* 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.
*/
#if !defined(__ASCENDC_INCLUDE_INTERNAL_HEADERS__)
#pragma message( \
"impl/adv_api/detail/select/selectwithbytesmask/selectwithbytesmask_3510_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/select/selectwithbytesmask.h\"\" and use public functions or variables defined in interface headers files.")
#define __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#define __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SELECT_SELECTWITHBYTESMASK_SELECTWITHBYTESMASK_C310_IMPL_H__
#endif
#ifndef LIB_SELECT_SELECT_WITH_BYTES_MASK_C310_IMPL_H
#define LIB_SELECT_SELECT_WITH_BYTES_MASK_C310_IMPL_H
#include "kernel_tensor.h"
#include "kernel_basic_intf.h"
#include "kernel_utils.h"
namespace AscendC {
template <typename T, typename U, CMPMODE cmpMode>
__simd_callee__ inline void RegTensorToMaskReg(
Reg::RegTensor<U>& vMaskReg0, Reg::RegTensor<U>& vMaskReg1, Reg::MaskReg& localMask0, Reg::MaskReg& maskReg0)
{
Reg::MaskReg maskReg1;
Reg::MaskReg localMask1;
if constexpr (sizeof(U) == 1) {
Reg::CompareScalar<uint8_t, cmpMode>(
localMask0, (Reg::RegTensor<uint8_t>&)vMaskReg0, static_cast<uint8_t>(0), maskReg0);
} else if constexpr (sizeof(T) == 2 && sizeof(U) == 4) {
Reg::MaskUnPack(maskReg1, maskReg0);
Reg::CompareScalar<U, cmpMode>(localMask0, vMaskReg0, static_cast<U>(0), maskReg1);
Reg::CompareScalar<U, cmpMode>(localMask1, vMaskReg1, static_cast<U>(0), maskReg1);
Reg::MaskDeInterleave<T>(localMask0, localMask1, localMask0, localMask1);
} else {
Reg::CompareScalar<U, cmpMode>(localMask0, vMaskReg0, static_cast<U>(0), maskReg0);
}
}
template <typename T, typename U, bool reverse = false>
__simd_vf__ inline void SelectWithBytesMaskPerAxisImpl(
__ubuf__ T* dstUb, __ubuf__ T* src0Ub, T src1, __ubuf__ U* maskUb, const uint32_t firstAxis,
const uint32_t srcLastAxis, const uint32_t maskLastAxis)
{
Reg::RegTensor<T> vSrcReg0;
Reg::RegTensor<T> vSrcReg1;
Reg::RegTensor<T> vDstReg;
Reg::RegTensor<U> vMaskReg0;
Reg::RegTensor<U> vMaskReg1;
Reg::MaskReg maskReg0;
Reg::MaskReg localMask0;
Reg::Duplicate(vSrcReg1, src1);
uint32_t sreg;
uint32_t sregLower = static_cast<uint32_t>(GetVecLen() / sizeof(T));
uint16_t repeatTimes = static_cast<uint16_t>(DivCeil(srcLastAxis, sregLower));
for (uint16_t loopH = 0; loopH < static_cast<uint16_t>(firstAxis); ++loopH) {
sreg = srcLastAxis;
for (uint16_t i = 0; i < repeatTimes; ++i) {
maskReg0 = Reg::UpdateMask<T>(sreg);
Reg::LoadAlign<T>(vSrcReg0, src0Ub + loopH * srcLastAxis + i * sregLower);
if constexpr (sizeof(T) == 2 && sizeof(U) == 1) {
Reg::LoadAlign<uint8_t, Reg::LoadDist::DIST_UNPACK_B8>(
(Reg::RegTensor<uint8_t>&)vMaskReg0,
(__ubuf__ uint8_t*)maskUb + loopH * maskLastAxis + i * sregLower);
} else if constexpr (sizeof(T) == 2 && sizeof(U) == 4) {
Reg::LoadAlign<U>(vMaskReg0, maskUb + loopH * maskLastAxis + i * sregLower);
Reg::LoadAlign<U>(vMaskReg1, maskUb + loopH * maskLastAxis + i * sregLower + sregLower / 2);
} else if constexpr (sizeof(T) == 4 && sizeof(U) == 1) {
Reg::LoadAlign<uint8_t, Reg::LoadDist::DIST_UNPACK4_B8>(
(Reg::RegTensor<uint8_t>&)vMaskReg0,
(__ubuf__ uint8_t*)maskUb + loopH * maskLastAxis + i * sregLower);
} else if constexpr (sizeof(T) == 4 && sizeof(U) == 2) {
Reg::LoadAlign<U, Reg::LoadDist::DIST_UNPACK_B16>(
vMaskReg0, maskUb + loopH * maskLastAxis + i * sregLower);
} else if constexpr (sizeof(T) == sizeof(U)) {
Reg::LoadAlign<U>(vMaskReg0, maskUb + loopH * maskLastAxis + i * sregLower);
}
if constexpr (!reverse) {
RegTensorToMaskReg<T, U, CMPMODE::EQ>(vMaskReg0, vMaskReg1, localMask0, maskReg0);
} else {
RegTensorToMaskReg<T, U, CMPMODE::NE>(vMaskReg0, vMaskReg1, localMask0, maskReg0);
}
Reg::Select(vDstReg, vSrcReg0, vSrcReg1, localMask0);
Reg::StoreAlign<T>(dstUb + loopH * srcLastAxis + i * sregLower, vDstReg, maskReg0);
}
}
}
template <typename T, typename U, bool reverse = false>
