* 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 kernel_operator_vec_cmpsel_intf_impl.h
* \brief
*/
#ifndef ASCENDC_MODULE_OPERATOR_VEC_CMPSEL_INTERFACE_IMPL_H
#define ASCENDC_MODULE_OPERATOR_VEC_CMPSEL_INTERFACE_IMPL_H
#include "kernel_tensor.h"
#include "kernel_check.h"
#include "kernel_struct_binary.h"
#include "kernel_struct_unary.h"
#if __NPU_ARCH__ == 1001
#include "dav_c100/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 2002
#include "dav_m200/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 2201
#include "dav_c220/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 3002
#include "dav_m300/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 3102
#include "dav_m310/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 3101
#include "dav_c310/kernel_operator_vec_cmp_impl.h"
#include "dav_c310/kernel_operator_vec_sel_impl.h"
#elif (__NPU_ARCH__ == 5102)
#include "dav_m510/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 3003
#include "dav_l300/kernel_operator_vec_cmpsel_impl.h"
#elif __NPU_ARCH__ == 3113
#include "dav_l311/kernel_operator_vec_cmpsel_impl.h"
#endif
#pragma begin_pipe(V)
namespace AscendC {
* Compare *
* ************************************************************************************************* */
* @ingroup Compare Level 0
* @brief Compare the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input LocalTensor
* @param [in] cmpMode compare mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.src0BlkStride src0 block stride
* @param [in] intriParams.src1BlkStride src1 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.src0RepStride src0 repeat stride
* @param [in] intriParams.src1RepStride src1 repeat stride
*/
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Compare(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, CMPMODE cmpMode, const uint64_t mask[], uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
using SrcPrimType = PrimT<T>;
using DstPrimType = PrimT<U>;
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float, dst: int8_t / uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float, int32_t>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float / int32_t, dst: int8_t / uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncVecBinaryCmp(dst, src0, src1, mask, repeatTime, repeatParams, "Compare")) {
ASCENDC_REPORT_CHECK_ERROR("Compare", KernelFuncType::MASK_BIT_MODE);
}
#endif
VcmpvImpl<SrcPrimType, DstPrimType, isSetMask>((__ubuf__ DstPrimType*)dst.GetPhyAddr(), (__ubuf__ SrcPrimType*)src0.GetPhyAddr(),
(__ubuf__ SrcPrimType*)src1.GetPhyAddr(), cmpMode, mask, repeatTime, repeatParams);
}
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Compare(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, CMPMODE cmpMode, const uint64_t mask, uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
using SrcPrimType = PrimT<T>;
using DstPrimType = PrimT<U>;
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float, dst: int8_t / uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float, int32_t>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float / int32_t, dst: int8_t / uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncVecBinaryCmp(dst, src0, src1, mask, repeatTime, repeatParams, "Compare")) {
ASCENDC_REPORT_CHECK_ERROR("Compare", KernelFuncType::MASK_COUNT_MODE);
}
#endif
VcmpvImpl<SrcPrimType, DstPrimType, isSetMask>((__ubuf__ DstPrimType*)dst.GetPhyAddr(), (__ubuf__ SrcPrimType*)src0.GetPhyAddr(),
(__ubuf__ SrcPrimType*)src1.GetPhyAddr(), cmpMode, mask, repeatTime, repeatParams);
}
template <typename T, bool isSetMask>
__aicore__ inline void Compare(const LocalTensor<T>& src0, const LocalTensor<T>& src1, CMPMODE cmpMode,
const uint64_t mask[], const BinaryRepeatParams& repeatParams)
{
using PrimType = PrimT<T>;
#if __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<PrimType, half, float>()), {KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, "
"current api support dtype combination is src: half / float.");});
#endif
