* Copyright (c) 2025 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.
*/
#include "kernel_operator.h"
using namespace AscendC;
namespace {
constexpr int32_t BUFFER_NUM = 2;
class KernelAdd {
public:
__aicore__ inline KernelAdd() {}
__aicore__ inline void Init(__gm__ float* srcA, __gm__ float* srcB, __gm__ float* dst, uint32_t totalSize)
{
blockIdx = GetBlockIdx();
blockDim = GetBlockNum();
elementsPerBlock = totalSize / blockDim;
tileLength = elementsPerBlock / TILE_NUM / BUFFER_NUM;
startIdx = blockIdx * elementsPerBlock;
srcAGlobal.SetGlobalBuffer(srcA + startIdx, elementsPerBlock);
srcBGlobal.SetGlobalBuffer(srcB + startIdx, elementsPerBlock);
dstGlobal.SetGlobalBuffer(dst + startIdx, elementsPerBlock);
pipe.InitBuffer(queueInA, BUFFER_NUM, tileLength * sizeof(float));
pipe.InitBuffer(queueInB, BUFFER_NUM, tileLength * sizeof(float));
pipe.InitBuffer(queueOut, BUFFER_NUM, tileLength * sizeof(float));
}
__aicore__ inline void Process()
{
int32_t loopCount = TILE_NUM * BUFFER_NUM;
for (int32_t i = 0; i < loopCount; i++) {
CopyIn(i);
Compute(i);
CopyOut(i);
}
}
private:
__aicore__ inline void CopyIn(int32_t progress)
{
LocalTensor<float> localA = queueInA.AllocTensor<float>();
LocalTensor<float> localB = queueInB.AllocTensor<float>();
DataCopy(localA, srcAGlobal[progress * tileLength], tileLength);
DataCopy(localB, srcBGlobal[progress * tileLength], tileLength);
queueInA.EnQue(localA);
queueInB.EnQue(localB);
}
__aicore__ inline void Compute(int32_t progress)
{
LocalTensor<float> localA = queueInA.DeQue<float>();
LocalTensor<float> localB = queueInB.DeQue<float>();
LocalTensor<float> localOut = queueOut.AllocTensor<float>();
Add(localOut, localA, localB, tileLength);
queueOut.EnQue<float>(localOut);
queueInA.FreeTensor(localA);
queueInB.FreeTensor(localB);
}
__aicore__ inline void CopyOut(int32_t progress)
{
LocalTensor<float> localOut = queueOut.DeQue<float>();
DataCopy(dstGlobal[progress * tileLength], localOut, tileLength);
queueOut.FreeTensor(localOut);
}
private:
TPipe pipe;
TQue<QuePosition::VECIN, BUFFER_NUM> queueInA;
TQue<QuePosition::VECIN, BUFFER_NUM> queueInB;
TQue<QuePosition::VECOUT, BUFFER_NUM> queueOut;
GlobalTensor<float> srcAGlobal;
GlobalTensor<float> srcBGlobal;
GlobalTensor<float> dstGlobal;
uint32_t blockIdx;
uint32_t blockDim;
uint32_t elementsPerBlock;
uint32_t startIdx;
uint32_t tileLength;
static constexpr int32_t TILE_NUM = 8;
};
}
extern "C" __global__ __aicore__ void AddKernel(__gm__ float* srcA, __gm__ float* srcB, __gm__ float* dst,
uint32_t totalSize)
{
KernelAdd op;
op.Init(srcA, srcB, dst, totalSize);
op.Process();
}
void AddDo(uint32_t blockDim, void* stream, float* srcA, float* srcB, float* dst, uint32_t totalSize)
{
AddKernel<<<blockDim, nullptr, stream>>>(srcA, srcB, dst, totalSize);
}