thread_group_tensor_slice_transfer_v6r1r2.hpp Source File

thread_group_tensor_slice_transfer_v6r1r2.hpp Source File#

Composable Kernel: thread_group_tensor_slice_transfer_v6r1r2.hpp Source File
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
#pragma once
 
#include "ck/utility/common_header.hpp"
#include "ck/tensor_description/tensor_descriptor.hpp"
#include "ck/tensor_description/tensor_descriptor_helper.hpp"
#include "ck/tensor_description/cluster_descriptor.hpp"
#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp"
 
namespace ck {
 
// this version does following things to avoid scratch memory issue
// 1. Use StaticallyIndexedArray instead of C array for thread buffer
// 2. ThreadwiseTensorSliceTransfer_v3 does not keep reference to tensor descriptor
// 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
template <typename ThreadGroup,
          typename ElementwiseOperation,
          typename SliceLengths,
          typename ThreadClusterLengths,
          typename ThreadClusterArrangeOrder,
          typename SrcData,
          typename DstData,
          typename SrcDesc,
          typename DstDesc,
          typename DimAccessOrder,
          index_t VectorDim,
          index_t ScalarPerVector,
          bool ThreadTransferSrcResetCoordinateAfterRun,
          bool ThreadTransferDstResetCoordinateAfterRun>
struct ThreadGroupTensorSliceTransfer_v6r1r2
{
    static constexpr index_t nDim = remove_reference_t<SrcDesc>::GetNumOfDimension();
 
    static constexpr auto thread_slice_lengths = SliceLengths{} / ThreadClusterLengths{};
 
    using Index = MultiIndex<nDim>;
 
    __device__ constexpr ThreadGroupTensorSliceTransfer_v6r1r2(
        const SrcDesc& src_desc,
        const Index& src_block_slice_origin,
        const DstDesc& dst_desc,
        const Index& dst_block_slice_origin,
        const ElementwiseOperation& element_op)
        : threadwise_transfer_(src_desc,
                               make_zero_multi_index<nDim>(),
                               dst_desc,
                               make_zero_multi_index<nDim>(),
                               element_op)
 
    {
        static_assert(nDim == remove_cvref_t<SrcDesc>::GetNumOfDimension() &&
                          nDim == remove_cvref_t<DstDesc>::GetNumOfDimension() &&
                          nDim == ThreadClusterLengths::Size() &&
                          nDim == ThreadClusterArrangeOrder::Size() &&
                          nDim == DimAccessOrder::Size(),
                      "wrong! nDim not consistent");
 
        static_assert(
            is_same<SliceLengths, decltype(thread_slice_lengths * ThreadClusterLengths{})>{},
            "wrong! threads should be mapped to cover entire slicing window");
 
        static_assert(ThreadGroup::GetNumOfThread() >= thread_cluster_desc_.GetElementSize(),
                      "wrong! ThreadGroup::GetNumOfThread() too small");
 
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
                make_multi_index(ThreadGroup::GetThreadId()));
 
            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
 
            threadwise_transfer_.SetSrcSliceOrigin(src_desc,
                                                   src_block_slice_origin + thread_data_idx_begin);
            threadwise_transfer_.SetDstSliceOrigin(dst_desc,
                                                   dst_block_slice_origin + thread_data_idx_begin);
        }
    }
 
    template <typename SrcBuffer, typename DstBuffer, InMemoryDataOperationEnum DstInMemOp>
    __device__ void Run(const SrcDesc& src_desc,
                        const SrcBuffer& src_buf,
                        const DstDesc& dst_desc,
                        DstBuffer& dst_buf)
    {
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            threadwise_transfer_.template Run<SrcBuffer, DstBuffer, DstInMemOp>(
                src_desc, src_buf, dst_desc, dst_buf);
        }
    }
 
    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc, const Index& step)
    {
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            threadwise_transfer_.MoveSrcSliceWindow(src_desc, step);
        }
    }
 
    __device__ void MoveDstSliceWindow(const DstDesc& dst_desc, const Index& step)
    {
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            threadwise_transfer_.MoveDstSliceWindow(dst_desc, step);
        }
    }
 
    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_block_slice_origin)
    {
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
                make_multi_index(ThreadGroup::GetThreadId()));
 
            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
 
            threadwise_transfer_.SetSrcSliceOrigin(src_desc,
                                                   src_block_slice_origin + thread_data_idx_begin);
        }
    }
 
    __device__ void SetDstSliceOrigin(const DstDesc& dst_desc, const Index& dst_block_slice_origin)
    {
        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
        {
            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
                make_multi_index(ThreadGroup::GetThreadId()));
 
            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
 
            threadwise_transfer_.SetDstSliceOrigin(dst_desc,
                                                   dst_block_slice_origin + thread_data_idx_begin);
        }
    }
 
    private:
    static constexpr auto thread_cluster_desc_ =
        make_cluster_descriptor(ThreadClusterLengths{}, ThreadClusterArrangeOrder{});
 
    using ThreadwiseTransfer =
        ThreadwiseTensorSliceTransfer_v6r1r2<SrcData,
                                             DstData,
                                             SrcDesc,
                                             DstDesc,
                                             ElementwiseOperation,
                                             decltype(thread_slice_lengths),
                                             DimAccessOrder,
                                             VectorDim,
                                             ScalarPerVector,
                                             ThreadTransferSrcResetCoordinateAfterRun,
                                             ThreadTransferDstResetCoordinateAfterRun>;
 
    ThreadwiseTransfer threadwise_transfer_;
};
 
} // namespace ck