Skip to content

X
X2Paddle

PaddlePaddle / X2Paddle
大约 1 年 前同步成功

通知 328
Star 698
Fork 167
X
X2Paddle
未验证 提交 8fee1a69 编写于 作者: J Jason 提交者： GitHub
浏览文件
操作

Merge pull request #160 from mamingjie-China/develop 

modify the codes in TensorFlow converting
上级 2bcb8fd3 bac4164d
显示空白变更内容
内联 并排
Showing with 233 addition and 102 deletion
x2paddle/op_mapper/tf_op_mapper.py
浏览文件 @ 8fee1a69
...@@ -168,7 +168,11 @@ class TFOpMapper(OpMapper): ...@@ -168,7 +168,11 @@ class TFOpMapper(OpMapper):
x_input = y x_input = y
y_input = x y_input = x
x_shape = y.out_shapes[0] x_shape = y.out_shapes[0]
if len(x_shape) == 0:
x_shape = [1]
y_shape = x.out_shapes[0] y_shape = x.out_shapes[0]
if len(y_shape) == 0:
y_shape = [1]
else: else:
if len(x_shape) == 1 and len(y_shape) == 4 and x_shape[ if len(x_shape) == 1 and len(y_shape) == 4 and x_shape[
0] == y_shape[-1] and y_shape.count(-1) < 1: 0] == y_shape[-1] and y_shape.count(-1) < 1:
......
x2paddle/op_mapper/tf_op_mapper_nhwc.py
浏览文件 @ 8fee1a69
...@@ -121,6 +121,25 @@ class TFOpMapperNHWC(OpMapper): ...@@ -121,6 +121,25 @@ class TFOpMapperNHWC(OpMapper):
pd_param_name = list(param.values())[0] pd_param_name = list(param.values())[0]
tf_param = node.get_attr(tf_param_name) tf_param = node.get_attr(tf_param_name)
attr[pd_param_name] = tf_param attr[pd_param_name] = tf_param
if len(input.out_shapes[0]) == 4 and op_info[0] != 'shape':
attr1 = {"perm": [0, 3, 1, 2]}
node.fluid_code.add_layer('transpose',
inputs=input,
output=node,
param_attr=attr1)
input = node
node.fluid_code.add_layer(op_info[0],
inputs=input,
output=node,
param_attr=attr)
input = node
attr2 = {"perm": [0, 2, 3, 1]}
node.fluid_code.add_layer('transpose',
inputs=input,
output=node,
param_attr=attr2)
else:
node.fluid_code.add_layer(op_info[0], node.fluid_code.add_layer(op_info[0],
inputs=input, inputs=input,
output=node, output=node,
...@@ -149,7 +168,11 @@ class TFOpMapperNHWC(OpMapper): ...@@ -149,7 +168,11 @@ class TFOpMapperNHWC(OpMapper):
x_input = y x_input = y
y_input = x y_input = x
x_shape = y.out_shapes[0] x_shape = y.out_shapes[0]
if len(x_shape) == 0:
x_shape = [1]
y_shape = x.out_shapes[0] y_shape = x.out_shapes[0]
if len(y_shape) == 0:
y_shape = [1]
else: else:
raise Exception("Unexpected situation happend") raise Exception("Unexpected situation happend")
...@@ -193,6 +216,25 @@ class TFOpMapperNHWC(OpMapper): ...@@ -193,6 +216,25 @@ class TFOpMapperNHWC(OpMapper):
output="y_tmp", output="y_tmp",
param_attr=attr) param_attr=attr)
y_input = "y_tmp" y_input = "y_tmp"
if len(x_shape) == 4 and len(y_shape) == 4:
node.fluid_code.add_layer("transpose",
inputs=x_input,
output=x_input,
param_attr={'perm': [0, 3, 1, 2]})
node.fluid_code.add_layer("transpose",
inputs=y_input,
output=y_input,
param_attr={'perm': [0, 3, 1, 2]})
inputs = {"x": x_input, "y": y_input}
node.fluid_code.add_layer(op_type,
inputs=inputs,
output=node,
param_attr=None)
node.fluid_code.add_layer("transpose",
inputs=node,
output=node,
param_attr={'perm': [0, 2, 3, 1]})
else:
inputs = {"x": x_input, "y": y_input} inputs = {"x": x_input, "y": y_input}
node.fluid_code.add_layer(op_type, node.fluid_code.add_layer(op_type,
inputs=inputs, inputs=inputs,
...@@ -978,9 +1020,7 @@ class TFOpMapperNHWC(OpMapper): ...@@ -978,9 +1020,7 @@ class TFOpMapperNHWC(OpMapper):
if pad_mode == "SAME": if pad_mode == "SAME":
if node.tf_data_format == "NHWC": if node.tf_data_format == "NHWC":
print(out_shape)
out_shape = [out_shape[i] for i in [0, 3, 1, 2]] out_shape = [out_shape[i] for i in [0, 3, 1, 2]]
print(out_shape)
for i in range(4): for i in range(4):
if out_shape[i] < 0: if out_shape[i] < 0:
out_shape[i] = 999999 out_shape[i] = 999999
......
x2paddle/optimizer/tf_optimizer.py
浏览文件 @ 8fee1a69
...@@ -232,84 +232,35 @@ class TFOptimizer(object): ...@@ -232,84 +232,35 @@ class TFOptimizer(object):
'act'] 'act']
node.fluid_code.clear() node.fluid_code.clear()
self.graph.remove_node(node.layer_name) self.graph.remove_node(node.layer_name)
self.graph.identity_map[node.layer_name] = input.layer_name
def remove_transpose(self): def remove_transpose(self):
graph_copy = cp.deepcopy(self.graph) graph_copy = cp.deepcopy(self.graph)
nhwc_insensitive_ops = [ nhwc_insensitive_ops = [
'Relu', 'Relu6', 'Abs', 'Sigmoid', 'Exp', 'Rsqrt', 'swish_f32', 'Relu', 'Relu6', 'Abs', 'Sigmoid', 'Exp', 'Rsqrt', 'swish_f32',
'LeakyRelu', 'Cast' 'LeakyRelu', 'Cast', 'Tanh'
] ]
elementwise_ops = [ elementwise_ops = [
'Sub', 'Add', 'RealDiv', 'Maximum', 'Mul', 'FloorDiv', 'Sub', 'Add', 'RealDiv', 'Maximum', 'Mul', 'FloorDiv',
'GreaterEqual' 'GreaterEqual'
] ]
for node_name in self.graph.topo_sort:
node = graph_copy.get_node(node_name)
if node is None:
continue
if node.layer_type in nhwc_insensitive_ops:
graph_copy.remove_node(node_name)
optimize_ops = [ optimize_ops = [
'Conv2D', 'MaxPool', 'FusedBatchNorm', 'DepthwiseConv2dNative', 'Conv2D', 'MaxPool', 'FusedBatchNorm', 'DepthwiseConv2dNative',
'AvgPool', 'Pad', 'Conv2DBackpropInput', 'ResizeNearestNeighbor', 'AvgPool', 'Pad', 'Conv2DBackpropInput', 'ResizeNearestNeighbor',
'ResizeBilinear', "Placeholder" 'ResizeBilinear', "Placeholder"
] ]
can_be_optimized_ops = [
'Conv2D', 'MaxPool', 'FusedBatchNorm', 'DepthwiseConv2dNative',
'AvgPool', 'Pad', 'Conv2DBackpropInput', 'ResizeNearestNeighbor',
'ResizeBilinear', "Placeholder", 'Relu', 'Relu6', 'Abs', 'Sigmoid',
'Exp', 'Rsqrt', 'swish_f32', 'LeakyRelu', 'Cast', 'Tanh'
]
for node_name in self.graph.topo_sort: for node_name in self.graph.topo_sort:
node = graph_copy.get_node(node_name) node = graph_copy.get_node(node_name)
if node is None: if node is None:
continue continue
if node.layer_type in elementwise_ops: if node.layer_type in can_be_optimized_ops:
is_nhwc = True
for in_name in node.inputs:
in_node = graph_copy.get_node(in_name)
if hasattr(in_node, "is_nhwc"):
if not in_node.is_nhwc:
is_nhwc = False
else:
if len(in_node.fluid_code.layers) < 2:
is_nhwc = False
continue
if in_node.fluid_code.layers[
-1].op != "transpose" or in_node.fluid_code.layers[
-1].param_attr["perm"] != [0, 2, 3, 1]:
is_nhwc = False
continue
node.is_nhwc = is_nhwc
for i in range(len(self.graph.topo_sort)):
node_name = self.graph.topo_sort[-1 * i - 1]
node = graph_copy.get_node(node_name)
if node is None:
continue
if node.layer_type in elementwise_ops:
can_be_removed = True
if len(node.fluid_code.layers) > 1:
can_be_removed = False
if not node.is_nhwc:
can_be_removed = False
for out_name in node.outputs:
out_node = graph_copy.get_node(out_name)
if hasattr(out_node, "is_nhwc"):
if not out_node.is_nhwc:
can_be_removed = False
else:
if len(out_node.fluid_code.layers) < 2:
can_be_removed = False
break
if out_node.fluid_code.layers[
0].op != "transpose" or out_node.fluid_code.layers[
0].param_attr["perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
node.can_be_removed = can_be_removed
for node_name in self.graph.topo_sort:
node = graph_copy.get_node(node_name)
if node is None:
continue
if node.layer_type in optimize_ops:
if node.fluid_code.layers[ if node.fluid_code.layers[
-1].op != "transpose" or node.fluid_code.layers[ -1].op != "transpose" or node.fluid_code.layers[
-1].param_attr["perm"] != [0, 2, 3, 1]: -1].param_attr["perm"] != [0, 2, 3, 1]:
...@@ -327,6 +278,9 @@ class TFOptimizer(object): ...@@ -327,6 +278,9 @@ class TFOptimizer(object):
0].param_attr["perm"] != [0, 3, 1, 2]: 0].param_attr["perm"] != [0, 3, 1, 2]:
can_be_removed = False can_be_removed = False
break break
elif out_node.layer_type in elementwise_ops:
can_be_removed = False
break
