ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-19 20:40:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add dynamic elementwise op (#1426)

Browse Source 
* Add dynamic elementwise op

Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>

* CI issues fix

* Custom parameter value for dynamic functions - Comments addressed

---------

Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>
Co-authored-by: ThruptiRajLakshmanaGowda <tlakshma@amd.com>
This commit is contained in:
Bartłomiej Kocot
2024-10-26 15:22:37 +02:00
committed by GitHub
parent 54f0e6f4bb
commit 31bf253aeb
32 changed files with 2194 additions and 170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/CMakeLists.txt Normal file
View File

@@ -0,0 +1,8 @@
# ONLY XDL_KERNELS
set(GROUPED_CONV2D_FWD_DYNAMIC_OP
xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instance.cpp)
add_instance_library(device_grouped_conv2d_fwd_dynamic_op_instance ${GROUPED_CONV2D_FWD_DYNAMIC_OP})

55
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
ck::Tuple<>,
NHWGK,
BF16,
BF16,
ck::Tuple<>,
BF16,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

55
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
ck::Tuple<>,
NHWGK,
F16,
F16,
ck::Tuple<>,
F16,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

55
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
ck::Tuple<>,
NHWGK,
F32,
F32,
ck::Tuple<>,
F32,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

54
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instance.cpp Normal file
View File

@@ -0,0 +1,54 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
ck::Tuple<>,
NHWGK,
int8_t,
int8_t,
ck::Tuple<>,
int8_t,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

8
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/CMakeLists.txt Normal file
View File

@@ -0,0 +1,8 @@
# ONLY XDL_KERNELS
set(GROUPED_CONV3D_FWD_DYNAMIC_OP
xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp)
add_instance_library(device_grouped_conv3d_fwd_dynamic_op_instance ${GROUPED_CONV3D_FWD_DYNAMIC_OP})

55
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
BF16,
BF16,
ck::Tuple<>,
BF16,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

55
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
F16,
F16,
ck::Tuple<>,
F16,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

55
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp Normal file
View File

@@ -0,0 +1,55 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
F32,
F32,
ck::Tuple<>,
F32,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

54
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp Normal file
View File

@@ -0,0 +1,54 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_dynamic_op_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
int8_t,
int8_t,
ck::Tuple<>,
int8_t,
PassThrough,
PassThrough,
DynamicUnaryOp>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<3,
NDHWGC,
GKZYXC,
Tuple<>,
NDHWGK,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck