mirror of https://github.com/alibaba/MNN.git
110 lines
4.4 KiB
C++
110 lines
4.4 KiB
C++
//
|
|
// GeometryELU.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2020/07/23.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "geometry/GeometryComputer.hpp"
|
|
#include "core/OpCommonUtils.hpp"
|
|
#include "geometry/GeometryComputerUtils.hpp"
|
|
|
|
namespace MNN {
|
|
|
|
static void initTensor(std::shared_ptr<Tensor> tensor, Tensor* input) {
|
|
TensorUtils::copyShape(input, tensor.get(), true, true);
|
|
}
|
|
|
|
class GeometryELU : public GeometryComputer {
|
|
public:
|
|
virtual bool onCompute(const Op* op, const std::vector<Tensor*>& inputs, const std::vector<Tensor*>& outputs, Context& context, CommandBuffer& res) const override {
|
|
MNN_ASSERT(1 == inputs.size());
|
|
MNN_ASSERT(1 == outputs.size());
|
|
auto input = inputs[0];
|
|
auto output = outputs[0];
|
|
// ELU : y = x > 0 ? x : alpha * (exp(x) - 1)
|
|
// exp + sub + mul : y1 = alhpa * (exp(x) - 1)
|
|
// exp
|
|
std::shared_ptr<Tensor> expValue(new Tensor);
|
|
{
|
|
initTensor(expValue, input);
|
|
auto cmd = GeometryComputerUtils::makeUnary(UnaryOpOperation_EXP, input, expValue.get());
|
|
res.extras.emplace_back(expValue);
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
// sub
|
|
std::shared_ptr<Tensor> subValue(new Tensor);
|
|
{
|
|
auto oneConst = context.allocConst(op, {}, halide_type_of<float>());
|
|
oneConst->host<float>()[0] = 1.0;
|
|
initTensor(subValue, input);
|
|
auto cmd = GeometryComputerUtils::makeBinary(BinaryOpOperation_SUB, expValue.get(), oneConst.get(), subValue.get());
|
|
res.extras.emplace_back(subValue);
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
// mul
|
|
std::shared_ptr<Tensor> mulValue(new Tensor);
|
|
{
|
|
auto alphaConst = context.allocConst(op, {}, halide_type_of<float>());
|
|
float alpha = 0.0;
|
|
if (op->type() == OpType_ELU) {
|
|
alpha = op->main_as_ELU()->alpha();
|
|
} else if (op->type() == OpType_Selu){
|
|
alpha = op->main_as_Selu()->alpha() *
|
|
op->main_as_Selu()->scale();
|
|
}
|
|
alphaConst->host<float>()[0] = alpha;
|
|
initTensor(mulValue, input);
|
|
auto cmd = GeometryComputerUtils::makeBinary(BinaryOpOperation_MUL, subValue.get(), alphaConst.get(), mulValue.get());
|
|
res.extras.emplace_back(mulValue);
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
// compare + select : y = x > 0 ? x : y1
|
|
// compare
|
|
std::shared_ptr<Tensor> compValue(new Tensor);
|
|
{
|
|
auto zeroConst = context.allocConst(op, {}, halide_type_of<float>());
|
|
zeroConst->host<float>()[0] = 0;
|
|
TensorUtils::copyShape(input, compValue.get(), true, true);
|
|
compValue->buffer().type = halide_type_of<int32_t>();
|
|
auto cmd = GeometryComputerUtils::makeBinary(BinaryOpOperation_GREATER, input, zeroConst.get(), compValue.get());
|
|
res.extras.emplace_back(compValue);
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
std::shared_ptr<Tensor> scaleValue(new Tensor);
|
|
{
|
|
if (op->type() == OpType_Selu) {
|
|
auto scaleConst = context.allocConst(op, {}, halide_type_of<float>());
|
|
float scale = op->main_as_Selu()->scale();
|
|
scaleConst->host<float>()[0] = scale;
|
|
initTensor(scaleValue, input);
|
|
auto cmd = GeometryComputerUtils::makeBinary(BinaryOpOperation_MUL, input, scaleConst.get(), scaleValue.get());
|
|
res.extras.emplace_back(scaleValue);
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
}
|
|
// select
|
|
{
|
|
flatbuffers::FlatBufferBuilder builder;
|
|
OpBuilder opBuilder(builder);
|
|
opBuilder.add_type(OpType_Select);
|
|
builder.Finish(opBuilder.Finish());
|
|
auto y0 = op->type() == OpType_ELU ? input : scaleValue.get();
|
|
auto cmd = GeometryComputerUtils::makeCommand(builder, {compValue.get(), y0, mulValue.get()}, {output});
|
|
res.command.emplace_back(std::move(cmd));
|
|
}
|
|
return true;
|
|
}
|
|
};
|
|
|
|
static void _create() {
|
|
std::shared_ptr<GeometryComputer> comp(new GeometryELU);
|
|
GeometryComputer::registerGeometryComputer(comp, {OpType_ELU, OpType_Selu});
|
|
}
|
|
|
|
REGISTER_GEOMETRY(GeometryELU, _create);
|
|
|
|
} // namespace MNN
|
|
|