* 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 lerp.h
* \brief
*/
#ifndef __LERP_H__
#define __LERP_H__
#include "kernel_operator.h"
#include "kernel_tiling/kernel_tiling.h"
#include "lerp_tiling_data.h"
#include "lerp_tiling_key.h"
namespace NsLerp {
using namespace AscendC;
constexpr int32_t BUFFER_NUM = 2;
template<typename TYPE_START, typename TYPE_END, typename TYPE_WEIGHT, typename TYPE_Y, bool IsExistBigCore>
class Lerp {
public:
__aicore__ inline Lerp() {}
__aicore__ inline void Init(GM_ADDR start, GM_ADDR end, GM_ADDR weight, GM_ADDR y,
uint64_t smallCoreDataNum, uint64_t bigCoreDataNum,
uint64_t bigCoreLoopNum, uint64_t smallCoreLoopNum,
uint64_t ubPartDataNum, uint64_t smallCoreTailDataNum,
uint64_t bigCoreTailDataNum, uint64_t tailBlockNum)
{
ASSERT(GetBlockNum() != 0 && "block dim can not be zero!");
uint64_t coreNum = GetBlockIdx();
uint64_t globalBufferIndex = bigCoreDataNum * GetBlockIdx();
this->ubPartDataNum = ubPartDataNum;
if constexpr (IsExistBigCore)
{
if (coreNum < tailBlockNum)
{
this->coreDataNum = bigCoreDataNum;
this->tileNum = bigCoreLoopNum;
this->tailDataNum = bigCoreTailDataNum;
}
else
{
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex -= (bigCoreDataNum - smallCoreDataNum) * (GetBlockIdx() - tailBlockNum);
}
}
else
{
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex = smallCoreDataNum * GetBlockIdx();
}
startGm.SetGlobalBuffer((__gm__ TYPE_START*)start + globalBufferIndex, this->coreDataNum);
endGm.SetGlobalBuffer((__gm__ TYPE_END*)end + globalBufferIndex, this->coreDataNum);
weightGm.SetGlobalBuffer((__gm__ TYPE_WEIGHT*)weight + globalBufferIndex, this->coreDataNum);
yGm.SetGlobalBuffer((__gm__ TYPE_Y*)y + globalBufferIndex, this->coreDataNum);
pipe.InitBuffer(inQueueStart, BUFFER_NUM, this->ubPartDataNum * sizeof(TYPE_START));
pipe.InitBuffer(inQueueEnd, BUFFER_NUM, this->ubPartDataNum * sizeof(TYPE_END));
pipe.InitBuffer(inQueueWeight, BUFFER_NUM, this->ubPartDataNum * sizeof(TYPE_WEIGHT));
pipe.InitBuffer(outQueueY, BUFFER_NUM, this->ubPartDataNum * sizeof(TYPE_Y));
if constexpr (!std::is_same_v<TYPE_START, float>)
{
pipe.InitBuffer(tmp1, this->ubPartDataNum * sizeof(float));
pipe.InitBuffer(tmp2, this->ubPartDataNum * sizeof(float));
}
}
__aicore__ inline void Process()
{
int32_t loopCount = this->tileNum;
this->processDataNum = this->ubPartDataNum;
for (int32_t i = 0; i < loopCount-1; i++)
{
CopyIn(i);
Compute(i);
CopyOut(i);
}
this->processDataNum = this->tailDataNum;
CopyIn(loopCount-1);
Compute(loopCount-1);
CopyOut(loopCount-1);
}
private:
__aicore__ inline void CopyIn(int32_t progress)
{
LocalTensor<TYPE_START> startLocal = inQueueStart.AllocTensor<TYPE_START>();
LocalTensor<TYPE_END> endLocal = inQueueEnd.AllocTensor<TYPE_END>();
LocalTensor<TYPE_WEIGHT> weightLocal = inQueueWeight.AllocTensor<TYPE_WEIGHT>();
DataCopy(startLocal, startGm[progress * this->ubPartDataNum], this->processDataNum);
DataCopy(endLocal, endGm[progress * this->ubPartDataNum], this->processDataNum);
DataCopy(weightLocal, weightGm[progress * this->ubPartDataNum], this->processDataNum);
inQueueStart.EnQue(startLocal);
inQueueEnd.EnQue(endLocal);
inQueueWeight.EnQue(weightLocal);
}
__aicore__ inline void Compute(int32_t progress)
{
LocalTensor<TYPE_START> startLocal = inQueueStart.DeQue<TYPE_START>();
LocalTensor<TYPE_END> endLocal = inQueueEnd.DeQue<TYPE_END>();
LocalTensor<TYPE_WEIGHT> weightLocal = inQueueWeight.DeQue<TYPE_WEIGHT>();
LocalTensor<TYPE_Y> yLocal = outQueueY.AllocTensor<TYPE_Y>();
if constexpr (std::is_same_v<TYPE_START, float>)
{
Sub(yLocal, endLocal, startLocal, this->processDataNum);
Mul(yLocal, weightLocal, yLocal, this->processDataNum);
Add(yLocal, yLocal, startLocal, this->processDataNum);
}
else if constexpr (std::is_same_v<TYPE_START, half> || std::is_same_v<TYPE_START, bfloat16_t>)
{
auto fstart = tmp1.Get<float>();
auto fend = tmp2.Get<float>();
Cast(fstart, startLocal, RoundMode::CAST_NONE, this->processDataNum);
Cast(fend, endLocal, RoundMode::CAST_NONE, this->processDataNum);
Sub(fend, fend, fstart, this->processDataNum);
Cast(fstart, weightLocal, RoundMode::CAST_NONE, this->processDataNum);
Mul(fend, fstart, fend, this->processDataNum);
Cast(fstart, startLocal, RoundMode::CAST_NONE, this->processDataNum);
Add(fend, fend, fstart, this->processDataNum);
Cast(yLocal, fend, RoundMode::CAST_RINT, this->processDataNum);
}
outQueueY.EnQue<TYPE_Y>(yLocal);
inQueueStart.FreeTensor(startLocal);
inQueueEnd.FreeTensor(endLocal);
inQueueWeight.FreeTensor(weightLocal);
}
__aicore__ inline void CopyOut(int32_t progress)
{
LocalTensor<TYPE_Y> yLocal = outQueueY.DeQue<TYPE_Y>();
DataCopy(yGm[progress * this->ubPartDataNum], yLocal, this->processDataNum);
outQueueY.FreeTensor(yLocal);
}
private:
TPipe pipe;
TQue<QuePosition::VECIN, BUFFER_NUM> inQueueStart;
TQue<QuePosition::VECIN, BUFFER_NUM> inQueueEnd;
TQue<QuePosition::VECIN, BUFFER_NUM> inQueueWeight;
TQue<QuePosition::VECOUT, BUFFER_NUM> outQueueY;
TBuf<QuePosition::VECCALC> tmp1;
TBuf<QuePosition::VECCALC> tmp2;
GlobalTensor<TYPE_START> startGm;
GlobalTensor<TYPE_END> endGm;
GlobalTensor<TYPE_WEIGHT> weightGm;
GlobalTensor<TYPE_Y> yGm;
uint64_t coreDataNum;
uint64_t tileNum;
uint64_t ubPartDataNum;
uint64_t tailDataNum;
uint64_t processDataNum;
};
}
#endif
