* 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 cosh_common_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/math/cosh/cosh_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/math/cosh.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_COSH_COSH_COMMON_IMPL_H__
#endif
#ifndef IMPL_MATH_COSH_COSH_COMMON_IMPL_H
#define IMPL_MATH_COSH_COSH_COMMON_IMPL_H
#include "kernel_tensor.h"
#include "kernel_basic_intf.h"
#include "kernel_pop_stack_buffer.h"
#include "../../common/check.h"
#ifdef ASCENDC_CPU_DEBUG
#include "../../api_check/kernel_check/math/cosh/cosh_check.h"
#endif
#include "../../api_check/kernel_api_check.h"
#if defined(__NPU_ARCH__) && __NPU_ARCH__ == 2201
#include "cosh_v220_impl.h"
#elif defined(__NPU_ARCH__) && __NPU_ARCH__ == 2002
#include "cosh_v200_impl.h"
#endif
namespace AscendC {
constexpr float SCALAR_LN2 = -0.69314718055994530941723212145818;
constexpr float SCALAR_BROAD_CAST = 0.25;
const uint8_t COSH_HALF_CALC_PROCEDURE = 6;
const uint8_t COSH_FLOAT_CALC_PROCEDURE = 2;
template <typename T>
__aicore__ inline void CoshCompute(
const LocalTensor<T>& dst, const LocalTensor<T>& src, const LocalTensor<T>& tmpBuffer, uint32_t calSize)
{
UnaryRepeatParams unaryParams;
BinaryRepeatParams binaryParams;
LocalTensor<T> tmpBuffer2 = tmpBuffer[calSize];
Adds<T, false>(tmpBuffer2, src, static_cast<T>(SCALAR_LN2), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Exp<T, false>(tmpBuffer, tmpBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Duplicate<T, false>(dst, static_cast<T>(SCALAR_BROAD_CAST), MASK_PLACEHOLDER, 1, 1, DEFAULT_REPEAT_STRIDE);
PipeBarrier<PIPE_V>();
Div<T, false>(tmpBuffer2, dst, tmpBuffer, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Add<T, false>(dst, tmpBuffer, tmpBuffer2, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
}
template <>
__aicore__ inline void CoshCompute<half>(
const LocalTensor<half>& dst, const LocalTensor<half>& src, const LocalTensor<half>& tmpBuffer, uint32_t calSize)
{
UnaryRepeatParams unaryParams;
BinaryRepeatParams binaryParams;
const LocalTensor<float>& tmpFloatBuffer1 = tmpBuffer.ReinterpretCast<float>();
const LocalTensor<float>& tmpFloatBuffer2 = tmpFloatBuffer1[calSize];
const LocalTensor<float>& tmpFloatBuffer3 = tmpFloatBuffer2[calSize];
Cast<float, half, false>(
tmpFloatBuffer3, src, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
Adds<float, false>(tmpFloatBuffer2, tmpFloatBuffer3, SCALAR_LN2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Exp<float, false>(tmpFloatBuffer1, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Duplicate<float, false>(tmpFloatBuffer3, SCALAR_BROAD_CAST, MASK_PLACEHOLDER, 1, 1, DEFAULT_REPEAT_STRIDE);
PipeBarrier<PIPE_V>();
Div<float, false>(tmpFloatBuffer2, tmpFloatBuffer3, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Add<float, false>(tmpFloatBuffer3, tmpFloatBuffer1, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
CoshCast(dst, tmpFloatBuffer3);
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void CoshImpl(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const LocalTensor<uint8_t>& sharedTmpBuffer,
const uint32_t calCount)
{
if ASCEND_IS_AIC {
return;
}
CHECK_FUNC_HIGHLEVEL_API(Cosh, (T, isReuseSource), (dstTensor, srcTensor, sharedTmpBuffer, calCount));
const uint32_t bufferSize = sharedTmpBuffer.GetSize();
const uint32_t tmpBufferSize = bufferSize / sizeof(T);
CheckTmpBufferSize(tmpBufferSize, 0, bufferSize);
LocalTensor<T> tmpBuffer = sharedTmpBuffer.ReinterpretCast<T>();
uint32_t calSize = 0;
if constexpr (sizeof(T) == sizeof(half)) {
calSize = tmpBufferSize / COSH_HALF_CALC_PROCEDURE / ONE_BLK_SIZE * ONE_BLK_SIZE;
} else {
calSize = tmpBufferSize / COSH_FLOAT_CALC_PROCEDURE / ONE_BLK_SIZE * ONE_BLK_SIZE;
}
CheckTmpBufferSize(calSize, 0, bufferSize);
const uint32_t round = calCount / calSize;
const uint32_t tail = calCount % calSize;
SetMaskCount();
SetVectorMask<T, MaskMode::COUNTER>(0, calSize);
uint32_t offset = 0;
for (uint32_t i = 0; i < round; i++) {
CoshCompute(dstTensor[offset], srcTensor[offset], tmpBuffer, calSize);
offset = offset + calSize;
}
if (tail != 0) {
SetVectorMask<T, MaskMode::COUNTER>(0, tail);
CoshCompute(dstTensor[offset], srcTensor[offset], tmpBuffer, calSize);
}
SetMaskNorm();
ResetMask();
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void CoshImpl(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const uint32_t calCount)
{
if ASCEND_IS_AIC {
return;
}
LocalTensor<uint8_t> sharedTmpBuffer;
bool ans = PopStackBuffer<uint8_t, TPosition::LCM>(sharedTmpBuffer);
ASCENDC_ASSERT((ans), { KERNEL_LOG(KERNEL_ERROR, "PopStackBuffer Error!"); });
CoshImpl<T, isReuseSource>(dstTensor, srcTensor, sharedTmpBuffer, calCount);
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_COSH_COSH_COMMON_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_COSH_COSH_COMMON_IMPL_H__
#endif