* 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 tan_common_impl.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/math/tan/tan_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/tan.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_TAN_TAN_COMMON_IMPL_H__
#endif
#ifndef IMPL_MATH_TAN_TAN_COMMON_IMPL_H
#define IMPL_MATH_TAN_TAN_COMMON_IMPL_H
#include "kernel_basic_intf.h"
#include "kernel_tensor.h"
#include "kernel_pop_stack_buffer.h"
#include "../../common/check.h"
#ifdef ASCENDC_CPU_DEBUG
#include "../../api_check/kernel_check/math/tan/tan_check.h"
#endif
#include "../../api_check/kernel_api_check.h"
#if defined(__NPU_ARCH__) && __NPU_ARCH__ == 2201
#include "tan_v220_impl.h"
#elif defined(__NPU_ARCH__) && __NPU_ARCH__ == 2002
#include "tan_v200_impl.h"
#endif
namespace AscendC {
constexpr uint32_t TAN_HALF_CALC_PROCEDURE = 10;
constexpr uint32_t TAN_FLOAT_CALC_PROCEDURE = 4;
constexpr float PI_FOR_X_TODIV = 0.3183098733425140380859375;
constexpr float KPI_FIRS_PI_MULS = 0.0009670257568359375;
constexpr float PI_V2 = 3.140625;
constexpr float PI_DOWN = 1.57079637050628662109375;
constexpr float PI_DOWN_NEG = -1.57079637050628662109375;
constexpr float KPI_TWI_PI_MULS = 6.2771141529083251953125e-7;
constexpr float PI_RESDOWN_ADDS = 0.00000004371139000189375;
constexpr float PI_RESDOWN_ADDS_NEG = -0.00000004371139000189375;
constexpr float KPI_THIR_PI_MULS = 1.21644916362129151821136474609375e-10;
constexpr float KPI_FOR_PI_MULS = -1.0291767438275201129727065563201904296875e-13;
constexpr float TAN_RES_MULTI_SCA = 0.0698520831551998762793;
constexpr float TAN_RES_ADDICT_UP = -6.8711573651634203789;
constexpr float TAN_2ADDS = 61.20362572811089435388;
constexpr float TAN_3ADDS = -24.8048928861126769186219;
__aicore__ inline void KPI_0(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(dstTensor, roundTensor, PI_V2, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(srcTensor, srcTensor, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
}
__aicore__ inline void KPI_1(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(dstTensor, roundTensor, KPI_FIRS_PI_MULS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(srcTensor, srcTensor, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(resTensor1, srcTensor, PI_DOWN, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(resTensor2, srcTensor, PI_DOWN_NEG, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
}
__aicore__ inline void KPI_2(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(dstTensor, roundTensor, KPI_TWI_PI_MULS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(srcTensor, srcTensor, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor1, resTensor1, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor2, resTensor2, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(resTensor1, resTensor1, PI_RESDOWN_ADDS_NEG, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(resTensor2, resTensor2, PI_RESDOWN_ADDS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
}
__aicore__ inline void KPI_3(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(dstTensor, roundTensor, KPI_THIR_PI_MULS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(srcTensor, srcTensor, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor1, resTensor1, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor2, resTensor2, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
}
__aicore__ inline void KPI_4(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(dstTensor, roundTensor, KPI_FOR_PI_MULS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(srcTensor, srcTensor, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor1, resTensor1, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Sub<float, false>(resTensor2, resTensor2, dstTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
}
__aicore__ inline void TanRound(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
k=round(x/�), x0=x-k�, x0�?-�/2, �/2)
�=�_0+�_1+�_2+�_3+�_4 achieve final precision compensation.
Final solution锛?
k = round(x * invpi)
x -= k * pi_0
x -= k * pi_1
down1 = x + pio2_high // pi/2 + x
down2 = x - pio2_high // x - pi/2
x -= k * pi_2
down1 -= k * pi_2
down2 -= k * pi_2
x -= k * pi_3
down1 -= k * pi_3
down2 -= k * pi_3
x -= k * pi_4
down1 -= k * pi_4
down2 -= k * pi_4
*/
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Muls<float, false>(roundTensor, srcTensor, PI_FOR_X_TODIV, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
TanCast(roundTensor, roundTensor, RoundMode::CAST_RINT);
KPI_0(dstTensor, srcTensor, roundTensor);
KPI_1(dstTensor, srcTensor, roundTensor, resTensor1, resTensor2);
KPI_2(dstTensor, srcTensor, roundTensor, resTensor1, resTensor2);
KPI_3(dstTensor, srcTensor, roundTensor, resTensor1, resTensor2);
KPI_4(dstTensor, srcTensor, roundTensor, resTensor1, resTensor2);
}
__aicore__ inline void TanPolynomialApproximation(
const LocalTensor<float>& dstTensor, const LocalTensor<float>& srcTensor, const LocalTensor<float>& roundTensor,
const LocalTensor<float>& resTensor1, const LocalTensor<float>& resTensor2)
{
tan(x) = xP(x) / ((�/2 - x)(�/2 + x)Q(x))
P(x) = (x^2 * R0 + R1) * x^2 + R2
Q(x) = x^2 * R3
R0 = 0.0698520831551998762793
R1 = -6.8711573651634203789
R2 = 61.20362572811089435388
R3 = -24.8048928861126769186219
*/
const UnaryRepeatParams unaryParams;
const BinaryRepeatParams binParams;
Mul<float, false>(roundTensor, srcTensor, srcTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Muls<float, false>(dstTensor, roundTensor, TAN_RES_MULTI_SCA, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(dstTensor, dstTensor, TAN_RES_ADDICT_UP, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(dstTensor, dstTensor, roundTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(dstTensor, dstTensor, TAN_2ADDS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(dstTensor, dstTensor, srcTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Adds<float, false>(roundTensor, roundTensor, TAN_3ADDS, MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(roundTensor, roundTensor, resTensor1, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Mul<float, false>(roundTensor, roundTensor, resTensor2, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
Div<float, false>(dstTensor, dstTensor, roundTensor, MASK_PLACEHOLDER, 1, binParams);
PipeBarrier<PIPE_V>();
}
template <typename T>
__aicore__ inline void TanCompute(
const LocalTensor<T>& dstTensor, const LocalTensor<T>& srcTensor, const LocalTensor<uint8_t>& tmpBuffer,
uint32_t calSize)
{
const UnaryRepeatParams unaryParams;
const LocalTensor<T>& tmpTensor1 = tmpBuffer.ReinterpretCast<float>();
const LocalTensor<T>& tmpTensor2 = tmpTensor1[calSize];
const LocalTensor<T>& tmpTensor3 = tmpTensor2[calSize];
const LocalTensor<T>& tmpTensor4 = tmpTensor3[calSize];
Adds<T, false>(tmpTensor4, srcTensor, static_cast<float>(0.0), MASK_PLACEHOLDER, 1, unaryParams);
PipeBarrier<PIPE_V>();
TanRound(dstTensor, tmpTensor4, tmpTensor1, tmpTensor2, tmpTensor3);
TanPolynomialApproximation(dstTensor, tmpTensor4, tmpTensor1, tmpTensor2, tmpTensor3);
}
template <>
__aicore__ inline void TanCompute(
const LocalTensor<half>& dstTensor, const LocalTensor<half>& srcTensor, const LocalTensor<uint8_t>& tmpBuffer,
uint32_t calSize)
{
const LocalTensor<float>& tempTensorConv = tmpBuffer.ReinterpretCast<float>();
const LocalTensor<float>& tmpTensor1 = tempTensorConv[calSize];
const LocalTensor<float>& tmpTensor2 = tmpTensor1[calSize];
const LocalTensor<float>& tmpTensor3 = tmpTensor2[calSize];
const LocalTensor<float>& tmpTensor4 = tmpTensor3[calSize];
Cast<float, half, false>(
tmpTensor1, srcTensor, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, DEFAULT_REPEAT_STRIDE, HALF_DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
TanRound(tempTensorConv, tmpTensor1, tmpTensor2, tmpTensor3, tmpTensor4);
TanPolynomialApproximation(tempTensorConv, tmpTensor1, tmpTensor2, tmpTensor3, tmpTensor4);
Cast<half, float, false>(
dstTensor, tempTensorConv, RoundMode::CAST_NONE, MASK_PLACEHOLDER, 1,
{1, 1, HALF_DEFAULT_REPEAT_STRIDE, DEFAULT_REPEAT_STRIDE});
PipeBarrier<PIPE_V>();
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void TanImpl(
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(Tan, (T, isReuseSource), (dstTensor, srcTensor, sharedTmpBuffer, calCount));
uint32_t tmpBufferSize = sharedTmpBuffer.GetSize();
uint32_t splitCount = tmpBufferSize / sizeof(T);
if constexpr (sizeof(T) == sizeof(half)) {
splitCount = splitCount / TAN_HALF_CALC_PROCEDURE / ONE_BLK_SIZE * ONE_BLK_SIZE;
} else {
splitCount = splitCount / TAN_FLOAT_CALC_PROCEDURE / ONE_BLK_SIZE * ONE_BLK_SIZE;
}
CheckTmpBufferSize(splitCount, 0, tmpBufferSize);
const uint32_t loopCount = calCount / splitCount;
const uint32_t calcTail = calCount % splitCount;
SetMaskCount();
SetVectorMask<T, MaskMode::COUNTER>(0, splitCount);
uint32_t offset = 0;
for (uint32_t i = 0; i < loopCount; ++i) {
TanCompute(dstTensor[i * splitCount], srcTensor[i * splitCount], sharedTmpBuffer, 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, MaskMode::COUNTER>(0, calcTail);
TanCompute(dstTensor[loopCount * splitCount], srcTensor[loopCount * splitCount], sharedTmpBuffer, tailCount);
}
SetMaskNorm();
ResetMask();
}
template <typename T, bool isReuseSource = false>
__aicore__ inline void TanImpl(
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!"); });
TanImpl<T, isReuseSource>(dstTensor, srcTensor, sharedTmpBuffer, calCount);
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_TAN_TAN_COMMON_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_MATH_TAN_TAN_COMMON_IMPL_H__
#endif