* 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.
*/
#include "onnx_common.h"
#include "op_nn_proto_extend.h"
using namespace std;
using namespace ge;
using ge::Operator;
namespace domi {
enum OnnxDataType {
UNDEFINED = 0,
FLOAT = 1,
UINT_8 = 2,
INT_8 = 3,
UINT_16 = 4,
INT_16 = 5,
INT_32 = 6,
INT_64 = 7,
STRING = 8,
BOOL = 9,
FLOAT_16 = 10,
DOUBLE = 11,
UINT_32 = 12,
UINT_64 = 13,
BFLOAT_16 = 16
};
static DataType GetGeDataType(int32_t data_type)
{
static DataType onnxToGeDataType[UINT_64 + 1] = {DT_UNDEFINED, DT_FLOAT, DT_UINT8, DT_INT8, DT_UINT16,
DT_INT16, DT_INT32, DT_INT64, DT_STRING, DT_BOOL,
DT_FLOAT16, DT_DOUBLE, DT_UINT32, DT_UINT64};
if (data_type == BFLOAT_16) {
return DT_BF16;
} else if (data_type > UINT_64 || data_type < 0) {
return DT_UNDEFINED;
}
return onnxToGeDataType[data_type];
}
static uint8_t* ParseTensorValue(const ge::onnx::TensorProto& tp)
{
const uint8_t* data = nullptr;
auto data_type = tp.data_type();
OP_LOGD("ConstantOfShape", "Datatype[%d.]", data_type);
switch (data_type) {
case ge::onnx::TensorProto::DataType::TensorProto_DataType_INT64:
data = reinterpret_cast<const uint8_t*>(tp.int64_data().data());
break;
case ge::onnx::TensorProto::DataType::TensorProto_DataType_INT32:
data = reinterpret_cast<const uint8_t*>(tp.int32_data().data());
break;
case ge::onnx::TensorProto::DataType::TensorProto_DataType_FLOAT:
data = reinterpret_cast<const uint8_t*>(tp.float_data().data());
break;
case ge::onnx::TensorProto::DataType::TensorProto_DataType_DOUBLE:
data = reinterpret_cast<const uint8_t*>(tp.double_data().data());
break;
default:
OP_LOGE("ConstantOfShape", "Datatype[%d] don't support.", data_type);
}
return const_cast<uint8_t*>(data);
}
static Status ParseParamsConstantOfShape(const Message* op_src, ge::Operator& op_dest)
{
const ge::onnx::NodeProto* node = dynamic_cast<const ge::onnx::NodeProto*>(op_src);
if (node == nullptr) {
OP_LOGE("ConstantOfShape", "Dynamic cast op_src to NodeProto failed.");
return FAILED;
}
op_dest.DynamicInputRegister("args", 1);
op_dest.DynamicOutputRegister("output", 1);
op_dest.SetAttr("original_type", "ai.onnx::11::ConstantOfShape");
ge::TensorDesc tensorDesc;
vector<int64_t> dims = {};
ge::Shape shape(dims);
tensorDesc.SetShape(shape);
tensorDesc.SetDataType(DT_FLOAT);
tensorDesc.SetFormat(ge::FORMAT_NCHW);
tensorDesc.SetOriginShape(shape);
tensorDesc.SetOriginFormat(ge::FORMAT_NCHW);
size_t size = sizeof(float);
uint8_t* data = nullptr;
vector<float> data_dim = {0};
data = reinterpret_cast<uint8_t*>(data_dim.data());
for (const auto& attr : node->attribute()) {
if (attr.name() == "value" && attr.type() == ge::onnx::AttributeProto::TENSOR) {
if (attr.t().raw_data() != "") {
auto value = const_cast<char*>(attr.t().raw_data().data());
data = reinterpret_cast<uint8_t*>(value);
} else {
data = ParseTensorValue(attr.t());
}
DataType datatype0 = GetGeDataType(attr.t().data_type());
size = GetSizeByDataType(datatype0);
tensorDesc.SetDataType(datatype0);
}
}
const ge::Tensor valueTensor(tensorDesc, data, size);
op_dest.SetAttr("value", valueTensor);
op_dest.SetAttr("name", node->name());
return SUCCESS;
}
static Status ParseOpToGraphConstantOfShape(const Operator& op, Graph& graph)
{
std::string ori_name;
if (op.GetAttr("name", ori_name) != SUCCESS) {
OP_LOGE(GetOpName(op).c_str(), "get name from op failed.");
return FAILED;
}
auto data0 = op::Data((ori_name + "_data0").c_str()).set_attr_index(0);
ge::Tensor value;
if (op.GetAttr("value", value) != SUCCESS) {
OP_LOGE("ConstantOfShape", "get value from op failed");
return FAILED;
}
auto data1 = op::Const((ori_name + "_data1").c_str()).set_attr_value(value);
auto fill = op::Fill((ori_name + "_Fill").c_str()).set_input_dims(data0).set_input_value(data1);
std::vector<Operator> inputs{data0};
std::vector<std::pair<Operator, std::vector<size_t> > > output_indexs;
output_indexs.emplace_back(fill, vector<std::size_t>{0});
graph.SetInputs(inputs).SetOutputs(output_indexs);
return SUCCESS;
}
REGISTER_CUSTOM_OP("PartitionedCall")
.FrameworkType(ONNX)
.OriginOpType({ge::AscendString("ai.onnx::8::ConstantOfShape"), ge::AscendString("ai.onnx::9::ConstantOfShape"),
ge::AscendString("ai.onnx::10::ConstantOfShape"), ge::AscendString("ai.onnx::11::ConstantOfShape"),
ge::AscendString("ai.onnx::12::ConstantOfShape"), ge::AscendString("ai.onnx::13::ConstantOfShape"),
ge::AscendString("ai.onnx::14::ConstantOfShape"), ge::AscendString("ai.onnx::15::ConstantOfShape"),
ge::AscendString("ai.onnx::16::ConstantOfShape"), ge::AscendString("ai.onnx::17::ConstantOfShape"),
ge::AscendString("ai.onnx::18::ConstantOfShape")})
.ParseParamsFn(ParseParamsConstantOfShape)
.ParseOpToGraphFn(ParseOpToGraphConstantOfShape)
.ImplyType(ImplyType::TVM);
}
