[New Feature] Support Paddle PIR Visualization 🚀

.pre-commit-config.yaml

@@ -11,7 +11,7 @@ repos:
       - id: check-yaml
       - id: check-symlinks
       - id: destroyed-symlinks
-  - repo: https://gitlab.com/pycqa/flake8
+  - repo: https://github.com/pycqa/flake8
     rev: 3.8.4
     hooks:
       - id: flake8

demo/components/pir_translate.py

@@ -0,0 +1,28 @@
+import paddle
+from paddle import ir
+
+from visualdl import LogWriter
+
+paddle.enable_static()
+
+main_program, start_program = (
+    paddle.static.Program(),
+    paddle.static.Program(),
+)
+with paddle.static.program_guard(main_program, start_program):
+    x = paddle.static.data("x", [1, 64, 64, 8], dtype="float32")
+    y = paddle.static.data("y", [1, 64, 64, 8], dtype="float32")
+    divide_out = paddle.divide(x, y)
+    tanh_out = paddle.tanh(divide_out)
+    conv2d = paddle.nn.Conv2D(8, 32, 1, bias_attr=False, data_format='NHWC')
+    batch_norm = paddle.nn.BatchNorm(32, act='relu', data_layout='NHWC')
+    out = batch_norm(conv2d(tanh_out))
+
+newir_program = ir.translate_to_new_ir(main_program.desc)
+
+with LogWriter(logdir="./log/program_test/") as writer:
+    writer.add_graph(
+        model=newir_program,
+        input_spec=[paddle.static.InputSpec([-1, 1, 28, 28], 'float32')],
+        verbose=True,
+        is_pir=True)

visualdl/component/graph/exporter.py

@@ -17,19 +17,24 @@
 import tempfile
 
 from .graph_component import analyse_model
+from .graph_component import analyse_pir
 from .utils import create_opname_scope
 from .utils import print_model
 
 
-def translate_graph(model, input_spec, verbose=True):
-    import paddle
+def translate_graph(model, input_spec, verbose=True, **kwargs):
+    is_pir = kwargs.get('is_pir', False)
     with tempfile.TemporaryDirectory() as tmp:
-        model._full_name = '{}[{}]'.format(model.__class__.__name__, "model")
-        create_opname_scope(model)
-        model = paddle.jit.to_static(model, input_spec)
-        paddle.jit.save(model, os.path.join(tmp, 'temp'))
-        model_data = open(os.path.join(tmp, 'temp.pdmodel'), 'rb').read()
-        result = analyse_model(model_data)
+        if (not is_pir):
+            model._full_name = '{}[{}]'.format(model.__class__.__name__,
+                                               "model")
+            create_opname_scope(model)
+            model = paddle.jit.to_static(model, input_spec)
+            paddle.jit.save(model, os.path.join(tmp, 'temp'))
+            model_data = open(os.path.join(tmp, 'temp.pdmodel'), 'rb').read()
+            result = analyse_model(model_data)
+        else:
+            result = analyse_pir(model)
     if verbose:
         print_model(result)
     result = json.dumps(result, indent=2)

visualdl/component/graph/graph_component.py

@@ -17,6 +17,8 @@
 import pathlib
 import re
 
+from . import utils
+
 _graph_version = '1.0.0'
 
 
@@ -73,9 +75,8 @@ def create_non_leaf_nodes(parent_node_name, child_node_name, all_ops,
     if parent_node_name == '/':  # root node
         return
     else:
-        create_non_leaf_nodes(
-            os.path.dirname(parent_node_name), parent_node_name, all_ops,
-            general_children_dict)
+        create_non_leaf_nodes(os.path.dirname(parent_node_name),
+                              parent_node_name, all_ops, general_children_dict)
 
 
 def construct_edges(var_name, all_ops, all_vars, all_edges):
@@ -298,8 +299,8 @@ def analyse_model(model_pb):  # noqa: C901
 
     all_op_names = list(all_ops.keys())
     for op_name in all_op_names:
-        create_non_leaf_nodes(
-            os.path.dirname(op_name), op_name, all_ops, general_children_dict)
+        create_non_leaf_nodes(os.path.dirname(op_name), op_name, all_ops,
+                              general_children_dict)
 
