Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
gaoqiong
composable_kernel
Commits
bc5b84b1
Commit
bc5b84b1
authored
Dec 04, 2023
by
Artur Wojcik
Browse files
Merge branch 'develop' into uif2-initial
parents
acef6cc7
bc4bf9bd
Changes
66
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
50 additions
and
50 deletions
+50
-50
Config.cmake.in
Config.cmake.in
+1
-1
Jenkinsfile
Jenkinsfile
+1
-1
client_example/01_gemm/CMakeLists.txt
client_example/01_gemm/CMakeLists.txt
+1
-1
client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
+6
-6
client_example/03_gemm_layernorm/CMakeLists.txt
client_example/03_gemm_layernorm/CMakeLists.txt
+2
-2
client_example/04_contraction/CMakeLists.txt
client_example/04_contraction/CMakeLists.txt
+5
-5
client_example/05_layernorm/CMakeLists.txt
client_example/05_layernorm/CMakeLists.txt
+2
-2
client_example/06_softmax/CMakeLists.txt
client_example/06_softmax/CMakeLists.txt
+1
-1
client_example/07_grouped_convnd_fwd/CMakeLists.txt
client_example/07_grouped_convnd_fwd/CMakeLists.txt
+2
-2
client_example/08_fused_attention/CMakeLists.txt
client_example/08_fused_attention/CMakeLists.txt
+2
-2
client_example/09_quantization/CMakeLists.txt
client_example/09_quantization/CMakeLists.txt
+7
-7
client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
+3
-3
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
+5
-5
client_example/12_elementwise_normalization/CMakeLists.txt
client_example/12_elementwise_normalization/CMakeLists.txt
+1
-1
client_example/13_batchnorm/CMakeLists.txt
client_example/13_batchnorm/CMakeLists.txt
+3
-3
client_example/14_instance_id/CMakeLists.txt
client_example/14_instance_id/CMakeLists.txt
+1
-1
client_example/15_convnd_bwd_data/CMakeLists.txt
client_example/15_convnd_bwd_data/CMakeLists.txt
+2
-2
client_example/15_gemm_add_multiply/CMakeLists.txt
client_example/15_gemm_add_multiply/CMakeLists.txt
+1
-1
client_example/15_reduce/CMakeLists.txt
client_example/15_reduce/CMakeLists.txt
+1
-1
client_example/16_convnd_fwd/CMakeLists.txt
client_example/16_convnd_fwd/CMakeLists.txt
+3
-3
No files found.
Config.cmake.in
View file @
bc5b84b1
@PACKAGE_INIT@
@PACKAGE_INIT@
set(_composable_kernel_supported_components device_operations utility)
set(_composable_kernel_supported_components device_
other_operations device_gemm_operations device_conv_operations device_mha_operations device_contraction_operations device_reduction_
operations utility)
foreach(_comp ${composable_kernel_FIND_COMPONENTS})
foreach(_comp ${composable_kernel_FIND_COMPONENTS})
if(NOT _comp IN_LIST _composable_kernel_supported_components)
if(NOT _comp IN_LIST _composable_kernel_supported_components)
...
...
Jenkinsfile
View file @
bc5b84b1
def
rocmnode
(
name
)
{
def
rocmnode
(
name
)
{
return
'rocmtest
&&
miopen && '
+
name
return
'
(
rocmtest
||
miopen
)
&& '
+
name
}
}
def
show_node_info
()
{
def
show_node_info
()
{
...
...
client_example/01_gemm/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_gemm gemm.cpp
)
add_executable
(
client_gemm gemm.cpp
)
target_link_libraries
(
client_gemm PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm PRIVATE composable_kernel::device_
other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_gemm PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm PRIVATE cxx_std_17
)
client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
View file @
bc5b84b1
...
@@ -2,28 +2,28 @@ add_custom_target(client_gemm_fastgelu_examples DEPENDS
...
@@ -2,28 +2,28 @@ add_custom_target(client_gemm_fastgelu_examples DEPENDS
client_gemm_add_add_fastgelu client_gemm_add_fastgelu client_gemm_fastgelu
)
client_gemm_add_add_fastgelu client_gemm_add_fastgelu client_gemm_fastgelu
)
add_executable
(
client_gemm_add_add_fastgelu gemm_add_add_fastgelu.cpp
)
add_executable
(
client_gemm_add_add_fastgelu gemm_add_add_fastgelu.cpp
)
target_link_libraries
(
client_gemm_add_add_fastgelu PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_add_fastgelu PRIVATE composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_add_add_fastgelu PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_add_fastgelu PRIVATE cxx_std_17
)
add_executable
(
client_gemm_add_fastgelu gemm_add_fastgelu.cpp
)
add_executable
(
client_gemm_add_fastgelu gemm_add_fastgelu.cpp
)
target_link_libraries
(
client_gemm_add_fastgelu PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_fastgelu PRIVATE composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_add_fastgelu PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_fastgelu PRIVATE cxx_std_17
)
add_executable
(
client_gemm_fastgelu gemm_fastgelu.cpp
)
add_executable
(
client_gemm_fastgelu gemm_fastgelu.cpp
)
target_link_libraries
(
client_gemm_fastgelu PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_fastgelu PRIVATE composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_fastgelu PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_fastgelu PRIVATE cxx_std_17
)
add_custom_target
(
client_gemm_fastgelu_generic_examples DEPENDS
add_custom_target
(
client_gemm_fastgelu_generic_examples DEPENDS
client_gemm_add_add_fastgelu_generic client_gemm_add_fastgelu_generic client_gemm_fastgelu_generic
)
client_gemm_add_add_fastgelu_generic client_gemm_add_fastgelu_generic client_gemm_fastgelu_generic
)
add_executable
(
client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp
)
add_executable
(
client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp
)
target_link_libraries
(
client_gemm_add_add_fastgelu_generic
PRIVATE
composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_add_fastgelu_generic composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_add_add_fastgelu_generic PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_add_fastgelu_generic PRIVATE cxx_std_17
)
add_executable
(
client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp
)
add_executable
(
client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp
)
target_link_libraries
(
client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_add_fastgelu_generic PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_fastgelu_generic PRIVATE cxx_std_17
)
add_executable
(
client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp
)
add_executable
(
client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp
)
target_link_libraries
(
client_gemm_fastgelu_generic PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_fastgelu_generic PRIVATE composable_kernel::device_
gemm_
operations
)
target_compile_features
(
client_gemm_fastgelu_generic PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_fastgelu_generic PRIVATE cxx_std_17
)
client_example/03_gemm_layernorm/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_gemm_add_add_layernorm_naive gemm_add_add_layernorm_naive.cpp
)
add_executable
(
client_gemm_add_add_layernorm_naive gemm_add_add_layernorm_naive.cpp
)
target_link_libraries
(
client_gemm_add_add_layernorm_naive PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_add_layernorm_naive PRIVATE composable_kernel::device_
gemm_operations composable_kernel::device_other_
operations
)
target_compile_features
(
client_gemm_add_add_reduce_normalize PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_add_reduce_normalize PRIVATE cxx_std_17
)
add_executable
(
client_gemm_add_relu_add_layernorm_welford gemm_add_relu_add_layernorm_welford.cpp
)
add_executable
(
client_gemm_add_relu_add_layernorm_welford gemm_add_relu_add_layernorm_welford.cpp
)
target_link_libraries
(
client_gemm_add_relu_add_layernorm_welford PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_relu_add_layernorm_welford PRIVATE composable_kernel::device_
gemm_operations composable_kernel::device_other_
operations
)
target_compile_features
(
client_gemm_add_relu_add_layernorm_welford PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_add_relu_add_layernorm_welford PRIVATE cxx_std_17
)
client_example/04_contraction/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_contraction_scale_fp32 contraction_scale_fp32.cpp
)
add_executable
(
client_contraction_scale_fp32 contraction_scale_fp32.cpp
)
target_link_libraries
(
client_contraction_scale_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_contraction_scale_fp32 PRIVATE composable_kernel::device_
other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_contraction_scale PRIVATE cxx_std_17
)
target_compile_features
(
client_contraction_scale PRIVATE cxx_std_17
)
add_executable
(
client_contraction_bilinear_fp32 contraction_bilinear_fp32.cpp
)
add_executable
(
client_contraction_bilinear_fp32 contraction_bilinear_fp32.cpp
)
target_link_libraries
(
client_contraction_bilinear_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_contraction_bilinear_fp32 PRIVATE composable_kernel::device_
other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_contraction_bilinear PRIVATE cxx_std_17
)
target_compile_features
(
client_contraction_bilinear PRIVATE cxx_std_17
)
add_executable
(
client_contraction_scale_fp64 contraction_scale_fp64.cpp
)
add_executable
(
client_contraction_scale_fp64 contraction_scale_fp64.cpp
)
target_link_libraries
(
client_contraction_scale_fp64 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_contraction_scale_fp64 PRIVATE composable_kernel::device_
other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_contraction_scale_fp64 PRIVATE cxx_std_17
)
target_compile_features
(
client_contraction_scale_fp64 PRIVATE cxx_std_17
)
add_executable
(
client_contraction_bilinear_fp64 contraction_bilinear_fp64.cpp
)
add_executable
(
client_contraction_bilinear_fp64 contraction_bilinear_fp64.cpp
)
target_link_libraries
(
client_contraction_bilinear_fp64 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_contraction_bilinear_fp64 PRIVATE composable_kernel::device_
other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_contraction_blinear_fp64 PRIVATE cxx_std_17
)
target_compile_features
(
client_contraction_blinear_fp64 PRIVATE cxx_std_17
)
add_executable
(
contraction_g1m2n3k1_add_xdl_fp16 contraction_g1m2n3k1_add_xdl_fp16.cpp
)
add_executable
(
contraction_g1m2n3k1_add_xdl_fp16 contraction_g1m2n3k1_add_xdl_fp16.cpp
)
target_link_libraries
(
contraction_g1m2n3k1_add_xdl_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
contraction_g1m2n3k1_add_xdl_fp16 PRIVATE composable_kernel::device_
other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
contraction_g1m2n3k1_add_xdl-fp16 PRIVATE cxx_std_17
)
target_compile_features
(
contraction_g1m2n3k1_add_xdl-fp16 PRIVATE cxx_std_17
)
client_example/05_layernorm/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_layernorm2d_fwd layernorm2d_fwd.cpp
)
add_executable
(
client_layernorm2d_fwd layernorm2d_fwd.cpp
)
target_link_libraries
(
client_layernorm2d_fwd PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_layernorm2d_fwd PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_layernorm2d_fwd PRIVATE cxx_std_17
)
target_compile_features
(
client_layernorm2d_fwd PRIVATE cxx_std_17
)
add_executable
(
client_layernorm4d_fwd layernorm4d_fwd.cpp
)
add_executable
(
client_layernorm4d_fwd layernorm4d_fwd.cpp
)
target_link_libraries
(
client_layernorm4d_fwd PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_layernorm4d_fwd PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_layernorm4d_fwd PRIVATE cxx_std_17
)
target_compile_features
(
client_layernorm4d_fwd PRIVATE cxx_std_17
)
client_example/06_softmax/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_softmax4d softmax4d.cpp
)
add_executable
(
client_softmax4d softmax4d.cpp
)
target_link_libraries
(
client_softmax4d PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_softmax4d PRIVATE composable_kernel::device_
other_operations composable_kernel::device_reduction_
operations
)
target_compile_features
(
client_softmax4d PRIVATE cxx_std_17
)
target_compile_features
(
client_softmax4d PRIVATE cxx_std_17
)
client_example/07_grouped_convnd_fwd/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp
)
add_executable
(
client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp
)
target_link_libraries
(
client_grouped_conv2d_fwd PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv2d_fwd PRIVATE composable_kernel::device_
conv_
operations
)
add_executable
(
client_grouped_conv1d_fwd grouped_conv1d_fwd.cpp
)
add_executable
(
client_grouped_conv1d_fwd grouped_conv1d_fwd.cpp
)
target_link_libraries
(
client_grouped_conv1d_fwd PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv1d_fwd PRIVATE composable_kernel::device_
conv_
operations
)
client_example/08_fused_attention/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_fused_attention fused_attention.cpp
)
add_executable
(
client_fused_attention fused_attention.cpp
)
target_link_libraries
(
client_fused_attention PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_fused_attention PRIVATE composable_kernel::device_
other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_fused_attention PRIVATE cxx_std_17
)
target_compile_features
(
client_fused_attention PRIVATE cxx_std_17
)
add_executable
(
client_fused_attention_bias fused_attention_bias.cpp
)
add_executable
(
client_fused_attention_bias fused_attention_bias.cpp
)
target_link_libraries
(
client_fused_attention_bias PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_fused_attention_bias PRIVATE composable_kernel::device_
other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_fused_attention_bias PRIVATE cxx_std_17
)
target_compile_features
(
client_fused_attention_bias PRIVATE cxx_std_17
)
client_example/09_quantization/CMakeLists.txt
View file @
bc5b84b1
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_executable
(
client_conv2d_fwd_bias_tanh_perchannel_quantization conv2d_fwd_bias_tanh_perchannel_quantization.cpp
)
add_executable
(
client_conv2d_fwd_bias_tanh_perchannel_quantization conv2d_fwd_bias_tanh_perchannel_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_bias_tanh_perchannel_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_bias_tanh_perchannel_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_bias_tanh_perchangel_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_bias_tanh_perchangel_quantization PRIVATE cxx_std_17
)
add_executable
(
client_conv2d_fwd_bias_relu_perchannel_quantization conv2d_fwd_bias_relu_perchannel_quantization.cpp
)
add_executable
(
client_conv2d_fwd_bias_relu_perchannel_quantization conv2d_fwd_bias_relu_perchannel_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE cxx_std_17
)
add_executable
(
client_conv2d_fwd_bias_tanh_perlayer_quantization conv2d_fwd_bias_tanh_perlayer_quantization.cpp
)
add_executable
(
client_conv2d_fwd_bias_tanh_perlayer_quantization conv2d_fwd_bias_tanh_perlayer_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE cxx_std_17
)
add_executable
(
client_conv2d_fwd_bias_relu_perlayer_quantization conv2d_fwd_bias_relu_perlayer_quantization.cpp
)
add_executable
(
client_conv2d_fwd_bias_relu_perlayer_quantization conv2d_fwd_bias_relu_perlayer_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE cxx_std_17
)
add_executable
(
client_conv2d_fwd_perchannel_quantization conv2d_fwd_perchannel_quantization.cpp
)
add_executable
(
client_conv2d_fwd_perchannel_quantization conv2d_fwd_perchannel_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_perchannel_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_perchannel_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_perchannel_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_perchannel_quantization PRIVATE cxx_std_17
)
add_executable
(
client_conv2d_fwd_perlayer_quantization conv2d_fwd_perlayer_quantization.cpp
)
add_executable
(
client_conv2d_fwd_perlayer_quantization conv2d_fwd_perlayer_quantization.cpp
)
target_link_libraries
(
client_conv2d_fwd_perlayer_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv2d_fwd_perlayer_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_conv2d_fwd_perlayer_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_conv2d_fwd_perlayer_quantization PRIVATE cxx_std_17
)
add_executable
(
client_gemm_quantization gemm_quantization.cpp
)
add_executable
(
client_gemm_quantization gemm_quantization.cpp
)
target_link_libraries
(
client_gemm_quantization PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_quantization PRIVATE composable_kernel::device_
conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_
operations
)
target_compile_features
(
client_gemm_quantization PRIVATE cxx_std_17
)
target_compile_features
(
client_gemm_quantization PRIVATE cxx_std_17
)
endif
()
endif
()
client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp
)
add_executable
(
client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp
)
target_link_libraries
(
client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_
conv_
operations
)
add_executable
(
client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp
)
add_executable
(
client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp
)
target_link_libraries
(
client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_
conv_
operations
)
add_executable
(
client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp
)
add_executable
(
client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp
)
target_link_libraries
(
client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 PRIVATE composable_kernel::device_
conv_
operations
)
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_fp16.cpp
)
target_link_libraries
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_
conv_
operations
)
target_compile_features
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE cxx_std_17
)
target_compile_features
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE cxx_std_17
)
add_executable
(
client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp
)
target_link_libraries
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_
conv_
operations
)
target_compile_features
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE cxx_std_17
)
target_compile_features
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE cxx_std_17
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_
conv_
operations
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE cxx_std_17
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE cxx_std_17
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_
conv_
operations
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE cxx_std_17
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE cxx_std_17
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE composable_kernel::device_
conv_
operations
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE cxx_std_17
)
target_compile_features
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE cxx_std_17
)
