* 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 reglu_common_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/activation/reglu/reglu_common_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/activation/reglu.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REGLU_COMMON_IMPL_H__
#endif
#ifndef IMPL_ACTIVATION_REGLU_REGLU_COMMON_IMPL_H
#define IMPL_ACTIVATION_REGLU_REGLU_COMMON_IMPL_H
#include "kernel_basic_intf.h"
#include "kernel_tensor.h"
#include "kernel_pop_stack_buffer.h"
#include "kernel_tiling/kernel_tiling.h"
#ifdef ASCENDC_CPU_DEBUG
#include "../../api_check/kernel_check/activation/reglu/reglu_check.h"
#endif
#include "../../api_check/kernel_api_check.h"
#if defined(__NPU_ARCH__) && __NPU_ARCH__ == 2201
#include "reglu_v220_impl.h"
#elif defined(__NPU_ARCH__) && __NPU_ARCH__ == 2002
#include "reglu_v200_impl.h"
#endif
namespace AscendC {
const uint8_t REGLU_HALF_CALC_PROCEDURE = 3;
const uint32_t REGLU_TEMP_BUFFER_OFFSET = 2U;
__aicore__ inline void Compute(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor0, const LocalTensor<float>& srcTensor1)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binaryParams;
Maxs<float, false>(dstTensor, srcTensor1, static_cast<float>(0), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(dstTensor, srcTensor0, dstTensor, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
}
template <typename T>
__aicore__ inline void ReGluCompute(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor0, const LocalTensor<T>& srcTensor1,
const LocalTensor<float>& tmpTensor, const uint32_t splitSize)
{
const LocalTensor<float>& x0CastBuffer = tmpTensor;
const LocalTensor<float>& x1CastBuffer = tmpTensor[splitSize];
const LocalTensor<float>& yCastBuffer = tmpTensor[splitSize * REGLU_TEMP_BUFFER_OFFSET];
Cast<float, T, false>(
x0CastBuffer, srcTensor0, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
Cast<float, T, false>(
x1CastBuffer, srcTensor1, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
Compute(yCastBuffer, x0CastBuffer, x1CastBuffer);
ReGluCast(dstTensor, yCastBuffer);
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void ReGluImpl(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor0, const LocalTensor<T>& srcTensor1,
const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t calCount)
{
CHECK_FUNC_HIGHLEVEL_API(ReGlu, (T, isReuseSource), (dstTensor, srcTensor0, srcTensor1, sharedTmpBuffer, calCount));
ASCENDC_ASSERT((calCount <= srcTensor0.GetSize()), { KERNEL_LOG(KERNEL_ERROR, "calCount must <= srcSize!"); });
ASCENDC_ASSERT(
((TPosition)dstTensor.GetPosition() == TPosition::VECIN ||
(TPosition)dstTensor.GetPosition() == TPosition::VECOUT ||
(TPosition)dstTensor.GetPosition() == TPosition::VECCALC),
{ KERNEL_LOG(KERNEL_ERROR, "dst position not support, just support position is VECIN, VECOUT, VECCALC."); });
SetMaskCount();
if constexpr (IsSameType<T, float>::value) {
SetVectorMask<T, MaskMode::COUNTER>(0, calCount);
Compute(dstTensor, srcTensor0, srcTensor1);
} else {
uint32_t tmpBufferSize = sharedTmpBuffer.GetSize() / sizeof(float);
ASCENDC_ASSERT((tmpBufferSize > 0), { KERNEL_LOG(KERNEL_ERROR, "tmpBufferSize must > 0!"); });
LocalTensor<float> tmpBuffer;
tmpBuffer = sharedTmpBuffer.ReinterpretCast<float>();
uint32_t stackSize = 0;
stackSize = tmpBufferSize / REGLU_HALF_CALC_PROCEDURE / ONE_BLK_SIZE * ONE_BLK_SIZE;
ASCENDC_ASSERT((stackSize > 0), { KERNEL_LOG(KERNEL_ERROR, "stackSize must > 0!"); });
const uint32_t round = calCount / stackSize;
const uint32_t tail = calCount % stackSize;
SetVectorMask<T, MaskMode::COUNTER>(0, stackSize);
uint32_t offset = 0;
for (uint32_t i = 0; i < round; i++) {
ReGluCompute(dstTensor[offset], srcTensor0[offset], srcTensor1[offset], tmpBuffer, stackSize);
offset = offset + stackSize;
}
if (tail != 0) {
SetVectorMask<T, MaskMode::COUNTER>(0, tail);
ReGluCompute(dstTensor[offset], srcTensor0[offset], srcTensor1[offset], tmpBuffer, stackSize);
}
}
SetMaskNorm();
ResetMask();
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void ReGluImpl(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor0, const LocalTensor<T>& srcTensor1,
const uint32_t calCount)
{
LocalTensor<uint8_t> sharedTmpBuffer;
bool hasStackBuffer = PopStackBuffer<uint8_t, TPosition::LCM>(sharedTmpBuffer);
ASCENDC_ASSERT((hasStackBuffer), { KERNEL_LOG(KERNEL_ERROR, "PopStackBuffer Error!"); });
ReGluImpl<T, isReuseSource>(dstTensor, srcTensor0, srcTensor1, sharedTmpBuffer, calCount);
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REGLU_COMMON_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_REGLU_COMMON_IMPL_H__
#endif