* Copyright (c) 2026 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file kernel_grouped_matmul add.h
* \brief
*/
#ifndef MATMUL_KERNEL_KERNEL_GROUPED_MATMUL_ADD_H
#define MATMUL_KERNEL_KERNEL_GROUPED_MATMUL_ADD_H
#define ASCENDC_CUBE_ONLY
#if ASC_DEVKIT_MAJOR >= 9
#include "kernel_basic_intf.h"
#else
#include "kernel_operator.h"
#endif
#include "kernel_operator_list_tensor_intf.h"
#include "lib/matmul_intf.h"
#include "../utils/common_utils.h"
#include "../utils/layout_utils.h"
#include "../utils/tuple_utils.h"
#include "../utils/coord_utils.h"
#include "../utils/tensor_utils.h"
#include "../utils/status_utils.h"
#include "../block/block_grouped_matmul_builder.h"
#include "../epilogue/block_epilogue_empty.h"
#include "../block/block_scheduler_utils.h"
#include "../block/block_scheduler_grouped_matmul_aswt.h"
namespace Cgmct {
namespace Gemm {
namespace Kernel {
constexpr uint64_t BLOCK_BYTE_SIZE = 32UL;
constexpr uint64_t M_VALUE = 0UL;
constexpr uint64_t N_VALUE = 1UL;
constexpr uint64_t K_VALUE = 2UL;
constexpr uint64_t NUM_ZERO = 0UL;
constexpr uint64_t NUM_ONE = 1UL;
constexpr uint64_t NUM_TWO = 2UL;
constexpr uint64_t NUM_THREE = 3UL;
template <class ProblemShape_, class BlockMmadBuilder_, class BlockEpilogue_, class BlockScheduler_,
typename Enable_ = void>
class KernelGroupedMatmulAdd {
static_assert(AscendC::Std::always_false_v<BlockScheduler_>,
"KernelGroupedMatmulAdd is not implemented for this scheduler");
};
template <class ProblemShape_, class BlockMmadBuilder_, class BlockEpilogue_, class BlockScheduler_>
class KernelGroupedMatmulAdd<
ProblemShape_, BlockMmadBuilder_, BlockEpilogue_, BlockScheduler_,
AscendC::Std::enable_if_t<AscendC::Std::is_same_v<BlockScheduler_, GroupedMatmulAswtScheduler>>> {
public:
__aicore__ inline KernelGroupedMatmulAdd() {}
__aicore__ inline ~KernelGroupedMatmulAdd() {}
using BlockEpilogue = BlockEpilogue_;
using BlockMmadBuilder = BlockMmadBuilder_;
using ProblemShape = ProblemShape_;
using BlockScheduler = BlockScheduler_;
static constexpr bool transA = BlockMmadBuilder::transA;
static constexpr bool transB = BlockMmadBuilder::transB;
static constexpr int64_t l1M = BlockMmadBuilder::l1M;
static constexpr int64_t l1N = BlockMmadBuilder::l1N;
static constexpr int64_t l1K = BlockMmadBuilder::l1K;
using BlockSchedulerOp =
typename Block::BlockSchedulerSelector<ProblemShape, typename BlockMmadBuilder::L1TileShape,
typename BlockMmadBuilder::L0TileShape, BlockScheduler, transA,
transB>::SchedulerOp;
using BlockMmadOp = typename BlockMmadBuilder::BlockMmadOp;
using BlockMmadArguments = typename BlockMmadBuilder::Arguments;
using BlockEpilogueArguments = typename BlockEpilogue::Arguments;
using BlockMmadParams = typename BlockMmadBuilder::Params;
using BlockEpilogueParams = typename BlockEpilogue::Params;
using AType = typename BlockMmadBuilder::AType;
using BType = typename BlockMmadBuilder::BType;
using CType = typename BlockMmadBuilder::CType;
using BiasType = typename BlockMmadBuilder::BiasType;
using TupleShape = AscendC::Shape<int64_t, int64_t, int64_t, int64_t>;
using BlockShape = AscendC::Shape<int64_t, int64_t, int64_t, int64_t>;
using BlockOffset = AscendC::Shape<int64_t, int64_t, int64_t>;
using CoordClass =
Coordinate<transA, transB, BlockMmadBuilder::formatA, BlockMmadBuilder::formatB, BlockMmadBuilder::formatC>;
AscendC::GlobalTensor<AType> aGlobal_;
AscendC::GlobalTensor<BType> bGlobal_;
AscendC::GlobalTensor<CType> cGlobal_;
AscendC::GlobalTensor<int64_t> groupListGm_;
BlockMmadParams blockMmadParams_{};
TupleShape problemShape_{};
BlockOffset baseOffset_{0, 0, 0};
int64_t preOffset_{0};
struct GMMAddTiling {
uint32_t groupNum;
uint32_t groupListType;
uint64_t mTailCnt;
uint64_t nTailCnt;
const TCubeTiling* __restrict matmulTiling;
__aicore__ GMMAddTiling() {}
__aicore__ GMMAddTiling(uint32_t groupNum_, uint32_t groupListType_, uint64_t mTailCnt_ = 1,
uint64_t nTailCnt_ = 1) :
groupNum(groupNum_),
groupListType(groupListType_), mTailCnt(mTailCnt_), nTailCnt(nTailCnt_)
{}
};
struct Arguments {
ProblemShape problemShape;
BlockMmadArguments mmadArgs;
BlockEpilogueArguments epilogueArgs;
GMMAddTiling gmmArgs;
Arguments() = default;
};
struct Params {
ProblemShape problemShape;
BlockMmadParams mmadParams;
BlockEpilogueParams epilogueParams;
GMMAddTiling gmmParams;
Params() = default;
};
__aicore__ inline int64_t GetSplitValueFromGroupList(uint64_t groupIdx, uint64_t groupListType)
{
int64_t splitValue = 0;
if (groupListType == 0) {
int64_t offset = static_cast<int64_t>(groupListGm_.GetValue(groupIdx));
splitValue = offset - preOffset_;
preOffset_ = offset;
} else {
splitValue = static_cast<int64_t>(groupListGm_.GetValue(groupIdx));
}
return splitValue;
}
__aicore__ inline void InitGlobalBuffer(const Params& params, uint64_t groupIdx)
{
aGlobal_.SetGlobalBuffer((__gm__ AType*)params.mmadParams.aGmAddr + Get<NUM_ZERO>(baseOffset_));
bGlobal_.SetGlobalBuffer((__gm__ BType*)params.mmadParams.bGmAddr + Get<NUM_ONE>(baseOffset_));
cGlobal_.SetGlobalBuffer((__gm__ CType*)params.mmadParams.cGmAddr + Get<NUM_TWO>(baseOffset_));
}
__aicore__ inline void UpdateOffset(uint64_t groupIdx)
{
if (groupIdx == 0) {
return;
}
int64_t m = Get<M_VALUE>(problemShape_);
int64_t n = Get<N_VALUE>(problemShape_);
int64_t k = Get<K_VALUE>(problemShape_);
Get<NUM_ZERO>(baseOffset_) = Get<NUM_ZERO>(baseOffset_) + m * k;
Get<NUM_ONE>(baseOffset_) = Get<NUM_ONE>(baseOffset_) + n * k;
Get<NUM_TWO>(baseOffset_) = Get<NUM_TWO>(baseOffset_) + m * n;
}
__aicore__ inline bool Init(const Params& params, uint64_t groupIdx)
{
UpdateOffset(groupIdx);
int64_t splitValue = GetSplitValueFromGroupList(groupIdx, params.gmmParams.groupListType);
Get<K_VALUE>(problemShape_) = splitValue;
if (splitValue <= 0) {
return false;
}
InitGlobalBuffer(params, groupIdx);
return true;
}
__aicore__ inline void InitParamsAndGm(const Params& params)
{
if (params.mmadParams.groupListGmAddr != nullptr) {
groupListGm_.SetGlobalBuffer(reinterpret_cast<__gm__ int64_t*>(params.mmadParams.groupListGmAddr));
}
Get<M_VALUE>(problemShape_) = params.gmmParams.matmulTiling->M;
Get<N_VALUE>(problemShape_) = params.gmmParams.matmulTiling->N;
cGlobal_.SetL2CacheHint(AscendC::CacheMode::CACHE_MODE_DISABLE);
}
__host_aicore__ static Status CheckShape(const ProblemShape& shape)
{
int64_t m = shape.m;
int64_t n = shape.n;
int64_t k = shape.k;
int64_t b = shape.b;
if (b > INT32_MAX) {
return Status::batchErrorExcceedsLimit;
}
if (m > INT32_MAX || n > INT32_MAX || k > INT32_MAX) {
return Status::mnkErrorExceedsLimit;
}
return Status::success;
}
__host_aicore__ static Status CanImplement(const Arguments& args)
{
CHECK_AND_RETURN(CheckShape(args.problemShape));
CHECK_AND_RETURN(BlockMmadBuilder::CanImplement(args.mmadArgs));
CHECK_AND_RETURN(BlockSchedulerOp::CanImplement(args.problemShape));
CHECK_AND_RETURN(BlockEpilogue::CanImplement(args.epilogueArgs));
return Status::success;
}
__host_aicore__ static size_t GetWorkspaceSize(ProblemShape shape, int64_t blockNum)
{
size_t workSpaceSize = 0;
workSpaceSize += BlockMmadBuilder::GetWorkspaceSize();
workSpaceSize += BlockEpilogue::GetWorkspaceSize(blockNum, l1M, l1N);
workSpaceSize += BlockSchedulerOp::GetWorkspaceSize(shape);
return workSpaceSize;
}
__host_aicore__ static Params InitParams(const Arguments& args, GM_ADDR workspace)
{
BlockMmadParams mmadParams = BlockMmadBuilder::InitParams(args.mmadArgs);
if constexpr (!AscendC::Std::is_same_v<BlockEpilogue, Block::BlockEpilogueEmpty>) {
mmadParams.cGmAddr = workspace;
}
BlockEpilogueParams epilogueParams = BlockEpilogue::InitParams(args.epilogueArgs, workspace);
Params params = {args.problemShape, mmadParams, epilogueParams, args.gmmArgs};
return params;
}
__host_aicore__ static int64_t GetBlockNum(ProblemShape shape)
{
return BlockSchedulerOp::GetBlockNum(shape);
}
__aicore__ inline void operator()(const Params& params)
{
if ASCEND_IS_AIV {
return;
}
BlockMmadOp blockMmadOp;
int64_t curBlockIdx = AscendC::GetBlockIdx();
int64_t blockNum = AscendC::GetBlockNum();
if (curBlockIdx >= blockNum || blockNum == 0) {
return;
}
InitParamsAndGm(params);
int32_t baseM = params.gmmParams.matmulTiling->baseM;
int32_t baseN = params.gmmParams.matmulTiling->baseN;
int32_t baseK = params.gmmParams.matmulTiling->baseK;
blockMmadOp.Init(const_cast<TCubeTiling* __restrict>(params.gmmParams.matmulTiling), GetTPipePtr());
for (uint64_t groupIdx = 0, count = 0; groupIdx < params.gmmParams.groupNum; groupIdx++) {
if (!Init(params, groupIdx)) {
continue;
}
int64_t m = Get<M_VALUE>(problemShape_);
int64_t n = Get<N_VALUE>(problemShape_);
int64_t k = Get<K_VALUE>(problemShape_);
CoordClass coord(m, n, k, baseM, baseN, baseK);
BlockSchedulerOp bs(m, n, k, baseM, baseN, baseK, curBlockIdx, blockNum, params.gmmParams.mTailCnt,
params.gmmParams.nTailCnt, true);
blockMmadOp.SetOrgShape(m, n, k);
uint64_t curCount = count + bs.GetTileNum();
uint64_t curBlock = curBlockIdx >= count ? curBlockIdx : curBlockIdx + blockNum;
for (; curBlock < curCount; curBlock += blockNum) {
BlockShape tileIdx = bs.GetTileIdx(curBlock, count);
BlockShape singleShape = bs.GetBlockShape(Get<NUM_ZERO>(tileIdx), Get<NUM_ONE>(tileIdx), curBlock,
BLOCK_BYTE_SIZE / sizeof(BType));
int64_t aOffset = coord.GetAOffset(Get<NUM_ZERO>(tileIdx), 0, 0, Get<NUM_TWO>(singleShape));
int64_t bOffset = coord.GetBOffset(Get<NUM_ONE>(tileIdx), 0, 0, BLOCK_BYTE_SIZE / sizeof(BType),
Get<NUM_THREE>(singleShape));
int64_t cOffset = coord.GetCOffset(Get<NUM_ZERO>(tileIdx), Get<NUM_ONE>(tileIdx), 0,
Get<NUM_TWO>(singleShape), Get<NUM_THREE>(singleShape));
if (Get<NUM_ZERO>(singleShape) <= 0 || Get<NUM_ONE>(singleShape) <= 0) {
continue;
}
blockMmadOp.SetSingleShape(Get<NUM_ZERO>(singleShape), Get<NUM_ONE>(singleShape), k);
blockMmadOp.SetTensorA(aGlobal_[aOffset], transA);
blockMmadOp.SetTensorB(bGlobal_[bOffset], transB);
blockMmadOp.IterateAll(cGlobal_[cOffset], 1);
}
count = curCount % blockNum;
}
}
};
}
}
}
#endif