CMakeLists.txt 8.1 KB
Newer Older
W
Wu Yi 已提交
1
include(operators)
D
dzhwinter 已提交
2

W
Wu Yi 已提交
3 4 5
# clean cache and pybind_file content first when rebuild
unset(GLOB_OP_LIB CACHE)
unset(OP_LIBRARY CACHE)
6 7 8 9
set(pybind_file ${PADDLE_BINARY_DIR}/paddle/fluid/pybind/pybind.h.tmp CACHE INTERNAL "pybind.h file")
set(pybind_file_final ${PADDLE_BINARY_DIR}/paddle/fluid/pybind/pybind.h)
file(WRITE ${pybind_file} "// Generated by the paddle/fluid/operators/CMakeLists.txt.  DO NOT EDIT!\n\n")

Q
qijun 已提交
10
add_subdirectory(math)
11
add_subdirectory(eigen)
W
Wu Yi 已提交
12 13 14 15 16 17 18 19
add_subdirectory(controlflow)
add_subdirectory(detection)
add_subdirectory(elementwise)
add_subdirectory(fused)
add_subdirectory(metrics)
add_subdirectory(optimizers)
add_subdirectory(reduce_ops)
add_subdirectory(sequence_ops)
T
tensor-tang 已提交
20
add_subdirectory(jit)
L
lidanqing 已提交
21 22 23
if(WITH_MKLDNN)
    add_subdirectory(mkldnn)
endif()
Q
QI JUN 已提交
24

T
tangwei12 已提交
25

T
typhoonzero 已提交
26
if(WITH_DISTRIBUTE)
27
    add_subdirectory(collective)
W
Wu Yi 已提交
28
endif()
29

T
tangwei12 已提交
30 31 32 33
if (WITH_PSCORE)
    add_subdirectory(pscore)
endif()

34 35
add_subdirectory(amp)

W
wopeizl 已提交
36
add_subdirectory(reader)
T
typhoonzero 已提交
37

W
Wu Yi 已提交
38 39
if (NOT WIN32)
    add_subdirectory(nccl)
S
sneaxiy 已提交
40 41
endif()

42
if (WITH_GPU AND TENSORRT_FOUND)
W
Wu Yi 已提交
43
    add_subdirectory(tensorrt)
44
endif()
45

D
denglin-github 已提交
46 47 48 49
if (WITH_DLNNE)
    add_subdirectory(dlnne)
endif()

石晓伟 已提交
50 51 52 53
if (WITH_LITE)
    add_subdirectory(lite)
endif()

W
Wilber 已提交
54
SET(OP_HEADER_DEPS xxhash executor)
石晓伟 已提交
55

56
if (WITH_GPU)
57 58 59
    if (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
        SET(OP_HEADER_DEPS ${OP_HEADER_DEPS} cub)
    endif()
60 61
endif()

62

A
Aurelius84 已提交
63
SET(OP_MKL_DEPS "")
64
if (NOT WITH_MKL OR NOT WITH_AVX)
A
Aurelius84 已提交
65 66 67
    SET(OP_MKL_DEPS ${OP_MKL_DEPS} match_matrix_tensor_op)
    SET(OP_MKL_DEPS ${OP_MKL_DEPS} var_conv_2d_op)
endif()
68
if(WITH_COVERAGE OR WIN32 OR WITH_NV_JETSON)
A
Aurelius84 已提交
69
    SET(OP_MKL_DEPS ${OP_MKL_DEPS} pyramid_hash_op)
K
Kevin 已提交
70 71
endif()

72 73 74 75 76
if(WITH_UNITY_BUILD)
    # Load Unity Build rules for operators in paddle/fluid/operators.
    include(unity_build_rule.cmake)
endif()

77
register_operators(EXCLUDES py_layer_op py_func_op warpctc_op dgc_op lstm_op run_program_op eye_op recurrent_op
T
tangwei12 已提交
78
        sync_batch_norm_op ${OP_MKL_DEPS} DEPS ${OP_HEADER_DEPS})
武毅 已提交
79

T
tangwei12 已提交
80
op_library(run_program_op SRCS run_program_op.cc run_program_op.cu.cc DEPS executor_cache ${OP_HEADER_DEPS})
81

