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
9b7093aa
Commit
9b7093aa
authored
Sep 21, 2023
by
Rostyslav Geyyer
Browse files
Merge branch 'develop' into lwpck-791
parents
4cb63955
bba085d2
Changes
109
Expand all
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
264 additions
and
326 deletions
+264
-326
library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
...r_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
...peration_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
+20
-24
library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
...operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
...ion_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
...ance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+5
-9
library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
...peration_instance/gpu/contraction_bilinear/CMakeLists.txt
+12
-15
library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
...r_operation_instance/gpu/contraction_scale/CMakeLists.txt
+12
-14
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
...sor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
+8
-21
library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
...c/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
+6
-14
library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
...ion_instance/gpu/elementwise_normalization/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
...ary/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+93
-107
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
...evice_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
+14
-14
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
...evice_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
+14
-14
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
...evice_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+14
-14
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
...m/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+22
-22
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
...m/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+22
-22
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
...m/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+22
-22
library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
...eration_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
...r_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
...n_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
+0
-2
No files found.
library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_batched_gemm_gemm_instance
add_instance_library
(
device_batched_gemm_gemm_instance
device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
View file @
9b7093aa
# ONLY DL_KERNELS
# ONLY DL_KERNELS
if
(
DL_KERNELS
)
set
(
BATCHED_GEMM_MULTID_INSTANCES
)
set
(
BATCHED_GEMM_MULTID_INSTANCES
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
)
endif
()
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
)
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
)
add_instance_library
(
device_batched_gemm_multi_d_instance
${
BATCHED_GEMM_MULTID_INSTANCES
}
)
list
(
APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
)
endif
()
add_instance_library
(
device_batched_gemm_multi_d_instance
${
BATCHED_GEMM_MULTID_INSTANCES
}
)
endif
()
library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_batched_gemm_reduce_instance
add_instance_library
(
device_batched_gemm_reduce_instance
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_batched_gemm_softmax_gemm_instance
add_instance_library
(
device_batched_gemm_softmax_gemm_instance
device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
View file @
9b7093aa
set
(
DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
)
set
(
DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
list
(
APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
)
device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
list
(
APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
)
device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
endif
()
device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
)
list
(
APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
)
list
(
APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
)
endif
()
add_instance_library
(
device_batched_gemm_softmax_gemm_permute_instance
${
DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
}
)
add_instance_library
(
device_batched_gemm_softmax_gemm_permute_instance
${
DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
View file @
9b7093aa
set
(
DEVICE_CONTRACTION_BILINEAR_INSTANCES
)
set
(
DEVICE_CONTRACTION_BILINEAR_INSTANCES
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
#float
#float
list
(
APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
list
(
APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
)
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
)
endif
()
#double
if
(
DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
#double
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
list
(
APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
)
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
)
endif
()
add_instance_library
(
device_contraction_bilinear_instance
${
DEVICE_CONTRACTION_BILINEAR_INSTANCES
}
)
add_instance_library
(
device_contraction_bilinear_instance
${
DEVICE_CONTRACTION_BILINEAR_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
View file @
9b7093aa
set
(
DEVICE_CONTRACTION_SCALE_INSTANCES
)
set
(
DEVICE_CONTRACTION_SCALE_INSTANCES
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
#float
#float
list
(
APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
list
(
APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
)
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
)
endif
()
#double
if
(
DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
#double
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
list
(
APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
)
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
)
endif
()
add_instance_library
(
device_contraction_scale_instance
${
DEVICE_CONTRACTION_SCALE_INSTANCES
}
)
add_instance_library
(
device_contraction_scale_instance
${
DEVICE_CONTRACTION_SCALE_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
View file @
9b7093aa
set
(
CONV2D_BWD_DATA_INSTANCES
)
set
(
CONV2D_BWD_DATA_INSTANCES
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
)
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
if
(
DL_KERNELS
)
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
)
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
endif
()
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
endif
()
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
)
if
(
DL_KERNELS
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
)
endif
()
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
)
if
(
DL_KERNELS
)
list
(
APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
)
endif
()
endif
()
add_instance_library
(
device_conv2d_bwd_data_instance
${
CONV2D_BWD_DATA_INSTANCES
}
)
add_instance_library
(
device_conv2d_bwd_data_instance
${
CONV2D_BWD_DATA_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
View file @
9b7093aa
set
(
DEVICE_CONV2D_FWD_INSTANCES
)
set
(
DEVICE_CONV2D_FWD_INSTANCES
)
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
)
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
endif
()
device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
)
device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
)
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
)
endif
()
add_instance_library
(
device_conv2d_fwd_instance
${
DEVICE_CONV2D_FWD_INSTANCES
}
)
add_instance_library
(
device_conv2d_fwd_instance
${
DEVICE_CONV2D_FWD_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_elementwise_normalization_instance
add_instance_library
(
device_elementwise_normalization_instance
device_elementwise_normalization_f16_instance.cpp
device_elementwise_normalization_f16_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
View file @
9b7093aa
set
(
GEMM_INSTANCES
)
set
(
GEMM_INSTANCES
)
if
(
DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
)
device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
)
device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
)
device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
)
if
(
DL_KERNELS
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
)
endif
()
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
if
(
DL_KERNELS
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
)
endif
()
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"int8"
OR NOT DEFINED DTYPES
)
if
(
DL_KERNELS
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
)
endif
()
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
)
endif
()
add_instance_library
(
device_gemm_instance
${
GEMM_INSTANCES
}
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
)
list
(
APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_gemm_instance
${
GEMM_INSTANCES
}
)
set
(
ENABLE_PIPELINE_V2_OPT OFF
)
set
(
ENABLE_PIPELINE_V2_OPT OFF
)
if
(
ENABLE_PIPELINE_V2_OPT
)
if
(
ENABLE_PIPELINE_V2_OPT
)
set
(
MAX_ILP_OPTS
set
(
MAX_ILP_OPTS
-mllvm
-mllvm
-amdgpu-enable-max-ilp-scheduling-strategy
-amdgpu-enable-max-ilp-scheduling-strategy
...
@@ -137,5 +123,5 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
...
@@ -137,5 +123,5 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
set_source_files_properties
(
device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
set_source_files_properties
(
device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
COMPILE_OPTIONS
"
${
MAX_ILP_OPTS
}
"
COMPILE_OPTIONS
"
${
MAX_ILP_OPTS
}
"
COMPILE_DEFINITIONS
"
${
WAVES_PER_EU_DEFS
}
;
${
IGLP_OPT_DEFS
}
"
)
COMPILE_DEFINITIONS
"
${
WAVES_PER_EU_DEFS
}
;
${
IGLP_OPT_DEFS
}
"
)
endif
(
ENABLE_PIPELINE_V2_OPT
)
endif
(
ENABLE_PIPELINE_V2_OPT
)
endif
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
View file @
9b7093aa
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
// MPerBlock=128, NPerBlock=128
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=64
// MPerBlock=64, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=16, NPerBlock=64
// MPerBlock=16, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=16
// MPerBlock=64, NPerBlock=16
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=64
// MPerBlock=8, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=8
// MPerBlock=64, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=8
// MPerBlock=8, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
// clang-format on
// clang-format on
>
;
>
;
...
...
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
View file @
9b7093aa
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
// MPerBlock=128, NPerBlock=128
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=64
// MPerBlock=64, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=16, NPerBlock=64
// MPerBlock=16, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=16
// MPerBlock=64, NPerBlock=16
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=64
// MPerBlock=8, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=8
// MPerBlock=64, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=8
// MPerBlock=8, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
// clang-format on
// clang-format on
>
;
>
;
...
...
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
View file @
9b7093aa
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
...
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// MPerBlock=128, NPerBlock=128
// MPerBlock=128, NPerBlock=128
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
4
,
4
>
,
S
<
4
,
4
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
2
,
8
>
,
S
<
2
,
8
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=64
// MPerBlock=64, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
64
,
8
,
2
,
4
,
4
,
1
,
S
<
4
,
2
>
,
S
<
8
,
1
>
,
S
<
4
,
1
,
2
,
2
>
,
S
<
2
,
1
,
32
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=16, NPerBlock=64
// MPerBlock=16, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
16
,
64
,
16
,
2
,
1
,
4
,
1
,
S
<
2
,
4
>
,
S
<
2
,
4
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=16
// MPerBlock=64, NPerBlock=16
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
16
,
16
,
2
,
4
,
1
,
1
,
S
<
4
,
2
>
,
S
<
4
,
2
>
,
S
<
4
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
16
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=64
// MPerBlock=8, NPerBlock=64
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
4
,
1
>
,
S
<
8
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
8
,
64
,
32
,
2
,
1
,
2
,
1
,
S
<
2
,
2
>
,
S
<
8
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
8
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=64, NPerBlock=8
// MPerBlock=64, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
4
,
1
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
64
,
64
,
8
,
32
,
2
,
2
,
1
,
1
,
S
<
8
,
2
>
,
S
<
2
,
2
>
,
S
<
8
,
1
,
4
,
2
>
,
S
<
4
,
1
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
32
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
// MPerBlock=8, NPerBlock=8
// MPerBlock=8, NPerBlock=8
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
2
,
1
>
,
S
<
4
,
1
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
,
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNPadding
,
8
,
8
,
8
,
4
,
2
,
2
,
1
,
1
,
S
<
1
,
2
>
,
S
<
1
,
4
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
1
,
8
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
4
,
1
,
2
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
1
>
// clang-format on
// clang-format on
>
;
>
;
...
...
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
View file @
9b7093aa
This diff is collapsed.
Click to expand it.
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
View file @
9b7093aa
This diff is collapsed.
Click to expand it.
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
View file @
9b7093aa
This diff is collapsed.
Click to expand it.
library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_gemm_add_add_fastgelu_instance
add_instance_library
(
device_gemm_add_add_fastgelu_instance
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_gemm_add_fastgelu_instance
add_instance_library
(
device_gemm_add_fastgelu_instance
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
)
)
endif
()
library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
View file @
9b7093aa
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_gemm_add_relu_add_layernorm_instance
add_instance_library
(
device_gemm_add_relu_add_layernorm_instance
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
)
)
endif
()
Prev
1
2
3
4
5
6
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