/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/ops/flatmm/kernel/flatmm_kernel.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/ops/flatmm/kernel/flatmm_kernel.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/ops/flatmm/kernel/flatmm_kernel.hpp Source File
Go to the documentation of this file.
 // Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 // SPDX-License-Identifier: MIT
  
 #pragma once
  
 #include <iostream>
 #include <string>
  
 #include "ck_tile/core.hpp"
 #include "ck_tile/ops/common.hpp"
 #include "ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_scheduler.hpp"
  
 namespace ck_tile {
 struct FlatmmProblem
 {
     CK_TILE_HOST FlatmmProblem() = default;
     CK_TILE_HOST FlatmmProblem(
         index_t M_, index_t N_, index_t K_, index_t stride_A_, index_t stride_B_, index_t stride_C_)
         : M(M_), N(N_), K(K_), stride_A(stride_A_), stride_B(stride_B_), stride_C(stride_C_)
     {
     }
  
     index_t M;
     index_t N;
     index_t K;
     index_t stride_A;
     index_t stride_B;
     index_t stride_C;
 };
  
 template <int SharedGranularityMN, int SharedGranularityK = 0, typename ScaleType_ = float>
 struct FlatmmScalePointer
 {
     using ScaleType                    = ScaleType_;
     static constexpr int GranularityMN = SharedGranularityMN;
     static constexpr int GranularityK  = SharedGranularityK;
  
     const ScaleType* ptr;
  
     CK_TILE_HOST_DEVICE FlatmmScalePointer() = default;
     CK_TILE_HOST_DEVICE FlatmmScalePointer(const ScaleType* ptr_) : ptr(ptr_) {}
     CK_TILE_HOST_DEVICE FlatmmScalePointer(const ScaleType* ptr_, [[maybe_unused]] index_t length_)
         : ptr(ptr_)
     {
     }
  
     CK_TILE_HOST_DEVICE FlatmmScalePointer operator+(index_t offset) const
     {
         FlatmmScalePointer ret;
         if constexpr(GranularityMN == 0)
         {
             ret.ptr = ptr + offset / GranularityK;
         }
         else
         {
             ret.ptr = ptr + offset / GranularityMN / GranularityK;
         }
         return ret;
     }
  
     CK_TILE_HOST_DEVICE ScaleType operator[](index_t i) const = delete;
 };
  
 template <int SharedGranularityMN, typename ScaleType_>
 struct FlatmmScalePointer<SharedGranularityMN, 0, ScaleType_>
 {
     using ScaleType                    = ScaleType_;
     static constexpr int GranularityMN = SharedGranularityMN;
     static constexpr int GranularityK  = 0;
  
     static_assert(GranularityMN != 0);
  
     const ScaleType* ptr;
     index_t length;
  
     CK_TILE_HOST_DEVICE FlatmmScalePointer() = default;
     CK_TILE_HOST_DEVICE FlatmmScalePointer(const ScaleType* ptr_) : ptr(ptr_), length(1) {}
     CK_TILE_HOST_DEVICE FlatmmScalePointer(const ScaleType* ptr_, index_t length_)
         : ptr(ptr_), length(length_)
     {
     }
  
     CK_TILE_HOST_DEVICE FlatmmScalePointer operator+(index_t offset) const
     {
         FlatmmScalePointer ret;
         if constexpr(GranularityMN == 1)
         {
             ret.ptr    = ptr + offset;
             ret.length = length - offset;
         }
         else
         {
             ret.ptr    = ptr + offset / GranularityMN;
             ret.length = length - offset / GranularityMN;
         }
         return ret;
     }
  
     CK_TILE_HOST_DEVICE ScaleType operator[](index_t i) const
     {
         // with additional oob check
         if constexpr(GranularityMN == 1)
             return i < length ? ptr[i] : 0;
         else
             return i / GranularityMN < length ? ptr[i / GranularityMN] : 0;
     }
 };
  
 // shared granularityMN = -1 means no scale
 template <typename ScaleType_>
 struct FlatmmScalePointer<-1, 0, ScaleType_>
 {
     using ScaleType                    = ScaleType_;
     static constexpr int GranularityMN = -1;
     static constexpr int GranularityK  = 0;
  
     const ScaleType* ptr = nullptr;
  
     CK_TILE_HOST_DEVICE constexpr FlatmmScalePointer() = default;
     CK_TILE_HOST_DEVICE constexpr FlatmmScalePointer(const ScaleType*) {}
     CK_TILE_HOST_DEVICE constexpr FlatmmScalePointer(const ScaleType*, index_t) {}
  
     CK_TILE_HOST_DEVICE constexpr FlatmmScalePointer operator+(index_t) const
     {
         return FlatmmScalePointer{};
     }
     CK_TILE_HOST_DEVICE constexpr ScaleType operator[](index_t) const
     {
         return 1; // alway return 1, it doesn't change the result
     }
 };
  
 template <index_t NumDTensor = 0>
 struct BaseFlatmmHostArgs
 {
     CK_TILE_HOST BaseFlatmmHostArgs() = default;
     CK_TILE_HOST BaseFlatmmHostArgs(const void* a_ptr_,
                                     const void* b_ptr_,
                                     const std::array<const void*, NumDTensor>& ds_ptr_,
                                     void* e_ptr_,
                                     index_t k_batch_,
                                     index_t M_,
                                     index_t N_,
                                     index_t K_,
                                     index_t stride_A_,
                                     index_t stride_B_,
                                     const std::array<index_t, NumDTensor>& stride_Ds_,
                                     index_t stride_E_)
         : a_ptr(a_ptr_),
           b_ptr(b_ptr_),
           ds_ptr(ds_ptr_),
           e_ptr(e_ptr_),
           M(M_),
           N(N_),
           K(K_),
           stride_A(stride_A_),
           stride_B(stride_B_),
           stride_Ds(stride_Ds_),
           stride_E(stride_E_),
           k_batch(k_batch_)
     {
     }
  
     const void* a_ptr;
     const void* b_ptr;
     const std::array<const void*, NumDTensor> ds_ptr;
     union
     {
         void* e_ptr;
         void* c_ptr;
     };
     index_t M;
     index_t N;
     index_t K;
     index_t stride_A;
     index_t stride_B;
     const std::array<index_t, NumDTensor> stride_Ds;
     union
     {
         index_t stride_E;
         index_t stride_C;
     };
  
     index_t k_batch;
 };
 template <class ScaleM       = FlatmmScalePointer<-1>,
           class ScaleN       = FlatmmScalePointer<-1>,
           index_t NumDTensor = 0>
 struct ScaleFlatmmHostArgs : public BaseFlatmmHostArgs<>
 {
     CK_TILE_HOST ScaleFlatmmHostArgs() = default;
     CK_TILE_HOST ScaleFlatmmHostArgs(const void* a_ptr_,
                                      const void* b_shuffle_ptr_,
                                      const std::array<const void*, NumDTensor>& ds_ptr_,
                                      void* c_ptr_,
                                      index_t k_batch_,
                                      index_t M_,
                                      index_t N_,
                                      index_t K_,
                                      index_t stride_A_,
                                      index_t stride_B_,
                                      const std::array<index_t, NumDTensor>& stride_Ds_,
                                      index_t stride_C_,
                                      ScaleM scale_m_ = nullptr,
                                      ScaleN scale_n_ = nullptr)
         : BaseFlatmmHostArgs(a_ptr_,
                              b_shuffle_ptr_,
                              ds_ptr_,
                              c_ptr_,
                              k_batch_,
                              M_,
                              N_,
                              K_,
                              stride_A_,
                              stride_B_,
                              stride_Ds_,
                              stride_C_),
           scale_m(scale_m_),
           scale_n(scale_n_)
     {
     }
     ScaleM scale_m = nullptr;
     ScaleN scale_n = nullptr;
 };
  
 template <int NumberTensor = 0>
 using FlatmmHostArgs =
     ScaleFlatmmHostArgs<FlatmmScalePointer<-1>, FlatmmScalePointer<-1>, NumberTensor>;
  
 template <class ScaleM, class ScaleN, index_t NumDTensor = 0>
 struct FlatmmKernelArgs
 {
     const void* a_ptr;
     // const void* b_shuffle_ptr;
     const void* b_ptr;
     const std::array<const void*, NumDTensor> ds_ptr;
     void* e_ptr;
     index_t M;
     index_t N;
     index_t K;
     index_t stride_A;
     index_t stride_B;
     std::array<index_t, NumDTensor> stride_Ds;
     index_t stride_E;
     index_t k_batch;
     ScaleM scale_m_ptr = nullptr;
     ScaleN scale_n_ptr = nullptr;
 };
  
 template <typename TilePartitioner_, typename FlatmmPipeline_, typename EpiloguePipeline_>
 struct FlatmmKernel
 {
     using TilePartitioner = remove_cvref_t<TilePartitioner_>;
     using FlatmmPipeline  = remove_cvref_t<FlatmmPipeline_>;
     using BlockGemmShape =
         remove_cvref_t<typename FlatmmPipeline::BlockGemmShape>; // TileFlatmmShape
     using EpiloguePipeline              = remove_cvref_t<EpiloguePipeline_>;
     using ALayout                       = remove_cvref_t<typename FlatmmPipeline::ALayout>;
     using BLayout                       = remove_cvref_t<typename FlatmmPipeline::BLayout>;
     using ELayout                       = remove_cvref_t<typename FlatmmPipeline::CLayout>;
     using DsLayout                      = remove_cvref_t<typename EpiloguePipeline::DsLayout>;
     using DsDataType                    = remove_cvref_t<typename EpiloguePipeline::DsDataType>;
     static constexpr index_t kBlockSize = FlatmmPipeline::BlockSize;
     static constexpr bool UsePersistentKernel = FlatmmPipeline::UsePersistentKernel;
  
     using ADataType = remove_cvref_t<typename FlatmmPipeline::ADataType>;
     using BDataType = remove_cvref_t<typename FlatmmPipeline::BDataType>;
     // Below type is actually accumulation data type - the output of block GEMM.
     using EDataType = remove_cvref_t<typename EpiloguePipeline::ODataType>;
  
     static constexpr index_t NumDTensor = DsDataType::size();
  
     static constexpr auto I0 = number<0>();
     static constexpr auto I1 = number<1>();
     static constexpr auto I2 = number<2>();
     static constexpr auto I3 = number<3>();
  
     static_assert(DsLayout::size() == DsDataType::size(),
                   "The size of DsLayout and DsDataType should be the same");
     // using KernelArgs = FlatmmKernelArgs<DsLayout::size()>;
  
     [[nodiscard]] CK_TILE_HOST static const std::string GetName()
     {
         // clang-format off
         return concat('_', "gemm", gemm_prec_str<ADataType, BDataType>, FlatmmPipeline::GetName());
         // clang-format on
     }
  
     CK_TILE_HOST static constexpr auto GridSize(index_t M, index_t N, index_t KBatch)
     {
         assert(!UsePersistentKernel);
         return dim3(TilePartitioner::GridSize(M, N), 1, KBatch);
     }
  
     template <class ScaleM, class ScaleN>
     CK_TILE_HOST static constexpr auto
     GridSize(const FlatmmKernelArgs<ScaleM, ScaleN, DsDataType::size()>& kargs)
     {
         if constexpr(UsePersistentKernel)
         {
             hipDeviceProp_t prop;
             int deviceId = 0; // default device
  
             constexpr int block_size = FlatmmKernel::BlockSize().x;
             int dync_smem_size       = 0;
             int maxActiveBlocksPerCU = 0;
  
             [[maybe_unused]] auto e = hipGetDeviceProperties(&prop, deviceId);
  
             e = hipOccupancyMaxActiveBlocksPerMultiprocessor(
                 &maxActiveBlocksPerCU,
                 reinterpret_cast<void*>(
                     kentry<1, FlatmmKernel, FlatmmKernelArgs<ScaleM, ScaleN, DsDataType::size()>>),
                 block_size,
                 dync_smem_size);
  
             const int persistent_block_size = prop.multiProcessorCount * maxActiveBlocksPerCU;
             const int total_work_tile_cnt   = TilePartitioner::GridSize(kargs.M, kargs.N);
  
             // std::cout << "maxActiveBlocksPerCU: " << maxActiveBlocksPerCU
             //           << ", persistent_block_size: " << persistent_block_size
             //           << ", total_work_tile_cnt: " << total_work_tile_cnt << std::endl;
  
             assert(kargs.k_batch == 1);
             return dim3(min(persistent_block_size, total_work_tile_cnt), 1, kargs.k_batch);
         }
         else
         {
             return dim3(TilePartitioner::GridSize(kargs.M, kargs.N), 1, kargs.k_batch);
         }
     }
  
     CK_TILE_HOST static constexpr auto BlockSize() { return dim3(kBlockSize); }
  
     template <class ScaleM, class ScaleN>
     CK_TILE_HOST static constexpr FlatmmKernelArgs<ScaleM, ScaleN, DsDataType::size()>
     MakeKernelArgs(const ScaleFlatmmHostArgs<ScaleM, ScaleN, DsDataType::size()>& hostArgs)
     {
         return {hostArgs.a_ptr,
                 hostArgs.b_ptr,
                 hostArgs.ds_ptr,
                 hostArgs.e_ptr,
                 hostArgs.M,
                 hostArgs.N,
                 hostArgs.K,
                 hostArgs.stride_A,
                 hostArgs.stride_B,
                 hostArgs.stride_Ds,
                 hostArgs.stride_E,
                 hostArgs.k_batch,
                 hostArgs.scale_m,
                 hostArgs.scale_n};
     }
  
     CK_TILE_HOST_DEVICE static constexpr index_t GetSmemPingSize()
     {
         return max(FlatmmPipeline::GetSmemSize(), EpiloguePipeline::GetSmemSize());
     }
     CK_TILE_HOST_DEVICE static constexpr index_t GetSmemPongSize()
     {
         return FlatmmPipeline::GetSmemSize();
     }
  
     struct SplitKBatchOffset
     {
         template <class KernelArgs>
         __device__ SplitKBatchOffset(const KernelArgs& kargs, const std::size_t k_id = blockIdx.z)
         {
             constexpr auto N1   = BlockGemmShape::WarpTile::at(number<1>{});
             constexpr auto K1   = BlockGemmShape::WarpTile::at(number<2>{});
             const index_t K_t   = kargs.k_batch * K1;
             const index_t KRead = (kargs.K + K_t - 1) / K_t * K1;
  
             if constexpr(std::is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
             {
                 a_k_split_offset = k_id * KRead;
             }
             else if constexpr(std::is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
             {
                 a_k_split_offset = k_id * KRead * kargs.stride_A;
             }
  
             if constexpr(std::is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
             {
                 b_k_split_offset = k_id * KRead * kargs.stride_B * N1;
             }
             else if constexpr(std::is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
             {
                 b_k_split_offset = k_id * KRead * N1;
             }
  
             if(k_id < static_cast<uint32_t>(kargs.k_batch - 1))
             {
                 splitted_k = KRead;
             }
             else
             {
                 splitted_k = kargs.K - KRead * (kargs.k_batch - 1);
             }
         }
  
         index_t a_k_split_offset;
         index_t b_k_split_offset;
         index_t splitted_k;
     };
  
     template <class KernelArgs>
     CK_TILE_HOST static bool IsSupportedArgument(const KernelArgs& kargs)
     {
         if constexpr(EpiloguePipeline::GetVectorSizeC() % 2 != 0 &&
                      is_any_of<EDataType, fp16_t, bf16_t>::value)
         {
             if(kargs.k_batch != 1)
             {
                 std::cerr << "Conditions not met for Kbatch >1 !" << std::endl;
                 return false;
             }
         }
         if constexpr(UsePersistentKernel)
         {
             if(kargs.k_batch != 1)
             {
                 std::cerr << "Persistent mode doesn't support Kbatch >1 !" << std::endl;
                 return false;
             }
         }
  
         if constexpr(std::is_same_v<ALayout, tensor_layout::gemm::RowMajor>)
         {
             if(kargs.K % TilePartitioner::KPerBlock != 0 && FlatmmPipeline::kPadK == false)
             {
                 std::cerr << "Can't support K that is not a multiple of KPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.K % FlatmmPipeline::GetVectorSizeA() != 0)
             {
                 std::cerr << "K is not a multiple of vector load size for A tensor!" << std::endl;
                 return false;
             }
         }
         else
         {
             if(kargs.M % TilePartitioner::MPerBlock != 0 && FlatmmPipeline::kPadM == false)
             {
                 std::cerr << "Can't support M that is not a multiple of MPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.M % FlatmmPipeline::GetVectorSizeA() != 0)
             {
                 std::cerr << "M is not a multiple of vector load size for A tensor!" << std::endl;
                 return false;
             }
         }
  
         if constexpr(std::is_same_v<BLayout, tensor_layout::gemm::RowMajor>)
         {
             if(kargs.N % TilePartitioner::NPerBlock != 0 && FlatmmPipeline::kPadN == false)
             {
                 std::cerr << "Can't support N that is not a multiple of NPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.N % FlatmmPipeline::GetVectorSizeB() != 0)
             {
                 std::cerr << "N is not a multiple of vector load size for B tensor!" << std::endl;
                 return false;
             }
         }
         else
         {
             if(kargs.K % TilePartitioner::KPerBlock != 0 && FlatmmPipeline::kPadK == false)
             {
                 std::cerr << "Can't support K that is not a multiple of KPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.K % FlatmmPipeline::GetVectorSizeB() != 0)
             {
                 std::cerr << "K is not a multiple of vector load size for B tensor!" << std::endl;
                 return false;
             }
         }
  
         bool DTesnorIsValid = {true};
         static_for<0, NumDTensor, 1>{}([&](auto index) {
             using DiLayout = remove_cvref_t<std::tuple_element_t<index.value, DsLayout>>;
             if(std::is_same_v<DiLayout, ELayout> == false)
             {
                 DTesnorIsValid = false;
             }
             if constexpr(std::is_same_v<DiLayout, tensor_layout::gemm::RowMajor>)
             {
                 if(kargs.N % TilePartitioner::NPerBlock != 0 && FlatmmPipeline::kPadN == false)
                 {
                     CK_TILE_ERROR("Can't support N for tensor D that is not a multiple of "
                                   "NPerBlock without padding!");
                     DTesnorIsValid = false;
                 }
                 if(kargs.N % EpiloguePipeline::GetVectorSizeD(index) != 0)
                 {
                     CK_TILE_ERROR("N is not a multiple of vector load size for D tensor!");
                     DTesnorIsValid = false;
                 }
             }
             else
             {
                 if(kargs.M % TilePartitioner::MPerBlock != 0 && FlatmmPipeline::kPadM == false)
                 {
                     CK_TILE_ERROR("Can't support M for tensor D that is not a multiple of "
                                   "MPerBlock without padding!");
  
                     DTesnorIsValid = false;
                 }
                 if(kargs.M % EpiloguePipeline::GetVectorSizeD(index) != 0)
                 {
                     CK_TILE_ERROR("M is not a multiple of vector load size for D tensor!");
                     DTesnorIsValid = false;
                 }
             }
         });
  
         if constexpr(std::is_same_v<ELayout, tensor_layout::gemm::RowMajor>)
         {
             if(kargs.N % TilePartitioner::NPerBlock != 0 && FlatmmPipeline::kPadN == false)
             {
                 std::cerr << "Can't support N that is not a multiple of NPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.N % EpiloguePipeline::GetVectorSizeC() != 0)
             {
                 std::cerr << "N is not a multiple of vector load size for C tensor!" << std::endl;
                 return false;
             }
         }
         else
         {
             if(kargs.M % TilePartitioner::MPerBlock != 0 && FlatmmPipeline::kPadM == false)
             {
                 std::cerr << "Can't support M that is not a multiple of MPerBlock"
                              " without padding!"
                           << std::endl;
                 return false;
             }
             if(kargs.M % EpiloguePipeline::GetVectorSizeC() != 0)
             {
                 std::cerr << "M is not a multiple of vector load size for C tensor!" << std::endl;
                 return false;
             }
         }
         return DTesnorIsValid;
     }
  
     template <typename KernelArgs>
     CK_TILE_DEVICE static auto MakeABlockWindow(const ADataType* a_ptr,
                                                 const KernelArgs& kargs,
                                                 const index_t k_size,
                                                 const index_t block_idx_m)
     {
         // Step 1: Create tensor view
         const auto& a_tensor_view = [&]() {
             if constexpr(std::is_same_v<ALayout, tensor_layout::gemm::RowMajor>)
             {
                 return make_naive_tensor_view<address_space_enum::global>(
                     a_ptr,
                     make_tuple(kargs.M, k_size),
                     make_tuple(kargs.stride_A, 1),
                     number<FlatmmPipeline::GetVectorSizeA()>{},
                     number<1>{});
             }
             else
             {
                 return make_naive_tensor_view<address_space_enum::global>(
                     a_ptr,
                     make_tuple(k_size, kargs.M),
                     make_tuple(kargs.stride_A, 1),
                     number<FlatmmPipeline::GetVectorSizeA()>{},
                     number<1>{});
             }
         }();
  
         // Step 2: Create padded view
         const auto& a_pad_view = [&]() {
             if constexpr(std::is_same_v<ALayout, tensor_layout::gemm::RowMajor>)
             {
                 return pad_tensor_view(a_tensor_view,
                                        make_tuple(number<TilePartitioner::MPerBlock>{},
                                                   number<TilePartitioner::KPerBlock>{}),
                                        sequence<false, FlatmmPipeline::kPadK>{});
             }
             else
             {
                 return pad_tensor_view(a_tensor_view,
                                        make_tuple(number<TilePartitioner::KPerBlock>{},
                                                   number<TilePartitioner::MPerBlock>{}),
                                        sequence<false, FlatmmPipeline::kPadM>{});
             }
         }();
  
         // Step 3: Create tile window
         if constexpr(std::is_same_v<ALayout, tensor_layout::gemm::RowMajor>)
         {
             return make_tile_window(a_pad_view,
                                     make_tuple(number<TilePartitioner::MPerBlock>{},
                                                number<TilePartitioner::KPerBlock>{}),
                                     {block_idx_m, 0});
         }
         else
         {
             return make_tile_window(a_pad_view,
                                     make_tuple(number<TilePartitioner::KPerBlock>{},
                                                number<TilePartitioner::MPerBlock>{}),
                                     {0, block_idx_m});
         }
     }
  
     template <typename KernelArgs>
     CK_TILE_DEVICE static auto MakeBFlatBlockWindow(const BDataType* b_flat_ptr,
                                                     const KernelArgs& kargs,
                                                     const index_t block_idx_n)
     {
         // Step 1: Create tensor view
         index_t kFlatK =
             FlatmmPipeline::flatKPerWarp * (kargs.K / BlockGemmShape::WarpTile::at(I2));
         index_t kFlatN = kargs.N * kargs.K / kFlatK;
  
         const auto& b_flat_tensor_view = make_naive_tensor_view<address_space_enum::global>(
             b_flat_ptr,
             make_tuple(kFlatN, kFlatK),
             make_tuple(kFlatK, 1),
             number<FlatmmPipeline::GetVectorSizeB()>{},
             number<1>{});
  
         // Step 2: No padding needed for b_flat
         // Step 3: Create tile window
         return make_tile_window(
             b_flat_tensor_view,
             make_tuple(number<FlatmmPipeline::flatNPerWarp>{},
                        number<FlatmmPipeline::flatKPerWarp>{}),
             {static_cast<int>(block_idx_n / BlockGemmShape::WarpTile::at(I1)), 0});
     }
  
     template <typename KernelArgs>
     CK_TILE_DEVICE static auto MakeDBlockWindows(const std::array<const void*, NumDTensor>& ds_ptr,
                                                  const KernelArgs& kargs,
                                                  const index_t block_idx_m,
                                                  const index_t block_idx_n)
     {
         // Step 1: Create tensor views
         const auto& ds_tensor_view = generate_tuple(
             [&](auto i) {
                 using DiLayout   = remove_cvref_t<std::tuple_element_t<i.value, DsLayout>>;
                 using DDataType_ = remove_cvref_t<std::tuple_element_t<i.value, DsDataType>>;
                 if constexpr(std::is_same_v<DiLayout, tensor_layout::gemm::RowMajor>)
                 {
                     return make_naive_tensor_view<address_space_enum::global>(
                         static_cast<const DDataType_*>(ds_ptr[i]),
                         make_tuple(kargs.M, kargs.N),
                         make_tuple(kargs.stride_Ds[i], 1),
                         number<EpiloguePipeline::GetVectorSizeD(i)>{},
                         number<1>{});
                 }
                 else
                 {
                     return make_naive_tensor_view<address_space_enum::global>(
                         static_cast<const DDataType_*>(ds_ptr[i]),
                         make_tuple(kargs.N, kargs.M),
                         make_tuple(kargs.stride_Ds[i], 1),
                         number<EpiloguePipeline::GetVectorSizeD(i)>{},
                         number<1>{});
                 }
             },
             number<NumDTensor>{});
  
         // Step 2: Create padded views
         const auto& ds_pad_view = generate_tuple(
             [&](auto i) {
                 using DiLayout = remove_cvref_t<std::tuple_element_t<i.value, DsLayout>>;
                 if constexpr(std::is_same_v<DiLayout, tensor_layout::gemm::RowMajor>)
                 {
                     return pad_tensor_view(ds_tensor_view[i],
                                            make_tuple(number<TilePartitioner::MPerBlock>{},
                                                       number<TilePartitioner::NPerBlock>{}),
                                            sequence<false, FlatmmPipeline::kPadN>{});
                 }
                 else
                 {
                     return pad_tensor_view(ds_tensor_view[i],
                                            make_tuple(number<TilePartitioner::NPerBlock>{},
                                                       number<TilePartitioner::MPerBlock>{}),
                                            sequence<false, FlatmmPipeline::kPadM>{});
                 }
             },
             number<NumDTensor>{});
  
         // Step 3: Create tile windows
         return generate_tuple(
             [&](auto i) {
                 using DiLayout = remove_cvref_t<std::tuple_element_t<i.value, DsLayout>>;
                 if constexpr(std::is_same_v<DiLayout, tensor_layout::gemm::RowMajor>)
                 {
                     return make_tile_window(ds_pad_view[i],
                                             make_tuple(number<TilePartitioner::MPerBlock>{},
                                                        number<TilePartitioner::NPerBlock>{}),
                                             {block_idx_m, block_idx_n});
                 }
                 else
                 {
                     return make_tile_window(ds_pad_view[i],
                                             make_tuple(number<TilePartitioner::NPerBlock>{},
                                                        number<TilePartitioner::MPerBlock>{}),
                                             {block_idx_n, block_idx_m});
                 }
             },
             number<NumDTensor>{});
     }
  
     template <memory_operation_enum DstInMemOp = memory_operation_enum::set, typename KernelArgs>
     CK_TILE_DEVICE static auto MakeEBlockWindow(EDataType* e_ptr,
                                                 const KernelArgs& kargs,
                                                 const index_t block_idx_m,
                                                 const index_t block_idx_n)
     {
         // Step 1: Create tensor view
         const auto& e_tensor_view = [&]() {
             if constexpr(std::is_same_v<ELayout, tensor_layout::gemm::RowMajor>)
             {
                 return make_naive_tensor_view<address_space_enum::global, DstInMemOp>(
                     e_ptr,
                     make_tuple(kargs.M, kargs.N),
                     make_tuple(kargs.stride_E, 1),
                     number<EpiloguePipeline::GetVectorSizeC()>{},
                     number<1>{});
             }
             else
             {
                 return make_naive_tensor_view<address_space_enum::global, DstInMemOp>(
                     e_ptr,
                     make_tuple(kargs.N, kargs.M),
                     make_tuple(kargs.stride_E, 1),
                     number<1>{},
                     number<1>{});
             }
         }();
  
         // Step 2: Create padded view
         const auto& e_pad_view = [&]() {
             if constexpr(std::is_same_v<ELayout, tensor_layout::gemm::RowMajor>)
             {
                 return pad_tensor_view(e_tensor_view,
                                        make_tuple(number<TilePartitioner::MPerBlock>{},
                                                   number<TilePartitioner::NPerBlock>{}),
                                        sequence<false, FlatmmPipeline::kPadN>{});
             }
             else
             {
                 return pad_tensor_view(e_tensor_view,
                                        make_tuple(number<TilePartitioner::MPerBlock>{},
                                                   number<TilePartitioner::NPerBlock>{}),
                                        sequence<FlatmmPipeline::kPadM, false>{});
             }
         }();
  
         // Step 3: Create tile window
         return make_tile_window(
             e_pad_view,
             make_tuple(number<TilePartitioner::MPerBlock>{}, number<TilePartitioner::NPerBlock>{}),
             {block_idx_m, block_idx_n});
     }
  
     template <typename KernelArgs>
     CK_TILE_DEVICE static auto MakeScaleMWindow(const KernelArgs& kargs,
                                                 const SplitKBatchOffset& splitk_batch_offset,
                                                 const index_t block_idx_m)
     {
         constexpr int ScaleGranularityM  = decltype(kargs.scale_m_ptr)::GranularityMN;
         constexpr int ScaleGranularityKA = decltype(kargs.scale_m_ptr)::GranularityK;
  
         auto scale_stride_m = ScaleGranularityM == 0 ? 0  // per-tensor scale
                                                      : 1; // per-token scale
  
         // Step 1: Create tensor view
         const auto scale_m_view = make_naive_tensor_view<address_space_enum::global>(
             kargs.scale_m_ptr.ptr,
             make_tuple(kargs.M / ScaleGranularityM,
                        ScaleGranularityKA == 0
                            ? 1
                            : (splitk_batch_offset.splitted_k / ScaleGranularityKA)),
             make_tuple(scale_stride_m, 0),
             number < ScaleGranularityM == 1 ? FlatmmPipeline::GetVectorSizeA() : 1 > {},
             number<1>{});
  
         // Step 2: Create tile window
         return make_tile_window(scale_m_view,
                                 make_tuple(number<TilePartitioner::MPerBlock>{},
                                            number < ScaleGranularityKA == 0
                                                ? TilePartitioner::NPerBlock
                                                : TilePartitioner::KPerBlock > {}),
                                 {block_idx_m, 0});
     }
  
     template <typename KernelArgs>
     CK_TILE_DEVICE static auto MakeScaleNWindow(const KernelArgs& kargs,
                                                 const SplitKBatchOffset& splitk_batch_offset,
                                                 const index_t block_idx_n)
     {
         constexpr int ScaleGranularityN  = decltype(kargs.scale_n_ptr)::GranularityMN;
         constexpr int ScaleGranularityKB = decltype(kargs.scale_n_ptr)::GranularityK;
  
         auto scale_stride_n = ScaleGranularityN == 0 ? 0  // per-tensor scale
                                                      : 1; // per-channel scale
  
         // Step 1: Create tensor view
         const auto scale_n_view = make_naive_tensor_view<address_space_enum::global>(
             kargs.scale_n_ptr.ptr,
             make_tuple(
                 ScaleGranularityKB == 0 ? 1 : (splitk_batch_offset.splitted_k / ScaleGranularityKB),
                 kargs.N / ScaleGranularityN),
             make_tuple(0, scale_stride_n),
             number < ScaleGranularityN == 1 ? FlatmmPipeline::GetVectorSizeB() : 1 > {},
             number<1>{});
  
         // Step 2: Create tile window
         return make_tile_window(scale_n_view,
                                 make_tuple(number < ScaleGranularityKB == 0
                                                ? TilePartitioner::MPerBlock
                                                : TilePartitioner::KPerBlock > {},
                                            number<TilePartitioner::NPerBlock>{}),
                                 {0, block_idx_n});
     }
  
     template <class ScaleM, class ScaleN, bool UseDefaultScheduler = true>
     CK_TILE_DEVICE static void
     RunFlatmm(const ADataType* a_ptr,
               const BDataType* b_flat_ptr,
               const std::array<const void*, NumDTensor>& ds_ptr,
               EDataType* e_ptr,
               void* smem_ptr_ping,
               void* smem_ptr_pong,
               const FlatmmKernelArgs<ScaleM, ScaleN, DsDataType::size()>& kargs,
               const SplitKBatchOffset& splitk_batch_offset,
               const index_t block_idx_m,
               const index_t block_idx_n)
     {
         // Create block windows using specialized methods
         const auto& a_block_window =
             MakeABlockWindow(a_ptr, kargs, splitk_batch_offset.splitted_k, block_idx_m);
         const auto& b_flat_block_window = MakeBFlatBlockWindow(b_flat_ptr, kargs, block_idx_n);
         const auto& ds_block_window = MakeDBlockWindows(ds_ptr, kargs, block_idx_m, block_idx_n);
         const auto& scale_m_window  = MakeScaleMWindow(kargs, splitk_batch_offset, block_idx_m);
         const auto& scale_n_window  = MakeScaleNWindow(kargs, splitk_batch_offset, block_idx_n);
  
         const index_t num_loop = TilePartitioner::GetLoopNum(splitk_batch_offset.splitted_k);
  
         // Run GEMM cooperatively by whole workgroup.
         const auto& c_block_tile = FlatmmPipeline{}.template operator()(
             a_block_window, b_flat_block_window, num_loop, smem_ptr_ping, smem_ptr_pong);
  
         // Run Epilogue Pipeline with k_batch dispatching
         if constexpr(ScaleM::GranularityMN != -1 || ScaleN::GranularityMN != -1)
         {
             if(kargs.k_batch == 1)
             {
                 auto e_block_window = MakeEBlockWindow<memory_operation_enum::set>(
                     e_ptr, kargs, block_idx_m, block_idx_n);
                 EpiloguePipeline{}
                     .template operator()<decltype(e_block_window),
                                          decltype(c_block_tile),
                                          decltype(ds_block_window)>(e_block_window,
                                                                     c_block_tile,
                                                                     ds_block_window,
                                                                     smem_ptr_ping,
                                                                     scale_m_window,
                                                                     scale_n_window);
             }
             else
             {
                 auto e_block_window = MakeEBlockWindow<memory_operation_enum::atomic_add>(
                     e_ptr, kargs, block_idx_m, block_idx_n);
                 EpiloguePipeline{}
                     .template operator()<decltype(e_block_window),
                                          decltype(c_block_tile),
                                          decltype(ds_block_window)>(e_block_window,
                                                                     c_block_tile,
                                                                     ds_block_window,
                                                                     smem_ptr_ping,
                                                                     scale_m_window,
                                                                     scale_n_window);
             }
         }
         else if(UseDefaultScheduler || (get_warp_id() == 0))
         {
             if(kargs.k_batch == 1)
             {
                 auto e_block_window = MakeEBlockWindow<memory_operation_enum::set>(
                     e_ptr, kargs, block_idx_m, block_idx_n);
                 EpiloguePipeline{}
                     .template operator()<decltype(e_block_window),
                                          decltype(c_block_tile),
                                          decltype(ds_block_window)>(
                         e_block_window, c_block_tile, ds_block_window, smem_ptr_ping);
             }
             else
             {
                 auto e_block_window = MakeEBlockWindow<memory_operation_enum::atomic_add>(
                     e_ptr, kargs, block_idx_m, block_idx_n);
                 EpiloguePipeline{}
                     .template operator()<decltype(e_block_window),
                                          decltype(c_block_tile),
                                          decltype(ds_block_window)>(
                         e_block_window, c_block_tile, ds_block_window, smem_ptr_ping);
             }
         }
     }
  
     template <class ScaleM, class ScaleN>
     CK_TILE_DEVICE void operator()(FlatmmKernelArgs<ScaleM, ScaleN, DsDataType::size()> kargs,
                                    int partition_idx = blockIdx.x) const
     {
         int total_work_tile_cnt = TilePartitioner::GridSize(kargs.M, kargs.N);
  
         do
         {
             const auto [iM, iN] =
                 TilePartitioner{kargs.M, kargs.N}.GetOutputTileIndex(partition_idx);
             const index_t i_m = amd_wave_read_first_lane(iM * TilePartitioner::MPerBlock);
             const index_t i_n = amd_wave_read_first_lane(iN * TilePartitioner::NPerBlock);
  
             const SplitKBatchOffset splitk_batch_offset(kargs);
             // options
             const ADataType* a_ptr =
                 static_cast<const ADataType*>(kargs.a_ptr) + splitk_batch_offset.a_k_split_offset;
             const BDataType* b_flat_ptr =
                 static_cast<const BDataType*>(kargs.b_ptr) + splitk_batch_offset.b_k_split_offset;
             EDataType* e_ptr = static_cast<EDataType*>(kargs.e_ptr);
  
             // allocate LDS
             __shared__ char smem_ptr_ping[GetSmemPingSize()];
             __shared__ char smem_ptr_pong[GetSmemPongSize()];
  
             if constexpr(!(EpiloguePipeline::GetVectorSizeC() % 2 != 0 &&
                            is_any_of<EDataType, fp16_t, bf16_t>::value))
             {
                 constexpr auto scheduler_type = (FlatmmPipeline::NumWaveGroups == 1);
                 RunFlatmm<ScaleM, ScaleN, scheduler_type>(a_ptr,
                                                           b_flat_ptr,
                                                           kargs.ds_ptr,
                                                           e_ptr,
                                                           smem_ptr_ping,
                                                           smem_ptr_pong,
                                                           kargs,
                                                           splitk_batch_offset,
                                                           i_m,
                                                           i_n);
             }
             partition_idx += gridDim.x;
         } while(UsePersistentKernel && partition_idx < total_work_tile_cnt);
     }
 };
  
 } // namespace ck_tile