ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 01:36:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Grouped conv bwd data NGCHW (#1967)

Browse Source 
* Grouped conv bwd data NGCHW

* fixes

* fix

* Improvements

* Fix

* Fix

* add client example
This commit is contained in:
Bartłomiej Kocot
2025-03-17 13:32:00 +01:00
committed by GitHub
parent 52b1cd7780
commit c2e4898b4b
26 changed files with 1351 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
View File

@@ -7,6 +7,9 @@ add_instance_library(
xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_f16_instance.cpp
xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_bf16_instance.cpp
xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_f32_instance.cpp
wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp

48
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_bf16_instance.cpp Normal file
View File

@@ -0,0 +1,48 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_ngkhw_gkyxc_ngchw_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_bf16_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

48
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_f16_instance.cpp Normal file
View File

@@ -0,0 +1,48 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_ngkhw_gkyxc_ngchw_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_f16_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

48
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_ngchw_gkyxc_ngkhw_f32_instance.cpp Normal file
View File

@@ -0,0 +1,48 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_ngkhw_gkyxc_ngchw_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_f32_instances<2,
NGKHW,
GKYXC,
Empty_Tuple,
NGCHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

3
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
View File

@@ -6,6 +6,9 @@ set(GROUPED_CONV3D_BWD_DATA
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_f16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_f32_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp

49
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_instance.cpp Normal file
View File

@@ -0,0 +1,49 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ngkdhw_gkzyxc_ngcdhw_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_bf16_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

49
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_f16_instance.cpp Normal file
View File

@@ -0,0 +1,49 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ngkdhw_gkzyxc_ngcdhw_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_f16_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

49
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ngcdhw_gkzyxc_ngkdhw_f32_instance.cpp Normal file
View File

@@ -0,0 +1,49 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_transpose_xdl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ngkdhw_gkzyxc_ngcdhw_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_transpose_xdl_f32_instances<3,
NGKDHW,
GKZYXC,
Empty_Tuple,
NGCDHW,
ConvBwdDataDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck