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
b93575ca
Commit
b93575ca
authored
Aug 28, 2023
by
Jing Zhang
Browse files
merge develop
parents
54df59bf
c8a8385f
Changes
352
Show whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
185 additions
and
118 deletions
+185
-118
example/17_convnd_bwd_data/CMakeLists.txt
example/17_convnd_bwd_data/CMakeLists.txt
+6
-2
example/18_batched_gemm_reduce/CMakeLists.txt
example/18_batched_gemm_reduce/CMakeLists.txt
+2
-0
example/20_grouped_conv_bwd_weight/CMakeLists.txt
example/20_grouped_conv_bwd_weight/CMakeLists.txt
+15
-11
example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
...d_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+12
-13
example/21_gemm_layernorm/CMakeLists.txt
example/21_gemm_layernorm/CMakeLists.txt
+2
-0
example/22_cgemm/CMakeLists.txt
example/22_cgemm/CMakeLists.txt
+15
-10
example/24_batched_gemm/CMakeLists.txt
example/24_batched_gemm/CMakeLists.txt
+16
-12
example/25_gemm_bias_e_permute/CMakeLists.txt
example/25_gemm_bias_e_permute/CMakeLists.txt
+4
-2
example/26_contraction/CMakeLists.txt
example/26_contraction/CMakeLists.txt
+8
-5
example/27_layernorm/CMakeLists.txt
example/27_layernorm/CMakeLists.txt
+4
-2
example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
+3
-1
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
+5
-3
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+25
-15
example/31_batched_gemm_gemm/CMakeLists.txt
example/31_batched_gemm_gemm/CMakeLists.txt
+12
-5
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
+22
-14
example/35_splitK_gemm/CMakeLists.txt
example/35_splitK_gemm/CMakeLists.txt
+16
-11
example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
+6
-5
example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
+6
-5
example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
+3
-1
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+3
-1
No files found.
example/17_convnd_bwd_data/CMakeLists.txt
View file @
b93575ca
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
@@ -7,5 +8,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
set
(
target 1
)
endif
()
endforeach
()
add_example_executable
(
example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp
)
target_link_libraries
(
example_convnd_bwd_data_dl_fp16 PRIVATE utility
)
if
(
DL_KERNELS
)
add_example_executable
(
example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp
)
target_link_libraries
(
example_convnd_bwd_data_dl_fp16 PRIVATE utility
)
endif
()
endif
()
example/18_batched_gemm_reduce/CMakeLists.txt
View file @
b93575ca
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
@@ -6,3 +7,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
set
(
target 1
)
endif
()
endforeach
()
endif
()
example/20_grouped_conv_bwd_weight/CMakeLists.txt
View file @
b93575ca
...
...
@@ -3,18 +3,22 @@ set(target 0)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_custom_target
(
example_grouped_conv_bwd_weight
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp
)
add_dependencies
(
example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp
)
add_dependencies
(
example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
example_grouped_conv_bwd_weight_xdl_bf16
)
add_dependencies
(
example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16
)
endif
()
set
(
target 1
)
endif
()
endforeach
()
add_custom_target
(
example_grouped_conv_bwd_weight_dl
)
add_example_executable
(
example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp
)
add_dependencies
(
example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
if
(
DL_KERNELS
)
add_custom_target
(
example_grouped_conv_bwd_weight_dl
)
add_example_executable
(
example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp
)
add_dependencies
(
example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16
)
endif
()
endif
()
\ No newline at end of file
example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
View file @
b93575ca
...
...
@@ -72,10 +72,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
// init to 0
wei_device_buf
.
SetZero
();
std
::
array
<
ck
::
index_t
,
NDimSpatial
>
input_spatial_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
>
filter_spatial_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
>
output_spatial_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
input_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
input_strides
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
filter_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
weights_strides
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
output_lengths
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
+
3
>
output_strides
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
>
conv_filter_strides
{};
std
::
array
<
ck
::
index_t
,
NDimSpatial
>
conv_filter_dilations
{};
...
...
@@ -84,10 +85,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
auto
range_copy
=
[](
const
auto
&
from
,
auto
to
)
{
std
::
copy
(
begin
(
from
),
end
(
from
),
to
);
};
range_copy
(
conv_param
.
input_spatial_lengths_
,
begin
(
input_spatial_lengths
));
range_copy
(
conv_param
.
filter_spatial_lengths_
,
begin
(
filter_spatial_lengths
));
range_copy
(
conv_param
.
output_spatial_lengths_
,
begin
(
output_spatial_lengths
));
range_copy
(
in_g_n_c_wis_desc
.
GetLengths
(),
begin
(
input_lengths
));
range_copy
(
in_g_n_c_wis_desc
.
GetStrides
(),
begin
(
input_strides
));
range_copy
(
wei_g_k_c_xs_desc
.
GetLengths
(),
begin
(
filter_lengths
));
range_copy
(
wei_g_k_c_xs_desc
.
GetStrides
(),
begin
(
weights_strides
));
range_copy
(
out_g_n_k_wos_desc
.
GetLengths
(),
begin
(
output_lengths
));
range_copy
(
out_g_n_k_wos_desc
.
GetStrides
(),
begin
(
output_strides
));
range_copy
(
conv_param
.
conv_filter_strides_
,
begin
(
conv_filter_strides
));
range_copy
(
conv_param
.
conv_filter_dilations_
,
begin
(
conv_filter_dilations
));
...
...
@@ -100,14 +102,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
auto
argument
=
conv
.
MakeArgument
(
static_cast
<
InDataType
*>
(
in_device_buf
.
GetDeviceBuffer
()),
static_cast
<
WeiDataType
*>
(
wei_device_buf
.
GetDeviceBuffer
()),
static_cast
<
OutDataType
*>
(
out_device_buf
.
GetDeviceBuffer
()),
conv_param
.
G_
,
conv_param
.
N_
,
conv_param
.
K_
,
conv_param
.
C_
,
input_spatial_lengths
,
filter_spatial_lengths
,
output_spatial_lengths
,
input_lengths
,
input_strides
,
filter_lengths
,
weights_strides
,
output_lengths
,
output_strides
,
conv_filter_strides
,
conv_filter_dilations
,
...
...
example/21_gemm_layernorm/CMakeLists.txt
View file @
b93575ca
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
@@ -9,3 +10,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
set
(
target 1
)
endif
()
endforeach
()
endif
()
example/22_cgemm/CMakeLists.txt
View file @
b93575ca
add_custom_target
(
example_cgemm_xdl
)
add_example_executable
(
example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp
)
add_example_executable
(
example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp
)
add_dependencies
(
example_cgemm_xdl example_cgemm_xdl_bf16
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp
)
add_dependencies
(
example_cgemm_xdl example_cgemm_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp
)
add_example_executable
(
example_cgemm_xdl_int8 cgemm_xdl_int8.cpp
)
add_dependencies
(
example_cgemm_xdl
example_cgemm_xdl_bf16
example_cgemm_xdl_fp16
example_cgemm_xdl_fp32
example_cgemm_xdl_int8
)
add_dependencies
(
example_cgemm_xdl example_cgemm_xdl_fp32
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_cgemm_xdl_int8 cgemm_xdl_int8.cpp
)
add_dependencies
(
example_cgemm_xdl example_cgemm_xdl_int8
)
endif
()
if
(
USE_BITINT_EXTENSION_INT4
)
add_example_executable
(
example_cgemm_xdl_int4 cgemm_xdl_int4.cpp
)
add_dependencies
(
example_cgemm_xdl example_cgemm_xdl_int4
)
...
...
example/24_batched_gemm/CMakeLists.txt
View file @
b93575ca
add_custom_target
(
example_batched_gemm_xdl
)
add_example_executable
(
example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp
)
add_example_executable
(
example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp
)
add_example_executable
(
example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp
)
add_dependencies
(
example_batched_gemm_xdl
example_batched_gemm_xdl_fp32
example_batched_gemm_xdl_fp16
example_batched_gemm_xdl_bfp16
example_batched_gemm_xdl_int8
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp
)
add_dependencies
(
example_batched_gemm_xdl example_batched_gemm_xdl_fp32
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp
)
add_dependencies
(
example_batched_gemm_xdl example_batched_gemm_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp
)
add_dependencies
(
example_batched_gemm_xdl example_batched_gemm_xdl_bfp16
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp
)
add_dependencies
(
example_batched_gemm_xdl example_batched_gemm_xdl_int8
)
endif
()
if
(
USE_BITINT_EXTENSION_INT4
)
add_example_executable
(
example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp
)
add_dependencies
(
example_batched_gemm_xdl example_batched_gemm_xdl_int4
)
...
...
example/25_gemm_bias_e_permute/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
)
add_example_executable
(
example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
)
add_example_executable
(
example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
)
endif
()
example/26_contraction/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp
)
add_example_executable
(
example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp
)
add_example_executable
(
example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp
)
add_example_executable
(
example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp
)
add_example_executable
(
example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp
)
add_example_executable
(
example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp
)
endif
()
example/27_layernorm/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_layernorm_fp16 layernorm_fp16.cpp
)
add_example_executable
(
example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_layernorm_fp16 layernorm_fp16.cpp
)
add_example_executable
(
example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp
)
endif
()
example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp
)
endif
()
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp
)
if
(
GPU_TARGETS MATCHES
"gfx1100"
OR GPU_TARGETS MATCHES
"gfx1101"
OR GPU_TARGETS MATCHES
"gfx1102"
)
if
(
GPU_TARGETS MATCHES
"gfx1100"
OR GPU_TARGETS MATCHES
"gfx1101"
OR GPU_TARGETS MATCHES
"gfx1102"
)
add_example_executable
(
example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp
)
endif
()
endif
()
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @
b93575ca
...
...
@@ -5,23 +5,29 @@ set(target 0)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list1 AND target EQUAL 0
)
add_custom_target
(
example_grouped_conv_fwd_multiple_d
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16
)
add_example_executable
(
example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8
)
endif
()
if
(
USE_BITINT_EXTENSION_INT4
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4
)
endif
()
# USE_BITINT_EXTENSION_INT4
add_example_executable
(
example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp
)
add_dependencies
(
example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16
)
set
(
target 1
)
endif
()
endforeach
()
...
...
@@ -29,8 +35,12 @@ endforeach()
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list2 AND target EQUAL 0
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp
)
endif
()
set
(
target 1
)
endif
()
endforeach
()
example/31_batched_gemm_gemm/CMakeLists.txt
View file @
b93575ca
...
...
@@ -3,10 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list1 AND target EQUAL 0
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp
)
endif
()
if
(
USE_BITINT_EXTENSION_INT4
)
add_example_executable
(
example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp
)
endif
(
USE_BITINT_EXTENSION_INT4
)
...
...
@@ -15,5 +20,7 @@ foreach(gpu IN LISTS GPU_TARGETS)
endforeach
()
if
(
NOT GPU_TARGETS MATCHES
"gfx94"
AND NOT GPU_TARGETS MATCHES
"gfx1"
)
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp
)
endif
()
endif
()
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
)
add_example_executable
(
example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
)
add_example_executable
(
example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
)
add_example_executable
(
example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
)
endif
()
add_custom_target
(
example_gemm_scale_softmax_gemm
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16
)
add_dependencies
(
example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16
)
endif
()
example/35_splitK_gemm/CMakeLists.txt
View file @
b93575ca
...
...
@@ -3,17 +3,22 @@ set(target 0)
foreach
(
gpu IN LISTS GPU_TARGETS
)
if
(
gpu IN_LIST gpu_list AND target EQUAL 0
)
add_custom_target
(
example_splitK_gemm_xdl
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp
)
add_dependencies
(
example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp
)
add_dependencies
(
example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp
)
add_dependencies
(
example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp
)
add_dependencies
(
example_splitK_gemm_xdl
example_splitK_gemm_xdl_fp32
example_splitK_gemm_xdl_fp16
example_splitK_gemm_xdl_bfp16
example_splitK_gemm_xdl_int8
)
add_dependencies
(
example_splitK_gemm_xdl example_splitK_gemm_xdl_int8
)
endif
()
if
(
USE_BITINT_EXTENSION_INT4
)
add_example_executable
(
example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp
)
add_dependencies
(
example_splitK_gemm_xdl example_splitK_gemm_xdl_int4
)
...
...
example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
View file @
b93575ca
...
...
@@ -33,6 +33,7 @@ using ADataType = BF16;
using
BDataType
=
BF16
;
using
AccDataType
=
F32
;
using
CDataType
=
F32
;
using
ComputeType
=
BF16
;
using
ALayout
=
Row
;
using
BLayout
=
Col
;
...
...
@@ -46,11 +47,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
DeviceGemmInstance
=
ck
::
tensor_operation
::
device
::
DeviceGemmXdlSplitKCShuffle
// clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
<
ADataType
,
BDataType
,
CDataType
,
AccDataType
,
ALayout
,
BLayout
,
CLayout
,
AElementOp
,
BElementOp
,
CElementOp
,
GemmDefault
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
3
,
8
,
8
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
;
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
Compute|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
Type|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
<
ADataType
,
BDataType
,
CDataType
,
AccDataType
,
ALayout
,
BLayout
,
CLayout
,
AElementOp
,
BElementOp
,
CElementOp
,
GemmDefault
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
3
,
8
,
8
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
,
ComputeType
>
;
// clang-format on
#include "run_splitK_gemm_example.inc"
...
...
example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
View file @
b93575ca
...
...
@@ -30,6 +30,7 @@ using ADataType = int8_t;
using
BDataType
=
int8_t
;
using
AccDataType
=
int32_t
;
using
CDataType
=
int32_t
;
using
ComputeType
=
int8_t
;
using
ALayout
=
Row
;
using
BLayout
=
Col
;
...
...
@@ -43,11 +44,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
DeviceGemmInstance
=
ck
::
tensor_operation
::
device
::
DeviceGemmXdlSplitKCShuffle
// clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
<
ADataType
,
BDataType
,
CDataType
,
AccDataType
,
ALayout
,
BLayout
,
CLayout
,
AElementOp
,
BElementOp
,
CElementOp
,
GemmDefault
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
16
,
16
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
3
,
16
,
16
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
;
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
Compute|
//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
Type|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
<
ADataType
,
BDataType
,
CDataType
,
AccDataType
,
ALayout
,
BLayout
,
CLayout
,
AElementOp
,
BElementOp
,
CElementOp
,
GemmDefault
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
16
,
16
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
3
,
16
,
16
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
,
ComputeType
>
;
// clang-format on
#include "run_splitK_gemm_example.inc"
...
...
example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
View file @
b93575ca
add_example_executable
(
example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_example_executable
(
example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
)
endif
()
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
View file @
b93575ca
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942
)
set
(
target 0
)
foreach
(
gpu IN LISTS GPU_TARGETS
)
...
...
@@ -11,3 +12,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
set
(
target 1
)
endif
()
endforeach
()
endif
()
Prev
1
2
3
4
5
6
7
…
18
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
