* 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 asinh_common_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/math/asinh/asinh_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/asinh.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ASINH_ASINH_COMMON_IMPL_H__
#endif
#ifndef IMPL_MATH_ASINH_ASINH_COMMON_IMPL_H
#define IMPL_MATH_ASINH_ASINH_COMMON_IMPL_H
#include "kernel_tensor.h"
#include "kernel_basic_intf.h"
#include "kernel_pop_stack_buffer.h"
#include "kernel_tiling/kernel_tiling.h"
#include "../../common/check.h"
#ifdef ASCENDC_CPU_DEBUG
#include "../../api_check/kernel_check/math/asinh/asinh_check.h"
#endif
#include "../../api_check/kernel_api_check.h"
namespace AscendC {
constexpr uint32_t ASINH_HALF_CALC_PROC = 3;
constexpr uint32_t ASINH_FLOAT_CALC_PROC = 3;
constexpr float ASINH_ONE = 1;
constexpr float ASINH_FP16_MAX = 32768;
constexpr float ASINH_FP16_MIN = 3.0517578125e-05;
constexpr float ASINH_FP32_MAX = 4611686018427387904;
constexpr float ASINH_FP32_MIN = 2.168404344971009e-19;
constexpr uint32_t ASINH_STRIDE_DIGITS = 2;
template <typename T, bool isReuseSource = false>
__aicore__ inline void AsinhImpl(const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor)
{
if ASCEND_IS_AIC {
return;
}
LocalTensor<uint8_t> sharedTmpBuffer;
bool ret = PopStackBuffer<uint8_t, TPosition::LCM>(sharedTmpBuffer);
ASCENDC_ASSERT((ret), { KERNEL_LOG(KERNEL_ERROR, "PopStackBuffer Error!"); });
AsinhImpl<T, isReuseSource>(dstTensor, srcTensor, sharedTmpBuffer, srcTensor.GetSize());
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void AsinhImpl(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const uint32_t calCount)
{
if ASCEND_IS_AIC {
return;
}
LocalTensor<uint8_t> sharedTmpBuffer;
bool ret = PopStackBuffer<uint8_t, TPosition::LCM>(sharedTmpBuffer);
ASCENDC_ASSERT((ret), { KERNEL_LOG(KERNEL_ERROR, "PopStackBuffer Error!"); });
AsinhImpl<T, isReuseSource>(dstTensor, srcTensor, sharedTmpBuffer, calCount);
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void AsinhImpl(
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(Asinh, (T, isReuseSource), (dstTensor, srcTensor, sharedTmpBuffer, calCount));
uint32_t tmpBufferSize = sharedTmpBuffer.GetSize();
uint32_t splitCount = tmpBufferSize / sizeof(float);
LocalTensor<float> tmpBuffer = sharedTmpBuffer.ReinterpretCast<float>();
tmpBuffer.SetSize(sharedTmpBuffer.GetSize() / sizeof(float));
if constexpr (sizeof(T) == sizeof(half)) {
splitCount = splitCount / ASINH_HALF_CALC_PROC / ONE_BLK_SIZE * ONE_BLK_SIZE;
;
} else {
splitCount = splitCount / ASINH_FLOAT_CALC_PROC / ONE_BLK_SIZE * ONE_BLK_SIZE;
;
}
CheckTmpBufferSize(splitCount, 0, tmpBufferSize);
uint32_t loopCount = calCount / splitCount;
uint32_t calcTail = calCount % splitCount;
SetMaskCount();
SetVectorMask<T>(0, splitCount);
for (uint32_t i = 0; i < loopCount; ++i) {
AsinhCompute(dstTensor[i * splitCount], srcTensor[i * splitCount], tmpBuffer, splitCount);
}
if (calcTail > 0) {
uint32_t tailCount = calcTail / ONE_BLK_SIZE * ONE_BLK_SIZE;
tailCount = (calcTail % ONE_BLK_SIZE == 0) ? tailCount : (tailCount + ONE_BLK_SIZE);
SetVectorMask<T>(0, calcTail);
AsinhCompute(dstTensor[loopCount * splitCount], srcTensor[loopCount * splitCount], tmpBuffer, tailCount);
}
SetMaskNorm();
ResetMask();
}
template <typename T>
__aicore__ inline void AsinhGetSign(
const LocalTensor<T>& dst, const LocalTensor<T>& src, const LocalTensor<T>& denominator)
{
UnaryRepeatParams unaryParams;
BinaryRepeatParams binaryParams;
constexpr float kFpMax = sizeof(T) == sizeof(float) ? ASINH_FP32_MAX : ASINH_FP16_MAX;
constexpr float kFpMin = sizeof(T) == sizeof(float) ? ASINH_FP32_MIN : ASINH_FP16_MIN;
Muls<T, false>(dst, src, static_cast<T>(kFpMax), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Abs<T, false>(denominator, dst, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Adds<T, false>(denominator, denominator, static_cast<T>(kFpMin), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Div<T, false>(dst, dst, denominator, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
}
template <typename T>
__aicore__ inline void AsinhCompute(
const LocalTensor<T>& dst, const LocalTensor<T>& src, const LocalTensor<float>& tmpBuffer, uint32_t calCount)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binaryParams;
LocalTensor<float> tmpFloatBuffer1 = tmpBuffer;
LocalTensor<float> tmpFloatBuffer3 = tmpFloatBuffer1[calCount];
LocalTensor<float> tmpFloatBuffer2 = tmpFloatBuffer1[calCount * 2];
Abs<T, false>(tmpFloatBuffer1, src, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<T, false>(tmpFloatBuffer2, tmpFloatBuffer1, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Adds<T, false>(tmpFloatBuffer2, tmpFloatBuffer2, static_cast<T>(ASINH_ONE), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sqrt<T, false>(tmpFloatBuffer2, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Add<T, false>(tmpFloatBuffer2, tmpFloatBuffer2, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Ln<T, false>(tmpFloatBuffer2, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
AsinhGetSign(tmpFloatBuffer1, src, tmpFloatBuffer3);
PipeBarrier<PIPE_V>();
Mul<T, false>(dst, tmpFloatBuffer2, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
}
template <>
__aicore__ inline void AsinhCompute(
const LocalTensor<half>& dst, const LocalTensor<half>& src, const LocalTensor<float>& tmpBuffer, uint32_t calCount)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binaryParams;
LocalTensor<float> tmpFloatBuffer1 = tmpBuffer;
LocalTensor<float> tmpFloatBuffer3 = tmpFloatBuffer1[calCount];
LocalTensor<float> tmpFloatBuffer2 = tmpFloatBuffer1[calCount * 2];
Cast<float, half, false>(
tmpFloatBuffer1, src, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
Abs<float, false>(tmpFloatBuffer1, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(tmpFloatBuffer2, tmpFloatBuffer1, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(
tmpFloatBuffer2, tmpFloatBuffer2, static_cast<half>(ASINH_ONE), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sqrt<float, false>(tmpFloatBuffer2, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Add<float, false>(tmpFloatBuffer2, tmpFloatBuffer2, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Ln<float, false>(tmpFloatBuffer2, tmpFloatBuffer2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Cast<float, half, false>(
tmpFloatBuffer1, src, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
AsinhGetSign(tmpFloatBuffer1, tmpFloatBuffer1, tmpFloatBuffer3);
PipeBarrier<PIPE_V>();
Mul<float, false>(tmpFloatBuffer2, tmpFloatBuffer2, tmpFloatBuffer1, MASK_PLACEHOLDER, 1, binaryParams);
PipeBarrier<PIPE_V>();
Cast<half, float, false>(
dst, tmpFloatBuffer2, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE / ASINH_STRIDE_DIGITS, DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ASINH_ASINH_COMMON_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_ASINH_ASINH_COMMON_IMPL_H__
#endif