* 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.
*/
* \file tensor_move.h
* \brief
*/
#ifndef TENSOR_MOVE_H_
#define TENSOR_MOVE_H_
#include "kernel_operator.h"
namespace TensorMove {
using namespace AscendC;
constexpr int64_t DB_BUFFER = 2;
template <typename T>
class TensorMoveKernel {
public:
__aicore__ inline TensorMoveKernel(){};
__aicore__ inline void Init(GM_ADDR x, GM_ADDR y, GM_ADDR workspace, const TensorMoveTilingData &tilingData);
__aicore__ inline void Process();
private:
__aicore__ inline void ParseTilingData(const TensorMoveTilingData &tilingData);
__aicore__ inline void CopyIn(int64_t offset, int64_t dataLen);
__aicore__ inline void CopyOut(int64_t offset, int64_t dataLen);
private:
TPipe pipe_;
TQueBind<QuePosition::VECIN, QuePosition::VECOUT, DB_BUFFER> dataQueue_;
GlobalTensor<T> xGm_;
GlobalTensor<T> yGm_;
int64_t blockIdx_ = 0;
int64_t blockOffset_ = 0;
int64_t totalCoreNum_;
int64_t usedCoreNum_;
int64_t ubFactor_;
int64_t tailBlockTailUbFactor_;
int64_t blockFactor_;
int64_t tailBlockFactor_;
int64_t bufferSize_;
};
template <typename T>
__aicore__ inline void TensorMoveKernel<T>::Init(
GM_ADDR x, GM_ADDR y, GM_ADDR workspace, const TensorMoveTilingData &tilingData)
{
blockIdx_ = GetBlockIdx();
ParseTilingData(tilingData);
blockOffset_ = GetBlockIdx() * blockFactor_ * ubFactor_;
xGm_.SetGlobalBuffer((__gm__ T *)(x) + blockOffset_);
yGm_.SetGlobalBuffer((__gm__ T *)(y) + blockOffset_);
bufferSize_ = ubFactor_ * sizeof(T);
pipe_.InitBuffer(dataQueue_, DB_BUFFER, bufferSize_);
}
template <typename T>
__aicore__ inline void TensorMoveKernel<T>::ParseTilingData(const TensorMoveTilingData &tilingData)
{
totalCoreNum_ = tilingData.totalCoreNum;
usedCoreNum_ = tilingData.usedCoreNum;
ubFactor_ = tilingData.ubFactor;
tailBlockTailUbFactor_ = tilingData.tailBlockTailUbFactor;
blockFactor_ = tilingData.blockFactor;
tailBlockFactor_ = tilingData.tailBlockFactor;
}
template <typename T>
__aicore__ inline void TensorMoveKernel<T>::CopyIn(int64_t offset, int64_t dataLen)
{
DataCopyExtParams inParams = {
static_cast<uint16_t>(1),
static_cast<uint32_t>(dataLen * sizeof(T)),
static_cast<uint32_t>(0),
static_cast<uint32_t>(0),
static_cast<uint32_t>(0)
};
DataCopyPadExtParams<T> padParams = { false, static_cast<uint8_t>(0), static_cast<uint8_t>(0), static_cast<T>(0) };
LocalTensor<T> xLocal = dataQueue_.AllocTensor<T>();
DataCopyPad(xLocal, xGm_[offset], inParams, padParams);
dataQueue_.EnQue(xLocal);
}
template <typename T>
__aicore__ inline void TensorMoveKernel<T>::CopyOut(int64_t offset, int64_t dataLen)
{
DataCopyExtParams outParams = {
static_cast<uint16_t>(1),
static_cast<uint32_t>(dataLen * sizeof(T)),
static_cast<uint32_t>(0),
static_cast<uint32_t>(0),
static_cast<uint32_t>(0)
};
LocalTensor<T> yLocal = dataQueue_.DeQue<T>();
DataCopyPad(yGm_[offset], yLocal, outParams);
dataQueue_.FreeTensor(yLocal);
}
template <typename T>
__aicore__ inline void TensorMoveKernel<T>::Process()
{
if (blockIdx_ >= usedCoreNum_) {
return;
}
int64_t loopSize = blockFactor_;
if (blockIdx_ == usedCoreNum_ - 1) {
loopSize = tailBlockFactor_;
}
int64_t offset = 0;
for (int64_t idx = 0; idx < loopSize - 1; idx++) {
offset = idx * ubFactor_;
CopyIn(offset, ubFactor_);
CopyOut(offset, ubFactor_);
}
offset = (loopSize - 1) * ubFactor_;
int64_t dataLen = ubFactor_;
if (blockIdx_ == usedCoreNum_ - 1) {
dataLen = tailBlockTailUbFactor_;
}
CopyIn(offset, dataLen);
CopyOut(offset, dataLen);
}
}
#endif