build graph, and model runs

.gitignore

 place_parser
 data/asap7
+*/__pycache__
+*.log

build_graph/create_graph.py

@@ -827,6 +827,7 @@ def create_graph6(data_dir, parsed_libs_dir, parsed_sdf_dir, save_dir, block_nam
       # NOTE: this assume that the path has only buffers inserted, 
       # and each buffer has only one input and one output
       delay += path.value # is it right??
+    arc.delay = delay
 
   rebuild = False
   if rebuild:
@@ -865,7 +866,7 @@ def create_graph6(data_dir, parsed_libs_dir, parsed_sdf_dir, save_dir, block_nam
       faninfanout[pin_id] = 1
     
     # find primary_io_pin
-    types = {'inport2d', 'inport2clk', 'inport2pin', 'inport2outport'}
+    types = {'inport2d', 'inport2clk', 'inport2in', 'inport2outport'}
     if arc.type in types:
       pin_id = place_pin2id_map[arc.src]
       is_primary_io_pin[pin_id] = 1
@@ -895,17 +896,30 @@ def create_graph6(data_dir, parsed_libs_dir, parsed_sdf_dir, save_dir, block_nam
     'inport2d', 'inport2clk', 'inport2in', 
     'q2outport', 'out2outport', 'inport2outport', 
     'q2in', 'out2in', 'out2d', 'q2d']
-  # type net_in is a reverse of net_out's src and dst
   
   # build DGL edge types
   for arc_type, types in dgl_edge_types.items():
     edges[arc_type] = []
     for merge_type in types:
       edges[arc_type] += edges[merge_type]
+  # type net_in is a reverse of net_out's src and dst
+  edges['net_in'] = []
+  for edge in edges['net_out']:
+    edges['net_in'].append((edge[1], edge[0]))
+
+  dgl_graph = {}
+  for arc_type in edges:
+    dgl_graph[('node', arc_type, 'node')] = edges[arc_type]
+  g = dgl.heterograph(dgl_graph)
+
+  """
   g = dgl.heterograph({
+    # for training
     ('node', 'net_out', 'node'): edges['net_out'],
+    ('node', 'net_in', 'node'): edges['net_in'],
     ('node', 'cell_out', 'node'): edges['cell_out'],
   })
+  """
 
   # test
   try:
@@ -957,12 +971,171 @@ def create_graph6(data_dir, parsed_libs_dir, parsed_sdf_dir, save_dir, block_nam
   ######## End build DGL graph ########
 
   ######## fill data into graph ########
+  # node data
+  num_pins = pin_cnt
+  valid_pin_list = place_pins_list
+
+  src_pin2path_map_place = dict()
+  for arc in timing_graph.all_arcs:
+    if not arc.src in src_pin2path_map_place:
+      src_pin2path_map_place[arc.src] = []
+    src_pin2path_map_place[arc.src].append(arc)
+
+  net_delays_at_fanin = np.zeros((num_pins, 4))
+  for edge in edges['net_out']:
+    src = edge[0]
+    dst = edge[1]
+    src_pin = place_pins_list[src]
+    dst_pin = place_pins_list[dst]
+    paths = src_pin2path_map_place[src_pin]
+    flag = False
+    for path in paths:
+      if path.dst == dst_pin:
+        net_delays_at_fanin[dst] = np.maximum(path.delay, net_delays_at_fanin[dst])
+        flag = True
+        break
+    if not flag:
+      pdb.set_trace()
+      assert(False)
+
+  pin_loc = np.zeros((num_pins, 4))
+  for i in range(num_pins):
+    pin_name = valid_pin_list[i]
+    inst_name = pin_name.split('/')[0]
+    place_node_id = node_name2id_map[inst_name]
+    place_node_x = node_x[place_node_id]
+    place_node_y = node_y[place_node_id]
+    place_node_pins = node2pin_map[place_node_id]
+    if inst_name == pin_name: # port
+      assert(place_node_pins.size == 1)
+      place_node_pin_id = place_node_pins[0]
+    else:
+      num_node_pins = place_node_pins.size
+      for pi in range(num_node_pins):
+        place_node_pin_id = place_node_pins[pi]
+        place_node_pin_name = pin_names_raw[place_node_pin_id].decode()
+        if pin_name.split('/')[1] == place_node_pin_name:
+          break
+    place_pin_x_off = pin_offset_x[place_node_pin_id]
+    place_pin_y_off = pin_offset_y[place_node_pin_id]
+    pin_x = place_node_x + place_pin_x_off
+    pin_y = place_node_y + place_pin_y_off
+    pin_loc[i] = np.array([pin_x - xl, pin_y - yl, xh - pin_x, yh - pin_y])
+
+  ndata = dict()
+  ndata["n_rats"] = torch.zeros(num_pins, 4)
+  ndata["n_ats"] = torch.zeros(num_pins, 4)
+  ndata["n_slews"] = torch.zeros(num_pins, 4)
+  ndata["n_net_delays"] = torch.from_numpy(net_delays_at_fanin).float()
+  ndata["nf"] = torch.zeros(num_pins, 10)
+  ndata["n_is_timing_endpt"] = torch.from_numpy(is_timing_endpoint).float()
+  for pi in range(num_pins):
+    pin_name = valid_pin_list[pi]
+    ndata["n_rats"][pi] = torch.from_numpy(pin_time[pin_name].rat)
+    ndata["n_ats"][pi] = torch.from_numpy(pin_time[pin_name].at)
+    ndata["n_slews"][pi] = torch.from_numpy(pin_time[pin_name].slew)
+
+    ndata['nf'][pi][0] = is_primary_io_pin[pi]
+    ndata['nf'][pi][1] = faninfanout[pi]
+    ndata['nf'][pi][2:6] = torch.from_numpy(pin_loc[pi])
+    # cap data
+    if is_primary_io_pin[pi]:
+      ndata["nf"][pi][6:10] = torch.zeros(4)
+    else:
+      inst = pin_name.split('/')[0]
+      pin_name_str = pin_name.split('/')[1]
+      celltype = inst2libcell_map[inst]
+      cap_data = np.zeros(4)
+      cap_data[0] = pin_caps['fast'][celltype][pin_name_str]['rise_capacitance']
+      cap_data[1] = pin_caps['fast'][celltype][pin_name_str]['fall_capacitance']
+      cap_data[2] = pin_caps['slow'][celltype][pin_name_str]['rise_capacitance']
+      cap_data[3] = pin_caps['slow'][celltype][pin_name_str]['fall_capacitance']
+      ndata["nf"][pi][6:10] = torch.from_numpy(cap_data)
+
+  # edge data
+  edges_features = dict()
+  edges_features['net_out'] = []
+  edges_features['net_in'] = []
+  edges_features['cell_out'] = []
+  for edge in edges['net_out']:
+    src = edge[0]
+    dst = edge[1]
+    srcx = pin_loc[src][0] + xl
+    srcy = pin_loc[src][1] + yl
+    dstx = pin_loc[dst][0] + xl
+    dsty = pin_loc[dst][1] + yl
+    ef = np.array([dstx - srcx, dsty - srcy])
+    edges_features['net_out'].append(ef)
+    edges_features['net_in'].append(-ef)
+  for edge in edges['cell_out']:
+    src = edge[0]
+    dst = edge[1]
+    src_pin = valid_pin_list[src]
+    dst_pin = valid_pin_list[dst]
+    inst = src_pin.split('/')[0]
+    assert(inst == dst_pin.split('/')[0])
+    path_key = dst_pin.split('/')[1] + '/' + src_pin.split('/')[1]
+    celltype = inst2libcell_map[inst]
+    ef = np.zeros(512)
+    ef[:4*15] = lib_data['fast'][celltype][path_key][:4*15]
+    ef[4*15 : 2*4*15] = lib_data['slow'][celltype][path_key][:4*15]
+    ef[2*4*15 : 2*4*15+4*49] = lib_data['fast'][celltype][path_key][-4*49:]
+    ef[-4*49:] = lib_data['slow'][celltype][path_key][-4*49:]
+    edges_features['cell_out'].append(ef)
+
+  edges_delay = dict()
+  edges_delay['net_out'] = []
+  edges_delay['cell_out'] = []
+  for edge in edges['net_out']:
+    src = edge[0]
+    dst = edge[1]
+    src_pin = valid_pin_list[src]
+    dst_pin = valid_pin_list[dst]
+    paths = src_pin2path_map_place[src_pin]
+    flag = False
+    for path in paths:
+      if path.dst == dst_pin:
+        flag = True
+        edges_delay['net_out'].append(path.delay)
+        break
+    if not flag:
+      pdb.set_trace()
+      assert(False)
+  for edge in edges['cell_out']:
+    src = edge[0]
+    dst = edge[1]
+    src_pin = valid_pin_list[src]
+    dst_pin = valid_pin_list[dst]
+    paths = src_pin2path_map_place[src_pin]
+    flag = False
+    for path in paths:
+      if path.dst == dst_pin:
+        flag = True
+        edges_delay['cell_out'].append(path.delay)
+        break
+    if not flag:
+      pdb.set_trace()
+      assert(False)
+
+  g = g_hetero
+  g.ndata['n_rats'] = ndata['n_rats'].float()
+  g.ndata['n_ats']  = ndata['n_ats'].float()
+  g.ndata['n_slews'] = ndata['n_slews'].float()
+  g.ndata['nf'] = ndata['nf'].float()
+  g.ndata['n_is_timing_endpt'] = ndata['n_is_timing_endpt'].float()
+
+  g.edges['cell_out'].data['e_delay'] = torch.tensor(edges_delay['cell_out']).float()
+  g.edges['cell_out'].data['ef'] = torch.tensor(edges_features['cell_out']).float()
+  g.edges['net_out'].data['e_delay'] = torch.tensor(edges_delay['net_out']).float()
+  g.edges['net_out'].data['ef'] = torch.tensor(edges_features['net_out']).float()
+  g.edges['net_in'].data['ef'] = torch.tensor(edges_features['net_in']).float()
+
   ######## End fill data into graph ########
 
   ######## save graph ########
+  dgl.save_graphs(os.path.join(save_dir, f'{block_name}.graph.bin'), [g])
   ######## End save graph ########
 
-
 if __name__ == '__main__':
     pdk = "asap7"
     
@@ -979,7 +1152,7 @@ if __name__ == '__main__':
     # blocks = "ethmac".split()
     # blocks = "mock-alu".split()
 
-    raw_data_dir = f"/cyberpi/OpenROAD-flow-scripts/flow/parse/{pdk}"
+    raw_data_dir = f"../data/{pdk}"
     techlib_dir = os.path.join(raw_data_dir, "techlib")
 
     do_parse_libs = 0
@@ -1027,13 +1200,6 @@ if __name__ == '__main__':
       sdf_dir = os.path.join(block_dir, 'parsed')
       lib_dir = os.path.join(techlib_dir, 'parsed_lib')
       
-      # NOTE: skip error blocks
-      # error_blocks = {'ethmac', 'ibex', 'jpeg', 'mock-alu'}
-      # if block in error_blocks:
-      #   print('skip')
-      #   continue
-
-      # check_legal(data_dir, lib_dir, sdf_dir, save_dir, block)
       create_graph6(data_dir, lib_dir, sdf_dir, save_dir, block)
 
 

build_graph/utils.py

@@ -400,6 +400,7 @@ def parse_libs(libs, save_dir):
       print(f'finish reading {liberty_file}')
 
   # pickle and save libdata and pin_cap
+  os.makedirs(save_dir, exist_ok=True)
   with open(os.path.join(save_dir, "lib_data.pkl"), 'wb') as f:
     pickle.dump(libdata, f)
     f.close()
@@ -465,7 +466,7 @@ class TimingArc:
     self.type = None
 
     self.buffers = [] # an ordered sequence of buffer, src->dst, (buf_in, buf_out)
-    self.delays = None  # delays from src->dst
+    self.delay = None  # delays from src->dst
 
 class TimingGraph:
   def __init__(self, inst2libcell_map):

build_model/load_graph.py

@@ -14,7 +14,7 @@ from model import PredModel
 
 pdk = "asap7"
 tag = "no_timing_opt"
-dir_prefix = f"/cyberpi/OpenROAD-flow-scripts/flow/parse/{pdk}/{tag}"
+dir_prefix = f"../data/{pdk}/{tag}"
     
 # blocks = "aes aes-mbff ethmac gcd ibex jpeg mock-alu uart".split()
 blocks = "aes aes-mbff gcd ibex jpeg uart".split()
@@ -33,20 +33,10 @@ groundtruth = True
 def gen_topo(g_hetero):
     torch.cuda.synchronize()
     time_s = time.time()
-    # na, nb = g_hetero.edges(etype='fanout2fanin', form='uv')
-    # ca, cb = g_hetero.edges(etype='in2out', form='uv')
-    # ffa, ffb = g_hetero.edges(etype='clk2q', form='uv')
-    # eda, edb = g_hetero.edges(etype='fanout2d', form='uv')
-    # g = dgl.graph((torch.cat([na, ca, ffa, eda]).cpu(), torch.cat([nb, cb, ffb, edb]).cpu()))
 
