frigate/docker/rocm/migraphx/onnx/parse_reduce_op.cpp
WhiteWolf84 7eefb89bf6 upload
2025-02-03 22:01:20 +01:00

234 lines
9.0 KiB
C++

/*
* The MIT License (MIT)
*
* Copyright (c) 2015-2024 Advanced Micro Devices, Inc. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include <migraphx/onnx/op_parser.hpp>
#include <migraphx/ranges.hpp>
#include <migraphx/instruction.hpp>
#include <migraphx/make_op.hpp>
#include <migraphx/onnx/checks.hpp>
namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace onnx {
template <typename Derived>
struct reduce_parser : op_parser<Derived>
{
instruction_ref parse_reduce_oper(const std::string& op_name,
const onnx_parser& parser,
onnx_parser::node_info info,
std::vector<instruction_ref> args) const
{
auto constant_axes = parse_constant_axes(args, info);
int noop_with_empty_axes =
parse_attribute<int>("noop_with_empty_axes", parser, info).value_or(0);
int keep_dims = parse_attribute<int>("keepdims", parser, info).value_or(1);
std::vector<int64_t> all_axes(args.front()->get_shape().ndim());
std::iota(all_axes.begin(), all_axes.end(), 0);
// Handle axes attribute, constant input axes, and missing both attribute and input cases
if(constant_axes.has_value())
{
if(noop_with_empty_axes != 0 and constant_axes->empty())
return args[0];
if(noop_with_empty_axes == 0 and constant_axes->empty())
constant_axes = all_axes;
auto reduce =
info.add_instruction(make_op(op_name, {{"axes", *constant_axes}}), args[0]);
if(keep_dims == 0)
return info.add_instruction(make_op("squeeze", {{"axes", *constant_axes}}), reduce);
return reduce;
}
// Handle variable input axes
if(keep_dims == 0)
MIGRAPHX_THROW("Keepdims not supported with runtime provided axes");
// Empty axes attribute indicates to the operator to look for axes in the inputs
// If the input axes are empty, the default behavior of reduce_op is to be an
// identity operator
auto reduce_op = make_op(op_name, {{"axes", {}}});
if(noop_with_empty_axes != 0)
return info.add_instruction(reduce_op, args);
if(args[1]->get_shape().dynamic())
{
auto reduce_input_axes = info.add_instruction(reduce_op, args);
auto all_axes_lit = info.add_literal(
literal{shape{shape::type_t::int64_type, {all_axes.size()}}, all_axes});
auto reduce_all_axes = info.add_instruction(reduce_op, args[0], all_axes_lit);
auto zero = info.add_literal(literal{shape{shape::type_t::int64_type}, {0u}});
auto axes_size = info.add_instruction(make_op("dimensions_of", {{"end", 1}}), args[1]);
auto is_axes_empty = info.add_instruction(make_op("equal"), axes_size, zero);
return info.add_instruction(
make_op("where"), is_axes_empty, reduce_all_axes, reduce_input_axes);
}
else if(args[1]->get_shape().elements() == 0)
{
auto all_axes_lit = info.add_literal(
literal{shape{shape::type_t::int64_type, {all_axes.size()}}, all_axes});
return info.add_instruction(reduce_op, args[0], all_axes_lit);
}
else
{
return info.add_instruction(reduce_op, args);
}
}
private:
template <typename T>
std::optional<T> parse_attribute(const std::string& attribute_name,
const onnx_parser& parser,
onnx_parser::node_info& info) const
{
if(not contains(info.attributes, attribute_name))
return std::nullopt;
return parser.parse_value(info.attributes[attribute_name]).at<T>();
}
std::optional<std::vector<int64_t>> parse_constant_axes(std::vector<instruction_ref>& args,
onnx_parser::node_info& info) const
{
std::vector<int64_t> axes;
if(args.size() == 2)
{
if(not args[1]->can_eval())
return std::nullopt;
args[1]->eval().visit([&](auto s) { axes.assign(s.begin(), s.end()); });
}
else if(contains(info.attributes, "axes"))
{
auto&& attr_axes = info.attributes["axes"].ints();
axes.assign(attr_axes.begin(), attr_axes.end());
}
return axes;
}
};
struct parse_reduce_op : reduce_parser<parse_reduce_op>
{
std::vector<op_desc> operators() const
{
return {{"ReduceMax", "reduce_max"},
{"ReduceMean", "reduce_mean"},
{"ReduceMin", "reduce_min"},
{"ReduceProd", "reduce_prod"},
{"ReduceSum", "reduce_sum"}};
}
instruction_ref parse(const op_desc& opd,
const onnx_parser& parser,
onnx_parser::node_info info,
std::vector<instruction_ref> args) const
{
return parse_reduce_oper(opd.op_name, parser, std::move(info), std::move(args));
}
};
struct parse_reduce_l1 : reduce_parser<parse_reduce_l1>
{
std::vector<op_desc> operators() const { return {{"ReduceL1"}}; }
instruction_ref parse(const op_desc& /*opd*/,
const onnx_parser& parser,
onnx_parser::node_info info,
std::vector<instruction_ref> args) const
{
args[0] = info.add_instruction(make_op("abs"), args[0]);
return parse_reduce_oper("reduce_sum", parser, std::move(info), std::move(args));
}
};
struct parse_reduce_l2 : reduce_parser<parse_reduce_l2>
{
std::vector<op_desc> operators() const { return {{"ReduceL2"}}; }
instruction_ref parse(const op_desc& /*opd*/,
const onnx_parser& parser,
const onnx_parser::node_info& info,
std::vector<instruction_ref> args) const
{
args[0] = info.add_instruction(make_op("mul"), args[0], args[0]);
auto sum_ins = parse_reduce_oper("reduce_sum", parser, info, std::move(args));
return info.add_instruction(make_op("sqrt"), sum_ins);
}
};
struct parse_reduce_log_sum : reduce_parser<parse_reduce_log_sum>
{
std::vector<op_desc> operators() const { return {{"ReduceLogSum"}}; }
instruction_ref parse(const op_desc& /*opd*/,
const onnx_parser& parser,
const onnx_parser::node_info& info,
std::vector<instruction_ref> args) const
{
auto sum_ins = parse_reduce_oper("reduce_sum", parser, info, std::move(args));
return info.add_instruction(make_op("log"), sum_ins);
}
};
struct parse_reduce_log_sum_exp : reduce_parser<parse_reduce_log_sum_exp>
{
std::vector<op_desc> operators() const { return {{"ReduceLogSumExp"}}; }
instruction_ref parse(const op_desc& /*opd*/,
const onnx_parser& parser,
const onnx_parser::node_info& info,
std::vector<instruction_ref> args) const
{
args[0] = info.add_instruction(make_op("exp"), args[0]);
auto sum_ins = parse_reduce_oper("reduce_sum", parser, info, std::move(args));
return info.add_instruction(make_op("log"), sum_ins);
}
};
struct parse_reduce_sum_square : reduce_parser<parse_reduce_sum_square>
{
std::vector<op_desc> operators() const { return {{"ReduceSumSquare"}}; }
instruction_ref parse(const op_desc& /*opd*/,
const onnx_parser& parser,
onnx_parser::node_info info,
std::vector<instruction_ref> args) const
{
args[0] = info.add_instruction(make_op("mul"), args[0], args[0]);
return parse_reduce_oper("reduce_sum", parser, std::move(info), std::move(args));
}
};
} // namespace onnx
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx