* 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 dynamic_quant_unalign_310p.h
* \brief
*/
#ifndef DYNAMIC_QUANT_UNALIGN_310P_H
#define DYNAMIC_QUANT_UNALIGN_310P_H
#include "kernel_operator.h"
#include "dynamic_quant_align_310p.h"
namespace DynamicQuantNDOpt {
#if __CCE_AICORE__ < 220
class DynamicQuantUnalign310P : public DynamicQuantAlign310p
{
public:
__aicore__ inline DynamicQuantUnalign310P()
{}
__aicore__ inline void Init(
GM_ADDR x, GM_ADDR y, GM_ADDR scale, GM_ADDR offset, const DynamicQuantTilingData* __restrict tilingData)
{
DynamicQuantAlign310p::InitParams(tilingData, offset);
DynamicQuantAlign310p::InitBuffer(x, y, scale, offset);
DynamicQuantAlign310p::InitQueue();
}
__aicore__ inline void Process()
{
DynamicQuantAlign310p::ProcessWithHeadDoublePipLineForAlign();
ProcessWithTailCopyInForUnalign();
DynamicQuantAlign310p::ProcessWithHeadLastComputeAndCopyOutForAlign();
ProcessWithTailComputeAndCopyOutForUnalign();
}
__aicore__ inline void CopyOutUnalign(uint64_t progress, uint32_t calCount)
{
AscendC::LocalTensor<int8_t> outLocal = outQueue_.DeQue<int8_t>();
AscendC::LocalTensor<float> scaleLocal = scaleQueue_.DeQue<float>();
AscendC::LocalTensor<float> offsetLocal;
if (asymmetric_) {
offsetLocal = offsetQueue_.DeQue<float>();
}
DataCopy(outGm_[progress * lenCopyRow_], outLocal, calCount);
uint32_t realNumRowAlign = (calCount / sizeH_ + DYNAMIC_QUANT_ALIGN_NUM_SCALE - 1) /
DYNAMIC_QUANT_ALIGN_NUM_SCALE * DYNAMIC_QUANT_ALIGN_NUM_SCALE;
DataCopy(scaleGm_[progress * numCopyRow_], scaleLocal, realNumRowAlign);
if (asymmetric_) {
DataCopy(offsetGm_[progress * numCopyRow_], offsetLocal, realNumRowAlign);
offsetQueue_.FreeTensor(offsetLocal);
}
outQueue_.FreeTensor(outLocal);
scaleQueue_.FreeTensor(scaleLocal);
}
__aicore__ inline void ProcessWithTailCopyInForUnalign()
{
if (isTailLoop_) {
DynamicQuantAlign310p::CopyIn(loopCount_, lenLastTail_);
}
}
__aicore__ inline void ProcessWithTailComputeAndCopyOutForUnalign()
{
if (isTailLoop_) {
if (asymmetric_) {
DynamicQuantAlign310p::ComputeAsymmetric(numLastTailRow_, sizeH_);
} else {
DynamicQuantAlign310p::Compute(numLastTailRow_, sizeH_);
}
CopyOutUnalign(loopCount_, lenLastTail_);
}
}
};
#endif
}
#endif