Conv transpose eltwiseadd bn fuse pass (#37800)

* conv_transpose_eltwiseadd_bn_fuse_pass

* change timeout

* add TIMEOUT

* add random num for group and dilation

* change PassCompat

paddle/fluid/framework/ir/conv_bn_fuse_pass.cc

@@ -608,7 +608,7 @@ ConvTransposeBNFusePass::ConvTransposeBNFusePass() {
       .IsOptional()
       .End()
       .AddAttr("groups")
-      .IsNumGE(1)
+      .IsNumEQ(1)
       .End()
       .AddAttr("dilations")
       .IsType<std::vector<int>>()
@@ -624,7 +624,7 @@ ConvTransposeBNFusePass::ConvTransposeBNFusePass() {
       .IsStringIn({"EXPLICIT", "SAME", "VALID"})
       .End()
       .AddAttr("data_format")
-      .IsStringIn({"NCHW", "NHWC", "AnyLayout"})
+      .IsStringIn({"NCHW", "AnyLayout"})
       .End();
 }
 
@@ -652,7 +652,7 @@ ConvTransposeEltwiseAddBNFusePass::ConvTransposeEltwiseAddBNFusePass() {
       .IsOptional()
       .End()
       .AddAttr("groups")
-      .IsNumGE(1)
+      .IsNumEQ(1)
       .End()
       .AddAttr("dilations")
       .IsType<std::vector<int>>()
@@ -668,7 +668,7 @@ ConvTransposeEltwiseAddBNFusePass::ConvTransposeEltwiseAddBNFusePass() {
       .IsStringIn({"EXPLICIT", "SAME", "VALID"})
       .End()
       .AddAttr("data_format")
-      .IsStringIn({"NCHW", "NHWC", "AnyLayout"})
+      .IsStringIn({"NCHW", "AnyLayout"})
       .End();
 }
 
@@ -738,3 +738,9 @@ REGISTER_PASS_CAPABILITY(conv_eltwiseadd_bn_fuse_pass)
             .LE("conv2d", 1)
             .LE("elementwise_add", 1)
             .EQ("batch_norm", 0));
+REGISTER_PASS_CAPABILITY(conv_transpose_eltwiseadd_bn_fuse_pass)
+    .AddCombination(
+        paddle::framework::compatible::OpVersionComparatorCombination()
+            .LE("conv2d_transpose", 2)
+            .LE("elementwise_add", 1)
+            .EQ("batch_norm", 0));

python/paddle/fluid/tests/unittests/ir/inference/CMakeLists.txt

@@ -88,5 +88,6 @@ endif()
 if (WITH_MKLDNN)
   set_tests_properties(test_mkldnn_depthwise_conv_pass PROPERTIES TIMEOUT 120)
   set_tests_properties(test_mkldnn_prelu_op PROPERTIES TIMEOUT 300)
+  set_tests_properties(test_conv_transpose_eltwiseadd_bn_fuse_pass PROPERTIES TIMEOUT 250)
 endif()
 endif()

python/paddle/fluid/tests/unittests/ir/inference/test_conv_transpose_eltwiseadd_bn_fuse_pass.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# 
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+# 
+#     http://www.apache.org/licenses/LICENSE-2.0
+# 
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from auto_scan_test import PassAutoScanTest, IgnoreReasons
+from program_config import TensorConfig, ProgramConfig, OpConfig
+import numpy as np
+import copy as cp
+import paddle.inference as paddle_infer
+from functools import partial
+from typing import Optional, List, Callable, Dict, Any, Set
+import unittest
+
+import hypothesis
+from hypothesis import given, settings, seed, example, assume, reproduce_failure
+import hypothesis.strategies as st
+
+
+class TestConvTransposeEltwiseaddBnFusePass(PassAutoScanTest):
+    '''
+    conv_input   conv_weight_var(persistable)
+      \       /
+         conv_op     
+          |
+      conv_out_var  elementwise_add_y 
+          |       /
+    elementwise_add
+          |
+    elementwise_add_out (bn_scale_var, bn_bias_var, bn_mean_var,bn_variance_var)
+                |            /
+                batch_norm_op
+                |            \
+                bn_out_var     (bn_mean_out_var, bn_variance_out_var,bn_saved_mean_var, bn_saved_variance_var)
+    '''
+
+    def test(self):
+        self.run_and_statis(
+            quant=False,
+            max_examples=150,
+            max_duration=250,
+            passes=["conv_transpose_eltwiseadd_bn_fuse_pass"])
+
+    def sample_program_config(self, draw):
+        # generate random number
+        random_batch_size = draw(st.integers(min_value=1, max_value=3))
+        random_channel = draw(st.integers(min_value=2, max_value=10))
+        random_input_dim1 = draw(st.integers(min_value=20, max_value=50))
+        random_input_dim2 = draw(st.integers(min_value=20, max_value=50))
+        random_groups = draw(st.integers(min_value=1, max_value=2))
+        random_dilations = draw(
+            st.lists(
+                st.integers(
+                    min_value=1, max_value=3), min_size=2, max_size=2))
+        random_strides = draw(
+            st.lists(
+                st.integers(
+                    min_value=1, max_value=4), min_size=2, max_size=2))
+        random_paddings = draw(
+            st.lists(
+                st.integers(
+                    min_value=0, max_value=4), min_size=2, max_size=2))
+        random_padding_algorithm = draw(
+            st.sampled_from(["EXPLICIT", "SAME", "VALID"]))
+        random_data_layout = draw(st.sampled_from(["NCHW", "NHWC"]))
+        random_use_mkldnn = draw(st.booleans())
+        random_output_size = []
+        random_filter = draw(
+            st.lists(
+                st.integers(
+                    min_value=1, max_value=4), min_size=2, max_size=2))
+        random_out_channel = draw(st.integers(min_value=20, max_value=25))
+        random_epsilon = draw(st.floats(min_value=0.0, max_value=0.001))
+
+        def generate_conv2d_Input():
+            shape = [random_input_dim1, random_input_dim2]
+            if random_data_layout == "NCHW":
+                shape.insert(0, random_channel * random_groups)
+                shape.insert(0, random_batch_size)
+            else:
+                shape.append(random_channel)
+                shape.insert(0, random_batch_size)
+            return np.random.random(shape).astype(np.float32)
+
+        def generate_conv2d_Filter():
+            shape = cp.copy(random_filter)
+            shape.insert(0, random_out_channel * random_groups)
+            shape.insert(0, random_channel * random_groups)
+            return np.random.random(shape).astype(np.float32)
+
+        def generate_elementwise_add_Y():
+            return np.random.random(
+                [random_out_channel * random_groups * random_groups]).astype(
+                    np.float32)
+
+        def generate_batch_norm_Scale():
+            return np.random.random(
+                [random_out_channel * random_groups * random_groups]).astype(
+                    np.float32)
+
+        def generate_batch_norm_Bias():
+            return np.random.random(
+                [random_out_channel * random_groups * random_groups]).astype(
+                    np.float32)
+
+        def generate_batch_norm_Mean():
+            return np.random.random(
+                [random_out_channel * random_groups * random_groups]).astype(
+                    np.float32)
+
+        def generate_batch_norm_Variance():
+            return np.random.random(
+                [random_out_channel * random_groups * random_groups]).astype(
+                    np.float32)
+
+        # define op
+        conv2d_op = OpConfig(
+            type="conv2d_transpose",
+            inputs={
+                "Input": ["conv2d_Input"],
+                "Filter": ["conv2d_Filter"],
+            },
+            outputs={"Output": ["conv2d_Out"], },
+            attrs={
+                'groups': random_groups,
+                'dilations': random_dilations,
+                'strides': random_strides,
+                'paddings': random_paddings,
+                'padding_algorithm': random_padding_algorithm,
+                'data_format': random_data_layout,
+                'output_size': random_output_size,
+                'output_padding': random_output_size,
+                'use_mkldnn': random_use_mkldnn,
+                'is_test': True,
+            })
+
+        elementwise_op = OpConfig(
+            type="elementwise_add",
+            inputs={
+                "X": ["conv2d_Out"],
+                "Y": ["elementwise_add_Y"],
+            },
+            outputs={"Out": ["elementwise_add_Out"], },
+            attrs={'axis': 1, })
+
+        batch_norm_op = OpConfig(
+            type="batch_norm",
+            inputs={
+                "X": ["elementwise_add_Out"],
+                "Scale": ["batch_norm_Scale"],
+                "Bias": ["batch_norm_Bias"],
+                "Mean": ["batch_norm_Mean"],
+                "Variance": ["batch_norm_Variance"],
+            },
+            outputs={
+                "Y": ["batch_norm_Y"],
+                "MeanOut": ["batch_norm_Mean"],
+                "VarianceOut": ["batch_norm_Variance"],
+                "SavedMean": ["batch_norm_SavedMean"],
+                "SavedVariance": ["batch_norm_SavedVariance"],
+                "ReserveSpace": ["batch_norm_ReserveSpace"],
+            },
+            attrs={
+                'epsilon': random_epsilon,
+                'is_test': True,
+                'trainable_statistics': False,
+                'data_layout': random_data_layout,
+                'use_mkldnn': random_use_mkldnn,
+            })
+
+        # define model_net
+        model_net = [conv2d_op, elementwise_op, batch_norm_op]
+
+        # set tensor
+        program_config = ProgramConfig(
+            ops=model_net,
+            inputs={
+                "conv2d_Input": TensorConfig(data_gen=generate_conv2d_Input),
+            },
+            weights={
+                "conv2d_Filter": TensorConfig(data_gen=generate_conv2d_Filter),
+                "elementwise_add_Y":
+                TensorConfig(data_gen=generate_elementwise_add_Y),
+                "batch_norm_Scale":
+                TensorConfig(data_gen=generate_batch_norm_Scale),
+                "batch_norm_Bias":
+                TensorConfig(data_gen=generate_batch_norm_Bias),
+                "batch_norm_Mean":
+                TensorConfig(data_gen=generate_batch_norm_Mean),
+                "batch_norm_Variance":
+                TensorConfig(data_gen=generate_batch_norm_Variance),
+            },
+            outputs=["batch_norm_Y"])
+
+        return program_config
+
+    def sample_predictor_configs(self, program_config):
+        # for mkldnn
+        config = self.create_inference_config()
+        if program_config.ops[2].attrs['use_mkldnn']:
+            config.enable_mkldnn()
+            yield config, ['conv2d_transpose', 'elementwise_add'], (1e-5, 1e-5)
+        # cpu
+        else:
+            yield config, ['conv2d_transpose', 'elementwise_add'], (1e-5, 1e-5)
+
+    def is_program_valid(self, program_config: ProgramConfig) -> bool:
+        attrs = [
+            program_config.ops[i].attrs
+            for i in range(len(program_config.ops))
+        ]
+
+        if attrs[0]['data_format'] == "NHWC":
+            return False
+
+        return True
+
+    def add_ignore_pass_case(self):
+        def teller1(program_config, predictor_config):
+            if program_config.ops[0].attrs['data_format'] == "NHWC":
+                return True
+            return False
+
+        def teller2(program_config, predictor_config):
+            if program_config.ops[0].attrs['groups'] != 1:
+                return True
+            return False
+
+        self.add_ignore_check_case(
+            teller1, IgnoreReasons.PASS_ACCURACY_ERROR,
+            "The output format of conv2d_transpose is wrong when data_format attribute is NHWC"
+        )
+        self.add_ignore_check_case(teller2, IgnoreReasons.PASS_ACCURACY_ERROR,
+                                   "there is diff when group >1 in this pass")