client_example/12_elementwise_normalization/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_elementwise_layernorm2d elementwise_layernorm2d.cpp
)
add_executable
(
client_elementwise_layernorm2d elementwise_layernorm2d.cpp
)
target_link_libraries
(
client_elementwise_layernorm2d PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_elementwise_layernorm2d PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_elementwise_layernorm2d PRIVATE cxx_std_17
)
target_compile_features
(
client_elementwise_layernorm2d PRIVATE cxx_std_17
)
client_example/13_batchnorm/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_batchnorm_fwd_nhwc batchnorm_fwd_nhwc.cpp
)
add_executable
(
client_batchnorm_fwd_nhwc batchnorm_fwd_nhwc.cpp
)
target_link_libraries
(
client_batchnorm_fwd_nhwc PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_batchnorm_fwd_nhwc PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_batchnorm_fwd_nhwc PRIVATE cxx_std_17
)
target_compile_features
(
client_batchnorm_fwd_nhwc PRIVATE cxx_std_17
)
add_executable
(
client_batchnorm_bwd_nhwc batchnorm_bwd_nhwc.cpp
)
add_executable
(
client_batchnorm_bwd_nhwc batchnorm_bwd_nhwc.cpp
)
target_link_libraries
(
client_batchnorm_bwd_nhwc PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_batchnorm_bwd_nhwc PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_batchnorm_bwd_nhwc PRIVATE cxx_std_17
)
target_compile_features
(
client_batchnorm_bwd_nhwc PRIVATE cxx_std_17
)
add_executable
(
client_batchnorm_infer_nhwc batchnorm_infer_nhwc.cpp
)
add_executable
(
client_batchnorm_infer_nhwc batchnorm_infer_nhwc.cpp
)
target_link_libraries
(
client_batchnorm_infer_nhwc PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_batchnorm_infer_nhwc PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_batchnorm_infer_nhwc PRIVATE cxx_std_17
)
target_compile_features
(
client_batchnorm_infer_nhwc PRIVATE cxx_std_17
)
client_example/14_instance_id/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_batchnorm_fwd_instance_id batchnorm_fwd_instance_id.cpp
)
add_executable
(
client_batchnorm_fwd_instance_id batchnorm_fwd_instance_id.cpp
)
target_link_libraries
(
client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_batchnorm_fwd_instance_id PRIVATE composable_kernel::device_
other_
operations
)
target_compile_features
(
client_batchnorm_fwd_instance_id PRIVATE cxx_std_17
)
target_compile_features
(
client_batchnorm_fwd_instance_id PRIVATE cxx_std_17
)
client_example/15_convnd_bwd_data/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp
)
add_executable
(
client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp
)
add_executable
(
client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp
)
add_executable
(
client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp
)
target_link_libraries
(
client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_
conv_
operations
)
target_link_libraries
(
client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_
conv_
operations
)
client_example/15_gemm_add_multiply/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_gemm_add_multiply gemm_add_multiply.cpp
)
add_executable
(
client_gemm_add_multiply gemm_add_multiply.cpp
)
target_link_libraries
(
client_gemm_add_multiply PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_gemm_add_multiply PRIVATE composable_kernel::device_gemm_operations
)
\ No newline at end of file
\ No newline at end of file
client_example/15_reduce/CMakeLists.txt
View file @
bc5b84b1
add_executable
(
client_reduce_nhwc_c reduce_nhwc_c.cpp
)
add_executable
(
client_reduce_nhwc_c reduce_nhwc_c.cpp
)
target_link_libraries
(
client_reduce_nhwc_c PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_reduce_nhwc_c PRIVATE composable_kernel::device_
reduction_
operations
)
client_example/16_convnd_fwd/CMakeLists.txt
View file @
bc5b84b1
if
((
DTYPES MATCHES
"fp16"
)
OR NOT DEFINED DTYPES
)
if
((
DTYPES MATCHES
"fp16"
)
OR NOT DEFINED DTYPES
)
add_executable
(
client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp
)
add_executable
(
client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp
)
target_link_libraries
(
client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_
conv_
operations
)
endif
()
endif
()
if
((
DTYPES MATCHES
"fp8"
)
OR NOT DEFINED DTYPES
)
if
((
DTYPES MATCHES
"fp8"
)
OR NOT DEFINED DTYPES
)
add_executable
(
client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp
)
add_executable
(
client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp
)
target_link_libraries
(
client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_
conv_
operations
)
endif
()
endif
()
if
((
DTYPES MATCHES
"fp32"
)
OR NOT DEFINED DTYPES
)
if
((
DTYPES MATCHES
"fp32"
)
OR NOT DEFINED DTYPES
)
add_executable
(
client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp
)
add_executable
(
client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp
)
target_link_libraries
(
client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_
conv_
operations
)
endif
()
endif
()
Prev
1
2
3
4
Next
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
