ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-13 02:27:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add comp and mem instances for all vanilla 2D grouped conv fwd types. Skipping "x2" and "part2" instance lists, can be added later without special names if necessary.

Browse Source
This commit is contained in:
kiefer
2025-08-29 13:36:31 +00:00
parent 382d6fea6b
commit 63f52e0618
22 changed files with 1525 additions and 265 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_comp_instance.hpp
View File

@@ -51,6 +51,35 @@ static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
static constexpr auto Interwave = BlockGemmPipelineScheduler::Interwave;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Compute friendly
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 256, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 256, 32, 8, 8, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 256, 32, 8, 8, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 64, 64, 8, 8, 32, 32, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 64, 128, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 64, 64, 64, 8, 8, 32, 32, 1, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
@@ -61,11 +90,29 @@ template <index_t NDimSpatial,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_f16_comp_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 16, 16, 4, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_int8_comp_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 32, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1>
// clang-format on
>;

174
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_cshufflev3_mem_instance.hpp Normal file
View File

@@ -0,0 +1,174 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_wmma_cshuffle_v3.hpp"
#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
#ifdef CK_ENABLE_FP8
using F8 = ck::f8_t;
#endif
#ifdef CK_ENABLE_BF8
using BF8 = ck::bf8_t;
#endif
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
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;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto ConvFwd1x1P0 = ConvolutionForwardSpecialization::Filter1x1Pad0;
static constexpr auto ConvFwd1x1S1P0 = ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
static constexpr auto ConvFwdOddC =
ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
static constexpr auto Interwave = BlockGemmPipelineScheduler::Interwave;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
BlockGemmPipelineScheduler BlkGemmPipeSched,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances =
std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// Latency friendly
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// // Memory friendly
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 16, 64, 8, 8, 16, 16, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 64, 64, 8, 8, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 64, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 128, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 16, 256, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF16, BF16, F32, BF16, DsDataTypes, BF16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
BlockGemmPipelineScheduler BlkGemmPipeSched,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances =
std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// // Memory friendly
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 16, 64, 8, 8, 16, 16, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 64, 64, 8, 8, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 64, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 128, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 16, 256, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
BlockGemmPipelineScheduler BlkGemmPipeSched,
typename DsDataTypes = Tuple<>,
typename OutElementOp = PassThrough>
using device_grouped_conv_fwd_wmma_cshufflev3_int8_mem_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MWmma| NWmma| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | WMMA| WMMA| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MWmmaPerWave| NWmmaPerWave| _MBlock_MWaveMPerWmma| ScalarPerVector| TODO: Other pipelines
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerWmma| _NWaveNPerWmma|
//########################################| | | | | | | | | | | | | | | | | TODO: ??| | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGroupedConvFwdMultipleABD_Wmma_CShuffle_V3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
// // Memory friendly
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 256, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 32, 64, 8, 8, 32, 32, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 128, 16, 64, 8, 8, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 32, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 64, 16, 64, 8, 8, 16, 16, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 128, 8, 8, 16, 16, 1, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 64, 16, 16, 64, 8, 8, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 32, 64, 8, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 64, 64, 8, 8, 16, 16, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 64, 64, 8, 8, 32, 32, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 16, 128, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 128, 32, 128, 64, 8, 8, 32, 32, 1, 2, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 16, 256, 64, 8, 8, 16, 16, 1, 4, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
// DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, int8_t, int8_t, int32_t, int8_t, DsDataTypes, int8_t, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 32, 256, 64, 8, 8, 32, 32, 1, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 16, 1, 16>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

66
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
View File

@@ -30,6 +30,8 @@
#include "grouped_convolution_forward_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_wmma_cshufflev3_merged_groups.inc"
#include "grouped_convolution_forward_comp_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_mem_inter_wmma_cshufflev3.inc"
#include "grouped_convolution_forward_mem_intra_wmma_cshufflev3.inc"
#endif
namespace ck {
@@ -804,14 +806,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_merged_groups_nhwgc_gkyxc_nhwgk_f16_instances(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_2x_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_part2_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_BF16
@@ -829,14 +829,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_merged_groups_nhwgc_gkyxc_nhwgk_bf16_instances(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_2x_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_part2_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_INT8
@@ -850,11 +848,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_merged_groups_nhwgc_gkyxc_nhwgk_int8_instances(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances(
op_ptrs);
}
#endif
}
@@ -876,13 +875,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_instances(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_2x_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_part2_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instances(
op_ptrs);
}
#endif
#ifdef CK_ENABLE_BF16
@@ -898,14 +894,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_16x16_instances(
op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_2x_instances(op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_part2_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instances(
// op_ptrs);
// add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instances(
// op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instances(
op_ptrs);
add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instances(
op_ptrs);
}
#endif
}

