Skip to content

GitLab

Commit 289f15de authored by aska-0096's avatar aska-0096
Browse files

Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into wmma_gemm

parents 9bd44685 d58b7f51
Hide whitespace changes
Inline Side-by-side
Showing with 982 additions and 888 deletions
library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp"
#include "ck/utility/data_type.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F64 = double;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename DyElementwiseOp>
using device_batchnorm_backward_f64_blockwise_instances = std::tuple<
// XDataType, DxDataType, DyDataType, AccDataType, ScaleDataType, DscaleDbiasDataType, MeanVarDataType, DyElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XDyDxVectorDim, XSrcVectorSize, DySrcVectorSize, DxDstVectorSize, ScaleSrcVectorSize, DscaleDbiasDstVectorSize, MeanVarSrcVectorSize
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename DyElementwiseOp>
using device_batchnorm_backward_f64_multiblock_instances =
std::tuple <
// XDataType, DxDataType, DyDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, DyElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XDyDxVectorDim, XSrcVectorSize, DySrcVectorSize, DxDstVectorSize, ScaleSrcDstVectorSize, BiasDstVectorSize, MeanVarSrcVectorSize
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 1, 1, 1>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormBwdImpl<F64, F64, F64, F64, F64, F64, F64, DyElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
void add_device_batchnorm_backward_rank_4_3_f64_instances(
std::vector<
std::unique_ptr<DeviceBatchNormBwd<F64, F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&
instances)
{
add_device_operation_instances(
instances, device_batchnorm_backward_f64_blockwise_instances<4, 3, PassThrough>{});
add_device_operation_instances(
instances, device_batchnorm_backward_f64_multiblock_instances<4, 3, PassThrough>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp"
#include "ck/utility/data_type.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 F32 = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_bf16_blockwise_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_bf16_multiblock_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<BF16, BF16, F32, BF16, BF16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
void add_device_batchnorm_forward_rank_4_3_bf16_instances(
std::vector<
std::unique_ptr<DeviceBatchNormFwd<BF16, BF16, F32, BF16, BF16, F32, PassThrough, 4, 3>>>&
instances)
{
add_device_operation_instances(
instances, device_batchnorm_forward_bf16_blockwise_instances<4, 3, PassThrough>{});
add_device_operation_instances(
instances, device_batchnorm_forward_bf16_multiblock_instances<4, 3, PassThrough>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp"
#include "ck/utility/data_type.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f16_blockwise_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f16_multiblock_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F16, F16, F32, F16, F16, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
void add_device_batchnorm_forward_rank_4_3_f16_instances(
std::vector<
std::unique_ptr<DeviceBatchNormFwd<F16, F16, F32, F16, F16, F32, PassThrough, 4, 3>>>&
instances)
{
add_device_operation_instances(
instances, device_batchnorm_forward_f16_blockwise_instances<4, 3, PassThrough>{});
add_device_operation_instances(
instances, device_batchnorm_forward_f16_multiblock_instances<4, 3, PassThrough>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp"
#include "ck/utility/data_type.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F32 = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f32_blockwise_instances = std::tuple<
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f32_multiblock_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F32, F32, F32, F32, F32, F32, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
void add_device_batchnorm_forward_rank_4_3_f32_instances(
std::vector<
std::unique_ptr<DeviceBatchNormFwd<F32, F32, F32, F32, F32, F32, PassThrough, 4, 3>>>&
instances)
{
add_device_operation_instances(
instances, device_batchnorm_forward_f32_blockwise_instances<4, 3, PassThrough>{});
add_device_operation_instances(
instances, device_batchnorm_forward_f32_multiblock_instances<4, 3, PassThrough>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp"
#include "ck/utility/data_type.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F64 = double;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f64_blockwise_instances = std::tuple<
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, false, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
// clang-format off
template <index_t Rank, index_t NumReduceDim, typename YElementwiseOp>
using device_batchnorm_forward_f64_multiblock_instances =
std::tuple <
// XDataType, YDataType, AccDataType, ScaleDataType, BiasDataType, MeanVarDataType, YElementwiseOp, Rank, NumReduceDim, UseMultiBlockInK, BLockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XSrcYDstVectorDim, XSrcVectorSize, YDstVectorSize, ScaleSrcVectorSize, BiasSrcVectorSize, MeanVarSrcDstVectorSize
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 128, 2, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 64, 4, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 32, 8, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 16, 16, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 8, 32, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 4, 64, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 2, 128, 2, 2, 1, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 1, 1, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 0, 2, 2, 1, 1, 1>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 2, 2, 2>,
