CMakeLists.txt

set(pass_file ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/paddle_inference_pass.h.tmp)
set(pass_file_final ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/paddle_inference_pass.h)
file(WRITE ${pass_file} "// Generated by the paddle/fluid/framework/ir/CMakeLists.txt.  DO NOT EDIT!\n\n")
file(APPEND ${pass_file} "\#pragma once\n")
file(APPEND ${pass_file} "\#include \"paddle/fluid/framework/ir/pass.h\"\n")

copy_if_different(${pass_file} ${pass_file_final})

add_subdirectory(fuse_optimizer_ops_pass)
add_subdirectory(memory_optimize_pass)
add_subdirectory(multi_devices_graph_pass)
if(NOT APPLE AND NOT WIN32 AND (WITH_GPU OR WITH_ROCM))
    add_subdirectory(fusion_group)
endif()

# Usage: pass_library(target inference) will append to paddle_inference_pass.h
unset(INFER_IR_PASSES CACHE) # clear the global variable
function(pass_library TARGET DEST)
    set(options "")
    set(oneValueArgs "")
    set(multiValueArgs SRCS DEPS DIR)
    set(targetPrefix "")

    cmake_parse_arguments(pass_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
    if(pass_library_DIR)
        cc_library(${TARGET} SRCS ${pass_library_DIR}/${TARGET}.cc DEPS graph_pattern_detector pass fuse_pass_base op_version_registry ${pass_library_DEPS})
    else()
        cc_library(${TARGET} SRCS ${TARGET}.cc DEPS graph_pattern_detector pass fuse_pass_base op_version_registry ${pass_library_DEPS})
    endif()

    # add more DEST here, such as train, dist and collect USE_PASS into a file automatically.
    if (${DEST} STREQUAL "base" OR ${DEST} STREQUAL "inference")
        if(NOT CMAKE_BUILD_TYPE STREQUAL "Release")
            message(STATUS "add pass ${TARGET} ${DEST}")
        endif()
        file(APPEND ${pass_file} "USE_PASS(${TARGET});\n")
        set(INFER_IR_PASSES ${INFER_IR_PASSES} ${TARGET} CACHE INTERNAL "")
    endif()
endfunction()

cc_library(node SRCS node.cc DEPS proto_desc)
cc_library(graph SRCS graph.cc DEPS node pretty_log)
cc_library(graph_helper SRCS graph_helper.cc DEPS graph)
cc_library(pass SRCS pass.cc DEPS graph node graph_helper)
cc_library(graph_traits SRCS graph_traits.cc DEPS graph)

SET(GRAPH_PATTERN_DETECTOR_DEPS graph graph_helper graph_traits)
if (WITH_TESTING)
    SET(GRAPH_PATTERN_DETECTOR_DEPS ${GRAPH_PATTERN_DETECTOR_DEPS} gtest)
endif(WITH_TESTING)
cc_library(graph_pattern_detector SRCS graph_pattern_detector.cc DEPS ${GRAPH_PATTERN_DETECTOR_DEPS})

cc_library(subgraph_detector SRCS subgraph_detector.cc DEPS graph_pattern_detector executor)
cc_library(fuse_pass_base SRCS fuse_pass_base.cc DEPS pass)
cc_library(placement_pass_base SRCS placement_pass_base.cc DEPS pass)

cc_library(coalesce_grad_tensor_pass SRCS coalesce_grad_tensor_pass.cc DEPS graph graph_helper)

