* 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.
*/
#ifndef LIB_NORMALIZATION_RMSNORM_H
#define LIB_NORMALIZATION_RMSNORM_H
#include "kernel_tensor.h"
#include "kernel_basic_intf.h"
#include "kernel_pop_stack_buffer.h"
#include "kernel_tiling/kernel_tiling.h"
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 2201 || __NPU_ARCH__ == 2002)
#include "../../../impl/adv_api/detail/normalization/rmsnorm/rmsnorm_common_impl.h"
#elif defined(__NPU_ARCH__) && (__NPU_ARCH__ == 3101 || __NPU_ARCH__ == 3003 || __NPU_ARCH__ == 3113)
#include "../../../impl/adv_api/detail/normalization/rmsnorm/rmsnorm_c310_impl.h"
#endif
#if defined(__NPU_ARCH__) && (__NPU_ARCH__ == 2201 || __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 3101 || \
__NPU_ARCH__ == 3003 || __NPU_ARCH__ == 3113)
namespace AscendC {
#pragma begin_pipe(V)
* \brief Root Mean Square Layer Normalization, for more info see https://arxiv.org/abs/1910.07467
*
* \note support data type: half and float
*
* \param [in] isBasicBlock: indicate whether enable basic block
* \param [out] dstLocal: output LocalTensor
* \param [in] srcLocal: input LocalTensor
* \param [in] gammaLocal: input gamma LocalTensor
* \param [in] sharedTmpBuffer: input temporary localTensor
* \param [in] epsilon: epslion value
* \param [in] tiling: tiling for RmsNorm interface
*/
template <typename T, bool isBasicBlock = false>
__aicore__ inline void RmsNorm(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const LocalTensor<T>& gammaLocal, const LocalTensor<uint8_t>& sharedTmpBuffer, const T epsilon,
const RmsNormTiling& tiling)
{
if ASCEND_IS_AIC {
return;
}
ASCENDC_ASSERT((IsSameType<T, half>::value || IsSameType<T, float>::value),
{ KERNEL_LOG(KERNEL_ERROR, "RmsNorm only support data type: float/half"); });
RmsNormAPI::RmsNormImpl<T, isBasicBlock>(dstLocal, srcLocal, gammaLocal, sharedTmpBuffer, epsilon, tiling);
}
* \brief Root Mean Square Layer Normalization, for more info see https://arxiv.org/abs/1910.07467
*
* \note support data type: half and float. This interface will allocate a tmp buffer.
* user should make sure that memory on UB is big enough for input tiling.
*
* \param [in] isBasicBlock: indicate whether enable basic block
* \param [out] dstLocal: output LocalTensor
* \param [in] srcLocal: input LocalTensor
* \param [in] gammaLocal: input gamma LocalTensor
* \param [in] epsilon: epslion value
* \param [in] tiling: tiling for RmsNorm interface
*/
template <typename T, bool isBasicBlock = false>
__aicore__ inline void RmsNorm(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal,
const LocalTensor<T>& gammaLocal, const T epsilon, const RmsNormTiling& tiling)
{
if ASCEND_IS_AIC {
return;
}
LocalTensor<uint8_t> stackBufer;
bool ret = PopStackBuffer<uint8_t, TPosition::LCM>(stackBufer);
ASCENDC_ASSERT((ret), { KERNEL_LOG(KERNEL_ERROR, "RmsNorm failed to apply for tmp buffer!"); });
RmsNorm<T, isBasicBlock>(dstLocal, srcLocal, gammaLocal, stackBufer, epsilon, tiling);
}
#pragma end_pipe
}
#endif
#endif
