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

Grouped conv fwd with direct load (#3082)

Browse Source 
* Grouped conv fwd with direct load

* fix

* fix

* Add IsSupported check

* Fix

* fix inductor
This commit is contained in:
Bartłomiej Kocot
2025-10-29 09:54:42 +01:00
committed by GitHub
parent 3052d7c9e6
commit 66bae4306c
27 changed files with 2165 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
View File

@@ -49,6 +49,8 @@ set(GROUPED_CONV2D_FWD
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_tf32_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instance.cpp
# NHWGC, GKYXC, NHWGK
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_mem_inter_instance.cpp

53
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,53 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

52
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,52 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

18
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_bnorm_clamp/CMakeLists.txt
View File

@@ -164,6 +164,15 @@ generate_sharded_instantiations(
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV2D_FWD_BIAS_BNORM_CLAMP
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_mem_intra_instances
@@ -173,6 +182,15 @@ generate_sharded_instantiations(
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV2D_FWD_BIAS_BNORM_CLAMP
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instances

56
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_bnorm_clamp/xdl/mem/device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.in Normal file
View File

@@ -0,0 +1,56 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances = std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
BF16,
BF16,
Tuple<BF16, BF16, BF16, BF16, BF16>,
BF16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>;
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
template <int Shards, int ShardIndex>
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances_shard(device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(instances, ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
ConvFwdDefault,
Intrawave,Tuple<BF16, BF16, BF16, BF16, BF16>, BiasNormalizeInInferClamp>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances, ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,Tuple<BF16, BF16, BF16, BF16, BF16>, BiasNormalizeInInferClamp>,
Shards,
ShardIndex>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

56
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_bnorm_clamp/xdl/mem/device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instance.in Normal file
View File

@@ -0,0 +1,56 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances = std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
F16,
F16,
Tuple<F16, F16, F16, F16, F16>,
F16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>;
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
template <int Shards, int ShardIndex>
void add_device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances_shard(device_grouped_conv2d_fwd_bias_bn_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(instances, ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
ConvFwdDefault,
Intrawave,Tuple<F16, F16, F16, F16, F16>, BiasNormalizeInInferClamp>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances, ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC, Tuple<NHWGK, NHWGK, NHWGK, NHWGK, NHWGK>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,Tuple<F16, F16, F16, F16, F16>, BiasNormalizeInInferClamp>,
Shards,
ShardIndex>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

2
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_clamp/CMakeLists.txt
View File

@@ -6,6 +6,7 @@ add_instance_library(device_grouped_conv2d_fwd_bias_clamp_instance
xdl/merged_groups/device_grouped_conv2d_fwd_bias_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_part2_instance.cpp
@@ -16,6 +17,7 @@ add_instance_library(device_grouped_conv2d_fwd_bias_clamp_instance
xdl/merged_groups/device_grouped_conv2d_fwd_bias_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_fp16_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_direct_load_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_part2_instance.cpp

57
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_clamp/xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,57 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
BF16,
BF16,
Tuple<BF16>,
BF16,
PassThrough,
PassThrough,
AddClamp>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
ConvFwdDefault,
Intrawave,
Tuple<BF16>,
AddClamp>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,
Tuple<BF16>,
AddClamp>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

57
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_bias_clamp/xdl/mem/device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,57 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_bias_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
F16,
F16,
Tuple<F16>,
F16,
PassThrough,
PassThrough,
AddClamp>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
ConvFwdDefault,
Intrawave,
Tuple<F16>,
AddClamp>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<NHWGK>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,
Tuple<F16>,
AddClamp>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

2
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_clamp/CMakeLists.txt
View File

@@ -6,6 +6,7 @@ add_instance_library(device_grouped_conv2d_fwd_clamp_instance
xdl/merged_groups/device_grouped_conv2d_fwd_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_part2_instance.cpp
@@ -16,6 +17,7 @@ add_instance_library(device_grouped_conv2d_fwd_clamp_instance
xdl/merged_groups/device_grouped_conv2d_fwd_clamp_xdl_merged_groups_nhwgc_gkyxc_nhwgk_fp16_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_direct_load_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_comp_part2_instance.cpp

57
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_clamp/xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,57 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_bf16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
BF16,
BF16,
Tuple<>,
BF16,
PassThrough,
PassThrough,
Clamp>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault,
Intrawave,
Tuple<>,
Clamp>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,
Tuple<>,
Clamp>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

57
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_clamp/xdl/mem/device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_fp16_direct_load_instance.cpp Normal file
View File

@@ -0,0 +1,57 @@
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_clamp_xdl_nhwgc_gkyxc_nhwgk_f16_direct_load_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
F16,
F16,
Tuple<>,
F16,
PassThrough,
PassThrough,
Clamp>>>& instances)
{
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwdDefault,
Intrawave,
Tuple<>,
Clamp>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_direct_load_instances<2,
NHWGC,
GKYXC,
Tuple<>,
NHWGK,
ConvFwd1x1S1P0,
Intrawave,
Tuple<>,
Clamp>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck