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
bd0f0686
Commit
bd0f0686
authored
Jul 09, 2022
by
Jing Zhang
Browse files
merge develop
parents
e9b1000f
63914743
Changes
382
Hide whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
867 additions
and
46 deletions
+867
-46
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/CMakeLists.txt
...r_operation_instance/gpu/convnd_bwd_weight/CMakeLists.txt
+19
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp
...evice_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp
...device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp
...device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp
+86
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
...ce_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
...ice_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
...ice_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+86
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
...conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
..._conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
+88
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
..._conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
+87
-0
library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
...on_instance/gpu/elementwise/device_normalize_instance.cpp
+3
-1
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
...gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
+7
-5
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
...ce/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+7
-5
No files found.
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/CMakeLists.txt
0 → 100644
View file @
bd0f0686
#device_convnd_bwd_weight_instance
set
(
DEVICE_CONVND_BWD_WEIGHT_INSTANCE_SOURCE
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp;
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp;
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp;
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp;
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp;
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp;
)
add_library
(
device_convnd_bwd_weight_instance OBJECT
${
DEVICE_CONVND_BWD_WEIGHT_INSTANCE_SOURCE
}
)
target_compile_features
(
device_convnd_bwd_weight_instance PUBLIC
)
set_target_properties
(
device_convnd_bwd_weight_instance PROPERTIES POSITION_INDEPENDENT_CODE ON
)
rocm_install
(
TARGETS device_convnd_bwd_weight_instance
)
clang_tidy_check
(
device_convnd_bwd_weight_instance
)
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
BF16
=
bhalf_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
using
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
void
add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
1
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
BF16
=
bhalf_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
using
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
void
add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f32_default_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
2
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f32_default_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
BF16
=
bhalf_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f16_default_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
using
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
8
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
1
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
4
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
2
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
8
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
8
>
// clang-format on
>
;
void
add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f16_default_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
0 → 100644
View file @
bd0f0686
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <numeric>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F32
=
float
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdWeightDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Default
;
static
constexpr
auto
ConvBwdWeightFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardWeightSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f32_default_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightDefault
,
3
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
using
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances
=
std
::
tuple
<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Num| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths|ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
64
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
8
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
16
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
4
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
1
,
true
,
S
<
1
,
4
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
32
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
,
DeviceConvndBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdWeightFilter1x1Stride1Pad0
,
3
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
1
,
4
,
8
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
2
,
true
,
S
<
1
,
4
,
16
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
2
,
1
,
3
>
,
2
,
4
,
4
,
true
,
1
,
1
,
S
<
1
,
16
,
1
,
4
>
,
4
>
// clang-format on
>
;
void
add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instances
(
std
::
vector
<
DeviceConvBwdWeightPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f32_default_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
View file @
bd0f0686
...
...
@@ -7,11 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_5ary_elementwise.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -44,6 +45,7 @@ void add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(
instances
,
device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -31,7 +31,7 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
device_gemm_dl_f16_f16_f16_km_kn_mn_instances
=
std
::
tuple
<
// clang-format off
// #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -39,12 +39,14 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_instances = std::tuple<
>
;
void
add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Col
,
Row
,
Row
,
F16
,
F16
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f16_f16_f16_km_kn_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -31,7 +31,7 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
device_gemm_dl_f16_f16_f16_km_nk_mn_instances
=
std
::
tuple
<
// clang-format off
// #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -39,12 +39,14 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_instances = std::tuple<
>
;
void
add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Col
,
Col
,
Row
,
F16
,
F16
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f16_f16_f16_km_nk_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -31,7 +31,7 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
device_gemm_dl_f16_f16_f16_mk_kn_mn_instances
=
std
::
tuple
<
// clang-format off
// #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
2
,
1
,
4
,
2
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -39,12 +39,14 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
>
;
void
add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Row
,
Row
,
Row
,
F16
,
F16
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f16_f16_f16_mk_kn_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -32,7 +32,7 @@ using device_gemm_dl_f16_f16_f16_mk_nk_mn_instances =
std
::
tuple
<
// clang-format off
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F16
,
F16
,
F16
,
F32
,
Row
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
2
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
8
,
1
,
1
,
2
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
2
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
2
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -40,12 +40,14 @@ using device_gemm_dl_f16_f16_f16_mk_nk_mn_instances =
>
;
void
add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Row
,
Col
,
Row
,
F16
,
F16
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f16_f16_f16_mk_nk_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -31,7 +31,7 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
device_gemm_dl_f32_f32_f32_km_kn_mn_instances
=
std
::
tuple
<
// clang-format off
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F32
,
F32
,
F32
,
F32
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
1
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
1
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
2
,
1
,
4
,
1
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -39,12 +39,14 @@ using device_gemm_dl_f32_f32_f32_km_kn_mn_instances = std::tuple<
>
;
void
add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Col
,
Row
,
Row
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f32_f32_f32_km_kn_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -32,7 +32,7 @@ using device_gemm_dl_f32_f32_f32_km_nk_mn_instances =
std
::
tuple
<
// clang-format off
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F32
,
F32
,
F32
,
F32
,
Col
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
1
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
1
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
8
,
1
,
1
,
1
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -40,12 +40,14 @@ using device_gemm_dl_f32_f32_f32_km_nk_mn_instances =
>
;
void
add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Col
,
Col
,
Row
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f32_f32_f32_km_nk_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -32,7 +32,7 @@ using device_gemm_dl_f32_f32_f32_mk_kn_mn_instances =
std
::
tuple
<
// clang-format off
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F32
,
F32
,
F32
,
F32
,
Row
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
1
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
1
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
2
,
1
,
4
,
1
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -40,12 +40,14 @@ using device_gemm_dl_f32_f32_f32_mk_kn_mn_instances =
>
;
void
add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Row
,
Row
,
Row
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f32_f32_f32_mk_kn_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
F16
=
ck
::
half_t
;
using
F32
=
float
;
...
...
@@ -32,7 +32,7 @@ using device_gemm_dl_f32_f32_f32_mk_nk_mn_instances =
std
::
tuple
<
// clang-format off
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
F32
,
F32
,
F32
,
F32
,
Row
,
Col
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
1
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
8
,
1
,
1
,
1
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
8
,
1
,
1
,
1
>
,
S
<
2
,
1
,
128
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
4
,
1
,
1
,
1
>
,
S
<
1
,
2
,
0
,
3
>
,
S
<
1
,
1
,
1
,
1
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -40,12 +40,14 @@ using device_gemm_dl_f32_f32_f32_mk_nk_mn_instances =
>
;
void
add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Row
,
Col
,
Row
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_f32_f32_f32_mk_nk_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
View file @
bd0f0686
...
...
@@ -7,12 +7,12 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/
add_
device_operation_instance.hpp"
namespace
ck
{
namespace
tensor_operation
{
namespace
device
{
namespace
device_gemm_
instance
{
namespace
instance
{
using
Row
=
ck
::
tensor_layout
::
gemm
::
RowMajor
;
using
Col
=
ck
::
tensor_layout
::
gemm
::
ColumnMajor
;
...
...
@@ -28,7 +28,7 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
using
device_gemm_dl_i8_i8_i8_km_kn_mn_instances
=
std
::
tuple
<
// clang-format off
// #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
a
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Sp
e
cialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
// #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDl
<
int8_t
,
int8_t
,
int8_t
,
int32_t
,
Col
,
Row
,
Row
,
PassThrough
,
PassThrough
,
PassThrough
,
GemmDefault
,
256
,
128
,
128
,
16
,
4
,
4
,
4
,
1
,
S
<
8
,
2
>
,
S
<
8
,
2
>
,
S
<
2
,
1
,
4
,
4
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
4
>
,
S
<
2
,
1
,
4
,
4
>
,
S
<
8
,
1
,
32
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
1
>
,
S
<
0
,
3
,
1
,
2
>
,
S
<
1
,
1
,
4
,
4
>
,
S
<
0
,
1
,
2
,
3
,
4
,
5
>
,
5
,
4
>
...
...
@@ -36,12 +36,14 @@ using device_gemm_dl_i8_i8_i8_km_kn_mn_instances = std::tuple<
>
;
void
add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances
(
std
::
vector
<
DeviceGemmPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
std
::
vector
<
std
::
unique_ptr
<
DeviceGemm
<
Col
,
Row
,
Row
,
int8_t
,
int8_t
,
int8_t
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_gemm_dl_i8_i8_i8_km_kn_mn_instances
{});
}
}
// namespace
device_gemm_
instance
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
Prev
1
…
7
8
9
10
11
12
13
14
15
…
20
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