DeviceBatchNormFwdImpl<F64, F64, F64, F64, F64, F64, YElementwiseOp, Rank, NumReduceDim, true, 256, 1, 256, 2, 2, 1, 1, 1, 1, 1, 1>
>;
// clang-format on
void add_device_batchnorm_forward_rank_4_3_f64_instances(
std::vector<
std::unique_ptr<DeviceBatchNormFwd<F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&
instances)
{
add_device_operation_instances(
instances, device_batchnorm_forward_f64_blockwise_instances<4, 3, PassThrough>{});
add_device_operation_instances(
instances, device_batchnorm_forward_f64_multiblock_instances<4, 3, PassThrough>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv1d_bwd_weight/CMakeLists.txt deleted 100644 → 0
View file @ 9bd44685
add_instance_library(device_conv1d_bwd_weight_instance
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp
device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp
)
library/src/tensor_operation_instance/gpu/conv1d_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_f32_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_bf16_f32_bf16_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_f32_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv1d_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f16_default_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv1d_bwd_weight/device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NWC = ck::tensor_layout::convolution::NWC;
using KXC = ck::tensor_layout::convolution::KXC;
using NWK = ck::tensor_layout::convolution::NWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
using device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 1, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<1,
NWC,
KXC,
NWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_c_shuffle_nwc_kxc_nwk_f32_default_instances{});
add_device_operation_instances(
instances, device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_1x1_s1_p0_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
View file @ 289f15de
......@@ -3,4 +3,8 @@ add_instance_library(device_conv2d_bwd_data_instance
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
)
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.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 InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataDefault, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 8, 2>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 2>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 1, 8, 2>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
InDataType,
WeiDataType,
OutDataType,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.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 InDataType = float;
using WeiDataType = float;
using OutDataType = float;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataDefault, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 1, 8, 1>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 1>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 1, 1, 1>, S<1, 1, 8, 1>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
InDataType,
WeiDataType,
OutDataType,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp 0 → 100644
View file @ 289f15de
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.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 InDataType = int8_t;
using WeiDataType = int8_t;
using OutDataType = int8_t;
using AccDataType = int32_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDataDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataDefault, 256, 128, 128, 16, 4, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<1, 1, 8, 4>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
using device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances = std::tuple<
// clang-format off
//#########################| NDim| InData| WeiData| OutData| AccData| In| Wei| Out| Convolution| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
//#########################| Spatial| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Forward| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
//#########################| | | | | | Operation| Operation| Operation| Specialization| | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceConvNdBwdDataNwcKxcNwk_Dl< 2, InDataType, WeiDataType, OutDataType, AccDataType, InElementOp, WeiElementOp, OutElementOp, ConvBwdDataFilter1x1Stride1Pad0, 256, 128, 128, 16, 4, 4, 4, 1, S<8, 2>, S<8, 2>, S<8, 1, 1, 4>, S<2, 1, 128, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 4>, S<1, 2, 0, 3>, S<1, 1, 1, 4>, S<1, 1, 8, 4>, S<16, 1, 16, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 8, 1>, S<0, 3, 1, 2>, S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>, 5, 4>
// clang-format on
>;
void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
std::vector<std::unique_ptr<DeviceConvBwdData<2,
NHWC,
KYXC,
NHWK,
InDataType,
WeiDataType,
OutDataType,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/CMakeLists.txt deleted 100644 → 0
View file @ 9bd44685
add_instance_library(device_conv2d_bwd_weight_instance
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
)
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_f32_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<2,
NHWC,
KYXC,
NHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_bf16_f32_bf16_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
// TODO: retire dedicated 2d version
#include "ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.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 F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
// TODO: retire dedicated 2d version
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances = std::tuple<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<2,
