* 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 as_strided.cpp
* \brief as_strided
*/
#include <cstdint>
#include "./arch35/as_strided_move_align.h"
#include "./arch35/as_strided_dual_cut.h"
#include "./arch35/as_strided.h"
#include "./arch35/as_strided_struct.h"
#include "./arch35/as_strided_zero_stride.h"
#include "./arch35/as_strided_gather.h"
#include "./arch35/as_strided_simt.h"
using namespace AsStrided;
#define AS_STRIDED_B8 1
#define AS_STRIDED_B16 2
#define AS_STRIDED_B32 4
#define AS_STRIDED_B64 8
#define AS_STRIDED_MOVE_ALIGN_B8 101
#define AS_STRIDED_MOVE_ALIGN_B16 102
#define AS_STRIDED_MOVE_ALIGN_B32 104
#define AS_STRIDED_MOVE_ALIGN_B64 108
#define AS_STRIDED_DUAL_CUT 200
#define ALL_STRIDEDS_ZERO_KEY 300
#define SIMT_KEY 400
#define AS_STRIDED_GATHER 500
#define EMPTY_TENSOR_KEY 1000
extern "C" __global__ __aicore__ void as_strided(GM_ADDR input, GM_ADDR outShape, GM_ADDR outStride,
GM_ADDR storageOffset, GM_ADDR output, GM_ADDR workspace, GM_ADDR tiling)
{
KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
REGISTER_NONE_TILING;
if (TILING_KEY_IS(AS_STRIDED_B8)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStrided<int8_t> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_B16)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStrided<half> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_B32)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStrided<float> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_B64)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStrided<int64_t> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_MOVE_ALIGN_B8)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStridedMoveAlign<int8_t> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_MOVE_ALIGN_B16)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStridedMoveAlign<half> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_MOVE_ALIGN_B32)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStridedMoveAlign<float> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_MOVE_ALIGN_B64)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
KernelAsStridedMoveAlign<int64_t> op;
op.Init(input, outShape, outStride, output, tilingData);
op.Process();
} else if (TILING_KEY_IS(AS_STRIDED_DUAL_CUT)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedTilingData, tilingData, tiling);
if constexpr (sizeof(DTYPE_X) == sizeof(int8_t)) {
KernelAsStridedDualCut<int8_t, AS_STRIDED_DUAL_CUT> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
} else {
KernelAsStridedDualCut<DTYPE_X, AS_STRIDED_DUAL_CUT> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
}
} else if (TILING_KEY_IS(ALL_STRIDEDS_ZERO_KEY)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedZeroStrideTilingData, tilingData, tiling);
if(sizeof(DTYPE_X) == sizeof(int8_t)) {
StridedIsZero<int8_t> op;
op.Init(input, output, &tilingData);
op.Process();
} else if(sizeof(DTYPE_X) == sizeof(half)) {
StridedIsZero<half> op;
op.Init(input, output, &tilingData);
op.Process();
} else if(sizeof(DTYPE_X) == sizeof(float)) {
StridedIsZero<float> op;
op.Init(input, output, &tilingData);
op.Process();
} else if(sizeof(DTYPE_X) == sizeof(int64_t)) {
StridedIsZero<int64_t> op;
op.Init(input, output, &tilingData);
op.Process();
}
} else if (TILING_KEY_IS(SIMT_KEY)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedSimtTilingData, tilingData, tiling);
if(sizeof(DTYPE_X) == sizeof(int8_t)) {
AsStridedSimt<int8_t> op;
op.Init(input, output, &tilingData);
op.Process(tiling);
} else if(sizeof(DTYPE_X) == sizeof(half)) {
AsStridedSimt<half> op;
op.Init(input, output, &tilingData);
op.Process(tiling);
} else if(sizeof(DTYPE_X) == sizeof(float)) {
AsStridedSimt<float> op;
op.Init(input, output, &tilingData);
op.Process(tiling);
} else if(sizeof(DTYPE_X) == sizeof(int64_t)) {
AsStridedSimt<int64_t> op;
op.Init(input, output, &tilingData);
op.Process(tiling);
}
} else if (TILING_KEY_IS(AS_STRIDED_GATHER)) {
GET_TILING_DATA_WITH_STRUCT(AsStridedWithGatherTilingData, tilingData, tiling);
if constexpr (sizeof(DTYPE_X) == sizeof(int8_t)) {
KernelAsStridedGather<uint8_t> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
} else if constexpr (sizeof(DTYPE_X) == sizeof(int16_t)) {
KernelAsStridedGather<uint16_t> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
} else if constexpr (sizeof(DTYPE_X) == sizeof(int32_t)) {
KernelAsStridedGather<uint32_t> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
} else if constexpr (sizeof(DTYPE_X) == sizeof(int64_t)) {
KernelAsStridedGather<uint64_t> op;
op.Init(input, outShape, outStride, output, &tilingData);
op.Process();
}
} else if (TILING_KEY_IS(EMPTY_TENSOR_KEY)){
GET_TILING_DATA_WITH_STRUCT(AsStridedEmptyTilingData, tilingData, tiling);
}
}
