* 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 fma.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_FMA_H__
#endif
#ifndef LIB_MATH_FMA_H
#define LIB_MATH_FMA_H
#include "kernel_tensor.h"
#include "include/adv_api/math/fma_utils.h"
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102)
#include "../../../impl/adv_api/detail/math/fma/fma_common_impl.h"
#endif
namespace AscendC {
#pragma begin_pipe(V)
* \ingroup Fma
* \brief compute Fma elementwisely
* \tparam T: half/float
* \param [out] dst: output LocalTensor
* \param [in] src0: input LocalTensor
* \param [in] src1: input LocalTensor
* \param [in] src2: input LocalTensor
* \param [in] sharedTmpBuffer: extra temporary shared space used for intermediate values among calculation process,
* whose required space size should refer to corresponding tiling API, which is defined at fma_tiling.h.
* Generally, the more space you allocate, the better performance you will achieve, and the performance
* reaches peak when buffer size is maximum(calculated by tiling function). Moreover, it is not guaranteed
* that the shared space will be cleared after usage, the data could be anything.
* \param [in] count: the number of elements to be processed.
* \note dst/src0/src1/src2 Tensor must be 32B aligned, and it doesn't allow dst/src0/src1/src2/sharedTmpBuffer tensor address overlap.
*/
template <const FmaConfig& config = DEFAULT_FMA_CONFIG, typename T>
__aicore__ inline void Fma(const LocalTensor<T>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, const LocalTensor<T>& src2, const LocalTensor<uint8_t>& sharedTmpBuffer,
const uint32_t count)
{
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102)
FmaImpl<config, T>(dst, src0, src1, src2, sharedTmpBuffer, count);
#endif
}
* \ingroup Fma
* \brief compute Fma elementwisely
* \tparam T: half/float
* \param [out] dst: output LocalTensor
* \param [in] src0: input LocalTensor
* \param [in] src1: input LocalTensor
* \param [in] src2: input LocalTensor
* \param [in] count: the number of elements to be processed.
* \note dst/src0/src1/src2 Tensor must be 32B aligned, and it doesn't allow dst/src0/src1/src2 tensor address overlap.
*/
template <const FmaConfig& config = DEFAULT_FMA_CONFIG, typename T>
__aicore__ inline void Fma(const LocalTensor<T>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, const LocalTensor<T>& src2, const uint32_t count)
{
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102)
FmaImpl<config, T>(dst, src0, src1, src2, count);
#endif
}
#pragma end_pipe
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_FMA_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_FMA_H__
#endif
