* 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 transdata_proto.h
* \brief
*/
#ifndef OP_PROTO_TRANSDATA_PROTO_H_
#define OP_PROTO_TRANSDATA_PROTO_H_
#include "graph/operator_reg.h"
namespace ge
{
* @brief Do format transfer for various data format.
* In general, the framework will insert it atomatically. \n
* @par Inputs:
* src: A Tensor. Support 2D ~ 6D. For all branches can be types: bfloat16, float16, float32, int32, int8, bool.
* For branches without padding also can be types: int16, int64, uint8, uint16, uint32, uint64. \n
* @par Attributes:
* @li src_format: A string source data format, can be "NHWC", "NCHW" etc.
* @li dst_format: A string target data format, can be "NCHW" etc.
* @li src_subformat: A optional int32 for source sub-format, default value is 0.
* @li dst_subformat: A optional int32 for target sub-format, default value is 0.
* @li groups: A optional int32, the N axis must be divisible by "groups". Defaults to 1. \n
* @par Outputs:
* dst: A Tensor. Support 2D ~ 6D. Has the same type as "src".
*\n
*\n
* The following value ranges must be met.
* '<===>' indicates that format is bidirectionly supported, either input or output.
* '===>' indicates that format is unbidirectionly supported, and the input and
* output data types must be correspond to each other. \n
*\n
*\n
| src_format <===> dst_format | dtype | C0 | groups |\n
| :-------------------------: | :--------------------------------: |:-----:| :----: |\n
| NCHW <====> NC1HWC0 | float32, int32,uint32 | 8,16 | 1 |\n
| NCHW <====> NC1HWC0 | bfloat16, float16 | 16 | 1 |\n
| NCHW <====> NC1HWC0 | int8, uint8 | 32 | 1 |\n
| NHWC <====> NC1HWC0 | float32, int32,uint32 | 8,16 | 1 |\n
| NHWC <====> NC1HWC0 | bfloat16, float16 | 16 | 1 |\n
| NHWC <====> NC1HWC0 | int8, uint8 | 32 | 1 |\n
| ND <====> FRACTAL_NZ | float32, int32,uint32 | 16 | 1 |\n
| ND <====> FRACTAL_NZ | bfloat16, float16 | 16 | 1 |\n
| ND <====> FRACTAL_NZ | int8, uint8 | 32 | 1 |\n
| ND <====> FRACTAL_NZ | float4_e2m1 | 64 | 1 |\n
| ND ====> FRACTAL_NZ | float4_e1m2 | 64 | 1 |\n
| NCHW <====> FRACTAL_Z | float32, int32,uint32 | 8,16 | 1 |\n
| NCHW <====> FRACTAL_Z | bfloat16, float16 | 16 | 1 |\n
| NCHW <====> FRACTAL_Z | int8, uint8 | 32 | 1 |\n
| HWCN <====> FRACTAL_Z | float32, int32,uint32 | 8,16 | 1 |\n
| HWCN <====> FRACTAL_Z | bfloat16, float16 | 16 | 1 |\n
| HWCN <====> FRACTAL_Z | int8, uint8 | 32 | 1 |\n
| NCDHW <====> NDC1HWC0 | float32, int32,uint32 | 8,16 | 1 |\n
| NCDHW <====> NDC1HWC0 | bfloat16, float16 | 16 | 1 |\n
| NCDHW <====> NDC1HWC0 | int8, uint8 | 32 | 1 |\n
| NDHWC <====> NDC1HWC0 | float32, int32,uint32 | 8,16 | 1 |\n
| NDHWC <====> NDC1HWC0 | bfloat16, float16 | 16 | 1 |\n
| NDHWC <====> NDC1HWC0 | int8, uint8 | 32 | 1 |\n
| NCDHW <====> FRACTAL_Z_3D | float32, int32,uint32 | 8,16 | 1 |\n
| NCDHW <====> FRACTAL_Z_3D | bfloat16, float16 | 16 | 1 |\n
| NCDHW <====> FRACTAL_Z_3D | int8, uint8 | 32 | 1 |\n
| DHWCN <====> FRACTAL_Z_3D | float32, int32,uint32 | 16 | 1 |\n
| DHWCN <====> FRACTAL_Z_3D | bfloat16, float16 | 16 | 1 |\n
| DHWCN <====> FRACTAL_Z_3D | int8, uint8 | 32 | 1 |\n
| NCHW <====> FRACTAL_Z | float32, uint32, int32 | 8 | >1 |\n
| NCHW <====> FRACTAL_Z | float16, bfloat16, uint16, int16 | 16 | >1 |\n
| HWCN ====> FRACTAL_Z | float16, bfloat16, uint16, int16 | 16 | >1 |\n
| NCDHW <====> FRACTAL_Z_3D | float32, uint32, int32 | 8 | >1 |\n
| NCDHW <====> FRACTAL_Z_3D | float16, bfloat16, uint16, int16 | 16 | >1 |\n
| FRACTAL_Z_3D ====> DHWCN | float32, uint32, int32 | 8 | >1 |\n
| FRACTAL_Z_3D ====> DHWCN | float16, bfloat16, uint16, int16 | 16 | >1 |\n
| NCHW ====> FRACTAL_Z_C04 | float16, bfloat16 | 16 | 1 |\n
| FRACTAL_Z_C04 ====> NCHW | float32 | 16 | 1 |\n
| ND <====> FRACTAL_NZ_C0_16 | float32, uint32, int32 | 16 | 1 |\n
| ND <====> FRACTAL_NZ_C0_32 | float4_e2m1 | 32 | 1 |\n
| FRACTAL_NZ_C0_2 ====> ND | float32, int32 | 2 | 1 |\n
| FRACTAL_NZ_C0_4 ====> ND | float32, int32 | 4 | 1 |\n
*\n
*
*/
REG_OP(TransData)
.INPUT(src, TensorType({TensorType::BasicType(), DT_FLOAT4_E1M2, DT_FLOAT4_E2M1}))
.OUTPUT(dst, TensorType({TensorType::BasicType(), DT_FLOAT4_E1M2, DT_FLOAT4_E2M1}))
.REQUIRED_ATTR(src_format, String)
.REQUIRED_ATTR(dst_format, String)
.ATTR(src_subformat, Int, 0)
.ATTR(dst_subformat, Int, 0)
.ATTR(groups, Int, 1)
.OP_END_FACTORY_REG(TransData)
}
#endif