* 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 gelu.h
* \brief
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_GELU_H__
#endif
#ifndef LIB_GELU_GELU_H
#define LIB_GELU_GELU_H
#include "kernel_tensor.h"
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102 || __NPU_ARCH__ == 3003 || \
__NPU_ARCH__ == 3113)
#include "../../../impl/adv_api/detail/activation/gelu/gelu_3510_impl.h"
#else
#include "../../../impl/adv_api/detail/activation/gelu/gelu_impl.h"
#endif
namespace AscendC {
#pragma begin_pipe(V)
* \brief Gelu(x) = x / (1 + e ^ (-1.59576912 * (x + 0.044715 * x ^ 3)))
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] sharedTmpBuffer, input local temporary Tensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void Gelu(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
GeluImpl<T, highPrecision, highPerformance>(dstLocal, srcLocal, sharedTmpBuffer, dataSize);
}
* \ingroup Gelu
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void Gelu(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal, const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
GeluImpl<T, highPrecision, highPerformance>(dstLocal, srcLocal, dataSize);
}
* \brief FasterGelu(x) = x / (1 + e ^ (-1.702 * x))
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] sharedTmpBuffer, input local temporary Tensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void FasterGelu(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102 || __NPU_ARCH__ == 3003 || \
__NPU_ARCH__ == 3113)
ASCENDC_ASSERT((IsSameType<T, half>::value || IsSameType<T, float>::value),
{ KERNEL_LOG(KERNEL_ERROR, "FasterGelu only support data type: float/half"); });
#endif
FasterGeluImpl<T, highPrecision, highPerformance>(dstLocal, srcLocal, sharedTmpBuffer, dataSize);
}
* \ingroup FasterGelu
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] sharedTmpBuffer, input local temporary Tensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void FasterGelu(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3510 || __NPU_ARCH__ == 5102 || __NPU_ARCH__ == 3003 || \
__NPU_ARCH__ == 3113)
ASCENDC_ASSERT((IsSameType<T, half>::value || IsSameType<T, float>::value),
{ KERNEL_LOG(KERNEL_ERROR, "FasterGelu only support data type: float/half"); });
#endif
FasterGeluImpl<T, highPrecision, highPerformance>(dstLocal, srcLocal, dataSize);
}
* \brief sgn(x) = (x + 0.000000000001) / |(x + 0.000000000001)|
* \brief fast_gelu_v2(x) = x * (sgn(x) * [(-0.1444) * (clip(|0.7071 * x|, max=1.769) - 1.769) ^ 2 + 0.5] + 0.5)
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] sharedTmpBuffer, input local temporary Tensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void FasterGeluV2(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const LocalTensor<uint8_t>& sharedTmpBuffer, const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
ASCENDC_ASSERT((IsSameType<T, half>::value || IsSameType<T, float>::value),
{ KERNEL_LOG(KERNEL_ERROR, "FasterGeluV2 only support data type: float/half"); });
FasterGeluV2Impl<T, highPrecision, highPerformance>(dstLocal, srcLocal, sharedTmpBuffer, dataSize);
}
* \ingroup FasterGeluV2
* \param [out] dstLocal, output LocalTensor
* \param [in] srcLocal, input LocalTensor
* \param [in] sharedTmpBuffer, input local temporary Tensor
* \param [in] dataSize, number of input data elements
* \param [in] highPrecision, whether to enable the high-precision interface to improve the calculation accuracy
* \param [in] highPerformance, whether to enable the high-performance interface to improve the computing efficiency
*/
template <typename T, bool highPrecision = false, bool highPerformance = false>
__aicore__ inline void FasterGeluV2(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const uint32_t dataSize)
{
if ASCEND_IS_AIC {
return;
}
ASCENDC_ASSERT((IsSameType<T, half>::value || IsSameType<T, float>::value),
{ KERNEL_LOG(KERNEL_ERROR, "FasterGeluV2 only support data type: float/half"); });
FasterGeluV2Impl<T, highPrecision, highPerformance>(dstLocal, srcLocal, dataSize);
}
#pragma end_pipe
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_GELU_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_GELU_H__
#endif