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncVecBinaryCmpRgt(src0, src1, mask, repeatParams, "Compare")) {
ASCENDC_REPORT_CHECK_ERROR("Compare", KernelFuncType::MASK_BIT_MODE);
}
#endif
VcmpImpl<PrimType, isSetMask>((__ubuf__ PrimType*)src0.GetPhyAddr(),
(__ubuf__ PrimType*)src1.GetPhyAddr(), cmpMode, mask, repeatParams);
}
template <typename T, bool isSetMask>
__aicore__ inline void Compare(const LocalTensor<T>& src0, const LocalTensor<T>& src1, CMPMODE cmpMode,
const uint64_t mask, const BinaryRepeatParams& repeatParams)
{
using PrimType = PrimT<T>;
#if __NPU_ARCH__ == 2002 || __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<PrimType, half, float>()), {KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, "
"current api support dtype combination is src: half / float.");});
#endif
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncVecBinaryCmpRgt(src0, src1, mask, repeatParams, "Compare")) {
ASCENDC_REPORT_CHECK_ERROR("Compare", KernelFuncType::MASK_COUNT_MODE);
}
#endif
VcmpImpl<PrimType, isSetMask>((__ubuf__ PrimType*)src0.GetPhyAddr(), (__ubuf__ PrimType*)src1.GetPhyAddr(), cmpMode,
mask, repeatParams);
}
* @ingroup Compare Level 2
* @brief Compare the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input LocalTensor
* @param [in] cmpMode compare mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T, typename U>
__aicore__ inline void Compare(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, CMPMODE cmpMode, uint32_t count)
{
using SrcPrimType = PrimT<T>;
using DstPrimType = PrimT<U>;
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float, dst: int8_t / uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<SrcPrimType, half, float, int32_t>() && SupportType<DstPrimType, int8_t, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compare, current api support dtype combination is src: "
"half / float / int32_t, dst: int8_t / uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryCmp(dst, src0, src1, count, "Compare")) {
ASCENDC_REPORT_CHECK_ERROR("Compare", KernelFuncType::CALCOUNT_MODE);
}
#endif
#if defined(__NPU_ARCH__) && ((__NPU_ARCH__ == 3101) || (__NPU_ARCH__ == 5102))
ASCENDC_ASSERT(((count * sizeof(T)) % ONE_REPEAT_BYTE_SIZE == 0),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check count elements size in Compare, current size "
"is %u, should be an integer multiple of 256.", count * sizeof(T));});
#endif
VcmpvImpl((__ubuf__ DstPrimType*)dst.GetPhyAddr(), (__ubuf__ SrcPrimType*)src0.GetPhyAddr(),
(__ubuf__ SrcPrimType*)src1.GetPhyAddr(), cmpMode, count);
}
template <typename T>
__aicore__ inline void GetCmpMask(const LocalTensor<T>& dst)
{
using PrimType = PrimT<T>;
#if __NPU_ARCH__ == 2201
if (g_coreType == AIC) {
return;
}
#endif
#if ASCENDC_CPU_DEBUG
constexpr uint64_t align16B = 16;
CheckTensorAlign<T>(dst, align16B, "dst", "GetCmpMask");
CheckTensorPos<T>(dst, Hardware::UB, "dst", "VECIN / VECCALC / VECOUT", "GetCmpMask");
#endif
GetCmpMaskImpl((__ubuf__ PrimType*)dst.GetPhyAddr());
}
template <typename T>
__aicore__ inline void SetCmpMask(const LocalTensor<T>& src)
{
using PrimType = PrimT<T>;
#if __NPU_ARCH__ == 2201
if (g_coreType == AIC) {
return;
}
#endif
#if ASCENDC_CPU_DEBUG
constexpr uint64_t align16B = 16;
CheckTensorAlign<T>(src, align16B, "src", "SetCmpMask");
CheckTensorPos<T>(src, Hardware::UB, "src", "VECIN / VECCALC / VECOUT", "SetCmpMask");
#endif
SetCmpMaskImpl((__ubuf__ PrimType*)src.GetPhyAddr());
}
* CompareScalar *
* ************************************************************************************************* */
* @ingroup Compare Level 0
* @brief Compare the size of a tensor and a scalar one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.srcBlkStride src0 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.srcRepStride src0 repeat stride
*/
template <typename T, typename U, bool isSetMask = true>
__aicore__ inline void CompareScalar(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, const uint64_t mask[], uint8_t repeatTime,
const UnaryRepeatParams& repeatParams)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, ONE_REPEAT_BYTE_SIZE / sizeof(T), repeatTime,
repeatParams, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::MASK_BIT_MODE);
}
#endif
VcmpvsImpl<T, U, isSetMask>((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar,
cmpMode, mask, repeatTime, repeatParams);
}
template <typename T, typename U, bool isSetMask = true>
__aicore__ inline void CompareScalar(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, const uint64_t mask, uint8_t repeatTime,
const UnaryRepeatParams& repeatParams)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, ONE_REPEAT_BYTE_SIZE / sizeof(T), repeatTime,
repeatParams, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::MASK_COUNT_MODE);
}
#endif
VcmpvsImpl<T, U, isSetMask>((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar,
cmpMode, mask, repeatTime, repeatParams);
}
* @ingroup CompareScalar Level 2
* @brief CompareScalar the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T, typename U>
__aicore__ inline void CompareScalar(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, uint32_t count)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in CompareScalar, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
ASCENDC_ASSERT(((count * sizeof(T)) % ONE_REPEAT_BYTE_SIZE == 0),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check count elements size in CompareScalar, current size "
"is %u, should be an integer multiple of 256.", count * sizeof(T));});
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, count, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::CALCOUNT_MODE);
}
#endif
VcmpvsImpl((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar, cmpMode, count);
}
* Compares *
* ************************************************************************************************* */
* @ingroup Compare Level 0
* @brief Compare the size of a tensor and a scalar one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.srcBlkStride src0 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.srcRepStride src0 repeat stride
*/
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Compares(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, const uint64_t mask[], uint8_t repeatTime,
const UnaryRepeatParams& repeatParams)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, ONE_REPEAT_BYTE_SIZE / sizeof(T), repeatTime,
repeatParams, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::MASK_BIT_MODE);
}
#endif
VcmpvsImpl<T, U, isSetMask>((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar,
cmpMode, mask, repeatTime, repeatParams);
}
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Compares(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, const uint64_t mask, uint8_t repeatTime,
const UnaryRepeatParams& repeatParams)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, ONE_REPEAT_BYTE_SIZE / sizeof(T), repeatTime,
repeatParams, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::MASK_COUNT_MODE);
}
#endif
VcmpvsImpl<T, U, isSetMask>((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar,
cmpMode, mask, repeatTime, repeatParams);
}
* @ingroup Compares Level 2
* @brief Compares the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T, typename U>
__aicore__ inline void Compares(const LocalTensor<U>& dst, const LocalTensor<T>& src0,
const T src1Scalar, CMPMODE cmpMode, uint32_t count)
{
#if __NPU_ARCH__ == 2002
ASCENDC_ASSERT((SupportType<T, half, float>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float, dst: uint8_t.");});
#elif __NPU_ARCH__ == 2201
ASCENDC_ASSERT((SupportType<T, half, float, int32_t>() && SupportType<U, uint8_t>()),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check dtype in Compares, current api support dtype combination is "
"src0: half / float / int32_t, dst: uint8_t.");});
#endif
ASCENDC_ASSERT(((count * sizeof(T)) % ONE_REPEAT_BYTE_SIZE == 0),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check count elements size in Compares, current size "
"is %u, should be an integer multiple of 256.", count * sizeof(T));});
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp(dst, src0, src1Scalar, count, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::CALCOUNT_MODE);
}
#endif
VcmpvsImpl((__ubuf__ U*)dst.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(), src1Scalar, cmpMode, count);
}
* Select *
* ************************************************************************************************* */
* @ingroup Select Level 0
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input LocalTensor
* @param [in] selMode select mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.src0BlkStride src0 block stride
* @param [in] intriParams.src1BlkStride src1 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.src0RepStride src0 repeat stride
* @param [in] intriParams.src1RepStride src1 repeat stride
*/
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, const LocalTensor<T>& src1, SELMODE selMode, uint64_t mask[],
uint8_t repeatTime, const BinaryRepeatParams& repeatParams)
{
using DataPrimType = PrimT<T>;
using MaskPrimType = PrimT<U>;
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncSelectVec(dst, selMask, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_BIT_MODE);
}
#endif
VselImpl((__ubuf__ DataPrimType*)dst.GetPhyAddr(), (__ubuf__ MaskPrimType*)selMask.GetPhyAddr(),
(__ubuf__ DataPrimType*)src0.GetPhyAddr(), (__ubuf__ DataPrimType*)src1.GetPhyAddr(), selMode, mask, repeatTime,
repeatParams);
}
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, const LocalTensor<T>& src1, SELMODE selMode, uint64_t mask,
uint8_t repeatTime, const BinaryRepeatParams& repeatParams)
{
using DataPrimType = PrimT<T>;
using MaskPrimType = PrimT<U>;
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFuncSelectVec(dst, selMask, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_COUNT_MODE);
}
#endif
VselImpl((__ubuf__ DataPrimType*)dst.GetPhyAddr(), (__ubuf__ MaskPrimType*)selMask.GetPhyAddr(),
(__ubuf__ DataPrimType*)src0.GetPhyAddr(), (__ubuf__ DataPrimType*)src1.GetPhyAddr(), selMode, mask, repeatTime,
repeatParams);
}
template <typename T, SELMODE selMode>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<T>& src0,
const LocalTensor<T>& src1, uint8_t repeatTime, const BinaryRepeatParams& repeatParams)
{
using PrimType = PrimT<T>;
SelectCal<PrimType, selMode>((__ubuf__ PrimType*)dst.GetPhyAddr(), (__ubuf__ PrimType*)src0.GetPhyAddr(),
(__ubuf__ PrimType*)src1.GetPhyAddr(), repeatTime, repeatParams);
}
template <typename T, typename U>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, uint8_t repeatTime, const BinaryRepeatParams& repeatParams)
{
using DataPrimType = PrimT<T>;
using MaskPrimType = PrimT<U>;
SelectCal<DataPrimType, MaskPrimType>((__ubuf__ DataPrimType*)dst.GetPhyAddr(),
(__ubuf__ MaskPrimType*)selMask.GetPhyAddr(), (__ubuf__ DataPrimType*)src0.GetPhyAddr(), repeatTime,
repeatParams);
}
* @ingroup Select Level 2
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input LocalTensor
* @param [in] selMode select mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T, typename U>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, const LocalTensor<T>& src1, SELMODE selMode, uint32_t count)
{
using DataPrimType = PrimT<T>;
using MaskPrimType = PrimT<U>;
#if ASCENDC_CPU_DEBUG
if (!CheckFuncSelectVec(dst, selMask, src0, src1, (int32_t)count, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::CALCOUNT_MODE);
}
#endif
VselImpl((__ubuf__ DataPrimType*)dst.GetPhyAddr(), (__ubuf__ MaskPrimType*)selMask.GetPhyAddr(),
(__ubuf__ DataPrimType*)src0.GetPhyAddr(), (__ubuf__ DataPrimType*)src1.GetPhyAddr(), selMode, count);
}
* @ingroup Select Level 0
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input number
* @param [in] selMode select mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.src0BlkStride src0 block stride
* @param [in] intriParams.src1BlkStride src1 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.src0RepStride src0 repeat stride
* @param [in] intriParams.src1RepStride src1 repeat stride
