mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-12 17:26:00 +00:00
Grouped convolution backward data WMMA v3 implementation (#3460)
* Added device level implementation for bwd_data_wmma_v3. * Added first instance of bwd_data_wmma_v3(f16). * Add support for bwd data in gridwise implementation Some changes are general for convolution and some are specific for bwd data. We need to generalize them once we have fwd, bwd data and bwd weight * Initial device implementation of bwd data * Remove unused template parameters in device impl * Add one instance for different layout initial check of device implementation * Add tests for splitk and for different layouts * Appended more instances to wmma_v3_f16. * Added conv_2d bf16 wmma_v3 instances. * Added conv_3d_bf16 wmma_v3_instances. * Added conv_3d_f16_wmma_v3_instances. * Added SplitN test cases for wmma. * Conv3d_bwd_data_scale_wmma_v3 instances. * Conv3d_bwd_data_bilinear_wmma_v3_instances * Renaming the device level instances file to common name , since it is defined for different DataTypes. * Renaming the instances and fixing typo * Added the test cases to regression test list * NCHW support for wmma_v3 * Examples for bf16 and f16 bwd_data_wmma_v3 * Added transpose conditons for device impl * fixing bugs * Added the gemm_args array implmentation * WIP debug conv bwd * fix splitk * Grouped gemm fix * Update CmakeLists with EOF * Added more instances for tests * Fixed the run time error in examples and removed 3d conv examples. * Fixed a typo. * Updated CmakeLists to removed the 3d convultion deleted files * Added print error statements for unsupoorted argument * Added the merge conflict related changes * Fixed compilation error * Fixed the InstanceFactory duplication error. * Removed the print statements and added logs to Arg function * All the merge conflict related errors resolved * Added d_tensor tests. * Added the missing example types of wmm_v3 * Merge error fix * Corrected the instance name * Reverted the bias relu change * Revereted the transpose load local change * Updated the regression test list with bwd_data_scale * Revert "Revereted the transpose load local change" This reverts commit 0b7281edb2bf008e407006690a00621174d9d19b. * Revert "Merge error fix" This reverts commit f3c85daa474b1b83d10c8a3ce077354e71d91a2b. * Reverting the local change * Added merge error fix * Build error fix due to merge conflicts * Added bias_relu example for wmma_v3 * Modified the main method in dtensor tests * Updated the dtensor tests to pick all the shapes * Updated the dtensor test shapes. * Updated the mem operations in tests. * Added reference func * Fixed typos in device impl * Added new header file and modified the include file for 3d tests * Renamed the test file and added reference func call. * clang format fix * Added ignore params * Modified device impl and tests * Removed debug print statements and updated dtensor test shapes * Fixing merge conflicts * Fixing more merge conflicts * Fixed copyrights * Updated the tuned instances to bilinear and scale. * Adding tuned instances to vanilla wmma_v3 * Removed all unused instances and modified test layouts. * Cleaned up all instances , reverted back fwd fp16 instances and updated tuned fp16 instances. * Fix clang format * Updated tuned f16/-genric instances * Formatting the instances file * Fixed copyrights and clang issues * Nonsense commit to force git to force * Removed the transpose instances * Added verified genric instances * Fixing namespace errors * Added todo for failing shapes * Formatting instance file * Fix instance list formatting * Removing unnecessary formats * Renamed the common file * Unification of xdl and wmma bwd_data tests * Updated Cmake * Added all layout types and deleted code. * Updated Cmake to add the condition to all tests. --------- Co-authored-by: Enrico Degregori <enrico@streamhpc.com> Co-authored-by: Anton Gorenko <anton@streamhpc.com> Co-authored-by: kiefer <kiefer.van.teutem@streamhpc.com>
This commit is contained in:
ApoorvaKalyani
@@ -0,0 +1,100 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle_v3.hpp"
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
using BF16 = ck::bhalf_t;
|
||||
using F16 = ck::half_t;
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using namespace ck::tensor_layout::convolution;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
|
||||
|
||||
static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
|
||||
|
||||
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
|
||||
ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
|
||||
|
||||
// bf16_bf16_f32_bf16
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_bilinear_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 32, 64, 32, 8, 8, 16, 16, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 32, 32, 128, 8, 8, 16, 16, 2, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 64, 32, 8, 8, 16, 16, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 64, 64, 32, 8, 8, 16, 16, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 32, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 64, 64, 32, 8, 8, 16, 16, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Tuple<BF16>, BF16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 256, 128, 128, 32, 8, 8, 16, 16, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 16>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