pass_library(graph_to_program_pass base)
pass_library(graph_viz_pass base)
pass_library(lock_free_optimize_pass base)
pass_library(fc_fuse_pass inference)
pass_library(map_matmul_to_mul_pass inference)
pass_library(attention_lstm_fuse_pass inference)
pass_library(fc_lstm_fuse_pass inference)
pass_library(embedding_fc_lstm_fuse_pass inference)
pass_library(fc_gru_fuse_pass inference)
pass_library(seq_concat_fc_fuse_pass inference)
pass_library(multi_batch_merge_pass base)
pass_library(conv_bn_fuse_pass inference)
pass_library(seqconv_eltadd_relu_fuse_pass inference)
pass_library(seqpool_concat_fuse_pass inference)
pass_library(seqpool_cvm_concat_fuse_pass inference)
pass_library(repeated_fc_relu_fuse_pass inference)
pass_library(squared_mat_sub_fuse_pass inference)
pass_library(is_test_pass base)
pass_library(conv_elementwise_add_act_fuse_pass inference)
pass_library(conv_elementwise_add2_act_fuse_pass inference)
pass_library(conv_elementwise_add_fuse_pass inference)
pass_library(conv_affine_channel_fuse_pass inference)
pass_library(transpose_flatten_concat_fuse_pass inference)
pass_library(identity_scale_op_clean_pass base)
pass_library(sync_batch_norm_pass base)
pass_library(runtime_context_cache_pass base)
pass_library(quant_conv2d_dequant_fuse_pass inference)
pass_library(shuffle_channel_detect_pass inference)
pass_library(delete_quant_dequant_op_pass inference)
pass_library(delete_quant_dequant_filter_op_pass inference)
pass_library(delete_dropout_op_pass inference)
pass_library(simplify_with_basic_ops_pass base)
pass_library(fc_elementwise_layernorm_fuse_pass base)
pass_library(skip_layernorm_fuse_pass base)
pass_library(multihead_matmul_fuse_pass inference)
pass_library(adaptive_pool2d_convert_global_pass inference)
pass_library(unsqueeze2_eltwise_fuse_pass inference)
pass_library(layer_norm_fuse_pass inference)
if(WITH_GPU OR WITH_ROCM)
    pass_library(cudnn_placement_pass base DEPS placement_pass_base)
    pass_library(embedding_eltwise_layernorm_fuse_pass inference)
endif()

if(WITH_MKLDNN)
    pass_library(mkldnn_placement_pass base DEPS placement_pass_base DIR mkldnn)
    pass_library(mkldnn_inplace_pass inference DEPS mkldnn_placement_pass op_registry elementwise_add_op gelu_op activation_op softmax_op softmax DIR mkldnn)
    pass_library(depthwise_conv_mkldnn_pass base DIR mkldnn)
    pass_library(conv_bias_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(conv_activation_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(conv_concat_relu_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(conv_elementwise_add_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(scale_matmul_fuse_pass inference DIR mkldnn)
    pass_library(cpu_bfloat16_placement_pass inference DIR mkldnn)
    pass_library(cpu_bfloat16_pass inference DIR mkldnn)
    pass_library(fc_mkldnn_pass inference DIR mkldnn)
    pass_library(interpolate_mkldnn_pass inference DIR mkldnn)
    pass_library(fc_act_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(cpu_quantize_placement_pass base DIR mkldnn)
    pass_library(cpu_quantize_pass inference DIR mkldnn)
    pass_library(cpu_quantize_squash_pass inference DIR mkldnn)
    pass_library(reshape_transpose_matmul_mkldnn_fuse_pass inference DIR mkldnn)
    pass_library(matmul_transpose_reshape_fuse_pass inference DIR mkldnn)
    pass_library(batch_norm_act_fuse_pass inference DIR mkldnn)
    pass_library(multi_gru_fuse_pass inference DIR mkldnn)
    pass_library(multi_gru_seq_fuse_pass inference DIR mkldnn)
endif()

cc_library(fuse_bn_act_pass SRCS fuse_bn_act_pass.cc DEPS pass graph_pattern_detector )
cc_library(fuse_bn_add_act_pass SRCS fuse_bn_add_act_pass.cc DEPS pass graph_pattern_detector )
cc_library(fuse_elewise_add_act_pass SRCS fuse_elewise_add_act_pass.cc DEPS pass graph_pattern_detector )
cc_library(fuse_relu_depthwise_conv_pass SRCS fuse_relu_depthwise_conv_pass.cc DEPS pass graph_pattern_detector )

set(GLOB_PASS_LIB ${PASS_LIBRARY} CACHE INTERNAL "Global PASS library")

cc_library(pass_builder SRCS pass_builder.cc DEPS pass)
cc_library(pass_test_util SRCS pass_test_util.cc DEPS graph pass)

