/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-6.4.3/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp Source File
Go to the documentation of this file.
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
  
 #pragma once
  
 #include <iostream>
 #include <sstream>
 #include <array>
  
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/device_reduce.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
  
 namespace ck {
 namespace tensor_operation {
 namespace device {
  
 template <typename InDataType,
           typename AccDataType,
           typename OutDataType,
           index_t Rank,
           index_t NumReduceDim,
           typename ReduceOperation,
           typename InElementwiseOperation,
           typename AccElementwiseOperation,
           InMemoryDataOperationEnum OutMemoryDataOperation,
           bool PropagateNan,
           bool OutputIndex,
           bool HaveIndexInputIfOutputIndex,
           index_t BlockSize,
           index_t MThreadClusterSize,
           index_t KThreadClusterSize,
           index_t MThreadSliceSize,
           index_t KThreadSliceSize,
           index_t InSrcVectorDim,
           index_t InSrcVectorSize,
           index_t OutDstVectorSize>
 struct DeviceReduceMultiBlock : public DeviceReduce<InDataType,
                                                     AccDataType,
                                                     OutDataType,
                                                     Rank,
                                                     NumReduceDim,
                                                     ReduceOperation,
                                                     InElementwiseOperation,
                                                     AccElementwiseOperation,
                                                     PropagateNan,
                                                     OutputIndex>
 {
     static_assert(Rank <= 12, "Bigger Rank size is not supported!");
     static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize,
                   "Invalid thread cluster size assignments!");
  
     static_assert(((InSrcVectorDim == 0 && MThreadSliceSize % InSrcVectorSize == 0) ||
                    (InSrcVectorDim == 1 && KThreadSliceSize % InSrcVectorSize == 0)) &&
                       (MThreadSliceSize % OutDstVectorSize == 0),
                   "Invalid thread slice sizes and/or vector sizes configuration, please check!");
  
     using IndexDataType = int32_t;
  
     static constexpr bool HaveIndexInput = OutputIndex && HaveIndexInputIfOutputIndex;
  
     static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
  
     static constexpr index_t NumSrcDim = Rank;
     static constexpr index_t NumDstDim = (NumInvariantDim == 0) ? 1 : NumInvariantDim;
     static constexpr bool reduceAllDim = (NumInvariantDim == 0);
  
     // So far, only AtomicAdd is considered, other Atomic Operation like AtomicMax can be added
     // later
     static constexpr bool use_multiblock =
         (OutMemoryDataOperation == InMemoryDataOperationEnum::AtomicAdd);
  
     static_assert(ck::reduce::InMemoryDataOperationSupportedOnDataType<OutMemoryDataOperation,
                                                                        OutDataType>::value,
                   "The OutDataType must support the specified OutMemoryDataOperation!");
  
     static_assert(!use_multiblock || (use_multiblock && !OutputIndex),
                   "MultiBlock reduction can only be used when outputing index is not required");
  
     static_assert(
         ReduceOperation::IsCompatibleInMemoryDataOperation(OutMemoryDataOperation),
         "The reduction accumulation operation must be compatible with the OutMemoryDataOperation!");
  
     static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
     static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
  
     static auto MakeSrc2dDescriptor(const std::array<index_t, Rank>& inLengths,
                                     const std::array<index_t, Rank>& inStrides,
                                     int blkGroupSize,
                                     int numBlockTileIteration)
     {
         const auto tupleSrcLengths =
             generate_tuple([&](auto I) { return inLengths[I]; }, Number<Rank>{});
         const auto tupleSrcStrides =
             generate_tuple([&](auto I) { return inStrides[I]; }, Number<Rank>{});
  
         const auto inDesc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
  
         const auto in_grid_desc_m_k = [&]() {
             if constexpr(reduceAllDim)
             {
                 const auto one_dim_inDesc = transform_tensor_descriptor(
                     inDesc,
                     make_tuple(make_merge_transform(tupleSrcLengths)),
                     make_tuple(typename arithmetic_sequence_gen<0, NumSrcDim, 1>::type{}),
                     make_tuple(Sequence<0>{}));
  
                 return transform_tensor_descriptor(one_dim_inDesc,
                                                    make_tuple(make_unmerge_transform(make_tuple(
                                                        1, one_dim_inDesc.GetLength(Number<0>{})))),
                                                    make_tuple(Sequence<0>{}),
                                                    make_tuple(Sequence<0, 1>{}));
             }
             else
             {
                 using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
                 using ReduceDims = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
  
                 const auto reduceDimLengths = generate_tuple(
                     [&](auto I) { return inLengths[NumInvariantDim + I]; }, Number<NumReduceDim>{});
                 const auto invariantDimLengths =
                     generate_tuple([&](auto I) { return inLengths[I]; }, Number<NumInvariantDim>{});
  
                 return transform_tensor_descriptor(
                     inDesc,
                     make_tuple(make_merge_transform(invariantDimLengths),
                                make_merge_transform(reduceDimLengths)),
                     make_tuple(InvariantDims{}, ReduceDims{}),
                     make_tuple(Sequence<0>{}, Sequence<1>{}));
             }
         }();
  
         const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
         const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
  
         const int reduceSizePerBlock = K_BlockTileSize * numBlockTileIteration;
         const auto inPad_M =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
         const auto inPad_K = reduceSizePerBlock * blkGroupSize - reduceLength;
  
         auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
             in_grid_desc_m_k,
             make_tuple(make_right_pad_transform(invariantLength, inPad_M),
                        make_right_pad_transform(reduceLength, inPad_K)),
             make_tuple(Sequence<0>{}, Sequence<1>{}),
             make_tuple(Sequence<0>{}, Sequence<1>{}));
  
         return (in_grid_desc_m_k_padded);
     };
  
     static auto MakeDst1dDescriptor(const std::array<index_t, NumDstDim>& outLengths,
                                     const std::array<index_t, NumDstDim>& outStrides)
     {
         const auto tupleDstLengths =
             generate_tuple([&](auto I) { return outLengths[I]; }, Number<NumDstDim>{});
         const auto tupleDstStrides =
             generate_tuple([&](auto I) { return outStrides[I]; }, Number<NumDstDim>{});
  
         auto outDesc = make_naive_tensor_descriptor(tupleDstLengths, tupleDstStrides);
  
         auto out_grid_desc_m = transform_tensor_descriptor(
             outDesc,
             make_tuple(make_merge_transform(tupleDstLengths)),
             make_tuple(typename arithmetic_sequence_gen<0, NumDstDim, 1>::type{}),
             make_tuple(Sequence<0>{}));
  
         const auto invariantLength = out_grid_desc_m.GetLength(Number<0>{});
  
         const auto outPad =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
  
         auto out_grid_desc_m_padded = transform_tensor_descriptor(
             out_grid_desc_m,
             make_tuple(make_right_pad_transform(invariantLength, outPad)),
             make_tuple(Sequence<0>{}),
             make_tuple(Sequence<0>{}));
         return (out_grid_desc_m_padded);
     };
  
     static auto MakeDst1dDescriptorForBufferSet(const std::array<index_t, NumDstDim>& outLengths,
                                                 const std::array<index_t, NumDstDim>& outStrides)
     {
         const auto tupleDstLengths =
             generate_tuple([&](auto I) { return outLengths[I]; }, Number<NumDstDim>{});
         const auto tupleDstStrides =
             generate_tuple([&](auto I) { return outStrides[I]; }, Number<NumDstDim>{});
  
         auto outDesc = make_naive_tensor_descriptor(tupleDstLengths, tupleDstStrides);
  
         auto out_grid_desc_m = transform_tensor_descriptor(
             outDesc,
             make_tuple(make_merge_transform(tupleDstLengths)),
             make_tuple(typename arithmetic_sequence_gen<0, NumDstDim, 1>::type{}),
             make_tuple(Sequence<0>{}));
  
         const auto length = out_grid_desc_m.GetLength(Number<0>{});
  
         const auto pad = math::integer_least_multiple(length, BlockSize) - length;
  
         auto out_grid_desc_m_padded =
             transform_tensor_descriptor(out_grid_desc_m,
                                         make_tuple(make_right_pad_transform(length, pad)),
                                         make_tuple(Sequence<0>{}),
                                         make_tuple(Sequence<0>{}));
         return (out_grid_desc_m_padded);
     };
  
     struct Argument : public BaseArgument
     {
         Argument(const std::array<index_t, Rank> inLengths,
                  const std::array<index_t, Rank> inStrides,
                  const std::array<index_t, NumDstDim> outLengths,
                  const std::array<index_t, NumDstDim> outStrides,
                  const std::array<int, NumReduceDim> reduceDims,
                  double alpha,
                  double beta,
                  const InDataType* in_dev,
                  const IndexDataType* in_index_dev,
                  OutDataType* out_dev,
                  IndexDataType* out_index_dev,
                  const InElementwiseOperation in_elementwise_op,
                  const AccElementwiseOperation acc_elementwise_op)
             : outLengths_{outLengths},
               outStrides_{outStrides},
               in_dev_{in_dev},
               in_index_dev_{in_index_dev},
               out_dev_{out_dev},
               out_index_dev_{out_index_dev},
               in_elementwise_op_{in_elementwise_op},
               acc_elementwise_op_{acc_elementwise_op}
         {
             if(Rank != inLengths.size() || Rank != inStrides.size() ||
                NumReduceDim != reduceDims.size())
             {
                 throw std::runtime_error(
                     "One of inLengths/inStrides/reduceDims has invalid size!"
                     "\nExpected size inLengths: " +
                     std::to_string(Rank) + ", inStrides: " + std::to_string(Rank) +
                     ", reduceDims: " + std::to_string(NumReduceDim) +
                     "\nBut have inLengths: " + std::to_string(inLengths.size()) +
                     ", inStrides: " + std::to_string(inStrides.size()) +
                     ", reduceDims: " + std::to_string(reduceDims.size()));
             }
  
             for(std::size_t i = 0; i < reduceDims.size(); ++i)
             {
                 if(reduceDims[i] < 0 || reduceDims[i] >= Rank)
                 {
                     throw std::runtime_error("Provided reduce dimension exceed input tensor Rank!"
                                              "\nHave reduceDims[" +
                                              std::to_string(i) +
                                              "]: " + std::to_string(reduceDims[i]));
                 }
             }
  
             inLengths_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(inLengths, reduceDims);
             inStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(inStrides, reduceDims);
  
             alpha_ = type_convert<AccDataType>(alpha);
             beta_  = type_convert<AccDataType>(beta);
  
             std::tie(invariant_total_length, reduce_total_length) =
                 get_2d_lengths<Rank, NumReduceDim>(inLengths_);
  
             if constexpr(NumInvariantDim == 0)
                 invariant_lowest_length = 1;
             else
                 invariant_lowest_length = inLengths_[NumInvariantDim - 1];
  
             reduce_lowest_length = inLengths_[Rank - 1];
  
             if constexpr(use_multiblock)
             {
  
                 int iterations = 1;
                 while(true)
                 {
                     int testBlkGroupSize =
                         (reduce_total_length + (K_BlockTileSize * iterations) - 1) /
                         (K_BlockTileSize * iterations);
  
                     // we want the blkGroupSize be not more than 128
                     if(testBlkGroupSize <= 128)
                         break;
  
                     iterations++;
                 };
  
                 blkGroupSize = (reduce_total_length + (K_BlockTileSize * iterations) - 1) /
                                (K_BlockTileSize * iterations);
  
                 numBlockTileIteration = iterations;
             }
             else
             {
                 blkGroupSize = 1;
                 numBlockTileIteration =
                     (reduce_total_length + K_BlockTileSize - 1) / K_BlockTileSize;
             };
  
             gridSize = math::integer_least_multiple(invariant_total_length, M_BlockTileSize) /
                        M_BlockTileSize * blkGroupSize;
  
             gridSize_pre =
                 math::integer_least_multiple(invariant_total_length, BlockSize) / BlockSize;
         }
  
         std::array<index_t, Rank> inLengths_;
         std::array<index_t, Rank> inStrides_;
         std::array<index_t, NumDstDim> outLengths_;
         std::array<index_t, NumDstDim> outStrides_;
  
         AccDataType alpha_;
         AccDataType beta_;
  
         const InDataType* in_dev_;
         const IndexDataType* in_index_dev_;
         OutDataType* out_dev_;
         IndexDataType* out_index_dev_;
  
         InElementwiseOperation in_elementwise_op_;
         AccElementwiseOperation acc_elementwise_op_;
  
         index_t invariant_lowest_length;
         index_t reduce_lowest_length;
         long_index_t invariant_total_length;
         long_index_t reduce_total_length;
  
         int blkGroupSize;
         int numBlockTileIteration;
         size_t gridSize;
  
         size_t gridSize_pre;
     };
  
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
             const auto in_grid_desc_m_k = DeviceReduceMultiBlock::MakeSrc2dDescriptor(
                 arg.inLengths_, arg.inStrides_, arg.blkGroupSize, arg.numBlockTileIteration);
             const auto out_grid_desc_m =
                 DeviceReduceMultiBlock::MakeDst1dDescriptor(arg.outLengths_, arg.outStrides_);
             const auto out_grid_desc_m_2 = DeviceReduceMultiBlock::MakeDst1dDescriptorForBufferSet(
                 arg.outLengths_, arg.outStrides_);
  
             using InGridDesc_M_K  = decltype(in_grid_desc_m_k);
             using OutGridDesc_M   = decltype(out_grid_desc_m);
             using OutGridDesc_M_2 = decltype(out_grid_desc_m_2);
  
             using GridwiseReduce = GridwiseReduction_mk_to_m_multiblock<InDataType,
                                                                         OutDataType,
                                                                         AccDataType,
                                                                         IndexDataType,
                                                                         InGridDesc_M_K,
                                                                         OutGridDesc_M,
                                                                         ReduceOperation,
                                                                         InElementwiseOperation,
                                                                         AccElementwiseOperation,
                                                                         OutMemoryDataOperation,
                                                                         PropagateNan,
                                                                         BlockSize,
                                                                         MThreadClusterSize,
                                                                         KThreadClusterSize,
                                                                         MThreadSliceSize,
                                                                         KThreadSliceSize,
                                                                         InSrcVectorDim,
                                                                         InSrcVectorSize,
                                                                         OutDstVectorSize>;
  
             const auto kernel_main = kernel_reduce_multiblock<GridwiseReduce,
                                                               OutputIndex,
                                                               HaveIndexInput,
                                                               InDataType,
                                                               OutDataType,
                                                               AccDataType,
                                                               int32_t,
                                                               InGridDesc_M_K,
                                                               OutGridDesc_M,
                                                               InElementwiseOperation,
                                                               AccElementwiseOperation>;
  
             float avg_time = 0;
  
             if constexpr(use_multiblock)
             {
                 const auto identityVal =
                     ck::reduce::GetIdentityValueForInMemoryDataOperation<OutDataType>(
                         OutMemoryDataOperation);
  
                 const auto kernel_pre =
                     kernel_buffer_set_value<BlockSize, OutDataType, OutGridDesc_M_2>;
  
                 avg_time += launch_and_time_kernel(stream_config,
                                                    kernel_pre,
                                                    dim3(arg.gridSize_pre),
                                                    dim3(BlockSize),
                                                    0,
                                                    out_grid_desc_m_2,
                                                    arg.out_dev_,
                                                    identityVal);
             };
  
             avg_time += launch_and_time_kernel(stream_config,
                                                kernel_main,
                                                dim3(arg.gridSize),
                                                dim3(BlockSize),
                                                0,
                                                in_grid_desc_m_k,
                                                out_grid_desc_m,
                                                arg.in_elementwise_op_,
                                                arg.acc_elementwise_op_,
                                                arg.blkGroupSize,
                                                arg.numBlockTileIteration,
                                                arg.alpha_,
                                                arg.in_dev_,
                                                arg.in_index_dev_,
                                                arg.beta_,
                                                arg.out_dev_,
                                                arg.out_index_dev_);
  
             return (avg_time);
         };
  
         float Run(const BaseArgument* p_arg,
                   const StreamConfig& stream_config = StreamConfig{}) override
         {
             return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
         };
     };
  
     static bool IsSupportedArgument(const Argument* pArg)
     {
         if constexpr(use_multiblock)
         {
             if(static_cast<float>(pArg->beta_) != 0.0f)
                 return (false);
         };
  
         if constexpr(InSrcVectorDim == 0)
         {
             if constexpr(NumInvariantDim == 0)
             {
                 return (false);
             }
             else
             {
                 if(pArg->inStrides_[NumInvariantDim - 1] != 1)
                     return (false);
  
                 if(pArg->invariant_lowest_length % InSrcVectorSize != 0)
                     return (false);
             };
         }
         else
         {
             if(pArg->inStrides_[Rank - 1] != 1)
                 return (false);
  
             if(pArg->reduce_lowest_length % InSrcVectorSize != 0)
                 return (false);
         };
  
         // To improve
         if(pArg->invariant_lowest_length % OutDstVectorSize != 0)
             return (false);
  
         if constexpr(use_multiblock)
         {
             // blkGroupSize of 1 should be handled by Blockwise path using
             // InMemoryDataOperationEnum::Set
             if(pArg->blkGroupSize == 1)
                 return (false);
  
             // This is very strong restriction, but needed to avoid some failure
             if(pArg->invariant_lowest_length % M_BlockTileSize != 0)
                 return (false);
         }
         else
         {
             // cases with very small reduce_total_length should be handled by ThreadWise kernel
             // if(pArg->reduce_total_length / KThreadSliceSize < 2)
             //     return (false);
         };
  
         return (true);
     }
  
     bool IsSupportedArgument(const BaseArgument* p_arg) override
     {
         return IsSupportedArgument(dynamic_cast<const Argument*>(p_arg));
     };
  
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const std::array<index_t, Rank> inLengths,
                         const std::array<index_t, Rank> inStrides,
                         const std::array<index_t, NumDstDim> outLengths,
                         const std::array<index_t, NumDstDim> outStrides,
                         const std::array<int, NumReduceDim> reduceDims,
                         double alpha,
                         double beta,
                         const void* in_dev,
                         const void* in_index_dev,
                         void* out_dev,
                         void* out_index_dev,
                         const InElementwiseOperation in_elementwise_op,
                         const AccElementwiseOperation acc_elementwise_op) override
     {
         return std::make_unique<Argument>(inLengths,
                                           inStrides,
                                           outLengths,
                                           outStrides,
                                           reduceDims,
                                           alpha,
                                           beta,
                                           static_cast<const InDataType*>(in_dev),
                                           static_cast<const IndexDataType*>(in_index_dev),
                                           static_cast<OutDataType*>(out_dev),
                                           static_cast<IndexDataType*>(out_index_dev),
                                           in_elementwise_op,
                                           acc_elementwise_op);
     };
  
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
     {
         return std::make_unique<Invoker>();
     };
  
     std::string GetTypeString() const override
     {
         auto str = std::stringstream();
  
         // clang-format off
         str << (OutMemoryDataOperation == InMemoryDataOperationEnum::Set? "DeviceReduceBlockWise<" : "DeviceReduceMultiBlock<") << BlockSize << ",";
         str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
         str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
         str << "InSrcVectorDim_" << InSrcVectorDim << "_InSrcVectorSize_" << InSrcVectorSize << "_OutDstVectorSize_" << OutDstVectorSize << ">";
         // clang-format on
  
         return str.str();
     }
 };
  
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck