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_ROCM
Commits
9f1b4276
Commit
9f1b4276
authored
Apr 04, 2024
by
Jakub Piasecki
Browse files
resolved conflicts
parents
711857c4
c7010716
Changes
198
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
43 additions
and
60 deletions
+43
-60
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16_xdl.cpp
.../test_batched_gemm_bias_softmax_gemm_permute_fp16_xdl.cpp
+0
-0
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16_xdl.cpp
...rmute/test_batched_gemm_softmax_gemm_permute_bf16_xdl.cpp
+0
-0
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16_xdl.cpp
...rmute/test_batched_gemm_softmax_gemm_permute_fp16_xdl.cpp
+0
-0
test/contraction/CMakeLists.txt
test/contraction/CMakeLists.txt
+9
-12
test/contraction/test_contraction_interface_xdl.cpp
test/contraction/test_contraction_interface_xdl.cpp
+0
-0
test/contraction/test_contraction_xdl.cpp
test/contraction/test_contraction_xdl.cpp
+0
-0
test/convnd_bwd_data/CMakeLists.txt
test/convnd_bwd_data/CMakeLists.txt
+3
-8
test/convnd_bwd_data/convnd_bwd_data_xdl.cpp
test/convnd_bwd_data/convnd_bwd_data_xdl.cpp
+0
-0
test/convnd_fwd/CMakeLists.txt
test/convnd_fwd/CMakeLists.txt
+3
-8
test/convnd_fwd/convnd_fwd_xdl.cpp
test/convnd_fwd/convnd_fwd_xdl.cpp
+0
-0
test/gemm_add/CMakeLists.txt
test/gemm_add/CMakeLists.txt
+16
-8
test/gemm_add/test_gemm_add_fastgelu_xdl.cpp
test/gemm_add/test_gemm_add_fastgelu_xdl.cpp
+1
-1
test/gemm_add/test_gemm_add_relu_xdl.cpp
test/gemm_add/test_gemm_add_relu_xdl.cpp
+1
-1
test/gemm_add/test_gemm_add_silu_xdl.cpp
test/gemm_add/test_gemm_add_silu_xdl.cpp
+1
-1
test/gemm_add/test_gemm_add_xdl.hpp
test/gemm_add/test_gemm_add_xdl.hpp
+0
-0
test/gemm_layernorm/CMakeLists.txt
test/gemm_layernorm/CMakeLists.txt
+6
-13
test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16_xdl.cpp
...m_layernorm/test_gemm_add_relu_add_layernorm_fp16_xdl.cpp
+0
-0
test/gemm_reduce/CMakeLists.txt
test/gemm_reduce/CMakeLists.txt
+1
-1
test/gemm_reduce/gemm_reduce_fp16_xdl.cpp
test/gemm_reduce/gemm_reduce_fp16_xdl.cpp
+0
-0
test/gemm_split_k/CMakeLists.txt
test/gemm_split_k/CMakeLists.txt
+2
-7
No files found.
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
→
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16
_xdl
.cpp
View file @
9f1b4276
File moved
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
→
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16
_xdl
.cpp
View file @
9f1b4276
File moved
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
→
test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16
_xdl
.cpp
View file @
9f1b4276
File moved
test/contraction/CMakeLists.txt
View file @
9f1b4276
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
if
((
DTYPES MATCHES
"fp32"
OR DTYPES MATCHES
"fp64"
)
OR NOT DEFINED DTYPES
)
add_gtest_executable
(
test_contraction test_contraction.cpp
)
target_link_libraries
(
test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance
)
add_gtest_executable
(
test_contraction_interface test_contraction_interface.cpp
)
target_link_libraries
(
test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance
)
set
(
target 1
)
endif
()
if
((
DTYPES MATCHES
"fp32"
OR DTYPES MATCHES
"fp64"
)
OR NOT DEFINED DTYPES
)
add_gtest_executable
(
test_contraction test_contraction_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance
)
endif
()
endforeach
()
add_gtest_executable
(
test_contraction_interface test_contraction_interface_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance
)
endif
()
endif
()
test/contraction/test_contraction_interface.cpp
→
test/contraction/test_contraction_interface
_xdl
.cpp
View file @
9f1b4276
File moved
test/contraction/test_contraction.cpp
→
test/contraction/test_contraction
_xdl
.cpp
View file @
9f1b4276
File moved
test/convnd_bwd_data/CMakeLists.txt
View file @
9f1b4276
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_gtest_executable
(
test_convnd_bwd_data convnd_bwd_data.cpp
)
add_gtest_executable
(
test_convnd_bwd_data convnd_bwd_data_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance
)
set
(
target 1
)
endif
()
endforeach
()
\ No newline at end of file
endif
()
test/convnd_bwd_data/convnd_bwd_data.cpp
→
test/convnd_bwd_data/convnd_bwd_data
_xdl
.cpp
View file @
9f1b4276
File moved
test/convnd_fwd/CMakeLists.txt
View file @
9f1b4276
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_gtest_executable
(
test_convnd_fwd convnd_fwd.cpp
)
add_gtest_executable
(
test_convnd_fwd convnd_fwd_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance
)
set
(
target 1
)
endif
()