NHWC,
KYXC,
NHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_default_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv2d_bwd_weight/device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.hpp"
// TODO: retire dedicated 2d version
#include "ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.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 F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NHWC = ck::tensor_layout::convolution::NHWC;
using KYXC = ck::tensor_layout::convolution::KYXC;
using NHWK = ck::tensor_layout::convolution::NHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f32_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 2, F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
using device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
// clang-format off
//#################################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#################################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#################################################################################| | | | | Operation| Operation| Operation| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#################################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 64, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 1, true, S<1, 4, 32, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 2, true, S<1, 4, 16, 1>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 4, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<2,
NHWC,
KYXC,
NHWK,
F32,
F32,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_c_shuffle_nhwc_kyxc_nhwk_f32_default_instances{});
add_device_operation_instances(
instances, device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances{});
add_device_operation_instances(
instances, device_conv_dedicated_2d_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv3d_bwd_weight/CMakeLists.txt deleted 100644 → 0
View file @ 9bd44685
add_instance_library(device_conv3d_bwd_weight_instance
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
)
library/src/tensor_operation_instance/gpu/conv3d_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 = bhalf_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NDHWC = ck::tensor_layout::convolution::NDHWC;
using KZYXC = ck::tensor_layout::convolution::KZYXC;
using NDHWK = ck::tensor_layout::convolution::NDHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
using device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
using device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_f32_bf16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 4>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, BF16, F32, BF16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 4>
// clang-format on
>;
void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_f32_bf16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<3,
NDHWC,
KZYXC,
NDHWK,
BF16,
F32,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_bf16_f32_bf16_instances{});
add_device_operation_instances(
instances,
device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/conv3d_bwd_weight/device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp deleted 100644 → 0
View file @ 9bd44685
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_convnd_bwd_weight_nwc_kxc_nwk_xdl_cshuffle.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 F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using NDHWC = ck::tensor_layout::convolution::NDHWC;
using KZYXC = ck::tensor_layout::convolution::KZYXC;
using NDHWK = ck::tensor_layout::convolution::NDHWK;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdWeightDefault =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
using device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f16_default_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
using device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances = std::tuple<
// clang-format off
//#########################################| Num| InData| WeiData| OutData| AccData| In| Wei| Out| ConvBackward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransfer| CBlockTransfer|
//#########################################| Dim| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Weight| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ClusterLengths| ScalarPerVector|
//#########################################| Spatial| | | | | Operation| Operation| Operation| Specialization| | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| MBlock_MPerBlock| NWaveNPerXdl|
//#########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | NBlock_NPerBlock| |
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 32, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 8>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 8, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 8>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 1, true, S<1, 4, 16, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 32, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, 1, 1, S<1, 16, 1, 4>, 8>,
DeviceConvNdBwdWeightNwcKxcNwk_Xdl_CShuffle< 3, F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 4, 4>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 2, true, S<1, 4, 8, 2>, S<0, 3, 1, 2>, S<0, 2, 1, 3>, 2, 8, 4, true, 1, 1, S<1, 16, 1, 4>, 8>
// clang-format on
>;
void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
std::vector<std::unique_ptr<DeviceConvBwdWeight<3,
NDHWC,
KZYXC,
NDHWK,
F16,
F16,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_conv3d_bwd_weight_xdl_c_shuffle_ndhwc_kzyxc_ndhwk_f16_default_instances{});
add_device_operation_instances(
instances, device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment