* 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 softmaxflashv3.h
* \brief Softmaxflashv3 api of AscendC
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SOFTMAXFLASHV3_H__
#endif
#ifndef LIB_SOFTMAX_SOFTMAXFLASHV3_H
#define LIB_SOFTMAX_SOFTMAXFLASHV3_H
#include "kernel_tensor.h"
#include "kernel_tiling/kernel_tiling.h"
#include "include/adv_api/activation/softmax_utils.h"
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 2201 || __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 3510 || __NPU_ARCH__ == 3002 || __NPU_ARCH__ == 5102)
#include "../../../impl/adv_api/detail/activation/softmax/softmax_common.h"
#include "../../../impl/adv_api/detail/activation/softmax/softmax_flashv3_base_impl.h"
#endif
#pragma begin_pipe(V)
namespace AscendC {
* \ingroup SoftmaxFlashV3
* \brief softmax calculation in FlashAttention-2 process:
* if isUpdate = false:
* rowMeanLocal = rowsum(baseK) / baseK, rowMeanGlobal = rowsum(x), mean = rowMeanGlobal,
* meanTmp = rowMeanGlobal - rowMeanLocal, x = x - meanTmp * (a / 1 - a),
* maxTmp = rowmax(x), shiftCurr = (rowMeanGlobal - mean) * (a / 1 - a), max = maxTmp + shiftCurr,
* y = exp(x-max), sum = rowsum(y)
* if isUpdate = true:
* rowMeanLocal = rowsum(baseK) / baseK, rowMeanGlobal = rowsum(x), mean = (rowMeanGlobal + inmean * (n - 1)) / n,
* meanTmp = rowMeanGlobal - rowMeanLocal, x = x - meanTmp * (a / 1 - a),
* maxTmp = rowmax(x), shiftCurr = (rowMeanGlobal - mean) * (a / 1 - a), shiftPrev = (inmean - mean) * (a / 1 - a), max = max(maxTmp + shiftCurr, inmax + shiftPrev),
* y = exp(x-max), sum = rowsum(y)
* expmax = exp(inmax - max + shiftPrev), sum = expmax * insum + sum
* \note support data type: T half, U float
* \param [out] dstTensor: output y
* \param [out] meanTensor: output mean
* \param [out] expSumTensor: output sum
* \param [out] maxTensor: output max
* \param [out] expMaxTensor: output expmax, only valid when isUpdate is true
* \param [in] srcTensor: input x
* \param [in] inMeanTensor: input last inmean
* \param [in] inExpSumTensor: input last insum
* \param [in] inMaxTensor: input last inmax
* \param [in] softmaxShapeInfo: input src shape
* \param [in] tiling: input softmaxtiling
* \param [in] isUpdate: whether to update the sum and expmax
* \param [in] isReuseSource: reserved param
* \param [in] isBasicBlock: if src shape[m,k] satisfy the condition(m%8 == 0 && k%64 == 0), you can set true to improve
* performance, but it is a reserved param when isDataFormatNZ = true
* \param [in] isDataFormatNZ: if the data format of input srcTensor is NZ
*/
template <typename T, typename U, bool isUpdate = false, bool isReuseSource = false, bool isBasicBlock = false,
bool isDataFormatNZ = false, const SoftmaxConfig& config = SOFTMAX_DEFAULT_CFG>
__aicore__ inline void SoftmaxFlashV3(const LocalTensor<T>& dstTensor, const LocalTensor<U>& meanTensor,
const LocalTensor<U>& expSumTensor, const LocalTensor<U>& maxTensor, const LocalTensor<T>& srcTensor,
const LocalTensor<T>& expMaxTensor, const LocalTensor<U>& inMeanTensor, const LocalTensor<U>& inExpSumTensor,
const LocalTensor<U>& inMaxTensor, const SoftMaxTiling& tiling, const SoftMaxParams& params)
{
if ASCEND_IS_AIC {
return;
}
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 2201 || __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 3510 || __NPU_ARCH__ == 3002 || __NPU_ARCH__ == 5102)
TRACE_START(TraceId::SoftmaxFlashV3);
SoftmaxFlashV3Impl<T, U, isUpdate, isReuseSource, isBasicBlock, isDataFormatNZ, config>(dstTensor, meanTensor, expSumTensor,
maxTensor, srcTensor, expMaxTensor, inMeanTensor, inExpSumTensor, inMaxTensor, tiling, params);
TRACE_STOP(TraceId::SoftmaxFlashV3);
#endif
}
* \ingroup SoftmaxFlashV3
* \brief softmax calculation in FlashAttention-2 process:
* if isUpdate = false:
* rowMeanLocal = rowsum(baseK) / baseK, rowMeanGlobal = rowsum(x), mean = rowMeanGlobal,
* meanTmp = rowMeanGlobal - rowMeanLocal, x = x - meanTmp * (a / 1 - a),
* maxTmp = rowmax(x), shiftCurr = (rowMeanGlobal - mean) * (a / 1 - a), max = maxTmp + shiftCurr,
* y = exp(x-max), sum = rowsum(y)
* if isUpdate = true:
* rowMeanLocal = rowsum(baseK) / baseK, rowMeanGlobal = rowsum(x), mean = (rowMeanGlobal + inmean * (n - 1)) / n,
* meanTmp = rowMeanGlobal - rowMeanLocal, x = x - meanTmp * (a / 1 - a),
* maxTmp = rowmax(x), shiftCurr = (rowMeanGlobal - mean) * (a / 1 - a), shiftPrev = (inmean - mean) * (a / 1 - a), max = max(maxTmp + shiftCurr, inmax + shiftPrev),
* y = exp(x-max), sum = rowsum(y)
* expmax = exp(inmax - max + shiftPrev), sum = expmax * insum + sum
* \note support data type: T half, U float
* \param [out] dstTensor: output y
* \param [out] meanTensor: output mean
* \param [out] expSumTensor: output sum
* \param [out] maxTensor: output max
* \param [out] expMaxTensor: output expmax, only valid when isUpdate is true
* \param [in] srcTensor: input x
* \param [in] inMeanTensor: input last inmean
* \param [in] inExpSumTensor: input last insum
* \param [in] inMaxTensor: input last inmax
* \param [in] softmaxShapeInfo: input src shape
* \param [in] tiling: input softmaxtiling
* \param [in] isUpdate: whether to update the sum and expmax
* \param [in] isReuseSource: reserved param
* \param [in] isBasicBlock: if src shape[m,k] satisfy the condition(m%8 == 0 && k%64 == 0), you can set true to improve
* performance, but it is a reserved param when isDataFormatNZ = true
* \param [in] isDataFormatNZ: if the data format of input srcTensor is NZ
*/
template <typename T, typename U, bool isUpdate = false, bool isReuseSource = false, bool isBasicBlock = false,
bool isDataFormatNZ = false, const SoftmaxConfig& config = SOFTMAX_DEFAULT_CFG>
__aicore__ inline void SoftmaxFlashV3(const LocalTensor<T>& dstTensor, const LocalTensor<U>& meanTensor,
const LocalTensor<U>& expSumTensor, const LocalTensor<U>& maxTensor, const LocalTensor<T>& srcTensor,
const LocalTensor<T>& expMaxTensor, const LocalTensor<U>& inMeanTensor, const LocalTensor<U>& inExpSumTensor,
const LocalTensor<U>& inMaxTensor, const LocalTensor<uint8_t>& sharedTmpBuffer, const SoftMaxTiling& tiling,
const SoftMaxParams& params)
{
if ASCEND_IS_AIC {
return;
}
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 2201 || __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 3510 || __NPU_ARCH__ == 3002 || __NPU_ARCH__ == 5102)
TRACE_START(TraceId::SoftmaxFlashV3);
SoftmaxFlashV3Impl<T, U, isUpdate, isReuseSource, isBasicBlock, isDataFormatNZ, config>(dstTensor, meanTensor, expSumTensor,
maxTensor, srcTensor, expMaxTensor, inMeanTensor, inExpSumTensor, inMaxTensor, sharedTmpBuffer, tiling, params);
TRACE_STOP(TraceId::SoftmaxFlashV3);
#endif
}
}
#pragma end_pipe
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SOFTMAXFLASHV3_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SOFTMAXFLASHV3_H__
#endif
