* 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 fill.h
* \brief
*/
#ifndef __FILL_H__
#define __FILL_H__
#include "kernel_operator.h"
#include "kernel_tiling/kernel_tiling.h"
#include "fill_tiling_data.h"
#include "fill_tiling_key.h"
namespace NsFill {
using namespace AscendC;
constexpr uint32_t BUFFER_NUM = 1;
template <typename T, bool IsExistBigCore>
class KernelFill {
public:
__aicore__ inline KernelFill() {
}
__aicore__ inline void Init(GM_ADDR dims, GM_ADDR value, GM_ADDR y, uint32_t smallCoreDataNum,
uint32_t bigCoreDataNum, uint32_t bigCoreLoopNum, uint32_t smallCoreLoopNum,
uint32_t ubPartDataNum, uint32_t smallCoreTailDataNum, uint32_t bigCoreTailDataNum,
uint32_t tailBlockNum) {
ASSERT(AscendC::GetBlockNum() != 0 && "block dim can not be zero!");
uint32_t blockIdx = AscendC::GetBlockIdx();
uint32_t globalBufferIndex = bigCoreDataNum * AscendC::GetBlockIdx();
this->ubPartDataNum = ubPartDataNum;
if constexpr (IsExistBigCore) {
if (blockIdx < tailBlockNum) {
this->coreDataNum = bigCoreDataNum;
this->tileNum = bigCoreLoopNum;
this->tailDataNum = bigCoreTailDataNum;
} else {
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex -= (bigCoreDataNum - smallCoreDataNum) * (AscendC::GetBlockIdx() - tailBlockNum);
}
} else {
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex = smallCoreDataNum * AscendC::GetBlockIdx();
}
this->value = *reinterpret_cast<__gm__ DTYPE_VALUE *>(value);
yGm.SetGlobalBuffer((__gm__ DTYPE_Y *)y + globalBufferIndex, this->coreDataNum);
if constexpr (std::is_same_v<T, int8_t> || std::is_same_v<T, bool>) {
pipe.InitBuffer(tmpBuf, this->ubPartDataNum * sizeof(half));
}
pipe.InitBuffer(outQueueOUT, BUFFER_NUM, this->ubPartDataNum * sizeof(DTYPE_Y));
}
__aicore__ inline void Process() {
int32_t loopCount = this->tileNum;
this->processDataNum = this->ubPartDataNum;
for (int32_t i = 0; i < loopCount - 1; i++) {
Compute(i);
CopyOut(i);
}
this->processDataNum = this->tailDataNum;
Compute(loopCount - 1);
CopyOut(loopCount - 1);
}
private:
__aicore__ inline void Compute(uint32_t progress) {
if constexpr (std::is_same_v<T, int8_t> || std::is_same_v<T, bool>) {
AscendC::LocalTensor<int8_t> outLocal = outQueueOUT.AllocTensor<int8_t>();
AscendC::LocalTensor<half> tmpLocal = tmpBuf.Get<half>();
AscendC::Duplicate<half>(tmpLocal, (half)(this->value), this->processDataNum);
Cast(outLocal, tmpLocal, AscendC::RoundMode::CAST_NONE, this->processDataNum);
outQueueOUT.EnQue<int8_t>(outLocal);
} else {
AscendC::LocalTensor<T> outLocal = outQueueOUT.AllocTensor<T>();
AscendC::Duplicate<T>(outLocal, this->value, this->processDataNum);
outQueueOUT.EnQue<T>(outLocal);
}
}
__aicore__ inline void CopyOut(uint32_t progress) {
AscendC::LocalTensor<DTYPE_VALUE> outLocal = outQueueOUT.DeQue<DTYPE_VALUE>();
AscendC::DataCopy(yGm[progress * this->ubPartDataNum], outLocal, this->processDataNum);
outQueueOUT.FreeTensor(outLocal);
}
private:
AscendC::TPipe pipe;
AscendC::TQue<AscendC::QuePosition::VECOUT, BUFFER_NUM> outQueueOUT;
AscendC::GlobalTensor<DTYPE_Y> yGm;
AscendC::TBuf<AscendC::TPosition::VECCALC> tmpBuf;
T value;
uint32_t coreDataNum = 0;
uint32_t tileNum = 0;
uint32_t ubPartDataNum = 0;
uint32_t tailDataNum = 0;
uint32_t processDataNum = 0;
};
template <bool IsExistBigCore>
class KernelFill1_INT64 {
public:
__aicore__ inline KernelFill1_INT64() {
}
__aicore__ inline void Init(GM_ADDR dims, GM_ADDR values, GM_ADDR y, uint32_t smallCoreDataNum,
uint32_t bigCoreDataNum, uint32_t bigCoreLoopNum, uint32_t smallCoreLoopNum,
uint32_t ubPartDataNum, uint32_t smallCoreTailDataNum, uint32_t bigCoreTailDataNum,
uint32_t tailBlockNum) {
ASSERT(AscendC::GetBlockNum() != 0 && "block dim can not be zero!");
uint32_t blockIdx = AscendC::GetBlockIdx();
uint32_t globalBufferIndex = bigCoreDataNum * AscendC::GetBlockIdx();
this->ubPartDataNum = ubPartDataNum;
if constexpr (IsExistBigCore) {
if (blockIdx < tailBlockNum) {
this->coreDataNum = bigCoreDataNum;
this->tileNum = bigCoreLoopNum;
this->tailDataNum = bigCoreTailDataNum;
} else {
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex -= (bigCoreDataNum - smallCoreDataNum) * (AscendC::GetBlockIdx() - tailBlockNum);
}
} else {
this->coreDataNum = smallCoreDataNum;
this->tileNum = smallCoreLoopNum;
this->tailDataNum = smallCoreTailDataNum;
globalBufferIndex = smallCoreDataNum * AscendC::GetBlockIdx();
}
xGm.SetGlobalBuffer((__gm__ int32_t *)values, 2);
yGm.SetGlobalBuffer((__gm__ int32_t *)y + globalBufferIndex, this->coreDataNum);
this->high = xGm.GetValue(1);
this->low = xGm.GetValue(0);
pipe.InitBuffer(outQueueOUT, BUFFER_NUM, this->ubPartDataNum * sizeof(int32_t));
}
__aicore__ inline void Process() {
int32_t loopCount = this->tileNum;
this->processDataNum = this->ubPartDataNum;
this->repeatTimes = (this->processDataNum + 63) / 64;
for (int32_t i = 0; i < loopCount - 1; i++) {
Compute(i);
CopyOut(i);
}
this->processDataNum = this->tailDataNum;
this->repeatTimes = (this->processDataNum + 63) / 64;
Compute(loopCount - 1);
CopyOut(loopCount - 1);
}
private:
__aicore__ inline void Compute(uint32_t progress) {
AscendC::LocalTensor<int32_t> outLocal = outQueueOUT.AllocTensor<int32_t>();
uint64_t mask2[2] = {0xAAAAAAAAAAAAAAAA, 0};
uint64_t mask1[2] = {0x5555555555555555, 0};
AscendC::Duplicate(outLocal, low, mask1, this->repeatTimes, 1, 8);
AscendC::Duplicate(outLocal, high, mask2, this->repeatTimes, 1, 8);
outQueueOUT.EnQue<int32_t>(outLocal);
}
__aicore__ inline void CopyOut(uint32_t progress) {
AscendC::LocalTensor<int32_t> outLocal = outQueueOUT.DeQue<int32_t>();
AscendC::DataCopy(yGm[progress * this->ubPartDataNum], outLocal, this->processDataNum);
outQueueOUT.FreeTensor(outLocal);
}
private:
AscendC::TPipe pipe;
AscendC::TQue<AscendC::QuePosition::VECOUT, BUFFER_NUM> outQueueOUT;
AscendC::GlobalTensor<int32_t> yGm;
AscendC::GlobalTensor<int32_t> xGm;
int64_t value = 0;
uint32_t coreDataNum = 0;
uint32_t tileNum = 0;
uint32_t ubPartDataNum = 0;
uint32_t tailDataNum = 0;
uint32_t processDataNum = 0;
int32_t high = 0;
int32_t low = 0;
uint32_t repeatTimes = 0;
};
}
#endif