endforeach
()
endif
()
test/convnd_fwd/convnd_fwd.cpp
→
test/convnd_fwd/convnd_fwd
_xdl
.cpp
View file @
9f1b4276
File moved
test/gemm_add/CMakeLists.txt
View file @
9f1b4276
add_gtest_executable
(
test_gemm_add test_gemm_add.hpp
)
target_link_libraries
(
test_gemm_add PRIVATE utility device_gemm_add_instance
)
add_gtest_executable
(
test_gemm_add test_gemm_add_xdl.hpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_add PRIVATE utility device_gemm_add_instance
)
endif
()
add_gtest_executable
(
test_gemm_add_relu test_gemm_add_relu.cpp
)
target_link_libraries
(
test_gemm_add_relu PRIVATE utility device_gemm_add_instance device_gemm_add_relu_instance
)
add_gtest_executable
(
test_gemm_add_relu test_gemm_add_relu_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_add_relu PRIVATE utility device_gemm_add_instance device_gemm_add_relu_instance
)
endif
()
add_gtest_executable
(
test_gemm_add_silu test_gemm_add_silu.cpp
)
target_link_libraries
(
test_gemm_add_silu PRIVATE utility device_gemm_add_instance device_gemm_add_silu_instance
)
add_gtest_executable
(
test_gemm_add_silu test_gemm_add_silu_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_add_silu PRIVATE utility device_gemm_add_instance device_gemm_add_silu_instance
)
endif
()
add_gtest_executable
(
test_gemm_add_fastgelu test_gemm_add_fastgelu.cpp
)
target_link_libraries
(
test_gemm_add_fastgelu PRIVATE utility device_gemm_add_instance device_gemm_add_fastgelu_instance
)
add_gtest_executable
(
test_gemm_add_fastgelu test_gemm_add_fastgelu_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_add_fastgelu PRIVATE utility device_gemm_add_instance device_gemm_add_fastgelu_instance
)
endif
()
test/gemm_add/test_gemm_add_fastgelu.cpp
→
test/gemm_add/test_gemm_add_fastgelu
_xdl
.cpp
View file @
9f1b4276
...
...
@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_fastgelu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add
_xdl
.hpp"
template
<
typename
Tuple
>
class
TestGemmAddFastgelu
:
public
TestGemmAdd
<
Tuple
>
...
...
test/gemm_add/test_gemm_add_relu.cpp
→
test/gemm_add/test_gemm_add_relu
_xdl
.cpp
View file @
9f1b4276
...
...
@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_relu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add
_xdl
.hpp"
template
<
typename
Tuple
>
class
TestGemmAddRelu
:
public
TestGemmAdd
<
Tuple
>
...
...
test/gemm_add/test_gemm_add_silu.cpp
→
test/gemm_add/test_gemm_add_silu
_xdl
.cpp
View file @
9f1b4276
...
...
@@ -4,7 +4,7 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_silu_impl.hpp"
#include "test_gemm_add.hpp"
#include "test_gemm_add
_xdl
.hpp"
template
<
typename
Tuple
>
class
TestGemmAddSilu
:
public
TestGemmAdd
<
Tuple
>
...
...
test/gemm_add/test_gemm_add.hpp
→
test/gemm_add/test_gemm_add
_xdl
.hpp
View file @
9f1b4276
File moved
test/gemm_layernorm/CMakeLists.txt
View file @
9f1b4276
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_custom_target
(
test_gemm_layernorm
)
add_gtest_executable
(
test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance
)
add_dependencies
(
test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16
)
set
(
target 1
)
endif
()
endif
()
endforeach
()
add_gtest_executable
(
test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16_xdl.cpp
)
if
(
result EQUAL 0
)
add_custom_target
(
test_gemm_layernorm
)
target_link_libraries
(
test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance
)
add_dependencies
(
test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16
)
endif
()
test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
→
test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16
_xdl
.cpp
View file @
9f1b4276
File moved
test/gemm_reduce/CMakeLists.txt
View file @
9f1b4276
add_test_executable
(
test_gemm_reduce_fp16 gemm_reduce_fp16.cpp
)
add_test_executable
(
test_gemm_reduce_fp16 gemm_reduce_fp16
_xdl
.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_reduce_fp16 PRIVATE utility device_gemm_reduce_instance
)
endif
()
\ No newline at end of file
test/gemm_reduce/gemm_reduce_fp16.cpp
→
test/gemm_reduce/gemm_reduce_fp16
_xdl
.cpp
View file @
9f1b4276
File moved
test/gemm_split_k/CMakeLists.txt
View file @
9f1b4276
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_gtest_executable
(
test_gemm_splitk test_gemm_splitk.cpp
)
add_gtest_executable
(
test_gemm_splitk test_gemm_splitk_xdl.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_splitk PRIVATE utility device_gemm_splitk_instance
)
set
(
target 1
)
endif
()
endforeach
()
Prev
1
…
5
6
7
8
9
10
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