4, 5, 4, 5, 5]
+# offset[:]   = [0, 0, 3, 6, 8, 9, 9]
+# edge[:]     = [5, 4, 5, 3, 5, 2, 5, 1, 4, 3, 4, 2, 4, 1, 3, 2, 2, 1]
+
+acc0.write(0x00018, neighbor.physical_address)
+acc0.write(0x00020, offset.physical_address)
+acc0.write(0x00028, edge1.physical_address)
+acc0.write(0x00030, 2*num_edge_batch)
+acc0.write(0x00038, progress.physical_address)
+
+acc1.write(0x00018, neighbor.physical_address)
+acc1.write(0x00020, offset.physical_address)
+acc1.write(0x00028, edge2.physical_address)
+acc1.write(0x00030, 2*num_edge_batch)
+acc1.write(0x00038, progress.physical_address)
+
+acc2.write(0x00018, neighbor.physical_address)
+acc2.write(0x00020, offset.physical_address)
+acc2.write(0x00028, edge3.physical_address)
+acc2.write(0x00030, 2*num_edge_batch)
+acc2.write(0x00038, progress.physical_address)
+
+acc3.write(0x00018, neighbor.physical_address)
+acc3.write(0x00020, offset.physical_address)
+acc3.write(0x00028, edge4.physical_address)
+acc3.write(0x00030, 2*num_edge_last_batch)
+acc3.write(0x00038, progress.physical_address)
+
+# for i in range(neighbor.size):
+#     print("neighbor[%d] = %d" % (i, neighbor[i]))
+
+# for i in range(offset.size):
+#     print("offset[%d] = %d" % (i, offset[i]))
+
+# for i in range(edge.size):
+#     print("edge[%d] = %d" % (i, edge[i]))
+
+t2 = time.time()
+t = t2 - t1
+print("Preparing input data time: ", str(t))
+
+acc0.write(0x00000, 1)
+acc1.write(0x00000, 1)
+acc2.write(0x00000, 1)
+acc3.write(0x00000, 1)
+
+isready = 0;
+while( isready != 6 ):
+    isready = acc0.read(0x00000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = acc1.read(0x00000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = acc2.read(0x00000)
+
+isready = 0;
+while( isready != 6 ):
+    isready = acc3.read(0x00000)
+
+t3 = time.time()
+t = t3 - t2
+#tbatch = tbatch + t
+#print("Computation finished")
+print("PL Time: ", str(t))
+
+result1 = acc0.read(0x00010)
+result2 = acc1.read(0x00010)
+result3 = acc2.read(0x00010)
+result4 = acc3.read(0x00010)
+
+print("Return value 1: ", result1)
+print("Return value 2: ", result2)
+print("Return value 3: ", result3)
+print("Return value 4: ", result4)
+print("Number of triangles: ", result1+result2+result3+result4)