* 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 "stub_ops.h"
#include "conversion/pad_v3/op_graph/pad_v3_proto.h"
#include "conversion/mirror_pad/op_graph/mirror_pad_proto.h"
using namespace std;
using namespace ge;
using ge::Operator;
namespace domi {
static Status parse_params_pad_v11(const Message* op_src, ge::Operator& op_dest)
{
const ge::onnx::NodeProto* node = reinterpret_cast<const ge::onnx::NodeProto*>(op_src);
if (nullptr == node) {
OP_LOGE(GetOpName(op_dest).c_str(), "Dynamic cast op_src to NodeProto failed.");
return FAILED;
}
std::string mode_value = "constant";
for (const auto& attr : node->attribute()) {
if (attr.name() == "mode" && attr.type() == ge::onnx::AttributeProto::STRING) {
mode_value = attr.s();
}
}
op_dest.SetAttr("paddings_contiguous", false);
op_dest.SetAttr("mode", mode_value);
return SUCCESS;
}
static Status parse_params_pad_v9(const Message* op_src, ge::Operator& op_dest)
{
const ge::onnx::NodeProto* node = reinterpret_cast<const ge::onnx::NodeProto*>(op_src);
if (node == nullptr) {
OP_LOGE(GetOpName(op_dest).c_str(), "Dynamic cast op_src to NodeProto failed.");
return FAILED;
}
op_dest.SetAttr("name", node->name());
op_dest.DynamicInputRegister("x", 1);
op_dest.DynamicOutputRegister("output", 1);
op_dest.SetAttr("original_type", "ai.onnx::9::Pad");
std::vector<int32_t> v_pads;
bool set_pads_flag = false;
float value = 0.0;
std::string mode_value = "constant";
for (const auto& attr : node->attribute()) {
if (attr.name() == "mode" && attr.type() == ge::onnx::AttributeProto::STRING) {
mode_value = attr.s();
} else if (attr.name() == "pads") {
set_pads_flag = true;
unsigned int len = attr.ints_size();
if (len & 1) {
OP_LOGE(
GetOpName(op_dest).c_str(),
"the length of pads must be even, such as [x1_begin, x2_begin...x1_end, x2_end,...]");
return FAILED;
}
unsigned int half = len / 2;
for (unsigned int i = 0; i < half; i++) {
v_pads.push_back(static_cast<int32_t>(attr.ints(i)));
v_pads.push_back(static_cast<int32_t>(attr.ints(i + half)));
}
} else if (attr.name() == "value" && attr.type() == ge::onnx::AttributeProto::FLOAT) {
value = attr.f();
}
}
if (!set_pads_flag) {
OP_LOGE(GetOpName(op_dest).c_str(), "Dynamic cast op_src to NodeProto failed.");
return FAILED;
}
std::vector<int64_t> dims = {(int64_t)v_pads.size()};
ge::Tensor tensor1 = Vec2Tensor(v_pads, dims, ge::DT_INT32);
op_dest.SetAttr("paddings", tensor1);
ge::Tensor scalar_const_value = CreateScalar(value, ge::DT_FLOAT);
op_dest.SetAttr("constant_values", scalar_const_value);
op_dest.SetAttr("mode", mode_value);
return SUCCESS;
}
static Status ParseOpToGraphPad(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 value1;
if (op.GetAttr("paddings", value1) != SUCCESS) {
return FAILED;
}
ge::Tensor value2;
if (op.GetAttr("constant_values", value2) != SUCCESS) {
return FAILED;
}
std::string mode = "constant";
std::string mode_reflect = "REFLECT";
op.GetAttr("mode", mode);
auto data1 = op::Const((ori_name + "_data1").c_str()).set_attr_value(value1);
std::vector<Operator> inputs{data0};
std::vector<std::pair<Operator, std::vector<size_t> > > output_indexs;
if (mode == "reflect") {
auto mirror_pad = op::MirrorPad((ori_name + "_MirrorPad").c_str())
.set_input_x(data0)
.set_input_paddings(data1)
.set_attr_mode(mode_reflect);
output_indexs.emplace_back(mirror_pad, vector<std::size_t>{0});
} else {
auto data2 = op::Const((ori_name + "_data2").c_str()).set_attr_value(value2);
auto pad_v3 = op::PadV3((ori_name + "_PadV3").c_str())
.set_input_x(data0)
.set_input_paddings(data1)
.set_input_constant_values(data2)
.set_attr_mode(mode);
output_indexs.emplace_back(pad_v3, 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::Pad"),
ge::AscendString("ai.onnx::9::Pad"),
ge::AscendString("ai.onnx::10::Pad")})
.ParseParamsFn(parse_params_pad_v9)
.ParseOpToGraphFn(ParseOpToGraphPad)
.ImplyType(ImplyType::TVM);
REGISTER_CUSTOM_OP("PadV3")
.FrameworkType(ONNX)
.OriginOpType({ge::AscendString("ai.onnx::11::Pad"),
ge::AscendString("ai.onnx::12::Pad"),
ge::AscendString("ai.onnx::13::Pad"),
ge::AscendString("ai.onnx::14::Pad"),
ge::AscendString("ai.onnx::15::Pad"),
ge::AscendString("ai.onnx::16::Pad"),
ge::AscendString("ai.onnx::17::Pad"),
ge::AscendString("ai.onnx::18::Pad")})
.ParseParamsFn(parse_params_pad_v11)
.ImplyType(ImplyType::TVM);
}