*/
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, T src1, SELMODE selMode, uint64_t mask[], uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFunVecBinaryScalar(dst, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_BIT_MODE);
}
#endif
CheckTensorPos<U>(selMask, Hardware::UB, "selMask", "VECIN / VECCALC / VECOUT", "Select");
VselImpl((__ubuf__ T*)dst.GetPhyAddr(), (__ubuf__ U*)selMask.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(),
src1, selMode, mask, repeatTime, repeatParams);
}
template <typename T, typename U, bool isSetMask>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, T src1, SELMODE selMode, uint64_t mask, uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFunVecBinaryScalar(dst, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_COUNT_MODE);
}
#endif
CheckTensorPos<U>(selMask, Hardware::UB, "selMask", "VECIN / VECCALC / VECOUT", "Select");
VselImpl((__ubuf__ T*)dst.GetPhyAddr(), (__ubuf__ U*)selMask.GetPhyAddr(), (__ubuf__ T*)src0.GetPhyAddr(),
src1, selMode, mask, repeatTime, repeatParams);
}
* @ingroup Select Level 2
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input number
* @param [in] selMode select mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T, typename U>
__aicore__ inline void Select(const LocalTensor<T>& dst, const LocalTensor<U>& selMask,
const LocalTensor<T>& src0, T src1, SELMODE selMode, uint32_t count)
{
using DataPrimType = PrimT<T>;
using MaskPrimType = PrimT<U>;
#if ASCENDC_CPU_DEBUG
if (!CheckFunVecBinaryScalar(dst, src0, src1, (int32_t)count, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::CALCOUNT_MODE);
}
#endif
CheckTensorPos<U>(selMask, Hardware::UB, "selMask", "VECIN / VECCALC / VECOUT", "Select");
VselImpl((__ubuf__ DataPrimType*)dst.GetPhyAddr(), (__ubuf__ MaskPrimType*)selMask.GetPhyAddr(),
(__ubuf__ DataPrimType*)src0.GetPhyAddr(), src1, selMode, count);
}
#if defined(__NPU_ARCH__) && ((__NPU_ARCH__ == 3101) || (__NPU_ARCH__ == 5102))
* CompareScalar *
* ************************************************************************************************* */
* @ingroup Compare Level 0
* @brief Compare the size of a tensor and a scalar one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.srcBlkStride src0 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.srcRepStride src0 repeat stride
*/
template <bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4, typename MaskType>
__aicore__ inline void CompareScalarCommon(const T2& dst, const T3& src0, const T4& src1, CMPMODE cmpMode,
MaskType mask, uint8_t repeatTime, const UnaryRepeatParams& repeatParams)
{
static_assert(!TypeUtils::IsInnerDefaultType<T3, T4>(), "One of src0 and src1 should be Tensor");
static_assert(SupportType<T2, LocalTensor<uint8_t>>());
using ActualU = typename T2::PrimType;
if constexpr(TypeUtils::IsLocalTensorType<T3, T4>()) {
static_assert(Std::is_same<T3, T4>::value);
using ActualT = typename T3::PrimType;
static_assert((config.scalarTensorIndex == 0 || config.scalarTensorIndex == 1), "scalarTensorIndex out of range");
VcmpvsImpl<ActualT, ActualU, isSetMask, config.scalarTensorIndex>((__ubuf__ ActualU*)dst.GetPhyAddr(),
(__ubuf__ ActualT*)src0.GetPhyAddr(), (__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, mask, repeatTime, repeatParams);
} else if constexpr(TypeUtils::IsLocalTensorType<T4>() && TypeUtils::IsInnerDefaultType<T3>()) {
using ActualT = typename T4::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), src0,
(__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, mask, repeatTime, repeatParams);
} else if constexpr(TypeUtils::IsLocalTensorType<T3>() && TypeUtils::IsInnerDefaultType<T4>()) {
using ActualT = typename T3::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), (__ubuf__ ActualT*)src0.GetPhyAddr(),
src1, cmpMode, mask, repeatTime, repeatParams);
}
}
template <typename T0 = BinaryDefaultType, typename T1 = BinaryDefaultType, bool isSetMask = true,
const BinaryConfig &config = DEFAULT_BINARY_CONFIG, typename T2, typename T3, typename T4>
__aicore__ inline void CompareScalar(const T2& dst, const T3& src0, const T4& src1, CMPMODE cmpMode,
const uint64_t mask[], uint8_t repeatTime, const UnaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, ONE_REPEAT_BYTE_SIZE / sizeof(T0), repeatTime,
repeatParams, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::MASK_BIT_MODE);
}
#endif
CompareScalarCommon<isSetMask, config>(dst, src0, src1, cmpMode, mask, repeatTime, repeatParams);
}
template <typename T0 = BinaryDefaultType, typename T1 = BinaryDefaultType, bool isSetMask = true,
const BinaryConfig &config = DEFAULT_BINARY_CONFIG, typename T2, typename T3, typename T4>
__aicore__ inline void CompareScalar(const T2& dst, const T3& src0, const T4& src1, CMPMODE cmpMode,
const uint64_t mask, uint8_t repeatTime, const UnaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, ONE_REPEAT_BYTE_SIZE / sizeof(T0), repeatTime,
repeatParams, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::MASK_COUNT_MODE);
}
#endif
CompareScalarCommon<isSetMask, config>(dst, src0, src1, cmpMode, mask, repeatTime, repeatParams);
}
* @ingroup CompareScalar Level 2
* @brief CompareScalar the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T0 = BinaryDefaultType, typename T1 = BinaryDefaultType, bool isSetMask = true,
const BinaryConfig &config = DEFAULT_BINARY_CONFIG, typename T2, typename T3, typename T4>
__aicore__ inline void CompareScalar(const T2& dst, const T3& src0,
const T4& src1, CMPMODE cmpMode, uint32_t count)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, count, "CompareScalar")) {
ASCENDC_REPORT_CHECK_ERROR("CompareScalar", KernelFuncType::CALCOUNT_MODE);
}
#endif
static_assert(!TypeUtils::IsInnerDefaultType<T3, T4>(), "One of src0 and src1 should be Tensor");
static_assert(SupportType<T2, LocalTensor<uint8_t>>());
using ActualU = typename T2::PrimType;
#if defined(__NPU_ARCH__) && ((__NPU_ARCH__ == 3101) || (__NPU_ARCH__ == 5102))
ASCENDC_ASSERT(((count * sizeof(T3)) % ONE_REPEAT_BYTE_SIZE == 0),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check count elements size in CompareScalar, current size "
"is %u, should be an integer multiple of 256.", count * sizeof(T3));});
#endif
if constexpr(TypeUtils::IsLocalTensorType<T3, T4>()) {
static_assert(Std::is_same<T3, T4>::value);
using ActualT = typename T3::PrimType;
static_assert((config.scalarTensorIndex == 0 || config.scalarTensorIndex == 1), "scalarTensorIndex out of range");
VcmpvsImpl<ActualT, ActualU, isSetMask, config.scalarTensorIndex>((__ubuf__ ActualU*)dst.GetPhyAddr(),
(__ubuf__ ActualT*)src0.GetPhyAddr(), (__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T4>() && TypeUtils::IsInnerDefaultType<T3>()) {
using ActualT = typename T4::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), src0,
(__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T3>() && TypeUtils::IsInnerDefaultType<T4>()) {
using ActualT = typename T3::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), (__ubuf__ ActualT*)src0.GetPhyAddr(),
src1, cmpMode, count);
}
}
* Compares *
* ************************************************************************************************* */
* @ingroup Compare Level 0
* @brief Compare the size of a tensor and a scalar one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.srcBlkStride src0 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.srcRepStride src0 repeat stride
*/
template <typename T0, typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Compares(const T2& dst, const T3& src0, const T4& src1, CMPMODE cmpMode,
const uint64_t mask[], uint8_t repeatTime, const UnaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, ONE_REPEAT_BYTE_SIZE / sizeof(T0), repeatTime,
repeatParams, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::MASK_BIT_MODE);
}
#endif
CompareScalarCommon<isSetMask, config>(dst, src0, src1, cmpMode, mask, repeatTime, repeatParams);
}
template <typename T0, typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Compares(const T2& dst, const T3& src0, const T4& src1, CMPMODE cmpMode,
const uint64_t mask, uint8_t repeatTime, const UnaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, ONE_REPEAT_BYTE_SIZE / sizeof(T0), repeatTime,
repeatParams, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::MASK_COUNT_MODE);
}
#endif
CompareScalarCommon<isSetMask, config>(dst, src0, src1, cmpMode, mask, repeatTime, repeatParams);
}
* @ingroup Compares Level 2
* @brief Compares the size of two tensors one by one. If true, the corresponding bit is 1, otherwise it is 0
* @param [out] dst output LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1Scalar input Scalar
* @param [in] cmpMode compare mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T0, typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Compares(const T2& dst, const T3& src0,
const T4& src1, CMPMODE cmpMode, uint32_t count)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFuncVecBinaryScalarCmp<config>(dst, src0, src1, count, "Compares")) {
ASCENDC_REPORT_CHECK_ERROR("Compares", KernelFuncType::CALCOUNT_MODE);
}
#endif
static_assert(!TypeUtils::IsInnerDefaultType<T3, T4>(), "One of src0 and src1 should be Tensor");
static_assert(SupportType<T2, LocalTensor<uint8_t>>());
using ActualU = typename T2::PrimType;
#if defined(__NPU_ARCH__) && ((__NPU_ARCH__ == 3101) || (__NPU_ARCH__ == 5102))
ASCENDC_ASSERT(((count * sizeof(T3)) % ONE_REPEAT_BYTE_SIZE == 0),
{KERNEL_LOG(KERNEL_ERROR, "Failed to check count elements size in Compares, current size "
"is %u, should be an integer multiple of 256.", count * sizeof(T3));});
#endif
if constexpr(TypeUtils::IsLocalTensorType<T3, T4>()) {
static_assert(Std::is_same<T3, T4>::value);
using ActualT = typename T3::PrimType;
static_assert((config.scalarTensorIndex == 0 || config.scalarTensorIndex == 1), "scalarTensorIndex out of range");
VcmpvsImpl<ActualT, ActualU, isSetMask, config.scalarTensorIndex>((__ubuf__ ActualU*)dst.GetPhyAddr(),
(__ubuf__ ActualT*)src0.GetPhyAddr(), (__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T4>() && TypeUtils::IsInnerDefaultType<T3>()) {
using ActualT = typename T4::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), src0,
(__ubuf__ ActualT*)src1.GetPhyAddr(), cmpMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T3>() && TypeUtils::IsInnerDefaultType<T4>()) {
using ActualT = typename T3::PrimType;
VcmpvsImpl<ActualT, ActualU, isSetMask>((__ubuf__ ActualU*)dst.GetPhyAddr(), (__ubuf__ ActualT*)src0.GetPhyAddr(),
src1, cmpMode, count);
}
}
* @ingroup Select Level 0
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input number
* @param [in] selMode select mode
* @param [in] mask[]/mask mask array/count
* @param [in] repeatTime repeat times
* @param [in] intriParams.dstBlkStride dst block stride
* @param [in] intriParams.src0BlkStride src0 block stride
* @param [in] intriParams.src1BlkStride src1 block stride
* @param [in] intriParams.dstRepStride dst repeat stride
* @param [in] intriParams.src0RepStride src0 repeat stride
* @param [in] intriParams.src1RepStride src1 repeat stride
*/
template <typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3,
typename T4, typename MaskType>
__aicore__ inline void SelectCommon(const T2& dst, const LocalTensor<T1>& selMask,
const T3& src0, const T4& src1, SELMODE selMode, MaskType mask, uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
CheckTensorPos<T1>(selMask, Hardware::UB, "selMask", "VECIN / VECCALC / VECOUT", "Select");
static_assert(!TypeUtils::IsInnerDefaultType<T3, T4>(), "One of src0 and src1 should be Tensor");
static_assert(TypeUtils::IsLocalTensorType<T2>(), "dst should be Tensor");
using ActualT = typename T2::PrimType;
if constexpr(TypeUtils::IsLocalTensorType<T3, T4>()) {
static_assert(Std::is_same<T3, T4>::value);
static_assert(Std::is_same<ActualT, typename T3::PrimType>::value);
static_assert((config.scalarTensorIndex == 0 || config.scalarTensorIndex == 1), "scalarTensorIndex out of range");
VselImpl<ActualT, T1, config.scalarTensorIndex>((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(),
(__ubuf__ ActualT*)src0.GetPhyAddr(), (__ubuf__ ActualT*)src1.GetPhyAddr(), selMode, mask, repeatTime, repeatParams);
} else if constexpr(TypeUtils::IsLocalTensorType<T4>() && TypeUtils::IsInnerDefaultType<T3>()) {
static_assert(Std::is_same<ActualT, typename T4::PrimType>::value);
VselImpl((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(), src0,
(__ubuf__ ActualT*)src1.GetPhyAddr(), selMode, mask, repeatTime, repeatParams);
} else if constexpr(TypeUtils::IsLocalTensorType<T3>() && TypeUtils::IsInnerDefaultType<T4>()) {
static_assert(Std::is_same<ActualT, typename T3::PrimType>::value);
VselImpl((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(), (__ubuf__ ActualT*)src0.GetPhyAddr(),
src1, selMode, mask, repeatTime, repeatParams);
}
}
template <typename T0, typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Select(const T2& dst, const LocalTensor<T1>& selMask,
const T3& src0, const T4& src1, SELMODE selMode, uint64_t mask[], uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFunVecBinaryScalar<config>(dst, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_BIT_MODE);
}
#endif
SelectCommon<T1, isSetMask, config>(dst, selMask, src0, src1, selMode, mask, repeatTime, repeatParams);
}
template <typename T0, typename T1, bool isSetMask, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Select(const T2& dst, const LocalTensor<T1>& selMask,
const T3& src0, const T4& src1, SELMODE selMode, uint64_t mask, uint8_t repeatTime,
const BinaryRepeatParams& repeatParams)
{
#if ASCENDC_CPU_DEBUG
MaskSetter::Instance().SetMask(isSetMask);
if (!CheckFunVecBinaryScalar<config>(dst, src0, src1, mask, repeatTime, repeatParams, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::MASK_COUNT_MODE);
}
#endif
SelectCommon<T1, isSetMask, config>(dst, selMask, src0, src1, selMode, mask, repeatTime, repeatParams);
}
* @ingroup Select Level 2
* @brief Select element according to the bit value of sel
* @param [out] dst output LocalTensor
* @param [in] selMask select mask LocalTensor
* @param [in] src0 input LocalTensor
* @param [in] src1 input number
* @param [in] selMode select mode
* @param [in] count number Number of data involved in calculation
*/
template <typename T0, typename T1, const BinaryConfig &config, typename T2, typename T3, typename T4>
__aicore__ inline void Select(const T2& dst, const LocalTensor<T1>& selMask,
const T3& src0, const T4& src1, SELMODE selMode, uint32_t count)
{
#if ASCENDC_CPU_DEBUG
if (!CheckFunVecBinaryScalar<config>(dst, src0, src1, count, "Select")) {
ASCENDC_REPORT_CHECK_ERROR("Select", KernelFuncType::CALCOUNT_MODE);
}
#endif
CheckTensorPos<T1>(selMask, Hardware::UB, "selMask", "VECIN / VECCALC / VECOUT", "Select");
static_assert(!TypeUtils::IsInnerDefaultType<T3, T4>(), "One of src0 and src1 should be Tensor");
static_assert(TypeUtils::IsLocalTensorType<T2>(), "dst should be Tensor");
using ActualT = typename T2::PrimType;
if constexpr(TypeUtils::IsLocalTensorType<T3, T4>()) {
static_assert(Std::is_same<T3, T4>::value);
static_assert(Std::is_same<ActualT, typename T3::PrimType>::value);
static_assert((config.scalarTensorIndex == 0 || config.scalarTensorIndex == 1), "scalarTensorIndex out of range");
VselImpl<ActualT, T1, config.scalarTensorIndex>((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(),
(__ubuf__ ActualT*)src0.GetPhyAddr(), (__ubuf__ ActualT*)src1.GetPhyAddr(), selMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T4>() && TypeUtils::IsInnerDefaultType<T3>()) {
static_assert(Std::is_same<ActualT, typename T4::PrimType>::value);
VselImpl((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(), src0,
(__ubuf__ ActualT*)src1.GetPhyAddr(), selMode, count);
} else if constexpr(TypeUtils::IsLocalTensorType<T3>() && TypeUtils::IsInnerDefaultType<T4>()) {
static_assert(Std::is_same<ActualT, typename T3::PrimType>::value);
VselImpl((__ubuf__ ActualT*)dst.GetPhyAddr(), (__ubuf__ T1*)selMask.GetPhyAddr(), (__ubuf__ ActualT*)src0.GetPhyAddr(),
src1, selMode, count);
}
}
#endif
}
#pragma end_pipe
#endif
