昇腾社区首页
中文
注册
开发者
下载
支持
积分兑换 NEW

batch场景

单次Matmul计算处理的shape比较小时，由于每次计算均涉及到内部的通信，可能会影响性能，该接口提供批量处理Matmul的功能，调用一次IterateBatch，可以计算出多个singleCoreM * singleCoreN大小的C矩阵。

如下的示例中，包含4个矩阵乘操作a*a、b*b、c*c、d*d, 需要单核上计算多个singleCoreM *singleCoreN，shape较小的情况可以使能BatchMatmul，批量处理。以BMK*BKN=BMN（相关格式参见IterateBatch）场景为例，如下图，一次IterateBatch可同时计算出A = a*a、B = b*b、C = c*c、D = d*d。

实例化Matmul时，需要通过MatmulType设置输入输出的Layout格式为NORMAL(BMNK的数据排布格式使用NORMAL表示)。Host侧Tiling时需使用SetBatchInfoForNormal设置A/B/C的M/N/K轴信息和A/B矩阵的BatchNum数。

如下示例完成aGM、bGM矩阵乘，结果保存到cGm上，其中aGM、bGM、cGM数据的layout格式均为NORMAL，左矩阵每次计算batchA个MK数据，右矩阵每次计算batchB个KN数据。 
#include "kernel_operator.h"
#include "lib/matmul_intf.h"

extern "C" __global__  __aicore__ void kernel_matmul_rpc_batch(GM_ADDR aGM, GM_ADDR bGM, GM_ADDR cGM, GM_ADDR biasGM, GM_ADDR tilingGM, GM_ADDR workspaceGM, uint32_t isTransposeAIn, uint32_t isTransposeBIn, int32_t batchA,  int32_t batchB)
{
    // 定义matmul type
    typedef MatmulType <AscendC::TPosition::GM, CubeFormat::ND, half, false, LayoutMode::NORMAL> aType;
    typedef MatmulType <AscendC::TPosition::GM, CubeFormat::ND, half, true, LayoutMode::NORMAL> bType;
    typedef MatmulType <AscendC::TPosition::GM, CubeFormat::ND, float, false, LayoutMode::NORMAL> cType;
    typedef MatmulType <AscendC::TPosition::GM, CubeFormat::ND, float> biasType;

    // 初始化tiling数据
    TCubeTiling tiling;
    auto tempTilingGM = (__gm__ uint32_t*)tilingGM;
    auto tempTiling = (uint32_t*)&tiling;
    for (int i = 0; i < sizeof(TCubeTiling) / sizeof(int32_t); ++i, ++tempTilingGM, ++tempTiling) {
        *tempTiling = *tempTilingGM;
    }

    // 初始化gm数据
    GlobalTensor<half> aGlobal;
    GlobalTensor<half> bGlobal;
    GlobalTensor<float> cGlobal;
    GlobalTensor<float> biasGlobal;
    int32_t sizeA = tiling.ALayoutInfoB * tiling.singleCoreM * tiling.singleCoreK * sizeof(A_T);
    int32_t sizeB = tiling.BLayoutInfoB * tiling.singleCoreK * tiling.singleCoreN * sizeof(B_T);
    int32_t sizeC = tiling.CLayoutInfoB * tiling.singleCoreM * tiling.singleCoreN * sizeof(C_T);
    int32_t sizebias = tiling.CLayoutInfoB * tiling.singleCoreN * sizeof(C_T);
    aGlobal.SetGlobalBuffer(reinterpret_cast<__gm__ half*>(aGM), sizeA);
    bGlobal.SetGlobalBuffer(reinterpret_cast<__gm__ half*>(bGM), sizeB);
    cGlobal.SetGlobalBuffer(reinterpret_cast<__gm__ float*>(cGM), sizeC);
    biasGlobal.SetGlobalBuffer(reinterpret_cast<__gm__ float*>(biasGM), sizebias);
    tiling.shareMode = 0;            
    tiling.shareL1Size = 512 * 1024;
    tiling.shareL0CSize = 128 * 1024;
    tiling.shareUbSize = 0;
    int offset_a=0, offset_b=0, offset_c=0, offset_bias=0;
    GlobalTensor<A_T> gm_a;
    gm_a.SetGlobalBuffer(const_cast<__gm__ A_T*>(aGlobal[offset_a].GetPhyAddr()), tiling.singleCoreM * tiling.singleCoreK);
    GlobalTensor<B_T> gm_b;
    gm_b.SetGlobalBuffer(const_cast<__gm__ B_T*>(bGlobal[offset_b].GetPhyAddr()), tiling.singleCoreK * tiling.singleCoreN);
    GlobalTensor<C_T> gm_c;
    gm_c.SetGlobalBuffer(const_cast<__gm__ C_T*>(cGlobal[offset_c].GetPhyAddr()), tiling.singleCoreM * tiling.singleCoreN) ;
    GlobalTensor<BiasT> gm_bias;
    gm_bias.SetGlobalBuffer(const_cast<__gm__ BiasT*>(biasGlobal[offset_bias].GetPhyAddr()), tiling.singleCoreN);
    // 创建Matmul实例
    constexpr MatmulConfig MM_CFG = GetNormalConfig(false, false, false, BatchMode::BATCH_LESS_THAN_L1);
    Matmul<aType, bType, cType, biasType, MM_CFG> mm1;
    TPipe pipe;
    g_cubeTPipePtr = &pipe;
    SetSysWorkspace(workspaceGM);
    REGIST_MATMUL_OBJ(&pipe, GetSysWorkSpacePtr(), mm1);
    mm1.Init(&tiling);
    mm1.SetTensorA(gm_a, isTransposeAIn);
    mm1.SetTensorB(gm_b, isTransposeBIn);
    if(tiling.isBias) {
        mm1.SetBias(gm_bias);
    }
    // 多batch Matmul计算
    mm1.IterateBatch(gm_c, batchA, batchB, false);
}