* 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_math_proto_extend.h"
#include "math/mul/op_graph/mul_proto.h"
#include "math/add/op_graph/add_proto.h"
#include "math/cast/op_graph/cast_proto.h"
#include "math/shape/op_graph/shape_proto.h"
using namespace ge;
namespace domi {
static constexpr int KDtypeNotProvided = -1;
static Status ParseParamsRandomUniformLike(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(GetOpName(op_dest).c_str(), "Failed to dynamically cast op source to NodeProto.");
return FAILED;
}
op_dest.DynamicInputRegister("x", 1);
op_dest.DynamicOutputRegister("y", 1);
op_dest.SetAttr("original_type", "ai.onnx::11::RandomUniformLike");
int dtype = KDtypeNotProvided;
float low = 0.0f;
float high = 1.0f;
int seed = 0;
bool has_dtype_attr = false;
for (const auto& attr : node->attribute()) {
if (attr.name() == "dtype") {
dtype = attr.i();
has_dtype_attr = true;
} else if (attr.name() == "high") {
high = attr.f();
} else if (attr.name() == "low") {
low = attr.f();
} else if (attr.name() == "seed") {
seed = (int)attr.f();
}
}
op_dest.SetAttr("name", node->name());
op_dest.SetAttr("high", high);
op_dest.SetAttr("low", low);
op_dest.SetAttr("seed", seed);
op_dest.SetAttr("dtype", dtype);
op_dest.SetAttr("has_dtype_attr", has_dtype_attr);
return SUCCESS;
}
static ge::DataType ResolveOutputDtype(const ge::Operator& op)
{
int dtype = KDtypeNotProvided;
op.GetAttr("dtype", dtype);
bool has_dtype_attr = false;
op.GetAttr("has_dtype_attr", has_dtype_attr);
if (has_dtype_attr) {
return GetOmDtypeFromOnnxDtype(dtype);
}
ge::TensorDesc input_desc = op.GetInputDesc(0);
ge::DataType in_dtype = input_desc.GetDataType();
if (in_dtype != ge::DT_UNDEFINED) {
return in_dtype;
}
OP_LOGW(GetOpName(op).c_str(), "RandomUniformLike has no dtype attr.");
return ge::DT_FLOAT;
}
static Status ParseOpToGraphRandomUniformLike(const ge::Operator& op, ge::Graph& graph)
{
std::string ori_name;
if (op.GetAttr("name", ori_name) != SUCCESS) {
OP_LOGE(GetOpName(op).c_str(), "Unable to retrieve identifier from operator.");
return FAILED;
}
float low = 0.0f;
op.GetAttr("low", low);
float high = 1.0f;
op.GetAttr("high", high);
ge::DataType out_dtype = ResolveOutputDtype(op);
const std::set<ge::DataType> supported_dtypes = {ge::DT_FLOAT, ge::DT_FLOAT16, ge::DT_DOUBLE};
if (supported_dtypes.find(out_dtype) == supported_dtypes.end()) {
OP_LOGE(GetOpName(op).c_str(),
"RandomUniformLike output dtype[%d] not supported, only support float/float16/double.",
static_cast<int>(out_dtype));
return FAILED;
}
int seed = 0;
op.GetAttr("seed", seed);
float delta = high - low;
ge::Tensor scalar_mean = CreateScalar(low, ge::DT_FLOAT);
ge::Tensor scalar_scale = CreateScalar(delta, ge::DT_FLOAT);
auto data0 = op::Data((ori_name + "_data0").c_str()).set_attr_index(0);
auto shape_op = op::Shape((ori_name + "_shape").c_str()).set_input_x(data0).set_attr_dtype(ge::DT_INT32);
auto random_op = op::RandomUniform((ori_name + "_random_uniform").c_str())
.set_input_shape(shape_op)
.set_attr_dtype(out_dtype)
.set_attr_seed(seed)
.set_attr_seed2(seed);
auto const_scale = op::Const((ori_name + "_const_scale").c_str()).set_attr_value(scalar_scale);
auto const_mean = op::Const((ori_name + "_const_mean").c_str()).set_attr_value(scalar_mean);
auto cast_mean = op::Cast((ori_name + "_cast_mean").c_str()).set_input_x(const_mean).set_attr_dst_type(out_dtype);
auto
cast_scale = op::Cast((ori_name + "_cast_scale").c_str()).set_input_x(const_scale).set_attr_dst_type(out_dtype);
auto mul_op = op::Mul((ori_name + "_mul").c_str()).set_input_x1(random_op).set_input_x2(cast_scale);
auto add_op = op::Add((ori_name + "_add").c_str()).set_input_x1(mul_op).set_input_x2(cast_mean);
std::vector<ge::Operator> inputs{data0};
std::vector<std::pair<ge::Operator, std::vector<size_t>>> outputs;
outputs.emplace_back(add_op, std::vector<std::size_t>{0});
graph.SetInputs(inputs).SetOutputs(outputs);
return SUCCESS;
}
REGISTER_CUSTOM_OP("PartitionedCall")
.FrameworkType(ONNX)
.OriginOpType(
{ge::AscendString("ai.onnx::8::RandomUniformLike"), ge::AscendString("ai.onnx::9::RandomUniformLike"),
ge::AscendString("ai.onnx::10::RandomUniformLike"), ge::AscendString("ai.onnx::11::RandomUniformLike"),
ge::AscendString("ai.onnx::12::RandomUniformLike"), ge::AscendString("ai.onnx::13::RandomUniformLike"),
ge::AscendString("ai.onnx::14::RandomUniformLike"), ge::AscendString("ai.onnx::15::RandomUniformLike"),
ge::AscendString("ai.onnx::16::RandomUniformLike"), ge::AscendString("ai.onnx::17::RandomUniformLike"),
ge::AscendString("ai.onnx::18::RandomUniformLike")})
.ParseParamsFn(ParseParamsRandomUniformLike)
.ParseOpToGraphFn(ParseOpToGraphRandomUniformLike)
.ImplyType(ImplyType::TVM);
}