cc_test(node_test SRCS node_test.cc DEPS node)
cc_test(pass_test SRCS pass_test.cc DEPS graph pass graph_helper)
cc_test(graph_test SRCS graph_test.cc DEPS graph graph_helper op_registry)
cc_test(graph_helper_test SRCS graph_helper_test.cc DEPS graph graph_helper op_registry)
cc_test(graph_to_program_pass_test SRCS graph_to_program_pass_test.cc DEPS graph_to_program_pass)
cc_test(test_graph_pattern_detector SRCS graph_pattern_detector_tester.cc DEPS graph_pattern_detector)
cc_test(test_fc_fuse_pass_cc SRCS fc_fuse_pass_tester.cc DEPS fc_fuse_pass framework_proto)
cc_test(test_fc_lstm_fuse_pass_cc SRCS fc_lstm_fuse_pass_tester.cc DEPS fc_lstm_fuse_pass framework_proto)
cc_test(test_fc_gru_fuse_pass_cc SRCS fc_gru_fuse_pass_tester.cc DEPS fc_gru_fuse_pass framework_proto)
cc_test(test_seqpool_concat_fuse_pass SRCS seqpool_concat_fuse_pass_tester.cc DEPS seqpool_concat_fuse_pass framework_proto)
cc_test(test_seqpool_cvm_concat_fuse_pass SRCS seqpool_cvm_concat_fuse_pass_tester.cc DEPS seqpool_cvm_concat_fuse_pass framework_proto)
cc_test(test_repeated_fc_relu_fuse_pass_cc SRCS repeated_fc_relu_fuse_pass_tester.cc DEPS repeated_fc_relu_fuse_pass framework_proto)
cc_test(test_is_test_pass SRCS is_test_pass_tester.cc DEPS is_test_pass)
cc_test(test_simplify_with_basic_ops_pass SRCS simplify_with_basic_ops_pass_tester.cc DEPS simplify_with_basic_ops_pass)
cc_test(test_fc_elementwise_layernorm_fuse_pass SRCS fc_elementwise_layernorm_fuse_pass_tester.cc DEPS fc_elementwise_layernorm_fuse_pass)
cc_test(test_skip_layernorm_fuse_pass SRCS skip_layernorm_fuse_pass_tester.cc DEPS skip_layernorm_fuse_pass)
cc_test(test_multihead_matmul_fuse_pass SRCS multihead_matmul_fuse_pass_tester.cc DEPS multihead_matmul_fuse_pass)
cc_test(test_conv_bn_fuse_pass_cc SRCS conv_bn_fuse_pass_tester.cc DEPS conv_bn_fuse_pass)
cc_test(test_adaptive_pool2d_convert_global_pass SRCS adaptive_pool2d_convert_global_pass_tester.cc DEPS adaptive_pool2d_convert_global_pass)
cc_test(test_unsqueeze2_eltwise_fuse_pass SRCS unsqueeze2_eltwise_fuse_pass_tester.cc DEPS unsqueeze2_eltwise_fuse_pass)
cc_test(test_layer_norm_fuse_pass_cc SRCS layer_norm_fuse_pass_tester.cc DEPS layer_norm_fuse_pass pass_test_util naive_executor)
if(WITH_GPU OR WITH_ROCM)
    cc_test(test_embedding_eltwise_layernorm_fuse_pass SRCS embedding_eltwise_layernorm_fuse_pass_tester.cc DEPS embedding_eltwise_layernorm_fuse_pass)
    cc_test(test_cudnn_placement_pass SRCS cudnn_placement_pass_tester.cc DEPS cudnn_placement_pass)
endif()
if(NOT WIN32)
    cc_test(test_sync_batch_norm_pass SRCS sync_batch_norm_pass_tester.cc DEPS sync_batch_norm_pass)
endif()
if (WITH_MKLDNN)
    cc_test(test_depthwise_conv_mkldnn_pass SRCS mkldnn/depthwise_conv_mkldnn_pass_tester.cc DEPS depthwise_conv_mkldnn_pass)
