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
e60c5aea
Unverified
Commit
e60c5aea
authored
Feb 15, 2024
by
Illia Silin
Committed by
GitHub
Feb 15, 2024
Browse files
Merge pull request #36 from ROCm/lwpck-1299
Initial MI350 enablement.
parents
29dcb956
63df00cd
Changes
46
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
23 additions
and
23 deletions
+23
-23
example/01_gemm/CMakeLists.txt
example/01_gemm/CMakeLists.txt
+1
-1
example/02_gemm_bilinear/CMakeLists.txt
example/02_gemm_bilinear/CMakeLists.txt
+1
-1
example/03_gemm_bias_relu/CMakeLists.txt
example/03_gemm_bias_relu/CMakeLists.txt
+1
-1
example/04_gemm_add_add_fastgelu/CMakeLists.txt
example/04_gemm_add_add_fastgelu/CMakeLists.txt
+2
-2
example/09_convnd_fwd/CMakeLists.txt
example/09_convnd_fwd/CMakeLists.txt
+1
-1
example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
...e/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+1
-1
example/14_gemm_quantization/CMakeLists.txt
example/14_gemm_quantization/CMakeLists.txt
+1
-1
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+1
-1
example/17_convnd_bwd_data/CMakeLists.txt
example/17_convnd_bwd_data/CMakeLists.txt
+1
-1
example/18_batched_gemm_reduce/CMakeLists.txt
example/18_batched_gemm_reduce/CMakeLists.txt
+1
-1
example/20_grouped_conv_bwd_weight/CMakeLists.txt
example/20_grouped_conv_bwd_weight/CMakeLists.txt
+1
-1
example/21_gemm_layernorm/CMakeLists.txt
example/21_gemm_layernorm/CMakeLists.txt
+1
-1
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+1
-1
example/31_batched_gemm_gemm/CMakeLists.txt
example/31_batched_gemm_gemm/CMakeLists.txt
+2
-2
example/35_splitK_gemm/CMakeLists.txt
example/35_splitK_gemm/CMakeLists.txt
+1
-1
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+1
-1
example/40_conv2d_fwd_quantization/CMakeLists.txt
example/40_conv2d_fwd_quantization/CMakeLists.txt
+1
-1
example/41_grouped_conv_conv_fwd/CMakeLists.txt
example/41_grouped_conv_conv_fwd/CMakeLists.txt
+2
-2
example/44_elementwise_permute/CMakeLists.txt
example/44_elementwise_permute/CMakeLists.txt
+1
-1
example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
+1
-1
No files found.
example/01_gemm/CMakeLists.txt
View file @
e60c5aea
...
@@ -59,7 +59,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
...
@@ -59,7 +59,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
add_example_executable
(
example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp
)
add_example_executable
(
example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp
)
add_example_dependencies
(
example_gemm_xdl example_gemm_xdl_fp8_bf8
)
add_example_dependencies
(
example_gemm_xdl example_gemm_xdl_fp8_bf8
)
list
(
APPEND gpu_list gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/02_gemm_bilinear/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list1 gfx1100 gfx1101 gfx1102
)
list
(
APPEND gpu_list1 gfx1100 gfx1101 gfx1102
)
list
(
APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list1 AND target EQUAL 0
)
if
(
gpu IN_LIST gpu_list1 AND target EQUAL 0
)
...
...
example/03_gemm_bias_relu/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/04_gemm_add_add_fastgelu/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
@@ -33,4 +33,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
...