// f16_f16_f32_f16
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_bilinear_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Tuple<F16>, F16, PassThrough, PassThrough, Bilinear, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,125 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle_v3.hpp"
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
using BF16 = ck::bhalf_t;
|
||||
using F16 = ck::half_t;
|
||||
using F32 = float;
|
||||
using BF8 = ck::bf8_t;
|
||||
using F8 = ck::f8_t;
|
||||
|
||||
using Empty_Tuple = ck::Tuple<>;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using namespace ck::tensor_layout::convolution;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
|
||||
|
||||
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
|
||||
ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_f16_16_16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 32, 64, 8, 8, 16, 16, 4, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 32, 64, 8, 8, 16, 16, 4, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 64, 32, 8, 8, 16, 16, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 32, 64, 32, 8, 8, 16, 16, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 64, 64, 8, 8, 16, 16, 8, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_bf16_16_16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 32, 32, 128, 8, 8, 16, 16, 2, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 32, 64, 32, 8, 8, 16, 16, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 64, 32, 32, 128, 8, 8, 16, 16, 2, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 64, 32, 8, 8, 16, 16, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, PassThrough, ConvSpec, true, true, 256, 128, 128, 32, 8, 8, 16, 16, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 16>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,102 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle_v3.hpp"
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
using BF16 = ck::bhalf_t;
|
||||
using F16 = ck::half_t;
|
||||
using F32 = float;
|
||||
|
||||
using Empty_Tuple = ck::Tuple<>;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using namespace ck::tensor_layout::convolution;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
using Scale = ck::tensor_operation::element_wise::Scale;
|
||||
|
||||
static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
|
||||
|
||||
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
|
||||
ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
|
||||
|
||||
// bf16_bf16_f32_bf16
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_scale_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 32, 64, 32, 8, 8, 16, 16, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 32, 32, 128, 8, 8, 16, 16, 2, 1, S<16, 4, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<16, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 64, 32, 8, 8, 16, 16, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 64, 64, 32, 8, 8, 16, 16, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 32, 32, 8, 8, 16, 16, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 32, 64, 32, 8, 8, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 64, 64, 32, 8, 8, 16, 16, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, BF16, BF16, F32, BF16, Empty_Tuple, BF16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 256, 128, 128, 32, 8, 8, 16, 16, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 16>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
// f16_f16_f32_f16
|
||||
template <index_t NDimSpatial,
|
||||
typename ALayout,
|
||||
typename BLayout,
|
||||
typename DsLayout,
|
||||
typename ELayout,
|
||||
ConvolutionBackwardDataSpecialization ConvSpec>
|
||||
using device_grouped_conv_bwd_data_wmma_v3_scale_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
// ##############################################| NDim| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle | CShuffle | CDEBlockTransfer| CDEBlockTransfer|
|
||||
// ##############################################| Spatial| | | | | Type| Type| Type| DataType| Type| Type| Operation| Operation| Operation| DataSpecialization| GemmM| GemmN| Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat | _MBlock_MPerBlock| ScalarPerVector|
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _NBlock_NPerBlock| |
|
||||
// ##############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
// generic instance
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, S<1,1,1>>,
|
||||
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 64, 64, 64, 32, 8, 8, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 4>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>,
|
||||
DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffleV3<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, F16, F16, F32, F16, Empty_Tuple, F16, PassThrough, PassThrough, Scale, ConvSpec, true, true, 128, 128, 128, 32, 8, 8, 16, 16, 8, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 8, 1, 1, 1, S<1, 16, 1, 8>, S<8,8,8>>
|
||||
// clang-format on
|
||||
>;
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -422,6 +422,7 @@ struct DeviceOperationInstanceFactory<
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef CK_ENABLE_INT8
|
||||
if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
|
||||
is_same_v<OutDataType, int8_t> && is_same_v<ComputeTypeA, int8_t> &&
|
||||
@@ -441,12 +442,27 @@ struct DeviceOperationInstanceFactory<
|
||||
is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
|
||||
is_same_v<ComputeTypeB, F16>)
|
||||
{
|
||||
add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_f16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_f16_16_16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
#ifdef CK_ENABLE_BF16
|
||||
if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
|
||||
is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
|
||||
is_same_v<ComputeTypeB, BF16>)
|
||||
{
|
||||
add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_bf16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_bf16_16_16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
#ifdef CK_ENABLE_INT8
|
||||
if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
|
||||
is_same_v<OutDataType, int8_t> && is_same_v<ComputeTypeA, int8_t> &&
|
||||
@@ -475,6 +491,7 @@ struct DeviceOperationInstanceFactory<
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef CK_ENABLE_INT8
|
||||
if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
|
||||
is_same_v<OutDataType, int8_t> && is_same_v<ComputeTypeA, int8_t> &&
|
||||
@@ -499,6 +516,21 @@ struct DeviceOperationInstanceFactory<
|
||||
op_ptrs);
|
||||
add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_f16_16_16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
#ifdef CK_ENABLE_BF16
|
||||
if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
|
||||
is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
|
||||
is_same_v<ComputeTypeB, BF16>)
|
||||
{
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
op_ptrs);
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_bf16_16_16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
#ifdef CK_ENABLE_INT8
|
||||
@@ -515,7 +547,6 @@ struct DeviceOperationInstanceFactory<
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
return op_ptrs;
|
||||
}
|
||||
};
|
||||
|
||||
@@ -15,6 +15,7 @@ namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
#ifdef CK_USE_XDL
|
||||
#ifdef CK_ENABLE_FP16
|
||||
void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
@@ -78,6 +79,42 @@ void add_device_grouped_conv3d_bwd_data_xdl_bilinear_ndhwgk_gkzyxc_ndhwgc_bf16_i
|
||||
PassThrough,
|
||||
Bilinear>>>& instances);
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifdef CK_USE_WMMA
|
||||
#ifdef CK_ENABLE_FP16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_bilinear_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Tuple<NDHWGC>,
|
||||
NDHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Tuple<F16>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Bilinear>>>& instances);
|
||||
#endif
|
||||
|
||||
#ifdef CK_ENABLE_BF16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_bilinear_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Tuple<NDHWGC>,
|
||||
NDHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Tuple<BF16>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Bilinear>>>& instances);
|
||||
#endif
|
||||
#endif
|
||||
|
||||
template <ck::index_t NumDimSpatial,
|
||||
typename OutLayout,
|
||||
typename WeiLayout,
|
||||
@@ -123,6 +160,8 @@ struct DeviceOperationInstanceFactory<
|
||||
static auto GetInstances()
|
||||
{
|
||||
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
|
||||
|
||||
#ifdef CK_USE_XDL
|
||||
if constexpr(NumDimSpatial == 3)
|
||||
{
|
||||
if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
|
||||
@@ -169,6 +208,38 @@ struct DeviceOperationInstanceFactory<
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef CK_USE_WMMA
|
||||
|
||||
if constexpr(NumDimSpatial == 3)
|
||||
{
|
||||
if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
|
||||
is_same_v<OutLayout, NDHWGK>)
|
||||
{
|
||||
#ifdef CK_ENABLE_FP16
|
||||
if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
|
||||
is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
|
||||
is_same_v<ComputeTypeB, F16>)
|
||||
{
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_bilinear_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef CK_ENABLE_BF16
|
||||
else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
|
||||
is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
|
||||
is_same_v<ComputeTypeB, BF16>)
|
||||
{
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_bilinear_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
return op_ptrs;
|
||||
}
|
||||
};
|
||||
|
||||
@@ -15,6 +15,7 @@ namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace instance {
|
||||
|
||||
#ifdef CK_USE_XDL
|
||||
#ifdef CK_ENABLE_FP16