if can_be_removed and len(node.fluid_code.layers) > 1: if can_be_removed and len(node.fluid_code.layers) > 1:
true_node = self.graph.get_node(node_name) true_node = self.graph.get_node(node_name)
if true_node.layer_type == "Placeholder": if true_node.layer_type == "Placeholder":
...@@ -346,8 +300,6 @@ class TFOptimizer(object): ...@@ -346,8 +300,6 @@ class TFOptimizer(object):
del true_node.fluid_code.layers[-1] del true_node.fluid_code.layers[-1]
for out_name in output_names: for out_name in output_names:
out_node = self.graph.get_node(out_name) out_node = self.graph.get_node(out_name)
if out_node.layer_type in elementwise_ops:
continue
out_node.fluid_code.layers[ out_node.fluid_code.layers[
1].inputs = out_node.fluid_code.layers[0].inputs 1].inputs = out_node.fluid_code.layers[0].inputs
del out_node.fluid_code.layers[0] del out_node.fluid_code.layers[0]
...@@ -357,43 +309,178 @@ class TFOptimizer(object): ...@@ -357,43 +309,178 @@ class TFOptimizer(object):
if node is None: if node is None:
continue continue
if node.layer_type in elementwise_ops: if node.layer_type in elementwise_ops:
if not node.can_be_removed: can_be_removed = True
if node.fluid_code.layers[
-1].op != "transpose" or node.fluid_code.layers[
-1].param_attr["perm"] != [0, 2, 3, 1]:
continue
can_be_removed = True
output_names = node.outputs
for out_name in output_names:
out_node = graph_copy.get_node(out_name)
if len(out_node.fluid_code.layers) < 3:
can_be_removed = False
break
if hasattr(out_node, "can_be_removed"):
if not out_node.can_be_removed:
can_be_removed = False
break
if out_node.layer_type in can_be_optimized_ops:
if out_node.fluid_code.layers[
0].op != "transpose" or out_node.fluid_code.layers[
0].param_attr["perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
elif out_node.layer_type in elementwise_ops:
if out_node.fluid_code.layers[
0].op != "transpose" and out_node.fluid_code.layers[
1].op != "transpose":
can_be_removed = False
break
if out_node.fluid_code.layers[0].op == "transpose":
if out_node.fluid_code.layers[0].param_attr[
"perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
if out_node.fluid_code.layers[1].op == "transpose":
if out_node.fluid_code.layers[1].param_attr[
"perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
if can_be_removed and len(node.fluid_code.layers) > 1:
true_node = self.graph.get_node(node_name) true_node = self.graph.get_node(node_name)
for i, in_name in enumerate(node.inputs): true_node.fluid_code.layers[
in_node = graph_copy.get_node(in_name) -2].output = true_node.fluid_code.layers[-1].output
if hasattr(in_node, "is_nhwc") and in_node.is_nhwc: del true_node.fluid_code.layers[-1]
if i == 0: for out_name in output_names:
l = Layer() out_node = self.graph.get_node(out_name)
l.op = "transpose" if out_node.layer_type in can_be_optimized_ops:
l.inputs = true_node.fluid_code.layers[ out_node.fluid_code.layers[
0].inputs["x"] 1].inputs = out_node.fluid_code.layers[0].inputs
l.param_attr = {"perm": [0, 2, 3, 1]} del out_node.fluid_code.layers[0]
l.output = "nhwc_" + l.inputs.layer_name elif out_node.layer_type in elementwise_ops:
true_node.fluid_code.layers[0].inputs[ if out_node.inputs[0] in node.layer_name:
"x"] = l.output if out_node.fluid_code.layers[
true_node.fluid_code.layers.insert(0, l) 1].op == 'transpose':
elif i == 1: out_node.fluid_code.layers[2].inputs[
l = Layer() 'x'] = out_node.fluid_code.layers[
l.op = "transpose" 0].inputs
l.inputs = true_node.fluid_code.layers[ del out_node.fluid_code.layers[0]
0].inputs["y"]
l.param_attr = {"perm": [0, 2, 3, 1]}
l.output = "nhwc_" + l.inputs.layer_name
true_node.fluid_code.layers[0].inputs[
"y"] = l.output
true_node.fluid_code.layers.insert(0, l)
else: else:
raise Exception("Unexpected situation happend") out_node.fluid_code.layers[1].inputs[
continue 'x'] = out_node.fluid_code.layers[
0].inputs
del out_node.fluid_code.layers[0]
elif out_node.inputs[1] in node.layer_name:
if out_node.fluid_code.layers[
1].op == 'transpose':
out_node.fluid_code.layers[2].inputs[
'y'] = out_node.fluid_code.layers[
1].inputs
del out_node.fluid_code.layers[1]
else: else:
for out_name in node.outputs: out_node.fluid_code.layers[1].inputs[
'y'] = out_node.fluid_code.layers[
0].inputs
del out_node.fluid_code.layers[0]
graph_copy = cp.deepcopy(self.graph)
for node_name in self.graph.topo_sort:
node = graph_copy.get_node(node_name)
if node is None or len(node.fluid_code.layers) < 2:
continue
if node.layer_type in can_be_optimized_ops and node.layer_type != "Placeholder":
if node.fluid_code.layers[
-1].op != "transpose" or node.fluid_code.layers[
-1].param_attr["perm"] != [0, 2, 3, 1]:
continue
can_be_removed = True
output_names = node.outputs
for out_name in output_names:
out_node = graph_copy.get_node(out_name)
if hasattr(out_node, "can_be_removed"):
if not out_node.can_be_removed:
can_be_removed = False
break
if len(out_node.fluid_code.layers) < 2:
can_be_removed = False
break
if out_node.layer_type in can_be_optimized_ops:
if out_node.fluid_code.layers[
0].op != "transpose" or out_node.fluid_code.layers[
0].param_attr["perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
elif out_node.layer_type in elementwise_ops:
if out_node.fluid_code.layers[
0].op != "transpose" and out_node.fluid_code.layers[
1].op != "transpose":
can_be_removed = False
break
if out_node.fluid_code.layers[
0].op == "expand" or out_node.fluid_code.layers[
1].op == "expand":
can_be_removed = False
break
if out_node.fluid_code.layers[0].op == "transpose":
if out_node.fluid_code.layers[0].param_attr[
"perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
if out_node.fluid_code.layers[1].op == "transpose":
if out_node.fluid_code.layers[1].param_attr[
"perm"] != [0, 3, 1, 2]:
can_be_removed = False
break
elif out_node.layer_type not in elementwise_ops and out_node.layer_type not in can_be_optimized_ops:
can_be_removed = False
break
if can_be_removed:
true_node = self.graph.get_node(node_name)
if len(true_node.fluid_code.layers) < 2:
continue
true_node.fluid_code.layers[
-2].output = true_node.fluid_code.layers[-1].output
del true_node.fluid_code.layers[-1]
for out_name in output_names:
out_node = self.graph.get_node(out_name) out_node = self.graph.get_node(out_name)
if out_node.layer_type not in elementwise_ops: if out_node.layer_type in can_be_optimized_ops:
assert out_node.fluid_code.layers[
0].op == "transpose", "unexpected situation happend"
out_node.fluid_code.layers[ out_node.fluid_code.layers[
1].inputs = out_node.fluid_code.layers[0].inputs 1].inputs = out_node.fluid_code.layers[0].inputs
del out_node.fluid_code.layers[0] del out_node.fluid_code.layers[0]
elif out_node.layer_type in elementwise_ops:
if out_node.inputs[0] in node.layer_name:
if out_node.fluid_code.layers[
1].op == 'transpose':
if out_node.fluid_code.layers[
2].op == 'transpose':
out_node.fluid_code.layers[3].inputs[
'x'] = out_node.fluid_code.layers[
0].inputs
else:
out_node.fluid_code.layers[2].inputs[
'x'] = out_node.fluid_code.layers[
0].inputs
del out_node.fluid_code.layers[0]
else:
out_node.fluid_code.layers[1].inputs[
'x'] = out_node.fluid_code.layers[
0].inputs
del out_node.fluid_code.layers[0]
elif out_node.inputs[1] in node.layer_name:
if out_node.fluid_code.layers[
1].op == 'transpose':
out_node.fluid_code.layers[2].inputs[
'y'] = out_node.fluid_code.layers[
1].inputs
del out_node.fluid_code.layers[1]
else:
out_node.fluid_code.layers[1].inputs[
'y'] = out_node.fluid_code.layers[
0].inputs
del out_node.fluid_code.layers[0]
def make_nchw_input_output(self): def make_nchw_input_output(self):
for i, name in enumerate(self.graph.input_nodes): for i, name in enumerate(self.graph.input_nodes):
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或