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
8f722700
Commit
8f722700
authored
Jul 17, 2022
by
Chao Liu
Browse files
update ckprofiler for conv bwd data
parent
a711f90b
Changes
49
Expand all
Show whitespace changes
Inline
Side-by-side
Showing
20 changed files
with
594 additions
and
957 deletions
+594
-957
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
...evice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+70
-2
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
...vice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+104
-39
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
...ice_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+1
-1
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
...ice_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+1
-1
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/CMakeLists.txt
...sor_operation_instance/gpu/conv3d_bwd_data/CMakeLists.txt
+14
-0
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
...e_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
+102
-0
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
...ce_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
+102
-0
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
...ce_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
+101
-0
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
...e_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
+99
-0
library/src/tensor_operation_instance/gpu/convnd_bwd_data/CMakeLists.txt
...sor_operation_instance/gpu/convnd_bwd_data/CMakeLists.txt
+0
-22
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
.../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
+0
-89
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
...a/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
+0
-91
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
...a/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
+0
-88
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
.../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
+0
-91
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
...vice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+0
-89
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
...evice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+0
-89
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
...evice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+0
-88
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
...vice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+0
-89
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
...e_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
+0
-89
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
...ce_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
+0
-89
No files found.
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
View file @
8f722700
This diff is collapsed.
Click to expand it.
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
View file @
8f722700
This diff is collapsed.
Click to expand it.
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
View file @
8f722700
...
@@ -6,7 +6,7 @@
...
@@ -6,7 +6,7 @@
#include "ck/ck.hpp"
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
//TODO: retire dedicated 2d version
//
TODO: retire dedicated 2d version
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
...
...
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
View file @
8f722700
...
@@ -6,7 +6,7 @@
...
@@ -6,7 +6,7 @@
#include "ck/ck.hpp"
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
//TODO: retire dedicated 2d version
//
TODO: retire dedicated 2d version
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/device/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
...
...
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/CMakeLists.txt
0 → 100644
View file @
8f722700
# device_conv3d_bwd_data_instance
set
(
DEVICE_CONV3D_BWD_DATA_INSTANCE_SOURCE
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp;
)
add_library
(
device_conv3d_bwd_data_instance OBJECT
${
DEVICE_CONV3D_BWD_DATA_INSTANCE_SOURCE
}
)
target_compile_features
(
device_conv3d_bwd_data_instance PUBLIC
)
set_target_properties
(
device_conv3d_bwd_data_instance PROPERTIES POSITION_INDEPENDENT_CODE ON
)
rocm_install
(
TARGETS device_conv3d_bwd_data_instance
)
clang_tidy_check
(
device_conv3d_bwd_data_instance
)
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
0 → 100644
View file @
8f722700
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.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
NDHWC
=
ck
::
tensor_layout
::
convolution
::
NDHWC
;
using
KZYXC
=
ck
::
tensor_layout
::
convolution
::
KZYXC
;
using
NDHWK
=
ck
::
tensor_layout
::
convolution
::
NDHWK
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvBwdData
<
3
,
NDHWC
,
KZYXC
,
NDHWK
,
BF16
,
BF16
,
BF16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
0 → 100644
View file @
8f722700
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.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
NDHWC
=
ck
::
tensor_layout
::
convolution
::
NDHWC
;
using
KZYXC
=
ck
::
tensor_layout
::
convolution
::
KZYXC
;
using
NDHWK
=
ck
::
tensor_layout
::
convolution
::
NDHWK
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvBwdData
<
3
,
NDHWC
,
KZYXC
,
NDHWK
,
F16
,
F16
,
F16
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
0 → 100644
View file @
8f722700
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.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
NDHWC
=
ck
::
tensor_layout
::
convolution
::
NDHWC
;
using
KZYXC
=
ck
::
tensor_layout
::
convolution
::
KZYXC
;
using
NDHWK
=
ck
::
tensor_layout
::
convolution
::
NDHWK
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvBwdData
<
3
,
NDHWC
,
KZYXC
,
NDHWK
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
0 → 100644
View file @
8f722700
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_nwc_kxc_nwk_xdl.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
{
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
NDHWC
=
ck
::
tensor_layout
::
convolution
::
NDHWC
;
using
KZYXC
=
ck
::
tensor_layout
::
convolution
::
KZYXC
;
using
NDHWK
=
ck
::
tensor_layout
::
convolution
::
NDHWK
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances
=
std
::
tuple
<
// clang-format off
//##############################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
//##############################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvNdBwdDataNwcKxcNwk_Xdl
<
3
,
int8_t
,
int8_t
,
int8_t
,
int32_t
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceConvBwdData
<
3
,
NDHWC
,
KZYXC
,
NDHWK
,
int8_t
,
int8_t
,
int8_t
,
PassThrough
,
PassThrough
,
PassThrough
>>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/CMakeLists.txt
deleted
100644 → 0
View file @
a711f90b
# device_convnd_bwd_data_instance
set
(
DEVICE_CONVND_BWD_DATA_INSTANCE_SOURCE
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp;
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp;
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp;
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp;
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp;
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp;
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp;
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp;
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp;
)
add_library
(
device_convnd_bwd_data_instance OBJECT
${
DEVICE_CONVND_BWD_DATA_INSTANCE_SOURCE
}
)
target_compile_features
(
device_convnd_bwd_data_instance PUBLIC
)
set_target_properties
(
device_convnd_bwd_data_instance PROPERTIES POSITION_INDEPENDENT_CODE ON
)
rocm_install
(
TARGETS device_convnd_bwd_data_instance
)
clang_tidy_check
(
device_convnd_bwd_data_instance
)
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_data_xdl_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv1d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_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_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_data_xdl_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv1d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_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_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_data_xdl_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv1d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_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_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
DataType
=
int8_t
;
using
AccType
=
int32_t
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
#if 1
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
#endif
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
1
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_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| CThreadTransfer| CThreadTransfer|
//##############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
1
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances
{});
add_device_operation_instances
(
instances
,
device_conv1d_bwd_data_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_data_xdl_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv2d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_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_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv2d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_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_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv2d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
4
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
4
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
4
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
4
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
4
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F32
,
F32
,
F32
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
4
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
4
,
4
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
4
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_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_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
DataType
=
int8_t
;
using
AccType
=
int32_t
;
template
<
ck
::
index_t
...
Is
>
using
S
=
ck
::
Sequence
<
Is
...
>
;
using
PassThrough
=
ck
::
tensor_operation
::
element_wise
::
PassThrough
;
static
constexpr
auto
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
2
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_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| CThreadTransfer| CThreadTransfer|
//##############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
256
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
256
,
4
,
16
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
128
,
4
,
16
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
64
,
128
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
64
,
4
,
16
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
128
,
64
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
256
,
64
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
128
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
128
,
32
,
128
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
64
,
32
,
4
,
16
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
16
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
DataType
,
DataType
,
DataType
,
AccType
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
2
,
64
,
32
,
64
,
4
,
16
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
16
,
16
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
16
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances
{});
add_device_operation_instances
(
instances
,
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv3d_bwd_data_xdl_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | ./ | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
BF16
,
BF16
,
BF16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_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_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
deleted
100644 → 0
View file @
a711f90b
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_convnd_bwd_data_xdl_ndhwc_kzyxc_ndhwk.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
ConvBwdDataDefault
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Default
;
static
constexpr
auto
ConvBwdDataFilter1x1Stride1Pad0
=
ck
::
tensor_operation
::
device
::
ConvolutionBackwardDataSpecialization
::
Filter1x1Stride1Pad0
;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataDefault
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
using
device_conv3d_bwd_data_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| CThreadTransfer| CThreadTransfer|
//#############################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Data| Dim| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//#############################################################################| | | | | Operation| Operation| Operation| Specialization|Spatial| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//#############################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
256
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
256
,
4
,
8
,
32
,
32
,
2
,
4
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
128
,
4
,
8
,
32
,
32
,
4
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
64
,
128
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
64
,
64
,
4
,
8
,
32
,
32
,
2
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
128
,
64
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
256
,
64
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
64
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
64
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
128
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
1
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
128
,
32
,
128
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
32
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
32
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
4
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
64
,
32
,
4
,
8
,
32
,
32
,
2
,
1
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
,
DeviceConvndBwdDataXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
<
F16
,
F16
,
F16
,
F32
,
PassThrough
,
PassThrough
,
PassThrough
,
ConvBwdDataFilter1x1Stride1Pad0
,
3
,
64
,
32
,
64
,
4
,
8
,
32
,
32
,
1
,
2
,
S
<
4
,
16
,
1
>
,
S
<
1
,
0
,
2
>
,
S
<
1
,
0
,
2
>
,
2
,
8
,
8
,
true
,
S
<
4
,
16
,
1
>
,
S
<
2
,
0
,
1
>
,
S
<
0
,
2
,
1
>
,
1
,
2
,
8
,
true
,
7
,
1
>
// clang-format on
>
;
void
add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances
(
std
::
vector
<
DeviceConvBwdDataPtr
<
PassThrough
,
PassThrough
,
PassThrough
>>&
instances
)
{
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances
{});
add_device_operation_instances
(
instances
,
device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances
{});
}
}
// namespace instance
}
// namespace device
}
// namespace tensor_operation
}
// namespace ck
Prev
1
2
3
Next
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
