CMakeLists.txt 4.1 KB
Newer Older
1
2
3
# ckProfiler
set(PROFILER_SOURCES
    profiler.cpp
4
    # profile_gemm.cpp
5
    profile_gemm_splitk.cpp
6
7
8
9
10
11
12
13
14
15
16
17
    # profile_gemm_bilinear.cpp
    # profile_gemm_bias_add_reduce.cpp
    # profile_gemm_add_add_fastgelu.cpp
    # profile_gemm_add_multiply.cpp
    # profile_gemm_add_fastgelu.cpp
    # profile_gemm_add_relu_add_layernorm.cpp
    # profile_gemm_fastgelu.cpp
    # profile_gemm_reduce.cpp
    # profile_batched_gemm.cpp
    # profile_batched_gemm_gemm.cpp
    # profile_batched_gemm_add_relu_gemm_add.cpp
    # profile_batched_gemm_reduce.cpp
18
    profile_grouped_gemm.cpp
19
20
21
22
23
24
25
26
27
28
29
30
31
32
    # profile_conv_fwd.cpp
    # profile_conv_fwd_bias_relu.cpp
    # profile_conv_fwd_bias_relu_add.cpp
    # profile_conv_bwd_data.cpp
    # profile_grouped_conv_fwd.cpp
    # profile_grouped_conv_bwd_weight.cpp
    # profile_reduce.cpp
    # profile_groupnorm.cpp
    # profile_layernorm.cpp
    # profile_softmax.cpp
    # profile_batchnorm_fwd.cpp
    # profile_batchnorm_bwd.cpp
    # profile_batchnorm_infer.cpp
    # profile_grouped_gemm_fastgelu.cpp
33
34
35
36
37
38
39
40
)

set(PROFILER_EXECUTABLE ckProfiler)

add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES})
target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)

target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
41
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
42
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
43
44
45
46
47
48
49
50
51
52
53
54
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
55
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
# target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
73
rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)