|
||||
void add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
@@ -78,6 +79,41 @@ void add_device_grouped_conv3d_bwd_data_xdl_scale_ndhwgk_gkzyxc_ndhwgc_bf16_inst
|
||||
PassThrough,
|
||||
Scale>>>& instances);
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifdef CK_USE_WMMA
|
||||
#ifdef CK_ENABLE_FP16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_scale_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Tuple<>,
|
||||
NDHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Tuple<>,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Scale>>>& instances);
|
||||
#endif
|
||||
#ifdef CK_ENABLE_BF16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_scale_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Tuple<>,
|
||||
NDHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Tuple<>,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Scale>>>& instances);
|
||||
#endif
|
||||
#endif
|
||||
|
||||
template <ck::index_t NumDimSpatial,
|
||||
typename OutLayout,
|
||||
typename WeiLayout,
|
||||
@@ -123,6 +159,7 @@ struct DeviceOperationInstanceFactory<
|
||||
static auto GetInstances()
|
||||
{
|
||||
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
|
||||
#ifdef CK_USE_XDL
|
||||
if constexpr(NumDimSpatial == 3)
|
||||
{
|
||||
if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
|
||||
@@ -168,6 +205,36 @@ struct DeviceOperationInstanceFactory<
|
||||
#endif
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef CK_USE_WMMA
|
||||
|
||||
if constexpr(NumDimSpatial == 3)
|
||||
{
|
||||
if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
|
||||
is_same_v<OutLayout, NDHWGK>)
|
||||
{
|
||||
#ifdef CK_ENABLE_FP16
|
||||
if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
|
||||
is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
|
||||
is_same_v<ComputeTypeB, F16>)
|
||||
{
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_scale_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
#ifdef CK_ENABLE_BF16
|
||||
else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
|
||||
is_same_v<OutDataType, BF16> && is_same_v<ComputeTypeA, BF16> &&
|
||||
is_same_v<ComputeTypeB, BF16>)
|
||||
{
|
||||
add_device_grouped_conv3d_bwd_data_wmma_v3_scale_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
op_ptrs);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
return op_ptrs;
|
||||
}
|
||||
|
||||
@@ -38,6 +38,34 @@ void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_insta
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
|
||||
NHWGK,
|
||||
GKYXC,
|
||||
Empty_Tuple,
|
||||
NHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Empty_Tuple,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_f16_16_16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
|
||||
NHWGK,
|
||||
GKYXC,
|
||||
Empty_Tuple,
|
||||
NHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Empty_Tuple,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
|
||||
NHWGK,
|
||||
@@ -65,7 +93,6 @@ void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_insta
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
GNDHWK,
|
||||
@@ -237,6 +264,99 @@ void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_ins
|
||||
PassThrough>>>& instances);
|
||||
#endif
|
||||
|
||||
#ifdef CK_ENABLE_BF16
|
||||
|
||||
void add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_bf16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
|
||||
NHWGK,
|
||||
GKYXC,
|
||||
Empty_Tuple,
|
||||
NHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Empty_Tuple,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv2d_bwd_data_wmma_v3_nhwgk_gkyxc_nhwgc_bf16_16_16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
|
||||
NHWGK,
|
||||
GKYXC,
|
||||
Empty_Tuple,
|
||||
NHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Empty_Tuple,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
#endif
|
||||
|
||||
// conv3dbwdData
|
||||
|
||||
#ifdef CK_ENABLE_BF16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Empty_Tuple,
|
||||
NDHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Empty_Tuple,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_bf16_16_16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Empty_Tuple,
|
||||
NDHWGC,
|
||||
BF16,
|
||||
BF16,
|
||||
Empty_Tuple,
|
||||
BF16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
#endif
|
||||
#ifdef CK_ENABLE_FP16
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_f16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Empty_Tuple,
|
||||
NDHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Empty_Tuple,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
|
||||
void add_device_grouped_conv3d_bwd_data_wmma_v3_ndhwgk_gkzyxc_ndhwgc_f16_16_16_instances(
|
||||
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
|
||||
NDHWGK,
|
||||
GKZYXC,
|
||||
Empty_Tuple,
|
||||
NDHWGC,
|
||||
F16,
|
||||
F16,
|
||||
Empty_Tuple,
|
||||
F16,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
PassThrough>>>& instances);
|
||||
#endif
|
||||
|
||||
} // namespace instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
|
||||
Reference in New Issue
Block a user