82 83
if (WITH_GPU OR WITH_ROCM)
    if(WITH_ROCM)
84
        op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale SRCS warpctc_op.cc warpctc_op.cu)
Q
qingqing01 已提交
85
    # warpctc_op needs cudnn 7 above
86
    elseif(${CUDNN_MAJOR_VERSION} VERSION_LESS 7)
87
        op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale SRCS warpctc_op.cc warpctc_op.cu)
W
Wu Yi 已提交
88 89
    else()
        op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale)
W
Wu Yi 已提交
90
    endif()
91 92
    op_library(sync_batch_norm_op)
    file(APPEND ${pybind_file} "USE_CUDA_ONLY_OP(sync_batch_norm);\n")
W
Wu Yi 已提交
93 94
else()
    op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale)
W
Wu Yi 已提交
95
endif()
T
tangwei12 已提交
96 97 98 99

op_library(lstm_op DEPS ${OP_HEADER_DEPS}  lstm_compute)
op_library(eye_op DEPS ${OP_HEADER_DEPS})
op_library(recurrent_op DEPS ${OP_HEADER_DEPS})
100

101
set(COMMON_OP_DEPS ${OP_HEADER_DEPS})
102

G
gongweibao 已提交
103
if (WITH_DGC)
104 105 106 107 108
    op_library(dgc_op DEPS dgc)
    file(APPEND ${pybind_file} "USE_CUDA_ONLY_OP(dgc);\n")
    set(COMMON_OP_DEPS ${COMMON_OP_DEPS} dgc)
endif()

109
cc_library(common_infer_shape_functions SRCS common_infer_shape_functions.cc DEPS operator)
A
Aurelius84 已提交
110

Y
yaoxuefeng 已提交
111 112
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} selected_rows_functor selected_rows
lod_tensor maxouting unpooling pooling lod_rank_table context_project
113
sequence_pooling segment_pooling executor device_memory_aligment generator)
P
peizhilin 已提交
114
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} dynload_warpctc)
X
xuezhong 已提交
115
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} sequence_padding sequence_scale cos_sim_functor memory jit_kernel_helper concat_and_split cross_entropy softmax vol2col im2col sampler sample_prob tree2col)
116
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} sequence2batch lstm_compute matrix_bit_code gru_compute activation_functions beam_search fc matrix_inverse)
Y
yaoxuefeng 已提交
117
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} box_wrapper boost ps_gpu_wrapper)
118
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} common_infer_shape_functions)
119
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} eigen_function)
120
if (WITH_GPU OR WITH_ROCM)
121
  set(COMMON_OP_DEPS ${COMMON_OP_DEPS} depthwise_conv prelu bert_encoder_functor)
W
Wu Yi 已提交
122
endif()
123
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} device_memory_aligment)
124
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} layer)
H
Huihuang Zheng 已提交
125
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} tensor_formatter)
126
set(COMMON_OP_DEPS ${COMMON_OP_DEPS} op_version_registry)
127 128 129
if (WITH_ASCEND)
  set(COMMON_OP_DEPS ${COMMON_OP_DEPS} ascend_wrapper)
endif()
130

131
if (WITH_ASCEND_CL)
132
  cc_test(assign_op_npu_test SRCS assign_op_npu_test.cc DEPS assign_op)
133 134 135 136
  cc_library(npu_op_runner SRCS npu_op_runner.cc DEPS operator npu_info)
  set(COMMON_OP_DEPS ${COMMON_OP_DEPS} npu_op_runner)
endif()

W
Wu Yi 已提交
137 138 139 140 141 142 143 144
# FIXME(typhoonzero): operator deps may not needed.
# op_library(lod_tensor_to_array_op DEPS lod_rank_table_op)
# op_library(array_to_lod_tensor_op DEPS lod_rank_table_op)
# op_library(unsqueeze_op DEPS reshape_op)
# op_library(squeeze_op DEPS reshape_op)
# op_library(flatten_op DEPS reshape_op)
# op_library(unstack_op DEPS stack_op)
# op_library(tensor_array_to_tensor_op DEPS concat_op)
B
baiyf 已提交
145

W
Wu Yi 已提交
146 147
set(OPERATOR_DEPS ${OPERATOR_DEPS} ${COMMON_OP_DEPS})
set(GLOB_OPERATOR_DEPS ${OPERATOR_DEPS} CACHE INTERNAL "Global Op dependencies")
L
Luo Tao 已提交
148

