* 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 axpy_3510_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/math/axpy/axpy_3510_impl.h is an internal header file and must not be used directly. Functions or variables defined in this file may be removed in the future. Please use \"#include \"adv_api/math/axpy.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_AXPY_AXPY_C310_IMPL_H__
#endif
#ifndef IMPL_MATH_AXPY_AXPY_C310_IMPL_H
#define IMPL_MATH_AXPY_AXPY_C310_IMPL_H
#include "kernel_tensor.h"
#include "kernel_basic_intf.h"
#include "../../common/check.h"
#ifdef ASCENDC_CPU_DEBUG
#include "../../api_check/kernel_check/math/axpy/axpy_check.h"
#endif
#include "../../api_check/kernel_api_check.h"
namespace AscendC {
namespace AxpyAPI {
constexpr Reg::CastTrait castTraitF162F32 = {
Reg::RegLayout::ZERO, Reg::SatMode::UNKNOWN, Reg::MaskMergeMode::ZEROING, RoundMode::UNKNOWN};
template <typename T, typename U>
__simd_vf__ inline void AxpyCompute(
__ubuf__ T* dst, __ubuf__ U* src, U scalarValue, uint32_t calCount, uint16_t repeatTimes, uint16_t oneRepSize,
uint32_t mainBlockCount, uint32_t tailCount, uint16_t offset, uint16_t singleMainBlockCtrl)
{
Reg::MaskReg mask, maskTail;
Reg::RegTensor<T> dstVreg;
Reg::RegTensor<U> srcVreg;
mask = Reg::UpdateMask<T>(mainBlockCount);
maskTail = Reg::UpdateMask<T>(tailCount);
if constexpr (IsSameType<U, half>::value && IsSameType<T, float>::value) {
Reg::RegTensor<float> tempSrcVreg;
for (uint16_t i = 0; i < repeatTimes; ++i) {
Reg::LoadAlign<U, Reg::LoadDist::DIST_UNPACK_B16>(srcVreg, src + i * oneRepSize);
Reg::Cast<float, U, castTraitF162F32>(tempSrcVreg, srcVreg, mask);
Reg::LoadAlign(dstVreg, dst + i * oneRepSize);
Reg::Axpy(dstVreg, tempSrcVreg, scalarValue, mask);
Reg::StoreAlign(dst + i * oneRepSize, dstVreg, mask);
Reg::LoadAlign<U, Reg::LoadDist::DIST_UNPACK_B16>(srcVreg, src + i * oneRepSize + offset);
Reg::Cast<float, U, castTraitF162F32>(tempSrcVreg, srcVreg, mask);
Reg::LoadAlign(dstVreg, dst + i * oneRepSize + offset);
Reg::Axpy(dstVreg, tempSrcVreg, scalarValue, mask);
Reg::StoreAlign(dst + i * oneRepSize + offset, dstVreg, mask);
}
for (uint16_t j = 0; j < singleMainBlockCtrl; ++j) {
Reg::LoadAlign<U, Reg::LoadDist::DIST_UNPACK_B16>(srcVreg, src + repeatTimes * oneRepSize * 2);
Reg::Cast<float, U, castTraitF162F32>(tempSrcVreg, srcVreg, mask);
Reg::LoadAlign(dstVreg, dst + repeatTimes * oneRepSize * 2);
Reg::Axpy(dstVreg, tempSrcVreg, scalarValue, mask);
Reg::StoreAlign(dst + repeatTimes * oneRepSize * 2, dstVreg, mask);
}
Reg::LoadAlign<U, Reg::LoadDist::DIST_UNPACK_B16>(
srcVreg, src + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize);
Reg::Cast<float, U, castTraitF162F32>(tempSrcVreg, srcVreg, maskTail);
Reg::LoadAlign(dstVreg, dst + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize);
Reg::Axpy(dstVreg, tempSrcVreg, scalarValue, maskTail);
Reg::StoreAlign(dst + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize, dstVreg, maskTail);
} else {
for (uint16_t i = 0; i < repeatTimes; ++i) {
Reg::LoadAlign(srcVreg, src + i * oneRepSize);
Reg::LoadAlign(dstVreg, dst + i * oneRepSize);
Reg::Axpy(dstVreg, srcVreg, scalarValue, mask);
Reg::StoreAlign(dst + i * oneRepSize, dstVreg, mask);
Reg::LoadAlign(srcVreg, src + i * oneRepSize + offset);
Reg::LoadAlign(dstVreg, dst + i * oneRepSize + offset);
Reg::Axpy(dstVreg, srcVreg, scalarValue, mask);
Reg::StoreAlign(dst + i * oneRepSize + offset, dstVreg, mask);
}
for (uint16_t j = 0; j < singleMainBlockCtrl; ++j) {
Reg::LoadAlign(srcVreg, src + repeatTimes * oneRepSize * 2);
Reg::LoadAlign(dstVreg, dst + repeatTimes * oneRepSize * 2);
Reg::Axpy(dstVreg, srcVreg, scalarValue, mask);
Reg::StoreAlign(dst + repeatTimes * oneRepSize * 2, dstVreg, mask);
}
Reg::LoadAlign(srcVreg, src + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize);
Reg::LoadAlign(dstVreg, dst + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize);
Reg::Axpy(dstVreg, srcVreg, scalarValue, maskTail);
Reg::StoreAlign(dst + repeatTimes * oneRepSize * 2 + singleMainBlockCtrl * oneRepSize, dstVreg, maskTail);
}
}
}
template <typename T, typename U, bool isReuseSource>
__aicore__ inline void AxpyImpl(
const LocalTensor<T>& dstLocal, const LocalTensor<U>& srcLocal, const U scalarValue,
const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t calCount)
{
CHECK_FUNC_HIGHLEVEL_API(Axpy, (T, U, isReuseSource), (dstLocal, srcLocal, scalarValue, sharedTmpBuffer, calCount));
CheckTensorPosition(sharedTmpBuffer, "sharedTmpBuffer", "VECIN, VECOUT, VECCALC");
CheckTensorPosition(dstLocal, "dstLocal", "VECIN, VECOUT, VECCALC");
CheckTensorPosition(srcLocal, "srcLocal", "VECIN, VECOUT, VECCALC");
CheckCalCount(calCount, "calCount", dstLocal, "dstLocal", "Axpy");
CheckCalCount(calCount, "calCount", srcLocal, "srcLocal", "Axpy");
static_assert(SupportType<T, half, float>(), "Axpy current dst data type is not supported on current device!");
static_assert(SupportType<U, half, float>(), "Axpy current src data type is not supported on current device!");
__ubuf__ T* dst = (__ubuf__ T*)dstLocal.GetPhyAddr();
__ubuf__ U* src = (__ubuf__ U*)srcLocal.GetPhyAddr();
constexpr uint16_t oneRepSize = GetVecLen() / sizeof(T);
const uint32_t mainBlockCount = oneRepSize;
uint32_t tailCount = calCount % oneRepSize;
uint16_t repeatTimes = calCount / oneRepSize;
if (tailCount == 0 && repeatTimes > 0) {
repeatTimes--;
tailCount += oneRepSize;
}
uint16_t repeatTimesUnRoll = repeatTimes / 2;
uint16_t singleMainBlockCtrl = repeatTimes % 2;
uint16_t offset = repeatTimesUnRoll * oneRepSize;
AxpyAPI::AxpyCompute<T, U>(
dst, src, scalarValue, calCount, repeatTimesUnRoll, oneRepSize, mainBlockCount, tailCount, offset,
singleMainBlockCtrl);
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_AXPY_AXPY_C310_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_AXPY_AXPY_C310_IMPL_H__
#endif
