* Copyright (c) 2026 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 trace_def.cpp
* \brief Operator definition for trace operator
*
* REG_OP(Trace) from canndev (SE section 5.2):
* INPUT(x): DT_COMPLEX128, DT_COMPLEX64, DT_DOUBLE, DT_FLOAT, DT_FLOAT16,
* DT_BOOL, DT_INT8, DT_UINT8, DT_INT16, DT_UINT16, DT_INT32,
* DT_UINT32, DT_INT64, DT_UINT64, DT_BF16 (15 types in REG_OP)
* OUTPUT(y): DT_COMPLEX128, DT_COMPLEX64, DT_DOUBLE, DT_FLOAT, DT_FLOAT16,
* DT_INT64, DT_UINT64, DT_BF16 (8 possible output types)
*
* Ascend950 SIMT implementation supports 13 dtype combinations (see SE §5.7):
* complex64->complex64, float32->float32, float16->float16, bfloat16->bfloat16,
* int8->int64, int16->int64, int32->int64, int64->int64,
* uint8->int64, uint16->int64, uint32->int64, uint64->uint64, bool->int64
*
* Hardware limitation note (review issue #1):
* SE §5.2 REG_OP 声明包含 DT_DOUBLE 和 DT_COMPLEX128,但 Ascend950 aicore
* 不支持 double precision 运算(编译器报错 "double precision operation is
* not allowed in aicore function"),且 complex128 在 AscendC SIMT DTYPE_MAP
* 中映射为 "unknown"(无对应 C++ 类型)。因此本算子在 Ascend950 上仅注册
* 13 种可编译的 dtype 组合,DT_DOUBLE 和 DT_COMPLEX128 暂不支持。
* 若需支持这两种类型,需在其他支持 double 的芯片(如 ascend910b)上实现,
* 或使用软件模拟(性能代价较大)。
*
* uint64 output type note (review issue #2):
* SE §5.2 REG_OP OUTPUT 声明包含 DT_UINT64,SE §5.5 推导规则说 uint64→int64。
* 按 SE §5.2 REG_OP 定义为准,Output 保留 DT_UINT64(与 REG_OP 一致)。
*/
#include "register/op_def_registry.h"
namespace ops {
class Trace : public OpDef {
public:
explicit Trace(const char* name) : OpDef(name)
{
this->Input("x")
.ParamType(REQUIRED)
.DataType({ge::DT_COMPLEX64,
ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_BF16,
ge::DT_INT8, ge::DT_INT16, ge::DT_INT32, ge::DT_INT64,
ge::DT_UINT8, ge::DT_UINT16, ge::DT_UINT32, ge::DT_UINT64,
ge::DT_BOOL})
.Format({ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND})
.UnknownShapeFormat({ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND})
.AutoContiguous();
this->Output("y")
.ParamType(REQUIRED)
.DataType({ge::DT_COMPLEX64,
ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_BF16,
ge::DT_INT64, ge::DT_INT64, ge::DT_INT64, ge::DT_INT64,
ge::DT_INT64, ge::DT_INT64, ge::DT_INT64, ge::DT_UINT64,
ge::DT_INT64})
.Format({ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND})
.UnknownShapeFormat({ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND, ge::FORMAT_ND,
ge::FORMAT_ND})
.AutoContiguous();
OpAICoreConfig aicoreConfig;
aicoreConfig.DynamicCompileStaticFlag(true)
.DynamicFormatFlag(false)
.DynamicRankSupportFlag(true)
.DynamicShapeSupportFlag(true)
.NeedCheckSupportFlag(false)
.PrecisionReduceFlag(true)
.ExtendCfgInfo("opFile.value", "trace_apt");
this->AICore().AddConfig("ascend950", aicoreConfig);
}
};
OP_ADD(Trace);
}