-    # na, nb = g_hetero.edges(etype='net_out', form='uv')
-    # ca, cb = g_hetero.edges(etype='cell_out', form='uv')
-    # g = dgl.graph((torch.cat([na, ca]).cpu(), torch.cat([nb, cb]).cpu()))
-
-    na, nb = g_hetero.edges(etype='fanout2fanin', form='uv')
-    ca, cb = g_hetero.edges(etype='in2out', form='uv')
-    pa, pb = g_hetero.edges(etype='_2outport', form='uv')
-    g = dgl.graph((torch.cat([na, ca, pa]).cpu(), torch.cat([nb, cb, pb]).cpu()))
+    na, nb = g_hetero.edges(etype='net_out', form='uv')
+    ca, cb = g_hetero.edges(etype='cell_out', form='uv')
+    g = dgl.graph((torch.cat([na, ca]).cpu(), torch.cat([nb, cb]).cpu()))
 
     topo = dgl.topological_nodes_generator(g)   # this seems like topo-sort
     ret = [t.cuda() for t in topo]
@@ -57,14 +47,17 @@ def gen_topo(g_hetero):
 def load_data():
     data = {}
     for block in blocks:
-        graph_path = os.path.join(dir_prefix, block, "parsed", f"{block}.2.graph.bin")
+        graph_path = os.path.join(dir_prefix, block, "parsed", f"{block}.graph.bin")
         g = dgl.load_graphs(graph_path)[0][0].to('cuda')
         topo, topo_time = gen_topo(g)
         ts = {
-            # 'fanin_nodes': g.edges(etype='fanout2fanin')[1].long(),
-            # 'fanout_nodes': g.edges(etype='fanout2fanin')[0].long(),
             'input_nodes': (g.ndata['nf'][:, 1] < 0.5).nonzero().flatten().type(torch.int64),
             'output_nodes': (g.ndata['nf'][:, 1] > 0.5).nonzero().flatten().type(torch.int64),
+            'output_nodes_nonpi': torch.logical_and(g.ndata['nf'][:, 1] > 0.5, g.ndata['nf'][:, 0] < 0.5).nonzero().flatten().type(torch.int64),
+            # pi/po: primary input/output
+            'pi_nodes': torch.logical_and(g.ndata['nf'][:, 1] > 0.5, g.ndata['nf'][:, 0] > 0.5).nonzero().flatten().type(torch.int64),
+            'po_nodes': torch.logical_and(g.ndata['nf'][:, 1] < 0.5, g.ndata['nf'][:, 0] > 0.5).nonzero().flatten().type(torch.int64),
+            'endpoints': (g.ndata['n_is_timing_endpt'] > 0.5).nonzero().flatten().type(torch.long),
             'topo': topo,
         }
         data[block] = g, ts

build_model/model.py

@@ -54,14 +54,6 @@ class NetConv(torch.nn.Module):
         with g.local_scope():
             g.ndata['nf'] = nf
 
-            # # fanin nodes
-            # g.update_all(self.edge_msg_i, fn.sum('efi', 'new_nf'), etype='fanout2fanin')
-            # # fanout nodes
-            # g.apply_edges(self.edge_msg_o, etype='fanin2fanout')
-            # g.update_all(fn.copy_e('efo1', 'efo1'), fn.sum('efo1', 'nfo1'), etype='fanin2fanout')
-            # g.update_all(fn.copy_e('efo2', 'efo2'), fn.max('efo2', 'nfo2'), etype='fanin2fanout')
-            # g.apply_nodes(self.node_reduce_o, ts['fanout_nodes'])
-            
             # input nodes
             g.update_all(self.edge_msg_i, fn.sum('efi', 'new_nf'), etype='net_out')
             # output nodes
@@ -155,16 +147,9 @@ class SignalProp(torch.nn.Module):
             # g.apply_nodes(self.node_skip_level_o, ts['pi_nodes'])
 
             def prop_net(nodes, groundtruth):
-                g.pull(nodes, functools.partial(self.edge_msg_net, groundtruth=groundtruth), fn.sum('efn', 'new_nf'), etype='fanout2fanin')
+                g.pull(nodes, functools.partial(self.edge_msg_net, groundtruth=groundtruth), fn.sum('efn', 'new_nf'), etype='net_out')
 
             def prop_cell(nodes, groundtruth):
-                # cell_edges = ['in2out', 'clk2q', 'fanout2d']
-                # for etype in cell_edges:
-                #     es = g.in_edges(nodes, etype=etype)
-                #     g.apply_edges(functools.partial(self.edge_msg_cell, groundtruth=groundtruth), es, etype=etype)
-                #     g.send_and_recv(es, fn.copy_e('efc1', 'efc1'), fn.sum('efc1', 'nfc1'), etype=etype)
-                #     g.send_and_recv(es, fn.copy_e('efc2', 'efc2'), fn.max('efc2', 'nfc2'), etype=etype)
-
                 etype = 'cell_out'
                 es = g.in_edges(nodes, etype=etype)
                 g.apply_edges(functools.partial(self.edge_msg_cell, groundtruth=groundtruth), es, etype=etype)
@@ -204,4 +189,3 @@ class PredModel(torch.nn.Module):
         nf1 = torch.cat([nf0, x], dim=1)
         nf2, cell_delays = self.prop(g, ts, nf1)
         return net_delays, cell_delays, nf2
-