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
930b2872
Commit
930b2872
authored
Oct 11, 2023
by
Harisankar Sadasivan
Browse files
best performing kernel for GEMV codex problem with M=1 with inverted B matrix
parents
a1e17d18
a4f72a31
Changes
365
Show whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
326 additions
and
198 deletions
+326
-198
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
...ped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
...3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
+56
-0
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
...uped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
...uped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
...ped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
...tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
...device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+34
-34
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
...grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+35
-21
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
...mm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+35
-51
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
...grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+25
-17
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
...mm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+25
-24
library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
...eration_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
+0
-2
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
...eration_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
+16
-8
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
...ouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
...ouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
+0
-0
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
...age_to_column/device_image_to_column_nhwc_1d_instance.cpp
+31
-9
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
...age_to_column/device_image_to_column_nhwc_2d_instance.cpp
+28
-5
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
...age_to_column/device_image_to_column_nhwc_3d_instance.cpp
+32
-9
library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
...tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
+3
-9
library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
...ensor_operation_instance/gpu/normalization/CMakeLists.txt
+6
-7
No files found.
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/
xdl/
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
0 → 100755
View file @
930b2872
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
void
add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceGroupedConvFwdMultipleD
<
3
,
NDHWGC
,
GKZYXC
,
Empty_Tuple
,
NDHWGK
,
F16
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
F8
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_grouped_conv_fwd_xdl_f16_comp_f8_instances
<
3
,
NDHWGC
,
GKZYXC
,
Empty_Tuple
,
NDHWGK
,
ConvFwdDefault
>
{});
add_device_operation_instances
(
instances
,
device_grouped_conv_fwd_xdl_f16_comp_f8_instances
<
3
,
NDHWGC
,
GKZYXC
,
Empty_Tuple
,
NDHWGK
,
ConvFwd1x1P0
>
{});
add_device_operation_instances
(
instances
,
device_grouped_conv_fwd_xdl_f16_comp_f8_instances
<
3
,
NDHWGC
,
GKZYXC
,
Empty_Tuple
,
NDHWGK
,
ConvFwd1x1S1P0
>
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/
xdl/
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/
xdl/
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/
xdl/
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
View file @
930b2872
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_grouped_gemm_instance
device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
...
...
@@ -9,4 +8,3 @@ add_instance_library(device_grouped_gemm_instance
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
)
endif
()
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
View file @
930b2872
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
View file @
930b2872
...
...
@@ -33,27 +33,41 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// a[m, k] * b[k, n] = e[m, n]
using
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances
=
std
::
tuple
<
// clang-format off
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Currently AK1 must equal BK1 !
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 256, 32, 8, 2, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 128, 32, 8, 2, 32, 32, 2, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 16,16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
256
,
32
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| PipelineVersion|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| |
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Pipline v1
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
24
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
192
,
32
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
// Pipline v2
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
24
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
192
,
32
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
// clang-format on
>
;
...
...
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
View file @
930b2872
...
...
@@ -31,57 +31,41 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
using
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances
=
std
::
tuple
<
// clang-format off
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Currently AK1 must equal BK1 !
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 2, 32, 32, 2, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 16,16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 2, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 8>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 2, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 32, 1, 4>, 8>,
// DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 2, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
256
,
32
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
192
,
64
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
48
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| PipelineVersion|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| |
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Pipline v1
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
24
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, PipelineVersion::v1>,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, PipelineVersion::v1>,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, PipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
192
,
32
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
256
,
32
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
192
,
64
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
48
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
// Pipline v2
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
24
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, PipelineVersion::v2>,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 2, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, PipelineVersion::v2>,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Row, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 32, 192, 32, 8, 8, 32, 32, 1, 3, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 24, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, PipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
192
,
32
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
1
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
4
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Row
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
1
,
3
,
2
>
,
S
<
0
,
1
,
3
,
2
>
,
2
,
2
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
PipelineVersion
::
v2
>
// clang-format on
>
;
...
...
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
View file @
930b2872
...
...
@@ -32,23 +32,31 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// a[m, k] * b[n, k] = e[m, n]
using
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances
=
std
::
tuple
<
// clang-format off
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
256
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
256
,
32
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| PipelineVersion|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| |
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Pipeline v1
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
256
,
64
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
32
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
32
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
// Pipeline v2
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
256
,
64
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
64
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
256
,
128
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
32
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
128
,
32
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
32
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
128
,
64
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
1
,
64
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v2
>
// clang-format on
>
;
...
...
library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
View file @
930b2872
...
...
@@ -31,30 +31,31 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
using
device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances
=
std
::
tuple
<
// clang-format off
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
256
,
32
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
192
,
64
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
48
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
128
,
32
,
8
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
128
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
192
,
32
,
32
,
8
,
8
,
32
,
32
,
3
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
192
,
32
,
8
,
8
,
32
,
32
,
1
,
3
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
128
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
256
,
32
,
8
,
8
,
32
,
32
,
1
,
4
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
64
,
32
,
8
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
64
,
32
,
32
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
32
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
//################################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| C| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| PipelineVersion|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| |
//################################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| |
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Pipeline v1
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
256
,
64
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
32
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, ck::PipelineVersion::v1>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v1
>
,
// Pipeline v2
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
256
,
64
,
8
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
64
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
256
,
128
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
32
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
32
,
128
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
128
,
32
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
//DeviceGroupedGemmXdlSplitKCShuffle< Row, Col, Empty_Tuple, Row, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding, 1, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 8, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, ck::PipelineVersion::v2>,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
128
,
64
,
64
,
64
,
8
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
16
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
8
>
,
8
,
ck
::
PipelineVersion
::
v2
>
,
DeviceGroupedGemmXdlSplitKCShuffle
<
Row
,
Col
,
Empty_Tuple
,
Row
,
F16
,
F16
,
F32
,
F16
,
Empty_Tuple
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmMNKPadding
,
1
,
64
,
32
,
64
,
64
,
8
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
S
<
1
,
8
,
8
,
1
>
,
S
<
0
,
2
,
1
,
3
>
,
S
<
0
,
2
,
1
,
3
>
,
3
,
8
,
8
,
1
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
,
ck
::
PipelineVersion
::
v2
>
// clang-format on
>
;
...
...
library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
View file @
930b2872
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
add_instance_library
(
device_grouped_gemm_fastgelu_instance
device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
)
endif
()
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
View file @
930b2872
add_instance_library
(
device_grouped_gemm_fixed_nk_instance
device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
set
(
GROUPED_GEMM_FIXED_NK_INSTANCES
)
device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
)
endif
()
device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
)
if
((
DTYPES MATCHES
"fp8"
AND DTYPES MATCHES
"fp16"
)
OR NOT DEFINED DTYPES
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
)
endif
()
if
((
DTYPES MATCHES
"int8"
AND DTYPES MATCHES
"fp16"
)
OR NOT DEFINED DTYPES
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
)
list
(
APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
)
endif
()
add_instance_library
(
device_grouped_gemm_fixed_nk_instance
${
GROUPED_GEMM_FIXED_NK_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f
p
8_f16_mk_kn_mn_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
→
library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f
p
8_f16_mk_nk_mn_instance.cpp
View file @
930b2872
File moved
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
View file @
930b2872
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/
image_to_column
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/
conv_tensor_rearrange
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
...
...
@@ -9,28 +9,50 @@ namespace tensor_operation {
namespace
device
{
namespace
instance
{
void
add_device_image_to_column_nhwc_1d_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
1
,
GNWC
,
BF16
,
BF16
>>>&
instances
)
using
namespace
ck
::
conv_tensor_rearrange_op
;
void
add_device_image_to_column_nwc_1d_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
1
,
GNWC
,
BF16
,
BF16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_BF16
add_device_operation_instances
(
instances
,
device_image_to_column_bf16_instances
<
1
,
GNWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_1d_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
1
,
GNWC
,
F16
,
F16
>>>&
instances
)
void
add_device_image_to_column_nwc_1d_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
1
,
GNWC
,
F16
,
F16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP16
add_device_operation_instances
(
instances
,
device_image_to_column_f16_instances
<
1
,
GNWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_1d_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
1
,
GNWC
,
F32
,
F32
>>>&
instances
)
void
add_device_image_to_column_nwc_1d_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
1
,
GNWC
,
F32
,
F32
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP32
add_device_operation_instances
(
instances
,
device_image_to_column_f32_instances
<
1
,
GNWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_1d_i8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
1
,
GNWC
,
int8_t
,
int8_t
>>>&
instances
)
void
add_device_image_to_column_nwc_1d_i8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
1
,
GNWC
,
int8_t
,
int8_t
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_INT8
add_device_operation_instances
(
instances
,
device_image_to_column_i8_instances
<
1
,
GNWC
>
{});
#else
ignore
=
instances
;
#endif
}
}
// namespace instance
...
...
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
View file @
930b2872
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/
image_to_column
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/
conv_tensor_rearrange
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
...
...
@@ -9,28 +9,51 @@ namespace tensor_operation {
namespace
device
{
namespace
instance
{
using
namespace
ck
::
conv_tensor_rearrange_op
;
void
add_device_image_to_column_nhwc_2d_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
2
,
GNHWC
,
BF16
,
BF16
>>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
2
,
GNHWC
,
BF16
,
BF16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_BF16
add_device_operation_instances
(
instances
,
device_image_to_column_bf16_instances
<
2
,
GNHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_2d_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
2
,
GNHWC
,
F16
,
F16
>>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
2
,
GNHWC
,
F16
,
F16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP16
add_device_operation_instances
(
instances
,
device_image_to_column_f16_instances
<
2
,
GNHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_2d_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
2
,
GNHWC
,
F32
,
F32
>>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
2
,
GNHWC
,
F32
,
F32
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP32
add_device_operation_instances
(
instances
,
device_image_to_column_f32_instances
<
2
,
GNHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_2d_i8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
2
,
GNHWC
,
int8_t
,
int8_t
>>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
2
,
GNHWC
,
int8_t
,
int8_t
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_INT8
add_device_operation_instances
(
instances
,
device_image_to_column_i8_instances
<
2
,
GNHWC
>
{});
#else
ignore
=
instances
;
#endif
}
}
// namespace instance
...
...
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
View file @
930b2872
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/
image_to_column
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/
conv_tensor_rearrange
/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
...
...
@@ -9,28 +9,51 @@ namespace tensor_operation {
namespace
device
{
namespace
instance
{
void
add_device_image_to_column_nhwc_3d_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
3
,
GNDHWC
,
BF16
,
BF16
>>>&
instances
)
using
namespace
ck
::
conv_tensor_rearrange_op
;
void
add_device_image_to_column_ndhwc_3d_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
3
,
GNDHWC
,
BF16
,
BF16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_BF16
add_device_operation_instances
(
instances
,
device_image_to_column_bf16_instances
<
3
,
GNDHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_3d_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
3
,
GNDHWC
,
F16
,
F16
>>>&
instances
)
void
add_device_image_to_column_ndhwc_3d_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
3
,
GNDHWC
,
F16
,
F16
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP16
add_device_operation_instances
(
instances
,
device_image_to_column_f16_instances
<
3
,
GNDHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_3d_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
3
,
GNDHWC
,
F32
,
F32
>>>&
instances
)
void
add_device_image_to_column_ndhwc_3d_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
3
,
GNDHWC
,
F32
,
F32
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_FP32
add_device_operation_instances
(
instances
,
device_image_to_column_f32_instances
<
3
,
GNDHWC
>
{});
#else
ignore
=
instances
;
#endif
}
void
add_device_image_to_column_nhwc_3d_i8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceImageToColumn
<
3
,
GNDHWC
,
int8_t
,
int8_t
>>>&
instances
)
void
add_device_image_to_column_ndhwc_3d_i8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvTensorRearrange
<
3
,
GNDHWC
,
int8_t
,
int8_t
,
ImageToColumn
>>>&
instances
)
{
#ifdef CK_ENABLE_INT8
add_device_operation_instances
(
instances
,
device_image_to_column_i8_instances
<
3
,
GNDHWC
>
{});
#else
ignore
=
instances
;
#endif
}
}
// namespace instance
...
...
library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
View file @
930b2872
set
(
DEVICE_MAXPOOL_BWD_INSTANCES
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"bf16"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_bf16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f32_instance.cpp
)
endif
()
list
(
APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp
device_max_pool_bwd_bf16_instance.cpp
device_max_pool_bwd_f32_instance.cpp
)
add_instance_library
(
device_max_pool_bwd_instance
${
DEVICE_MAXPOOL_BWD_INSTANCES
}
)
library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
View file @
930b2872
set
(
DEVICE_NORMALIZATION_INSTANCES
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f16_instance.cpp
list
(
APPEND DEVICE_NORMALIZATION_INSTANCES
device_layernorm2d_f16_instance.cpp
device_layernorm4d_f16_instance.cpp
device_groupnorm_f16_instance.cpp
device_groupnorm_swish_f16_instance.cpp
device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp32"
OR NOT DEFINED DTYPES
)
list
(
APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f32_instance.cpp
device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
device_layernorm2d_f32_instance.cpp
device_layernorm4d_f32_instance.cpp
device_groupnorm_f32_instance.cpp
device_groupnorm_swish_f32_instance.cpp
)
endif
()
add_instance_library
(
device_normalization_instance
${
DEVICE_NORMALIZATION_INSTANCES
}
)
Prev
1
…
12
13
14
15
16
17
18
19
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
