ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-16 19:09:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Shard several of the most costly targets. (#2266)

Browse Source 
* Shard several of the most costly targets.

Introduces a filter_tuple_by_modulo to break up tuples.

Drops build time of target from 21 minutes to under 14 minutes with 64
build processes, or 11 minutes with 128 build processes.

time ninja -j 64 device_grouped_conv3d_fwd_instance

* fix clang format

* Fix build errors in instantiation code.

I wasn't sure how to test the header-only instantiation code on my
initial commit. From Jenkins CI test results, I see that there is a
test target that depends on these headers:

ninja -j 128 test_grouped_convnd_fwd

This allowed me to test the build locally. I found three mistakes I
made, mostly related to early experiments on I tried on the code.
This was hard to find earlier because this PR is really too large.

I also discovered that there are five 2D convolution targets that now
dominate the compilation time. I will likely address those in a later
PR, rather than adding even more changes to this PR.

* Fix link errors from mismatched declarations.

Our pattern for instantiating MIOpen templates uses duplicate
declarations (instead of headers). This is fragile, and I didn't
notice that my last commit had a bunch of link errors. I fixed these
mistakes, and the bin/test_grouped_conv_fwd test target binary now links
correctly.

* Migrate the design to a code-generation approach.

Use a CMake function with template files to generate the source files for the
intantiating the kerenels and to generate the calling function.

* Shard the longest 2D convolution builds

Now that we have automated the shard instantiation, we can shard the 2D
convolution targets that take the longest to build. The target
test_grouped_conv2d_fwd now compiles in 15 minutes.

* Use PROJECT_SOURCE_DIR for submodule compatibility

I used CMAKE_SOURCE_DIR to refer to the top-level source directory in
the ShardInstantiation.cmake file, but this can cause issues with
git submodules.  Instead, we should use PROJECT_SOURCE_DIR to ensure
compatibility when this project is used as a submodule in another
project.

---------

Co-authored-by: illsilin <Illia.Silin@amd.com>

[ROCm/composable_kernel commit: 3a0cb27966]
This commit is contained in:
John Shumway
2025-06-13 03:58:50 -07:00
committed by GitHub
parent 10498656ef
commit ea36ae016e
42 changed files with 1325 additions and 827 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
View File

@@ -1,5 +1,5 @@
# XDL_DL_WMMA_KERNELS
add_instance_library(device_grouped_conv2d_fwd_instance
set(GROUPED_CONV2D_FWD
#xdl
# GNHWC, GKYXC, GNHWK
xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -19,8 +19,6 @@ add_instance_library(device_grouped_conv2d_fwd_instance
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_f32_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_int8_instance.cpp
# NGCHW, GKCYX, NGKHW
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f32_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_16x16_instance.cpp
xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_16x16_instance.cpp
@@ -46,12 +44,10 @@ add_instance_library(device_grouped_conv2d_fwd_instance
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp
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_int8_mem_intra_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
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.cpp
# NGCHW, GKCYX, NGKHW
xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_mem_intra_instance.cpp
@@ -71,7 +67,6 @@ add_instance_library(device_grouped_conv2d_fwd_instance
xdl/comp/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_comp_part2_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_comp_part2_instance.cpp
# NGCHW, GKCYX, NGKHW
xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f32_comp_instance.cpp
xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_2x_instance.cpp
@@ -105,3 +100,47 @@ add_instance_library(device_grouped_conv2d_fwd_instance
wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp
wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp
)
# Add generated files for sharded instantiations.
include(ShardInstantiation)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instances
TEMPLATE_FILE xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instance.in
NUM_SHARDS 16
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instances
TEMPLATE_FILE xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instance.in
NUM_SHARDS 16
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instances
TEMPLATE_FILE xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instance.in
NUM_SHARDS 16
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.in
NUM_SHARDS 21
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.in
NUM_SHARDS 21
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
add_instance_library(device_grouped_conv2d_fwd_instance ${GROUPED_CONV2D_FWD})

38
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/comp/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instance.in
View File

@@ -1,16 +1,14 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
// 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_comp_instance.hpp"
#include "ck/host_utility/device_prop.hpp"
#include "ck/utility/filter_tuple.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_ngchw_gkcyx_ngkhw_bf16_comp_instances(
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
@@ -22,19 +20,23 @@ void add_device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instances(
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
// 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_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instances_shard([[maybe_unused]]
device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_comp_instances& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>{});
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_comp_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

40
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instance.in
View File

@@ -3,13 +3,11 @@
#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_instance.hpp"
#include "ck/utility/filter_tuple.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_ngchw_gkcyx_ngkhw_bf16_instances(
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
@@ -21,32 +19,40 @@ void add_device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instances(
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
// 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_xdl_ngchw_gkcyx_ngkhw_bf16_instances_shard(
device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_bf16_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<2,
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>{});
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<2,
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1P0>{});
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<2,
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1S1P0>{});
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

64
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instance.in
View File

@@ -3,13 +3,11 @@
#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_instance.hpp"
#include "ck/utility/filter_tuple.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_ngchw_gkcyx_ngkhw_f16_instances(
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
@@ -21,32 +19,40 @@ void add_device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instances(
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
// 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_xdl_ngchw_gkcyx_ngkhw_f16_instances_shard(
device_grouped_conv2d_fwd_xdl_ngchw_gkcyx_ngkhw_f16_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>{});
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1P0>{});
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1S1P0>{});
ck::util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

66
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.cpp
View File

@@ -1,66 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, 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_int8_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

80
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.in Normal file
View File

@@ -0,0 +1,80 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, 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::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>;
// 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_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances_shard(
device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

66
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.cpp
View File

@@ -1,66 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, 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_int8_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

80
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.in Normal file
View File

@@ -0,0 +1,80 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, 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::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>;
// 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_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances_shard(
device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

109
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
View File

@@ -11,8 +11,6 @@ set(GROUPED_CONV3D_FWD
xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_16x16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_16x16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_16x16_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_16x16_instance.cpp
@@ -32,23 +30,13 @@ set(GROUPED_CONV3D_FWD
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_comp_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_2x_instance.cpp
xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_part2_instance.cpp
@@ -71,6 +59,99 @@ set(GROUPED_CONV3D_FWD
wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
)
# Add generated files for sharded instantiations.
include(ShardInstantiation)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instances
TEMPLATE_FILE xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instance.in
NUM_SHARDS 8
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances
TEMPLATE_FILE xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.in
NUM_SHARDS 8
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances
TEMPLATE_FILE xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instance.in
NUM_SHARDS 10
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/mem
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instances
TEMPLATE_FILE xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instances
TEMPLATE_FILE xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instances
TEMPLATE_FILE xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instances
TEMPLATE_FILE xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instance.in
NUM_SHARDS 12
SRC_LIST GROUPED_CONV3D_FWD
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
list(APPEND GROUPED_CONV3D_FWD

111
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.cpp
View File

@@ -1,111 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/host_utility/device_prop.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
if(ck::get_device_name() != "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
}
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

66
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.in Normal file
View File

@@ -0,0 +1,66 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instances_shard(
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instances& instances)
{
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

111
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.cpp
View File

@@ -1,111 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/host_utility/device_prop.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
if(ck::get_device_name() != "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_part2<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
}
if(ck::get_device_name() == "gfx950")
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances_2x<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>{});
}
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

65
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.in Normal file
View File

@@ -0,0 +1,65 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instances_shard(
device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instances& instances)
{
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NDHWGC,
GKZYXC,
Empty_Tuple,
NDHWGK,
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

54
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instance.cpp
View File

@@ -1,54 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_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_ngcdhw_gkczyx_ngkdhw_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

65
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instance.in Normal file
View File

@@ -0,0 +1,65 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_comp_instances& instances)
{
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

54
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instance.cpp
View File

@@ -1,54 +0,0 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_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_ngcdhw_gkczyx_ngkdhw_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

63
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/comp/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instance.in Normal file
View File

@@ -0,0 +1,63 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_comp_instances& instances)
{
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_comp_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

53
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instance.cpp
View File

@@ -1,53 +0,0 @@
// 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_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_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

53
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instance.in
View File

@@ -1,15 +1,14 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#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_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instances(
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
@@ -21,35 +20,43 @@ void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_insta
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_bf16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>{});
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

53
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.cpp
View File

@@ -1,53 +0,0 @@
// 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_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_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

53
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.in
View File

@@ -1,15 +1,14 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#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_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances(
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
@@ -21,35 +20,43 @@ void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instan
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
ConvFwdDefault>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
ConvFwd1x1P0>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
util::filter_tuple_by_modulo_t<device_grouped_conv_fwd_xdl_f16_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>{});
ConvFwd1x1S1P0>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_1of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 0>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_2of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 1>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_3of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 2>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_4of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 3>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_5of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 4>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_6of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 5>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_7of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 6>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

9
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance_8of8.cpp Normal file
View File

@@ -0,0 +1,9 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instance.inc"
namespace ck::tensor_operation::device::instance {
template void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances_sharded<8, 7>(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_instances& instances);
} // namespace ck::tensor_operation::device::instance

64
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instance.in Normal file
View File

@@ -0,0 +1,64 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

65
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instance.in Normal file
View File

@@ -0,0 +1,65 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_intra_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

69
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instance.in
View File

@@ -3,13 +3,11 @@
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances(
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
@@ -21,35 +19,44 @@ void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instan
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

75
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instance.in
View File

@@ -3,53 +3,60 @@
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_bf16_mem_inter_instances(
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
BF16,
BF16,
F16,
F16,
Empty_Tuple,
BF16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_bf16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f16_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

69
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instance.in
View File

@@ -3,13 +3,11 @@
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instances(
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
@@ -21,35 +19,44 @@ void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instan
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_inter_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Interwave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Interwave>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

69
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instance.in
View File

@@ -3,13 +3,11 @@
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
namespace ck::tensor_operation::device::instance {
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances(
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
@@ -21,35 +19,44 @@ void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instan
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>{});
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
} // namespace ck::tensor_operation::device::instance

62
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/mem/device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instance.inc Normal file
View File

@@ -0,0 +1,62 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>;
template <int Shards, int ShardIndex>
void add_device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances_shard(
device_grouped_conv3d_fwd_xdl_ngcdhw_gkczyx_ngkdhw_f32_mem_intra_instances& instances)
{
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwdDefault,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1P0,
Intrawave>,
Shards,
ShardIndex>{});
add_device_operation_instances(instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_xdl_f32_mem_instances<3,
NGCDHW,
GKCZYX,
Empty_Tuple,
NGKDHW,
ConvFwd1x1S1P0,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance