ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 01:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Post-merge cleanup for WMMA grouped conv fwd (#3468)

Browse Source 
* remove duplicate aliases

* Split scaleadd_ab instances for WMMA grouped conv fwd

* removed big shape from the test
This commit is contained in:
Wojciech Laskowski
2025-12-22 15:57:45 +01:00
committed by GitHub
parent 44f1b5c5de
commit a8aebb7a8e
13 changed files with 570 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/CMakeLists.txt
View File

@@ -9,8 +9,14 @@ set(GROUPED_CONV3D_FWD_SCALEADD_AB
xdl/device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
# WMMA CSHUFFLE V3
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part1.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part2.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part3.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part4.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part1.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part2.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part3.cpp
wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part4.cpp
)
add_instance_library(device_grouped_conv3d_fwd_scaleadd_ab_instance ${GROUPED_CONV3D_FWD_SCALEADD_AB})

32
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part1.cpp
View File

@@ -9,7 +9,7 @@ namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part1(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
@@ -25,25 +25,25 @@ void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndh
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part2.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part2(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<BF16, BF16>,
ck::Tuple<BF16, BF16>,
ck::Tuple<>,
BF16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part3.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part3(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<BF16, BF16>,
ck::Tuple<BF16, BF16>,
ck::Tuple<>,
BF16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instance_part4.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<BF16, BF16>,
ck::Tuple<BF16, BF16>,
ck::Tuple<>,
BF16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_bf16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

32
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part1.cpp
View File

@@ -9,7 +9,7 @@ namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances(
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances_part1(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
@@ -25,25 +25,25 @@ void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndh
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part1<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part2.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances_part2(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<F16, F16>,
ck::Tuple<F16, F16>,
ck::Tuple<>,
F16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part2<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part3.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances_part3(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<F16, F16>,
ck::Tuple<F16, F16>,
ck::Tuple<>,
F16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part3<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

52
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_scaleadd_ab/wmma/device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instance_part4.cpp Normal file
View File

@@ -0,0 +1,52 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_wmma_cshufflev3_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ck::Tuple<F16, F16>,
ck::Tuple<F16, F16>,
ck::Tuple<>,
F16,
ScaleAdd,
ScaleAdd,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_scaleadd_ab_f16_instances_part4<3,
NDHWGC,
GKZYXC,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck