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

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

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/ops/gemm_quant/kernel/grouped_gemm_quant_kernel.hpp Source File
Go to the documentation of this file.
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
  
 #pragma once
  
 #include "ck_tile/core/numeric/math.hpp"
 #include "ck_tile/core/utility/literals.hpp"
 #include "ck_tile/core/utility/type_traits.hpp"
 #include "ck_tile/host/stream_utils.hpp"
 #include "ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v3.hpp"
 #include "ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_scheduler.hpp"
 #include "ck_tile/ops/gemm_quant/kernel/gemm_quant_kernel.hpp"
 #include "ck_tile/host.hpp"
  
 #include <hip/hip_runtime.h>
  
 namespace ck_tile {
  
 struct QuantGroupedGemmHostArgs
 {
     CK_TILE_HOST QuantGroupedGemmHostArgs(const void* a_ptr_,
                                           const void* b_ptr_,
                                           void* e_ptr_,
                                           const void* aq_ptr_,
                                           const void* bq_ptr_,
                                           index_t k_batch_,
                                           index_t M_,
                                           index_t N_,
                                           index_t K_,
                                           index_t QK_A_,
                                           index_t QK_B_,
                                           index_t stride_A_,
                                           index_t stride_B_,
                                           index_t stride_E_,
                                           index_t stride_AQ_,
                                           index_t stride_BQ_)
         : a_ptr(a_ptr_),
           b_ptr(b_ptr_),
           aq_ptr(aq_ptr_),
           bq_ptr(bq_ptr_),
           e_ptr(e_ptr_),
           M(M_),
           N(N_),
           K(K_),
           QK_A(QK_A_),
           QK_B(QK_B_),
           stride_A(stride_A_),
           stride_B(stride_B_),
           stride_AQ(stride_AQ_),
           stride_BQ(stride_BQ_),
           stride_E(stride_E_),
           k_batch(k_batch_)
     {
     }
  
     const void* a_ptr;
     const void* b_ptr;
     const void* aq_ptr;
     const void* bq_ptr;
     union
     {
         void* e_ptr;
         void* c_ptr;
     };
  
     index_t M;
     index_t N;
     index_t K;
     index_t QK_A;
     index_t QK_B;
     index_t stride_A;
     index_t stride_B;
     index_t stride_AQ;
     index_t stride_BQ;
  
     union
     {
         index_t stride_E;
         index_t stride_C;
     };
  
     index_t k_batch;
 };
  
 using QuantGroupedGemmKernelArgs = QuantGemmKernelArgs;
  
 struct QuantGemmTransKernelArg
 {
     QuantGroupedGemmKernelArgs group_karg;
     ck_tile::index_t block_start;
     ck_tile::index_t block_end;
  
     QuantGemmTransKernelArg() = delete;
     QuantGemmTransKernelArg(QuantGroupedGemmKernelArgs&& karg, index_t bl_start, index_t bl_end)
         : group_karg{karg}, block_start{bl_start}, block_end{bl_end}
     {
     }
  
     QuantGemmTransKernelArg(QuantGroupedGemmKernelArgs&& karg)
         : group_karg{karg}, block_start{0}, block_end{0}
     {
     }
 };
  
 template <typename TilePartitioner_,
           typename GemmPipeline_,
           typename EpiloguePipeline_,
           QuantType QuantType_>
 struct QuantGroupedGemmKernel
 {
     using Base = QuantGemmKernel<TilePartitioner_, GemmPipeline_, EpiloguePipeline_, QuantType_>;
  
     using TilePartitioner  = remove_cvref_t<TilePartitioner_>;
     using GemmPipeline     = remove_cvref_t<GemmPipeline_>;
     using EpiloguePipeline = remove_cvref_t<EpiloguePipeline_>;
  
     using ALayout = remove_cvref_t<typename GemmPipeline::ALayout>;
     using BLayout = remove_cvref_t<typename GemmPipeline::BLayout>;
     using CLayout = remove_cvref_t<typename GemmPipeline::CLayout>;
  