149
cc_test(test_common_infer_shape_functions SRCS test_common_infer_shape_functions.cc DEPS common_infer_shape_functions ${COMMON_OP_DEPS} activation_op elementwise_add_op softmax_op softmax)
150
cc_test(gather_test SRCS gather_test.cc DEPS tensor)
151
cc_test(assign_op_test SRCS assign_op_test.cc DEPS assign_op)
Y
Yu Yang 已提交
152
cc_test(scatter_test SRCS scatter_test.cc DEPS tensor math_function)
Q
Qiao Longfei 已提交
153
cc_test(beam_search_decode_op_test SRCS beam_search_decode_op_test.cc DEPS lod_tensor)
Y
Yi Wang 已提交
154
cc_test(strided_memcpy_test SRCS strided_memcpy_test.cc DEPS tensor memory)
Y
Yu Yang 已提交
155
cc_test(save_load_op_test SRCS save_load_op_test.cc DEPS save_op load_op)
156
cc_test(save_load_combine_op_test SRCS save_load_combine_op_test.cc DEPS save_combine_op load_combine_op)
Z
Zeng Jinle 已提交
157
if (WITH_GPU)
158
    nv_test(dropout_op_test SRCS dropout_op_test.cc DEPS dropout_op tensor generator)
Z
Zeng Jinle 已提交
159
    nv_test(test_leaky_relu_grad_grad_functor SRCS test_leaky_relu_grad_grad_functor.cc test_leaky_relu_grad_grad_functor.cu DEPS tensor device_context eigen3)
160
    nv_test(feed_forward_test SRCS feed_forward_test.cu DEPS elementwise_add_op matmul_op tensor generator)
161 162 163
elseif(WITH_ROCM)
    hip_test(dropout_op_test SRCS dropout_op_test.cc DEPS dropout_op tensor generator)
    hip_test(test_leaky_relu_grad_grad_functor SRCS test_leaky_relu_grad_grad_functor.cc test_leaky_relu_grad_grad_functor.cu DEPS tensor device_context eigen3)
Z
Zeng Jinle 已提交
164 165 166
else()
    cc_test(test_leaky_relu_grad_grad_functor SRCS test_leaky_relu_grad_grad_functor.cc DEPS tensor device_context eigen3)
endif()
167

H
Huihuang Zheng 已提交
168
cc_library(tensor_formatter SRCS tensor_formatter.cc DEPS ${OP_HEADER_DEPS})
S
sneaxiy 已提交
169 170
if (WITH_PYTHON)
  cc_library(py_func_op SRCS py_func_op.cc DEPS op_registry python pybind)
171
  cc_library(py_layer_op SRCS py_layer_op.cc DEPS op_registry python pybind)
S
sneaxiy 已提交
172 173
endif()

174 175
if (WITH_ASCEND_CL)
  cc_test(range_op_npu_test SRCS range_op_npu_test.cc DEPS op_registry range_op scope device_context enforce executor)
176
  cc_test(expand_op_npu_test SRCS expand_op_npu_test.cc DEPS op_registry expand_op eigen_function scope device_context enforce executor compare_op)
177 178
endif()

W
Wu Yi 已提交
179
set(GLOB_OP_LIB ${OP_LIBRARY} CACHE INTERNAL "Global OP library")
180
add_subdirectory(benchmark)
181 182

cc_test(op_debug_string_test SRCS op_debug_string_test.cc DEPS elementwise_add_op)
183 184 185 186
if (WITH_ASCEND_CL)
    cc_test(transpose_op_npu_test SRCS transpose_op_npu_test.cc DEPS op_registry transpose_op scope device_context enforce executor)
endif()

187 188 189

if(WITH_MKLDNN)
include(mkldnn/inplace_op_tests.cmake)
190
include(mkldnn/caching_tests.cmake)
J
Jacek Czaja 已提交
191
include(mkldnn/nhwc_op_tests.cmake)
192
endif()
193 194 195 196 197

if(WITH_UNITY_BUILD)
    # Using Unity Build to compile operators, `register_operator` will cause
    # the unity library to lose some symbols.
    # The specified link dependency needs to be displayed here.
T
tangwei12 已提交
198
    target_link_libraries(paddle_operators_unity ${OP_HEADER_DEPS} ${COMMON_OP_DEPS})
199
endif()
200 201 202 203

if(WITH_ASCEND_CL)
cc_test(gelu_op_npu_test SRCS gelu_op_npu_test.cc DEPS op_registry gelu_op scope device_context enforce executor)
endif()
B
Baibaifan 已提交
204 205 206 207

if (WITH_GPU OR WITH_ASCEND_CL)
cc_test(copy_cross_scope_test SRCS copy_cross_scope_test.cc DEPS op_registry copy_cross_scope_op scope device_context enforce executor)
endif()
208 209

copy_if_different(${pybind_file} ${pybind_file_final})