275
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_comp_wmma_cshufflev3.inc
View File

@@ -1,5 +1,5 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
@@ -10,107 +10,51 @@ namespace instance {
// grouped conv2d forward, NHWGC/GKYXC/NHWGK
#ifdef CK_ENABLE_BF16
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NHWGC,
// GKYXC,
// Empty_Tuple,
// NHWGK,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_INT8
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_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);
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_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);
#endif
// grouped conv2d forward, NGCHW/GKCYX/NGKHW
@@ -128,78 +72,22 @@ void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_in
PassThrough,
PassThrough,
PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NGCHW,
// GKCYX,
// Empty_Tuple,
// NGKHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NGCHW,
// GKCYX,
// Empty_Tuple,
// NGKHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif // CK_ENABLE_FP16
#ifdef CK_ENABLE_BF16
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NGCHW,
// GKCYX,
// Empty_Tuple,
// NGKHW,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NGCHW,
// GKCYX,
// Empty_Tuple,
// NGKHW,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
// NGCHW,
// GKCYX,
// Empty_Tuple,
// NGKHW,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
@@ -250,34 +138,6 @@ void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_in
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_f16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_f16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
@@ -294,35 +154,6 @@ void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_in
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_bf16_comp_2x_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
// void
// add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_bf16_comp_part2_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
} // namespace instance

162
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_mem_inter_wmma_cshufflev3.inc Normal file
View File

@@ -0,0 +1,162 @@
// 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 {
// grouped conv2d forward, NHWGC/GKYXC/NHWGK
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_INT8
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_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);
#endif
// grouped conv2d forward, NGCHW/GKCYX/NGKHW
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
// grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_mem_inter_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Empty_Tuple,
// NDHWGK,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_mem_inter_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Empty_Tuple,
// NDHWGK,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
// grouped conv3d forward, NGCDHW/GKCZYX/NGKDHW
#ifdef CK_ENABLE_FP16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_f16_mem_inter_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_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>>>& instances);
#endif
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

162
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_mem_intra_wmma_cshufflev3.inc Normal file
View File

@@ -0,0 +1,162 @@
// 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 {
// grouped conv2d forward, NHWGC/GKYXC/NHWGK
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_INT8
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_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);
#endif
// grouped conv2d forward, NGCHW/GKCYX/NGKHW
#ifdef CK_ENABLE_FP16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
void add_device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
// grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_bf16_mem_intra_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Empty_Tuple,
// NDHWGK,
// BF16,
// BF16,
// Empty_Tuple,
// BF16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ndhwgc_gkzyxc_ndhwgk_f16_mem_intra_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// Empty_Tuple,
// NDHWGK,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
// grouped conv3d forward, NGCDHW/GKCZYX/NGKDHW
#ifdef CK_ENABLE_FP16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_ngcdhw_gkczyx_ngkdhw_f16_mem_intra_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NGCDHW,
// GKCZYX,
// Empty_Tuple,
// NGKDHW,
// F16,
// F16,
// Empty_Tuple,
// F16,
// PassThrough,
// PassThrough,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_BF16
// void add_device_grouped_conv3d_fwd_wmma_cshufflev3_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>>>& instances);
#endif
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