__aicore__ inline void SelectWithBytesMaskProcess(
const LocalTensor<T>& dst, const LocalTensor<T>& src0, T src1, const LocalTensor<U>& mask,
const SelectWithBytesMaskShapeInfo& info)
{
__ubuf__ T* src0Ub = (__ubuf__ T*)src0.GetPhyAddr();
__ubuf__ T* dstUb = (__ubuf__ T*)dst.GetPhyAddr();
__ubuf__ U* maskUb = (__ubuf__ U*)mask.GetPhyAddr();
const uint32_t firstAxis = static_cast<uint32_t>(info.firstAxis);
const uint32_t srcLastAxis = static_cast<uint32_t>(info.srcLastAxis);
const uint32_t maskLastAxis = static_cast<uint32_t>(info.maskLastAxis);
SelectWithBytesMaskPerAxisImpl<T, U, reverse>(dstUb, src0Ub, src1, maskUb, firstAxis, srcLastAxis, maskLastAxis);
}
template <typename T, typename U, bool isReuseMask, bool reverse = false>
__aicore__ inline __inout_pipe__(V) void SelectWithBytesMaskImpl(
const LocalTensor<T>& dst, const LocalTensor<T>& src0, T src1, const LocalTensor<U>& mask,
const LocalTensor<uint8_t>& sharedTmpBuffer, const SelectWithBytesMaskShapeInfo& info)
{
if ASCEND_IS_AIC {
return;
}
static_assert(SupportType<T, float, half>(), "Select do not support this type on current device");
static_assert(
SupportType<U, bool, uint8_t, int8_t, uint16_t, int16_t, uint32_t, int32_t>(),
"Select do not support this type on current device");
CheckTensorPos<T>(dst, Hardware::UB, "dst", "VECIN / VECCALC / VECOUT", "Select");
CheckTensorPos<T>(src0, Hardware::UB, "src", "VECIN / VECCALC / VECOUT", "Select");
CheckTensorPos<U>(mask, Hardware::UB, "mask", "VECIN / VECCALC / VECOUT", "Select");
CheckTensorPos<uint8_t>(sharedTmpBuffer, Hardware::UB, "sharedTmpBuffer", "VECIN / VECCALC / VECOUT", "Select");
ASCENDC_ASSERT((info.srcLastAxis * sizeof(T) % ONE_BLK_SIZE == 0), {
KERNEL_LOG(KERNEL_ERROR, "srcLastAxis should be 32B aligned, current srcLastAxis is %u", info.srcLastAxis);
});
ASCENDC_ASSERT((info.maskLastAxis * sizeof(U) % ONE_BLK_SIZE == 0), {
KERNEL_LOG(KERNEL_ERROR, "maskLastAxis should be 32B aligned, current maskLastAxis is %u", info.maskLastAxis);
});
ASCENDC_ASSERT((info.maskLastAxis % BLOCK_CUBE == 0), {
KERNEL_LOG(
KERNEL_ERROR, "maskLastAxis should be multiples of 16, current maskLastAxis is %u", info.maskLastAxis);
});
const uint32_t firstAxis = info.firstAxis;
const uint32_t srcLastAxis = info.srcLastAxis;
const uint32_t maskLastAxis = info.maskLastAxis;
const uint32_t srcSize = src0.GetSize();
ASCENDC_ASSERT((srcSize == firstAxis * srcLastAxis), {
KERNEL_LOG(KERNEL_ERROR, "ShapeInfo must match with src Tensor size.");
});
ASCENDC_ASSERT((mask.GetSize() == firstAxis * maskLastAxis), {
KERNEL_LOG(KERNEL_ERROR, "ShapeInfo must match with mask Tensor size.");
});
ASCENDC_ASSERT((maskLastAxis >= srcLastAxis), {
KERNEL_LOG(KERNEL_ERROR, "maskLastAxis must be greater than or equal to srcLastAxis.");
});
SelectWithBytesMaskProcess<T, U, reverse>(dst, src0, src1, mask, info);
}
template <typename T, typename U, bool isReuseMask = true>
__aicore__ inline void SelectWithBytesMask(
const LocalTensor<T>& dst, const LocalTensor<T>& src0, T src1, const LocalTensor<U>& mask,
const LocalTensor<uint8_t>& sharedTmpBuffer, const SelectWithBytesMaskShapeInfo& info)
{
SelectWithBytesMaskImpl<T, U, isReuseMask, false>(dst, src0, src1, mask, sharedTmpBuffer, info);
}
template <typename T, typename U, bool isReuseMask = true>
__aicore__ inline void SelectWithBytesMask(
const LocalTensor<T>& dst, T src0, const LocalTensor<T>& src1, const LocalTensor<U>& mask,
const LocalTensor<uint8_t>& sharedTmpBuffer, const SelectWithBytesMaskShapeInfo& info)
{
SelectWithBytesMaskImpl<T, U, isReuseMask, true>(dst, src1, src0, mask, sharedTmpBuffer, info);
}
}
#endif
#if defined(__UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SELECT_SELECTWITHBYTESMASK_SELECTWITHBYTESMASK_C310_IMPL_H__)
#undef __ASCENDC_INCLUDE_INTERNAL_HEADERS__
#undef __UNDEF_ASCENDC_INCLUDE_INTERNAL_HEADERS_SELECT_SELECTWITHBYTESMASK_SELECTWITHBYTESMASK_C310_IMPL_H__
#endif
