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
04240ec7
Commit
04240ec7
authored
Oct 20, 2023
by
Rostyslav Geyyer
Browse files
Add conv bwd weight client example
parent
1fd27d52
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
78 additions
and
2 deletions
+78
-2
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
+2
-0
client_example/11_grouped_conv_bwd_weight/common.hpp
client_example/11_grouped_conv_bwd_weight/common.hpp
+6
-2
client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
...wd_weight/grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
+70
-0
No files found.
client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
View file @
04240ec7
...
@@ -2,8 +2,10 @@ add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_f
...
@@ -2,8 +2,10 @@ add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_f
add_executable
(
client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp
)
add_executable
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
)
target_link_libraries
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations
)
target_link_libraries
(
client_grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8 PRIVATE composable_kernel::device_operations
)
client_example/11_grouped_conv_bwd_weight/common.hpp
View file @
04240ec7
...
@@ -85,7 +85,9 @@ template <ck::index_t NumDimSpatial,
...
@@ -85,7 +85,9 @@ template <ck::index_t NumDimSpatial,
typename
OutDataType
,
typename
OutDataType
,
typename
InLayout
,
typename
InLayout
,
typename
WeiLayout
,
typename
WeiLayout
,
typename
OutLayout
>
typename
OutLayout
,
typename
AComputeType
=
InDataType
,
typename
BComputeType
=
AComputeType
>
bool
run_grouped_conv_bwd_weight
(
bool
run_grouped_conv_bwd_weight
(
const
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>&
input_lengths
,
const
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>&
input_lengths
,
const
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>&
input_strides
,
const
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>&
input_strides
,
...
@@ -113,7 +115,9 @@ bool run_grouped_conv_bwd_weight(
...
@@ -113,7 +115,9 @@ bool run_grouped_conv_bwd_weight(
OutDataType
,
OutDataType
,
PassThrough
,
PassThrough
,
PassThrough
,
PassThrough
,
PassThrough
>
;
PassThrough
,
AComputeType
,
BComputeType
>
;
// get device op instances
// get device op instances
const
auto
op_ptrs
=
ck
::
tensor_operation
::
device
::
instance
::
DeviceOperationInstanceFactory
<
const
auto
op_ptrs
=
ck
::
tensor_operation
::
device
::
instance
::
DeviceOperationInstanceFactory
<
DeviceOp
>::
GetInstances
();
DeviceOp
>::
GetInstances
();
...
...
client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16_comp_bf8_fp8.cpp
0 → 100644
View file @
04240ec7
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
using
InDataType
=
ck
::
half_t
;
using
WeiDataType
=
ck
::
half_t
;
using
OutDataType
=
ck
::
half_t
;
using
InLayout
=
ck
::
tensor_layout
::
convolution
::
NDHWGC
;
using
WeiLayout
=
ck
::
tensor_layout
::
convolution
::
GKZYXC
;
using
OutLayout
=
ck
::
tensor_layout
::
convolution
::
NDHWGK
;
using
AComputeType
=
ck
::
bf8_t
;
using
BComputeType
=
ck
::
f8_t
;
static
constexpr
ck
::
index_t
NumDimSpatial
=
3
;
static
constexpr
ck
::
index_t
G
=
8
;
static
constexpr
ck
::
index_t
N
=
64
;
static
constexpr
ck
::
index_t
K
=
128
;
static
constexpr
ck
::
index_t
C
=
128
;
static
constexpr
ck
::
index_t
Z
=
3
;
static
constexpr
ck
::
index_t
Y
=
3
;
static
constexpr
ck
::
index_t
X
=
3
;
static
constexpr
ck
::
index_t
Di
=
28
;
static
constexpr
ck
::
index_t
Hi
=
28
;
static
constexpr
ck
::
index_t
Wi
=
3
;
static
constexpr
ck
::
index_t
Do
=
28
;
static
constexpr
ck
::
index_t
Ho
=
28
;
static
constexpr
ck
::
index_t
Wo
=
3
;
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
input_lengths
{
G
,
N
,
C
,
Di
,
Hi
,
Wi
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
filter_lengths
{
G
,
K
,
C
,
Z
,
Y
,
X
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
output_lengths
{
G
,
N
,
K
,
Do
,
Ho
,
Wo
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
input_strides
{
N
*
Di
*
Hi
*
Wi
*
C
,
Di
*
Hi
*
Wi
*
C
,
1
,
Hi
*
Wi
*
C
,
Wi
*
C
,
C
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
weights_strides
{
K
*
Z
*
Y
*
X
*
C
,
Z
*
Y
*
X
*
C
,
1
,
Y
*
X
*
C
,
X
*
C
,
C
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
+
3
>
output_strides
{
N
*
Do
*
Ho
*
Wo
*
K
,
Do
*
Ho
*
Wo
*
K
,
1
,
Ho
*
Wo
*
K
,
Wo
*
K
,
K
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
>
conv_filter_strides
{
1
,
1
,
1
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
>
conv_filter_dilations
{
1
,
1
,
1
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
>
input_left_pads
{
1
,
1
,
1
};
static
constexpr
std
::
array
<
ck
::
index_t
,
NumDimSpatial
>
input_right_pads
{
1
,
1
,
1
};
int
main
()
{
return
run_grouped_conv_bwd_weight
<
NumDimSpatial
,
InDataType
,
WeiDataType
,
OutDataType
,
InLayout
,
WeiLayout
,
OutLayout
,
AComputeType
,
BComputeType
>
(
input_lengths
,
input_strides
,
filter_lengths
,
weights_strides
,
output_lengths
,
output_strides
,
conv_filter_strides
,
conv_filter_dilations
,
input_left_pads
,
input_right_pads
)
?
EXIT_SUCCESS
:
EXIT_FAILURE
;
}
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