43
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
View File

@@ -128,7 +128,24 @@ set(GROUPED_CONV2D_FWD
wmma/merged_groups/device_grouped_conv2d_fwd_wmma_cshufflev3_merged_groups_ngchw_gkcyx_ngkhw_bf16_instance.cpp
wmma/merged_groups/device_grouped_conv2d_fwd_wmma_cshufflev3_merged_groups_ngchw_gkcyx_ngkhw_f16_instance.cpp
# comp
#mem
# NHWGC, GKYXC, NHWGK intra
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instance.cpp
# NHWGC, GKYXC, NHWGK inter
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instance.cpp
# NGCHW, GKCYX, NGKHW intra
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instance.cpp
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instance.cpp
# NGCHW, GKCYX, NGKHW inter
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instance.cpp
wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instance.cpp
#comp
# NHWGC, GKYXC, NHWGK
wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instance.cpp
wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instance.cpp
wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_instance.cpp
# NGCHW, GKCYX, NGKHW
wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_instance.cpp
)
@@ -176,6 +193,14 @@ generate_sharded_instantiations(
OUTPUT_DIR ${GENERATED_DIR}/xdl/comp
)
set(GENERATED_DIR ${CMAKE_CURRENT_BINARY_DIR}/generated)
generate_sharded_instantiations(
INSTANCES_NAME device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances
TEMPLATE_FILE wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instance.in
NUM_SHARDS 1
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/wmma/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
@@ -184,6 +209,14 @@ 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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances
TEMPLATE_FILE wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.in
NUM_SHARDS 1
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/wmma/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
@@ -191,4 +224,12 @@ generate_sharded_instantiations(
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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances
TEMPLATE_FILE wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.in
NUM_SHARDS 1
SRC_LIST GROUPED_CONV2D_FWD
OUTPUT_DIR ${GENERATED_DIR}/wmma/mem
)
add_instance_library(device_grouped_conv2d_fwd_instance ${GROUPED_CONV2D_FWD})

44
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instance.in Normal file
View File

@@ -0,0 +1,44 @@
// 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_wmma_cshufflev3_comp_instance.hpp"
#include "ck/host_utility/device_prop.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances =
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
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_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances_shard(
[[maybe_unused]] device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_comp_instances&
instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

2
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_comp_instance.cpp
View File

@@ -1,5 +1,5 @@
// 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_wmma_cshufflev3_comp_instance.hpp"

67
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instance.cpp Normal file
View File

@@ -0,0 +1,67 @@
// 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_wmma_cshufflev3_comp_instance.hpp"
#include "ck/host_utility/device_prop.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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

67
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instance.cpp Normal file
View File

@@ -0,0 +1,67 @@
// 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_wmma_cshufflev3_comp_instance.hpp"
#include "ck/host_utility/device_prop.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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

67
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/comp/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_instance.cpp Normal file
View File

@@ -0,0 +1,67 @@
// 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_wmma_cshufflev3_comp_instance.hpp"
#include "ck/host_utility/device_prop.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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_comp_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_wmma_cshufflev3_int8_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_int8_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_int8_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_int8_comp_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

40
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instance.cpp Normal file
View File

@@ -0,0 +1,40 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

40
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instance.cpp Normal file
View File

@@ -0,0 +1,40 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_ngchw_gkcyx_ngkhw_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

40
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instance.cpp Normal file
View File

@@ -0,0 +1,40 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

40
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instance.cpp Normal file
View File

@@ -0,0 +1,40 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_ngchw_gkcyx_ngkhw_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NGCHW,
GKCYX,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

70
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp Normal file
View File

@@ -0,0 +1,70 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

70
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp Normal file
View File

@@ -0,0 +1,70 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_bf16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

70
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instance.cpp Normal file
View File

@@ -0,0 +1,70 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Interwave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

70
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instance.cpp Normal file
View File

@@ -0,0 +1,70 @@
// 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_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_wmma_cshufflev3_f16_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

82
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.in Normal file
View File

@@ -0,0 +1,82 @@
// 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_wmma_cshufflev3_mem_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances_shard(
device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_wmma_cshufflev3_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_wmma_cshufflev3_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_wmma_cshufflev3_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_wmma_cshufflev3_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance

82
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/mem/device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.in Normal file
View File

@@ -0,0 +1,82 @@
// 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_wmma_cshufflev3_mem_instance.hpp"
#include "ck/utility/filter_tuple.hpp"
namespace ck::tensor_operation::device::instance {
using device_grouped_conv2d_fwd_wmma_cshufflev3_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_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances_shard(
device_grouped_conv2d_fwd_wmma_cshufflev3_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances& instances)
{
add_device_operation_instances(
instances,
ck::util::filter_tuple_by_modulo_t<
device_grouped_conv_fwd_wmma_cshufflev3_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_wmma_cshufflev3_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_wmma_cshufflev3_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_wmma_cshufflev3_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>,
Shards,
ShardIndex>{});
}
} // namespace ck::tensor_operation::device::instance