     using ADataType   = remove_cvref_t<typename GemmPipeline::ADataType>;
     using BDataType   = remove_cvref_t<typename GemmPipeline::BDataType>;
     using CDataType   = remove_cvref_t<typename EpiloguePipeline::ODataType>;
     using AccDataType = remove_cvref_t<typename EpiloguePipeline::AccDataType>;
  
     using AQDataType =
         remove_cvref_t<typename detail::get_aq_data_type_or<GemmPipeline, AccDataType>::type>;
     using BQDataType =
         remove_cvref_t<typename detail::get_bq_data_type_or<GemmPipeline, AccDataType>::type>;
  
     static constexpr auto kQuantType = QuantType_;
  
     static_assert(
         !is_detected<is_tuple, ALayout>::value && !is_detected<is_tuple, ADataType>::value,
         "ALayout and ADataType must be scalars. Multiple parameters are not currently supported.");
  
     static_assert(
         !is_detected<is_tuple, BLayout>::value && !is_detected<is_tuple, BDataType>::value,
         "BLayout and BDataType must be scalars. Multiple parameters are not currently supported.");
  
     static_assert(!is_detected<is_tuple, CLayout>::value &&
                       !is_detected<is_tuple, CDataType>::value,
                   "C/ELayout and C/EDataType must be scalars.");
  
     using OffsetTile1DPartitioner = OffsettedTile1DPartitioner<TilePartitioner>;
     using Kernel =
         QuantGroupedGemmKernel<TilePartitioner, GemmPipeline, EpiloguePipeline, kQuantType>;
  
     static constexpr index_t kBlockSize       = GemmPipeline::BlockSize;
     static constexpr bool UsePersistentKernel = GemmPipeline::UsePersistentKernel;
     static_assert(UsePersistentKernel == true, "UsePersistentKernel must be true");
  
     [[nodiscard]] CK_TILE_HOST static const std::string GetName()
     {
         // clang-format off
         using P_ = GemmPipeline;
  
         return concat('_', "gemm_grouped", gemm_prec_str<ADataType, BDataType>(),
                       concat('x', P_::MPerBlock, P_::NPerBlock, P_::KPerBlock),
                       concat('x', P_::GetVectorSizeA(), P_::GetVectorSizeB(), P_::GetVectorSizeC()),
                       concat('x', P_::kPadM, P_::kPadN, P_::kPadK),
                       (UsePersistentKernel ? "Persistent" : "NonPersistent"));
         // clang-format on
     }
  
     CK_TILE_HOST static auto
     GetWorkSpaceSize(const std::vector<QuantGroupedGemmHostArgs>& gemm_descs) -> std::size_t
     {
         return gemm_descs.size() * sizeof(QuantGemmTransKernelArg);
     }
  
     CK_TILE_HOST static auto GetWorkSpaceSize(index_t group_count) -> std::size_t
     {
         return group_count * sizeof(QuantGemmTransKernelArg);
     }
  
     CK_TILE_HOST static auto BlockSize() -> dim3
     {
         if(is_wave32())
         {
             return dim3(kBlockSize / 2);
         }
         else
         {
             return dim3(kBlockSize);
         }
     }
  
     CK_TILE_HOST static auto MaxOccupancyGridSize(const stream_config& s) -> dim3
     {
         using ConstantPointer  = const void CK_CONSTANT_ADDRESS_SPACE*;
         const auto kernel_func = kentry<1, Kernel, ConstantPointer, index_t>;
         int occupancy;
         HIP_CHECK_ERROR(
             hipOccupancyMaxActiveBlocksPerMultiprocessor(&occupancy, kernel_func, kBlockSize, 0));
         const int grid_size = get_available_compute_units(s) * occupancy;
         return dim3(grid_size, 1, 1);
     }
  
     CK_TILE_HOST static auto GridSize(const std::vector<QuantGroupedGemmHostArgs>& gemm_descs)
     {
         index_t grid_size = 0;
         for(const auto& it_desc : gemm_descs)
         {
             const auto local_grid_size = TilePartitioner::GridSize(it_desc.M, it_desc.N);
             grid_size += local_grid_size * it_desc.k_batch;
         }
         return dim3(grid_size, 1, 1);
     }
  
     CK_TILE_HOST static auto MakeKargs(const std::vector<QuantGroupedGemmHostArgs>& gemm_descs)
         -> std::vector<QuantGemmTransKernelArg>
     {
         std::vector<QuantGemmTransKernelArg> gemm_kernel_args_;
         index_t group_count = ck_tile::type_convert<ck_tile::index_t>(gemm_descs.size());
         index_t grid_size   = 0;
         gemm_kernel_args_.reserve(group_count);
  
         for(std::size_t i = 0; i < gemm_descs.size(); ++i)
         {
             const index_t M = gemm_descs[i].M;
             const index_t N = gemm_descs[i].N;
             const index_t K = gemm_descs[i].K;
  
             if(M == 0 || N == 0 || K == 0)
             {
                 continue;
             }
  
             const index_t stride_a = gemm_descs[i].stride_A;
             const index_t stride_b = gemm_descs[i].stride_B;
             const index_t stride_e = gemm_descs[i].stride_C;
  
             const index_t grid_size_grp = TilePartitioner::GridSize(M, N) * gemm_descs[i].k_batch;
  
             const index_t block_start = grid_size;
             const index_t block_end   = grid_size + grid_size_grp;
  
             grid_size += grid_size_grp;
  
             auto karg =
                 QuantGroupedGemmKernelArgs{type_convert<const ADataType*>(gemm_descs[i].a_ptr),
                                            type_convert<const BDataType*>(gemm_descs[i].b_ptr),
                                            type_convert<CDataType*>(gemm_descs[i].e_ptr),
                                            type_convert<const AQDataType*>(gemm_descs[i].aq_ptr),
                                            type_convert<const BQDataType*>(gemm_descs[i].bq_ptr),
                                            gemm_descs[i].k_batch,
                                            M,
                                            N,
                                            K,
                                            gemm_descs[i].QK_A,
                                            gemm_descs[i].QK_B,
                                            stride_a,
                                            stride_b,
                                            stride_e,
                                            gemm_descs[i].stride_AQ,
                                            gemm_descs[i].stride_BQ};
  
             gemm_kernel_args_.emplace_back(std::move(karg), block_start, block_end);
         }
  
         return gemm_kernel_args_;
     }
  
     CK_TILE_HOST static bool IsSupportedArgument(const std::vector<QuantGemmTransKernelArg>& kargs)
     {
         for(const auto& karg : kargs)
         {
             if(!Base::IsSupportedArgument(karg.group_karg))
             {
                 return false;
             }
         }
         return true;
     }
  
     CK_TILE_HOST_DEVICE static constexpr auto GetSmemSize() -> index_t
     {
         return max(GemmPipeline::GetSmemSize(), EpiloguePipeline::GetSmemSize());
     }
  
     CK_TILE_DEVICE void Run(const QuantGroupedGemmKernelArgs& kargs,
                             const tuple<index_t, index_t>& block_idx_2d,
                             const index_t block_idx_z) const
     {
         const auto [iM, iN] = block_idx_2d;
  
         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 typename Base::SplitKBatchOffset splitk_batch_offset(kargs, block_idx_z);
  
         // options
         const ADataType* a_ptr   = static_cast<const ADataType*>(kargs.a_ptr);
         const BDataType* b_ptr   = static_cast<const BDataType*>(kargs.b_ptr);
         const AQDataType* aq_ptr = static_cast<const AQDataType*>(kargs.aq_ptr);
         const BQDataType* bq_ptr = static_cast<const BQDataType*>(kargs.bq_ptr);
         CDataType* c_ptr         = static_cast<CDataType*>(kargs.c_ptr);
  
         static_assert(GemmPipeline::DoubleSmemBuffer == false,
                       "DoubleSmemBuffer needs to be false");
         // allocate LDS
         __shared__ char smem_ptr_0[GetSmemSize()];
  
         RunGemmWithPipelineSelection(
             a_ptr, b_ptr, aq_ptr, bq_ptr, c_ptr, smem_ptr_0, kargs, splitk_batch_offset, i_m, i_n);
     }
  
     CK_TILE_DEVICE static void
     RunGemmWithPipelineSelection(const ADataType* a_ptr,
                                  const BDataType* b_ptr,
                                  const AQDataType* aq_ptr,
                                  const BQDataType* bq_ptr,
                                  CDataType* c_ptr,
                                  void* smem_ptr_0,
                                  const QuantGroupedGemmKernelArgs& kargs,
                                  const typename Base::SplitKBatchOffset& splitk_batch_offset,
                                  const index_t block_idx_m,
                                  const index_t block_idx_n)
     {
         // Create Gemm tensor views, pad views and tile windows
         const auto& gemm_tensor_views_tuple =
             Base::template MakeGemmTensorViews<EpiloguePipeline::MemoryOperation>(
                 a_ptr, b_ptr, aq_ptr, bq_ptr, c_ptr, kargs, splitk_batch_offset);
  
         const auto& gemm_pad_views = Base::MakeGemmPadViews(gemm_tensor_views_tuple);
         auto gemm_tile_windows =
             Base::MakeGemmTileWindows(gemm_pad_views, block_idx_m, block_idx_n);
         const auto& a_block_window = gemm_tile_windows.at(Base::I0);
         const auto& b_block_window = gemm_tile_windows.at(Base::I2);
  
         // Get hot-loop and tail configuration
         const index_t num_loop = __builtin_amdgcn_readfirstlane(
             TilePartitioner::GetLoopNum(splitk_batch_offset.splitted_k));
         const bool has_hot_loop   = GemmPipeline::BlockHasHotloop(num_loop);
         const TailNumber tail_num = GemmPipeline::GetBlockLoopTailNum(num_loop);
  
         // Run GEMM pipeline
         const auto& c_block_tile = GemmPipeline{}.template operator()(
             a_block_window, b_block_window, num_loop, has_hot_loop, tail_num, smem_ptr_0);
         // Run Epilogue Pipeline
         auto& c_block_window = gemm_tile_windows.at(Base::I4);
         if constexpr(kQuantType == QuantType::RowColQuant)
         {
             const auto& aq_block_window = gemm_tile_windows.at(Base::I1);
             const auto& bq_block_window = gemm_tile_windows.at(Base::I3);
             EpiloguePipeline{}.template
             operator()<decltype(c_block_window), decltype(c_block_tile), decltype(c_block_window)>(
                 c_block_window,
                 c_block_tile,
                 c_block_window,
                 smem_ptr_0,
                 aq_block_window,
                 bq_block_window);
         }
     }
  
     // For persistent kernels
     template <bool U   = UsePersistentKernel,
               typename = std::enable_if_t<U>,
               typename = void> // extra template parameter to avoid redefinition
     CK_TILE_DEVICE void operator()(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
                                    const index_t group_count) const
     {
         const index_t grid_size  = ck_tile::get_grid_size();
         const auto gemm_desc_ptr = reinterpret_cast<const QuantGemmTransKernelArg*>(
             cast_pointer_to_generic_address_space(gemm_descs_const));
         index_t block_id      = ck_tile::get_block_1d_id(); // initial block_id
         index_t cum_grid_size = 0;
         for(index_t group_id = 0; group_id < group_count; ++group_id)
         {
             const auto& kargs      = gemm_desc_ptr[group_id].group_karg;
             const auto& k_batch    = kargs.k_batch;
             const auto block_start = cum_grid_size;
             cum_grid_size += TilePartitioner::GridSize(kargs.M, kargs.N) * k_batch;
             while(block_id < cum_grid_size)
             {
                 const auto grid_size_2d = TilePartitioner::GridSize(kargs.M, kargs.N);
                 const auto block_idx_2d = OffsetTile1DPartitioner::GetOffsetedTileIndex(
                     0, kargs.M, kargs.N, (block_id - block_start) % grid_size_2d);
                 Run(kargs, block_idx_2d, (block_id - block_start) / grid_size_2d);
                 block_id = block_id + grid_size; // advance to next block
                 // NOTE: this check is redundant but helps the compiler avoid spilling some VGPR
                 if(block_id >= cum_grid_size)
                 {
                     break; // exit the loop if all blocks are processed
                 }
             }
         }
     }
 };
  
 } // namespace ck_tile