Skip to content

Grouped conv_fwd_bias_bnorm_clamp instances and tests #3525

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
ApoorvaKalyani wants to merge 21 commits into
base: develop
Choose a base branch
from
Draft

Grouped conv_fwd_bias_bnorm_clamp instances and tests #3525

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
21 commits
Select commit Hold shift + click to select a range
efb1c23
Added bias_bnorm_clamp instances.
ApoorvaKalyani Oct 13, 2025
0c616d6
fwd_bias_bnorm_clamp comp instances
ApoorvaKalyani Oct 17, 2025
b286483
fwd_bias_bnorm_mem_inter and mem_intra instances
ApoorvaKalyani Oct 17, 2025
8b2ae23
fwd_bias_bnorm_merged_group_instances
ApoorvaKalyani Oct 17, 2025
7712702
fwd_bias_bnorm_clamp_conv3d_bf16 and f16 instances
ApoorvaKalyani Oct 17, 2025
2e2c38f
Device level changes for fwd_bias_bnorm_clamp
ApoorvaKalyani Oct 17, 2025
3a60216
Added the test to the regression test list.
ApoorvaKalyani Oct 17, 2025
85ebc6f
Removed the part 2 and 2x instances
ApoorvaKalyani Dec 23, 2025
9645a87
Removed the irrelevant checks in wmma
ApoorvaKalyani Dec 23, 2025
459aee0
Refactored the instances to adapt to new device implementation
ApoorvaKalyani Dec 29, 2025
69b350b
Updated the reference and include files
ApoorvaKalyani Dec 29, 2025
93bf907
enabling tests
ApoorvaKalyani Dec 31, 2025
2da4ff6
Added missing profiler
ApoorvaKalyani Jan 2, 2026
b76de8a
Updated tolerance.
ApoorvaKalyani Jan 5, 2026
76cc7b7
Changing the tolerance to int values
ApoorvaKalyani Jan 7, 2026
39bd224
fixed clang
ApoorvaKalyani Jan 7, 2026
edf10a2
Updated tolerance
ApoorvaKalyani Jan 9, 2026
202ef52
Created and updated the new seperate instance file
ApoorvaKalyani Jan 9, 2026
1ba986b
Clang fix
ApoorvaKalyani Jan 9, 2026
26abd65
Fixed formatting of Instance file
ApoorvaKalyani Jan 9, 2026
4be35b9
Added missing instance entry , deleted by mistake
ApoorvaKalyani Jan 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
263 changes: 263 additions & 0 deletions ...e/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_bias_bn_clamp_instance.hpp
View file
Open in desktop

Large diffs are not rendered by default.

7 changes: 4 additions & 3 deletions ...ration_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_instance.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,9 +24,10 @@ using Empty_Tuple = ck::Tuple<>;

using namespace ck::tensor_layout::convolution;

using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using AddClamp = ck::tensor_operation::element_wise::AddClamp;
using Clamp = ck::tensor_operation::element_wise::Clamp;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using AddClamp = ck::tensor_operation::element_wise::AddClamp;
using Clamp = ck::tensor_operation::element_wise::Clamp;
using BiasNormalizeInInferClamp = ck::tensor_operation::element_wise::BiasNormalizeInInferClamp;

static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
Expand Down
93 changes: 93 additions & 0 deletions ...ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bias_bnorm_clamp.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,10 @@
#include "grouped_convolution_forward_bias_bnorm_clamp_xdl.inc"
#endif

#ifdef CK_USE_WMMA
#include "grouped_convolution_forward_bias_bnorm_clamp_wmma_cshufflev3.inc"
#endif

namespace ck {
namespace tensor_operation {
namespace device {
Expand Down Expand Up @@ -279,6 +283,95 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
}
#endif // CK_USE_XDL

#ifdef CK_USE_WMMA
// layout NHWGC/GKYXC/NHWGK
if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
{
#ifdef CK_ENABLE_BF16
if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<WeiDataType, ck::bhalf_t> &&
is_same_v<OutDataType, ck::bhalf_t> &&
is_same_v<AComputeType, ck::bhalf_t> &&
is_same_v<BComputeType, ck::bhalf_t>)
{
add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part1(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part2(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part3(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part4(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
is_same_v<OutDataType, half_t> && is_same_v<AComputeType, half_t> &&
is_same_v<BComputeType, half_t>)
{
add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part1(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part2(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part3(
op_ptrs);

add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part4(
op_ptrs);
}
#endif
}
// layout NDHWGC/GKZYXC/NDHWGK
if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
{
#ifdef CK_ENABLE_BF16
if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
is_same_v<WeiDataType, ck::bhalf_t> &&
is_same_v<OutDataType, ck::bhalf_t> &&
is_same_v<AComputeType, ck::bhalf_t> &&
is_same_v<BComputeType, ck::bhalf_t>)
{
add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part1(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part2(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part3(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP16
if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
is_same_v<OutDataType, half_t> && is_same_v<AComputeType, half_t> &&
is_same_v<BComputeType, half_t>)
{
add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part1(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part2(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part3(
op_ptrs);

add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
op_ptrs);
}
#endif
}
#endif // CK_USE_WMMA

return op_ptrs;
}
};
Expand Down
261 changes: 261 additions & 0 deletions ...r_operation_instance/gpu/grouped_convolution_forward_bias_bnorm_clamp_wmma_cshufflev3.inc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,261 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.

#pragma once

namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {

#ifdef CK_ENABLE_BF16

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part1(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part2(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part3(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_instances_part4(
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>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part1(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16, BF16, BF16, BF16, BF16>,
BF16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);
void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part2(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16, BF16, BF16, BF16, BF16>,
BF16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part3(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16, BF16, BF16, BF16, BF16>,
BF16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_instances_part4(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
BF16,
BF16,
Tuple<BF16, BF16, BF16, BF16, BF16>,
BF16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

#endif

#ifdef CK_ENABLE_FP16

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part1(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part2(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part3(
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>>>& instances);

void add_device_grouped_conv2d_fwd_bias_bn_clamp_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_instances_part4(
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>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part1(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16, F16, F16, F16, F16>,
F16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part2(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16, F16, F16, F16, F16>,
F16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part3(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16, F16, F16, F16, F16>,
F16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

void add_device_grouped_conv3d_fwd_bias_bn_clamp_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_instances_part4(
std::vector<std::unique_ptr<
DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
Tuple<NDHWGK, NDHWGK, NDHWGK, NDHWGK, NDHWGK>,
NDHWGK,
F16,
F16,
Tuple<F16, F16, F16, F16, F16>,
F16,
PassThrough,
PassThrough,
BiasNormalizeInInferClamp>>>& instances);

#endif

} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
Loading