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
8ed29492
Commit
8ed29492
authored
Sep 21, 2023
by
Rostyslav Geyyer
Browse files
Merge branch 'develop' into lwpck-929
parents
560919ab
bba085d2
Changes
128
Hide whitespace changes
Inline
Side-by-side
Showing
8 changed files
with
42 additions
and
47 deletions
+42
-47
test/elementwise_normalization/CMakeLists.txt
test/elementwise_normalization/CMakeLists.txt
+3
-3
test/gemm/CMakeLists.txt
test/gemm/CMakeLists.txt
+14
-16
test/gemm_layernorm/CMakeLists.txt
test/gemm_layernorm/CMakeLists.txt
+5
-5
test/gemm_reduce/CMakeLists.txt
test/gemm_reduce/CMakeLists.txt
+3
-4
test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
...uped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+2
-2
test/grouped_gemm/CMakeLists.txt
test/grouped_gemm/CMakeLists.txt
+0
-2
test/normalization/CMakeLists.txt
test/normalization/CMakeLists.txt
+13
-11
test/reduce/CMakeLists.txt
test/reduce/CMakeLists.txt
+2
-4
No files found.
test/elementwise_normalization/CMakeLists.txt
View file @
8ed29492
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_custom_target
(
test_elementwise_normalization
)
add_
custom_target
(
test_elementwise_
normalization
)
add_
gtest_executable
(
test_elementwise_
layernorm_fp16 test_elementwise_layernorm_fp16.cpp
)
add_gtest_executable
(
test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance
)
target_link_libraries
(
test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance
)
add_dependencies
(
test_elementwise_normalization test_elementwise_layernorm_fp16
)
add_dependencies
(
test_elementwise_normalization test_elementwise_layernorm_fp16
)
endif
()
endif
()
\ No newline at end of file
test/gemm/CMakeLists.txt
View file @
8ed29492
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_test_executable
(
test_gemm_fp32 gemm_fp32.cpp
)
add_test_executable
(
test_gemm_fp32 gemm_fp32.cpp
)
target_link_libraries
(
test_gemm_fp32 PRIVATE utility
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_fp32 PRIVATE device_gemm_instance
)
target_link_libraries
(
test_gemm_fp32 PRIVATE
utility
device_gemm_instance
)
endif
()
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_test_executable
(
test_gemm_fp16 gemm_fp16.cpp
)
add_test_executable
(
test_gemm_fp16 gemm_fp16.cpp
)
target_link_libraries
(
test_gemm_fp16 PRIVATE utility
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_fp16 PRIVATE device_gemm_instance
)
target_link_libraries
(
test_gemm_fp16 PRIVATE
utility
device_gemm_instance
)
add_library
(
gemm_standalone_xdl_fp16_instances STATIC
add_library
(
gemm_standalone_xdl_fp16_instances STATIC
instance/gemm_f16_nn_instance.cpp
instance/gemm_f16_nn_instance.cpp
instance/gemm_f16_nt_instance.cpp
instance/gemm_f16_nt_instance.cpp
instance/gemm_f16_tn_instance.cpp
instance/gemm_f16_tn_instance.cpp
instance/gemm_wavelet_f16_tn_instance.cpp
instance/gemm_wavelet_f16_tn_instance.cpp
instance/gemm_f16_tt_instance.cpp
instance/gemm_f16_tt_instance.cpp
)
)
endif
()
add_test_executable
(
test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp
)
add_test_executable
(
test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp
)
target_link_libraries
(
test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility
)
if
(
result EQUAL 0
)
target_include_directories
(
test_gemm_standalone_xdl_fp16 PRIVATE instance/
)
target_link_libraries
(
test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility
)
target_include_directories
(
test_gemm_standalone_xdl_fp16 PRIVATE instance/
)
endif
()
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_test_executable
(
test_gemm_bf16 gemm_bf16.cpp
)
add_test_executable
(
test_gemm_bf16 gemm_bf16.cpp
)
target_link_libraries
(
test_gemm_bf16 PRIVATE utility
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_bf16 PRIVATE device_gemm_instance
)
target_link_libraries
(
test_gemm_bf16 PRIVATE
utility
device_gemm_instance
)
endif
()
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_test_executable
(
test_gemm_int8 gemm_int8.cpp
)
add_test_executable
(
test_gemm_int8 gemm_int8.cpp
)
target_link_libraries
(
test_gemm_int8 PRIVATE utility
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_int8 PRIVATE device_gemm_instance
)
target_link_libraries
(
test_gemm_int8 PRIVATE
utility
device_gemm_instance
)
endif
()
endif
()
\ No newline at end of file
test/gemm_layernorm/CMakeLists.txt
View file @
8ed29492
...
@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
...
@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_custom_target
(
test_gemm_layernorm
)
add_custom_target
(
test_gemm_layernorm
)
add_gtest_executable
(
test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp
)
add_gtest_executable
(
test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp
)
target_link_libraries
(
test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance
)
if
(
result EQUAL 0
)
add_dependencies
(
test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16
)
target_link_libraries
(
test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance
)
set
(
target 1
)
add_dependencies
(
test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16
)
endif
()
set
(
target 1
)
endif
()
endif
()
endif
()
endforeach
()
endforeach
()
test/gemm_reduce/CMakeLists.txt
View file @
8ed29492
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_test_executable
(
test_gemm_reduce_fp16 gemm_reduce_fp16.cpp
)
add_test_executable
(
test_gemm_reduce_fp16 gemm_reduce_fp16.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_gemm_reduce_fp16 PRIVATE utility
)
target_link_libraries
(
test_gemm_reduce_fp16 PRIVATE utility device_gemm_reduce_instance
)
target_link_libraries
(
test_gemm_reduce_fp16 PRIVATE device_gemm_reduce_instance
)
endif
()
endif
()
\ No newline at end of file
test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
View file @
8ed29492
...
@@ -33,8 +33,8 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
...
@@ -33,8 +33,8 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
bool
skip_case
(
const
ck
::
utils
::
conv
::
ConvParam
&
params
,
const
ck
::
index_t
split_k
)
bool
skip_case
(
const
ck
::
utils
::
conv
::
ConvParam
&
params
,
const
ck
::
index_t
split_k
)
{
{
// K or C
are odd is
supported only by DL kernel (only applies to fp16)
//
Odd
K or C
values are
supported only by DL kernel (only applies to fp16)
// DL kernel
is on
ly support
ed for
split_k=1
// DL kernel
current
ly support
s only `
split_k=1
`
if
constexpr
(
std
::
is_same_v
<
InDataType
,
ck
::
half_t
>
)
if
constexpr
(
std
::
is_same_v
<
InDataType
,
ck
::
half_t
>
)
{
{
if
(
split_k
!=
1
&&
(
params
.
K_
%
2
!=
0
||
params
.
C_
%
2
!=
0
))
if
(
split_k
!=
1
&&
(
params
.
K_
%
2
!=
0
||
params
.
C_
%
2
!=
0
))
...
...
test/grouped_gemm/CMakeLists.txt
View file @
8ed29492
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
@@ -13,4 +12,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
...
@@ -13,4 +12,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
set
(
target 1
)
set
(
target 1
)
endif
()
endif
()
endforeach
()
endforeach
()
endif
()
test/normalization/CMakeLists.txt
View file @
8ed29492
if
(
DTYPES MATCHES
"fp16"
OR DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_custom_target
(
test_normalization
)
add_custom_target
(
test_normalization
)
add_gtest_executable
(
test_layernorm2d_fp32 test_layernorm2d_fp32.cpp
)
endif
()
if
(
result EQUAL 0
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_gtest_executable
(
test_layernorm2d_fp32 test_layernorm2d_fp32.cpp
)
add_gtest_executable
(
test_groupnorm_fp32 test_groupnorm_fp32.cpp
)
target_link_libraries
(
test_layernorm2d_fp32 PRIVATE utility device_normalization_instance
)
target_link_libraries
(
test_layernorm2d_fp32 PRIVATE utility device_normalization_instance
)
target_link_libraries
(
test_groupnorm_fp32 PRIVATE utility device_normalization_instance
)
add_dependencies
(
test_normalization test_layernorm2d_fp32
)
add_dependencies
(
test_normalization test_layernorm2d_fp32
)
endif
()
add_gtest_executable
(
test_groupnorm_fp32 test_groupnorm_fp32.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_groupnorm_fp32 PRIVATE utility device_normalization_instance
)
add_dependencies
(
test_normalization test_groupnorm_fp32
)
add_dependencies
(
test_normalization test_groupnorm_fp32
)
endif
()
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_gtest_executable
(
test_layernorm2d_fp16 test_layernorm2d_fp16.cpp
)
add_gtest_executable
(
test_layernorm2d_fp16 test_layernorm2d_fp16.cpp
)
if
(
result EQUAL 0
)
add_gtest_executable
(
test_groupnorm_fp16 test_groupnorm_fp16.cpp
)
target_link_libraries
(
test_layernorm2d_fp16 PRIVATE utility device_normalization_instance
)
target_link_libraries
(
test_layernorm2d_fp16 PRIVATE utility device_normalization_instance
)
target_link_libraries
(
test_groupnorm_fp16 PRIVATE utility device_normalization_instance
)
add_dependencies
(
test_normalization test_layernorm2d_fp16
)
add_dependencies
(
test_normalization test_layernorm2d_fp16
)
endif
()
add_gtest_executable
(
test_groupnorm_fp16 test_groupnorm_fp16.cpp
)
if
(
result EQUAL 0
)
target_link_libraries
(
test_groupnorm_fp16 PRIVATE utility device_normalization_instance
)
add_dependencies
(
test_normalization test_groupnorm_fp16
)
add_dependencies
(
test_normalization test_groupnorm_fp16
)
endif
()
endif
()
test/reduce/CMakeLists.txt
View file @
8ed29492
add_test_executable
(
test_reduce_no_index reduce_no_index.cpp
)
add_test_executable
(
test_reduce_no_index reduce_no_index.cpp
)
add_test_executable
(
test_reduce_with_index reduce_with_index.cpp
)
add_test_executable
(
test_reduce_with_index reduce_with_index.cpp
)
target_link_libraries
(
test_reduce_no_index PRIVATE utility
)
target_link_libraries
(
test_reduce_no_index PRIVATE utility device_reduce_instance
)
target_link_libraries
(
test_reduce_no_index PRIVATE device_reduce_instance
)
target_link_libraries
(
test_reduce_with_index PRIVATE utility device_reduce_instance
)
target_link_libraries
(
test_reduce_with_index PRIVATE utility
)
target_link_libraries
(
test_reduce_with_index PRIVATE device_reduce_instance
)
Prev
1
…
3
4
5
6
7
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