    cc_test(test_conv_bias_mkldnn_fuse_pass SRCS mkldnn/conv_bias_mkldnn_fuse_pass_tester.cc DEPS conv_bias_mkldnn_fuse_pass naive_executor)
    cc_test(test_conv_activation_mkldnn_fuse_pass SRCS mkldnn/conv_activation_mkldnn_fuse_pass_tester.cc DEPS conv_activation_mkldnn_fuse_pass)
    cc_test(test_conv_concat_relu_mkldnn_fuse_pass SRCS mkldnn/conv_concat_relu_mkldnn_fuse_pass_tester.cc DEPS conv_concat_relu_mkldnn_fuse_pass)
    cc_test(test_conv_elementwise_add_mkldnn_fuse_pass SRCS mkldnn/conv_elementwise_add_mkldnn_fuse_pass_tester.cc DEPS conv_elementwise_add_mkldnn_fuse_pass pass_test_util)
    cc_test(test_fc_act_mkldnn_fuse_pass SRCS mkldnn/fc_act_mkldnn_fuse_pass_tester.cc DEPS fc_act_mkldnn_fuse_pass pass_test_util)
    cc_test(test_batch_norm_act_fuse_pass SRCS mkldnn/batch_norm_act_fuse_pass_tester.cc DEPS batch_norm_act_fuse_pass pass_test_util)
    set(TEST_CONV_BN_PASS_DEPS conv_bn_fuse_pass graph_to_program_pass conv_op conv_transpose_op math_function im2col vol2col batch_norm_op gelu_op activation_op elementwise_add_op concat_and_split naive_executor device_context)
if (WITH_GPU OR WITH_ROCM)
    set(TEST_CONV_BN_PASS_DEPS ${TEST_CONV_BN_PASS_DEPS} depthwise_conv)
endif()
    cc_test(test_conv_batch_norm_mkldnn_fuse_pass SRCS mkldnn/mkldnn_conv_bn_fuse_pass_tester.cc DEPS ${TEST_CONV_BN_PASS_DEPS})
    cc_test(test_scale_matmul_fuse_pass SRCS mkldnn/scale_matmul_fuse_pass_tester.cc DEPS scale_matmul_fuse_pass)
    cc_test(test_mkldnn_placement_pass SRCS mkldnn/mkldnn_placement_pass_tester.cc DEPS mkldnn_placement_pass)
    cc_test(test_mkldnn_inplace_pass SRCS mkldnn/mkldnn_inplace_pass_tester.cc DEPS mkldnn_inplace_pass)
    cc_test(test_cpu_quantize_placement_pass SRCS mkldnn/cpu_quantize_placement_pass_tester.cc DEPS cpu_quantize_placement_pass)
    cc_test(test_cpu_quantize_pass SRCS mkldnn/cpu_quantize_pass_tester.cc DEPS cpu_quantize_pass naive_executor)
    cc_test(test_cpu_quantize_squash_pass SRCS mkldnn/cpu_quantize_squash_pass_tester.cc DEPS cpu_quantize_squash_pass naive_executor)
    cc_test(test_reshape_transpose_matmul_mkldnn_fuse_pass SRCS mkldnn/reshape_transpose_matmul_mkldnn_fuse_pass_tester.cc DEPS reshape_transpose_matmul_mkldnn_fuse_pass)
    cc_test(test_matmul_transpose_reshape_fuse_pass SRCS mkldnn/matmul_transpose_reshape_fuse_pass_tester.cc DEPS matmul_transpose_reshape_fuse_pass)
    cc_test(test_cpu_bfloat16_placement_pass SRCS mkldnn/cpu_bfloat16_placement_pass_tester.cc DEPS cpu_bfloat16_placement_pass)
    cc_test(test_cpu_bfloat16_pass SRCS mkldnn/cpu_bfloat16_pass_tester.cc DEPS cpu_bfloat16_pass)
    cc_test(test_multi_gru_fuse_pass SRCS mkldnn/multi_gru_fuse_pass_tester.cc DEPS multi_gru_fuse_pass)
    cc_test(test_multi_gru_seq_fuse_pass SRCS mkldnn/multi_gru_seq_fuse_pass_tester.cc DEPS multi_gru_seq_fuse_pass)
endif ()