     # fill all non-leaf node's  'output_nodes' 'input_nodes' 'output_vars' 'input_vars'
     # post-order traverse tree
@@ -345,8 +346,9 @@ def analyse_model(model_pb):  # noqa: C901
     for src_node, to_node in all_edges.keys():
         all_ops[src_node]['edge_output_nodes'].append(to_node)
         all_ops[to_node]['edge_input_nodes'].append(src_node)
-        all_edges[(src_node, to_node)]['vars'] = list(
-            all_edges[(src_node, to_node)]['vars'])
+        all_edges[(src_node,
+                   to_node)]['vars'] = list(all_edges[(src_node,
+                                                       to_node)]['vars'])
         if len(all_edges[(src_node, to_node)]['vars']) > 1:
             all_edges[(src_node, to_node)]['label'] = str(
                 len(all_edges[(src_node, to_node)]['vars'])) + ' tensors'
@@ -361,3 +363,96 @@ def analyse_model(model_pb):  # noqa: C901
         'edges': list(all_edges.values())
     }
     return final_data
+
+
+def analyse_pir(program):
+    from paddle.utils.unique_name import generate
+
+    all_ops = {}
+    all_vars = {}
+    all_edges = {}
+    # vars info
+    for op in (program.global_block().ops):
+        var_name = utils.gen_var_name(op.results())
+        all_vars[var_name] = {}
+        all_vars[var_name]['name'] = var_name
+        attrs = op.results()[0].get_defining_op().attrs()
+
+        if 'place' in attrs:
+            attrs['place'] = str(attrs['place'])
+        attrs['dtype'] = op.result(0).dtype.name
+
+        all_vars[var_name]['shape'] = op.result(0).shape
+        all_vars[var_name]['type'] = op.result(0).dtype.name
+        all_vars[var_name]['dtype'] = op.result(0).dtype.name
+
+        all_vars[var_name]['value'] = []
+        all_vars[var_name]['persistable'] = op.result(0).is_persistable
+        all_vars[var_name]['attrs'] = attrs
+        all_vars[var_name]['from_node'] = ''
+        all_vars[var_name]['to_nodes'] = []
+
+    # ops info
+    for op in (program.global_block().ops):
+        op_name = generate(op.name())
+
+        if op.num_operands() > 0:
+            all_ops[op_name] = {}
+            all_ops[op_name]['name'] = op_name
+            all_ops[op_name]['show_name'] = op_name
+            all_ops[op_name]['type'] = op.result(0).dtype.name
+            all_ops[op_name]['dtype'] = op.result(0).dtype.name
+
+            all_ops[op_name]['input_vars'] = {}
+            all_ops[op_name]['output_vars'] = {}
+
+            all_ops[op_name]['is_leaf_node'] = True
+            now_var = utils.gen_var_name(op.results())
+            for source in op.operands_source():
+                input_name = utils.gen_var_name(source)
+                all_ops[op_name]['input_vars'][input_name] = [input_name]
+                all_vars[input_name]['to_nodes'].append(op_name)
+            all_vars[now_var]['from_node'] = op_name
+            all_ops[op_name]['output_vars'][now_var] = [now_var]
+
+            all_ops[op_name]['attrs'] = attrs
+            all_ops[op_name]['attr_types'] = attrs
+            all_ops[op_name]['children_node'] = []
+            all_ops[op_name]['input_nodes'] = []
+            all_ops[op_name]['output_nodes'] = []
+            all_ops[op_name]['edge_input_nodes'] = []
+            all_ops[op_name]['edge_output_nodes'] = []
+
+    # create '/' op
+    all_ops['/'] = {}
+    all_ops['/']['name'] = '/'
+    all_ops['/']['show_name'] = '/'
+    all_ops['/']['type'] = ''
+    all_ops['/']['attrs'] = {}
+    all_ops['/']['input_vars'] = {}
+    all_ops['/']['output_vars'] = {}
+    all_ops['/']['is_leaf_node'] = False
+    all_ops['/']['children_node'] = []
+    for node in all_ops:
+        if node != '/':
+            all_ops['/']['children_node'].append(node)
+
+    for variable_name in all_vars:
+        if all_vars[variable_name]['from_node'] == '':
+            continue
+        from_node_name = all_vars[variable_name]['from_node']
+        for to_node_name in all_vars[variable_name]['to_nodes']:
+            if to_node_name != from_node_name:
+                all_ops[from_node_name]['output_nodes'].append(to_node_name)
+                all_ops[to_node_name]['input_nodes'].append(from_node_name)
+
+    # edge info
+    # TODO(Difers):add edge info in future
+
+    final_data = {
+        'version': _graph_version,
+        'nodes': list(all_ops.values()),
+        'vars': list(all_vars.values()),
+        'edges': list(all_edges.values())
+    }
+    return final_data

visualdl/component/graph/utils.py

@@ -17,6 +17,27 @@
 
 _name_scope_stack = deque()
 
+# TODO(Difers): remove it when the new IR's "name" interface is available.
+var_name = {}
+var_idx = [0]
+
+
+def gen_var_name(ops):
+    if not isinstance(ops, list):
+        ops = [ops]
+    for op in ops:
+        var = op.get_defining_op()
+        if var in var_name:
+            return var_name[var]
+        else:
+            try:
+                name = op.name
+            except ValueError:
+                name = "tmp_var_" + str(var_idx[0])
+                var_idx[0] += 1
+            var_name[var] = name
+            return var_name[var]
+
 
 def _opname_creation_prehook(layer, inputs):
     from paddle.static import name_scope