@@ -33,4 +33,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
add_example_dependencies
(
example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_lds_direct_load_fp32
)
add_example_dependencies
(
example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_lds_direct_load_fp32
)
set
(
target 1
)
set
(
target 1
)
endif
()
endif
()
endforeach
()
endforeach
()
\ No newline at end of file
example/09_convnd_fwd/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/14_gemm_quantization/CMakeLists.txt
View file @
e60c5aea
# dlops
# dlops
add_example_executable
(
example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp
)
add_example_executable
(
example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp
)
# xdlops
# xdlops
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/17_convnd_bwd_data/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/18_batched_gemm_reduce/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/20_grouped_conv_bwd_weight/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
list
(
APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102
)
list
(
APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
example/21_gemm_layernorm/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
list
(
APPEND gpu_list2 gfx1100 gfx1101 gfx1102
)
list
(
APPEND gpu_list2 gfx1100 gfx1101 gfx1102
)
set
(
target 0
)
set
(
target 0
)
...
...
example/31_batched_gemm_gemm/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
...
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif
()
endif
()
endforeach
()
endforeach
()
if
(
NOT GPU_TARGETS MATCHES
"gfx94"
AND NOT GPU_TARGETS MATCHES
"gfx1"
)
if
(
NOT GPU_TARGETS MATCHES
"gfx94"
AND NOT GPU_TARGETS MATCHES
"gfx95"
AND NOT GPU_TARGETS MATCHES
"gfx1"
)
add_example_executable
(
example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp
)
add_example_executable
(
example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp
)
endif
()
endif
()
example/35_splitK_gemm/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
list
(
APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102
)
list
(
APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
example/40_conv2d_fwd_quantization/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
example/41_grouped_conv_conv_fwd/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
list
(
APPEND gpu_list2 gfx908 gfx90a
)
list
(
APPEND gpu_list2 gfx908 gfx90a
)
set
(
target 0
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
...
@@ -13,6 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
endif
()
endif
()
endforeach
()
endforeach
()
if
(
NOT GPU_TARGETS MATCHES
"gfx94"
AND NOT GPU_TARGETS MATCHES
"gfx1"
)
if
(
NOT GPU_TARGETS MATCHES
"gfx94"
AND NOT GPU_TARGETS MATCHES
"gfx95"
AND NOT GPU_TARGETS MATCHES
"gfx1"
)
add_example_executable
(
example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp
)
add_example_executable
(
example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp
)
endif
()
endif
()
example/44_elementwise_permute/CMakeLists.txt
View file @
e60c5aea
...
@@ -5,6 +5,6 @@ add_example_executable(example_elementwise_permute_4D_fp16_row elementwise_permu
...
@@ -5,6 +5,6 @@ add_example_executable(example_elementwise_permute_4D_fp16_row elementwise_permu
add_example_executable
(
example_elementwise_permute_4D_fp32_col elementwise_permute_4D_fp32_col.cpp
)
add_example_executable
(
example_elementwise_permute_4D_fp32_col elementwise_permute_4D_fp32_col.cpp
)
add_example_executable
(
example_elementwise_permute_4D_fp16_col elementwise_permute_4D_fp16_col.cpp
)
add_example_executable
(
example_elementwise_permute_4D_fp16_col elementwise_permute_4D_fp16_col.cpp
)
add_example_executable
(
example_elementwise_permute elementwise_permute.cpp
)
add_example_executable
(
example_elementwise_permute elementwise_permute.cpp
)
if
((
NOT GPU_TARGETS MATCHES
"gfx940"
)
AND
(
NOT GPU_TARGETS MATCHES
"gfx941"
)
AND
(
NOT GPU_TARGETS MATCHES
"gfx942"
))
if
((
NOT GPU_TARGETS MATCHES
"gfx940"
)
AND
(
NOT GPU_TARGETS MATCHES
"gfx941"
)
AND
(
NOT GPU_TARGETS MATCHES
"gfx942"
)
AND
(
NOT GPU_TARGETS MATCHES
"gfx950"
)
)
add_example_executable
(
example_elementwise_permute_3d elementwise_permute_3d.cpp
)
add_example_executable
(
example_elementwise_permute_3d elementwise_permute_3d.cpp
)
endif
()
endif
()
example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
View file @
e60c5aea
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
gfx950
)
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
)
...
...
Prev
1
2
3
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
