* 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 sum_3510_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/reduce/sum/sum_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/reduce/sum.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REDUCE_SUM_SUM_C310_IMPL_H__
#endif
#ifndef IMPL_REDUCE_SUM_C310_IMPL_H
#define IMPL_REDUCE_SUM_C310_IMPL_H
#include "kernel_basic_intf.h"
#include "kernel_tensor.h"
#include "include/adv_api/reduce/sum_utils.h"
namespace AscendC {
namespace SumInternal {
template <typename T>
__simd_vf__ inline void SumForOneRepeatTime(
__ubuf__ T* dstUb, __ubuf__ T* srcUb, const SumParams sumParams, uint32_t count, uint32_t offset)
{
uint32_t calCount;
Reg::MaskReg mask;
Reg::UnalignReg uregOut;
Reg::RegTensor<T> srcReg, dstReg;
for (int i = 0; i < sumParams.outter; i++) {
calCount = count;
mask = Reg::UpdateMask<T>(calCount);
Reg::LoadAlign(srcReg, srcUb + i * offset);
Reg::ReduceSum(dstReg, srcReg, mask);
Reg::StoreUnAlign<T, Reg::PostLiteral::POST_MODE_UPDATE>(dstUb, dstReg, uregOut, 1);
}
Reg::StoreUnAlignPost(dstUb, uregOut, 0);
}
template <typename T, bool isFirstRepeat>
__simd_vf__ inline void ReduceSumNextN(
__ubuf__ T* dstUb, __ubuf__ T* srcUb, const SumParams sumParams, uint32_t calCount, uint32_t repeatTimes,
uint32_t offset)
{
uint32_t count;
Reg::MaskReg mask;
Reg::UnalignReg uregIn;
Reg::RegTensor<T> srcReg, dstReg;
constexpr int32_t eleCountPerVL = GetVecLen() / sizeof(T);
for (uint16_t i = 0; i < sumParams.outter; i++) {
count = calCount;
auto dstTmpUb = dstUb + i * offset;
for (uint16_t j = 0; j < repeatTimes; j++) {
mask = Reg::UpdateMask<T>(count);
if constexpr (isFirstRepeat) {
Reg::LoadAlign(srcReg, srcUb + i * sumParams.inner + j * eleCountPerVL);
} else {
Reg::LoadAlign(srcReg, srcUb + i * offset + j * eleCountPerVL);
}
Reg::ReduceSum(dstReg, srcReg, mask);
Reg::StoreUnAlign<T, Reg::PostLiteral::POST_MODE_UPDATE>(dstTmpUb, dstReg, uregIn, 1);
}
Reg::StoreUnAlignPost(dstTmpUb, uregIn, 0);
}
Reg::LocalMemBar<Reg::MemType::VEC_STORE, Reg::MemType::VEC_LOAD>();
}
}
template <typename T, int32_t reduceDim = -1, bool isReuseSource = false, bool isBasicBlock = false>
__aicore__ inline void SumCheckParams(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const LocalTensor<uint8_t>& sharedTmpBuffer,
const SumParams& sumParams)
{
static_assert(SupportType<T, half, float>(), "current data type is not supported on current device!");
CheckTensorPos<T>(dstTensor, Hardware::UB, "dstTensor", "VECIN / VECCALC / VECOUT", "Sum");
CheckTensorPos<T>(srcTensor, Hardware::UB, "srcTensor", "VECIN / VECCALC / VECOUT", "Sum");
CheckTensorPos<uint8_t>(sharedTmpBuffer, Hardware::UB, "sharedTmpBuffer", "VECIN / VECCALC / VECOUT", "Sum");
constexpr uint32_t sumInnerAlignLen = 32;
ASCENDC_ASSERT((1 <= sumParams.n) && (sumParams.n <= sumParams.inner), {
KERNEL_LOG(KERNEL_ERROR, "The value of n must be greater than or equal to 1 and less than or equal to inner.");
});
ASCENDC_ASSERT((sumParams.inner * sizeof(T) % sumInnerAlignLen == 0), {
KERNEL_LOG(KERNEL_ERROR, "The value of inner * sizeof(T) must be an integer multiple of 32.");
});
}
template <typename T, int32_t reduceDim = -1, bool isReuseSource = false, bool isBasicBlock = false>
__aicore__ inline void SumCompute(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const LocalTensor<uint8_t>& sharedTmpBuffer,
const SumParams& sumParams)
{
if ASCEND_IS_AIC {
return;
}
SumCheckParams<T, reduceDim, isReuseSource, isBasicBlock>(dstTensor, srcTensor, sharedTmpBuffer, sumParams);
__ubuf__ T* dstUb = (__ubuf__ T*)dstTensor.GetPhyAddr();
__ubuf__ T* srcUb = (__ubuf__ T*)srcTensor.GetPhyAddr();
__ubuf__ T* sharedTmpBufferUb = (__ubuf__ T*)sharedTmpBuffer.GetPhyAddr();
constexpr int32_t eleCountPerVL = GetVecLen() / sizeof(T);
uint16_t repeatTimes = CeilDivision(sumParams.n, eleCountPerVL);
uint32_t loopRepeatTimes;
uint32_t calCount = sumParams.n;
uint32_t totalCnt = 1;
uint32_t dataSize = repeatTimes;
uint32_t offset = AlignUp(CeilDivision(sumParams.inner, eleCountPerVL), 32);
while (dataSize > 1) {
++totalCnt;
dataSize = CeilDivision(dataSize, eleCountPerVL);
}
if (repeatTimes == 1) {
SumInternal::SumForOneRepeatTime<T>(dstUb, srcUb, sumParams, sumParams.n, sumParams.inner);
return;
}
SumInternal::ReduceSumNextN<T, true>(sharedTmpBufferUb, srcUb, sumParams, calCount, repeatTimes, offset);
--totalCnt;
loopRepeatTimes = repeatTimes;
while (totalCnt != 0) {
calCount = loopRepeatTimes;
loopRepeatTimes = CeilDivision(loopRepeatTimes, eleCountPerVL);
if (totalCnt == 1) {
SumInternal::SumForOneRepeatTime<T>(dstUb, sharedTmpBufferUb, sumParams, calCount, offset);
} else {
SumInternal::ReduceSumNextN<T, false>(
sharedTmpBufferUb, sharedTmpBufferUb, sumParams, calCount, loopRepeatTimes, offset);
}
--totalCnt;
}
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REDUCE_SUM_SUM_C310_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REDUCE_SUM_SUM_C310_IMPL_H__
#endif
