wehub-resource-sync
135 lines
4.6 KiB
C++
135 lines
4.6 KiB
C++
//
|
|
// UnaryTorch.cpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2021/05/11.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include <stdio.h>
|
|
#include "torchOpConverter.hpp"
|
|
|
|
DECLARE_OP_CONVERTER(UnaryTorch);
|
|
|
|
MNN::OpType UnaryTorch::opType() {
|
|
return MNN::OpType_UnaryOp;
|
|
}
|
|
|
|
MNN::OpParameter UnaryTorch::type() {
|
|
return MNN::OpParameter_UnaryOp;
|
|
}
|
|
|
|
std::vector<int> UnaryTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void UnaryTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
static std::map<std::string, MNN::UnaryOpOperation> gMaps{
|
|
{"abs", MNN::UnaryOpOperation_ABS},
|
|
{"neg", MNN::UnaryOpOperation_NEG},
|
|
{"floor", MNN::UnaryOpOperation_FLOOR},
|
|
{"ceil", MNN::UnaryOpOperation_CEIL},
|
|
{"square", MNN::UnaryOpOperation_SQUARE},
|
|
{"sqrt", MNN::UnaryOpOperation_SQRT},
|
|
{"rsqrt", MNN::UnaryOpOperation_RSQRT},
|
|
{"exp", MNN::UnaryOpOperation_EXP},
|
|
{"log", MNN::UnaryOpOperation_LOG},
|
|
{"sin", MNN::UnaryOpOperation_SIN},
|
|
{"cos", MNN::UnaryOpOperation_COS},
|
|
{"tan", MNN::UnaryOpOperation_TAN},
|
|
{"asin", MNN::UnaryOpOperation_ASIN},
|
|
{"acos", MNN::UnaryOpOperation_ACOS},
|
|
{"atan", MNN::UnaryOpOperation_ATAN},
|
|
{"reciprocal", MNN::UnaryOpOperation_RECIPROCAL},
|
|
{"log1p", MNN::UnaryOpOperation_LOG1P},
|
|
{"bernoulli", MNN::UnaryOpOperation_BNLL},
|
|
{"acosh", MNN::UnaryOpOperation_ACOSH},
|
|
{"sinh", MNN::UnaryOpOperation_SINH},
|
|
{"asinh", MNN::UnaryOpOperation_ASINH},
|
|
{"atanh", MNN::UnaryOpOperation_ATANH},
|
|
{"sign", MNN::UnaryOpOperation_SIGN},
|
|
{"round", MNN::UnaryOpOperation_ROUND},
|
|
{"cosh", MNN::UnaryOpOperation_COSH},
|
|
{"erf", MNN::UnaryOpOperation_ERF},
|
|
{"erfc", MNN::UnaryOpOperation_ERFC},
|
|
{"erfinv", MNN::UnaryOpOperation_ERFINV},
|
|
{"expm1", MNN::UnaryOpOperation_EXPM1},
|
|
{"tanh", MNN::UnaryOpOperation_TANH},
|
|
{"sigmoid", MNN::UnaryOpOperation_SIGMOID},
|
|
{"hardswish", MNN::UnaryOpOperation_HARDSWISH},
|
|
{"gelu", MNN::UnaryOpOperation_GELU_STANDARD},
|
|
};
|
|
auto param = new MNN::UnaryOpT;
|
|
std::string opType = getRealOpType(node);
|
|
param->opType = gMaps[opType];
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(UnaryTorch, abs);
|
|
REGISTER_CONVERTER(UnaryTorch, neg);
|
|
REGISTER_CONVERTER(UnaryTorch, floor);
|
|
REGISTER_CONVERTER(UnaryTorch, ceil);
|
|
REGISTER_CONVERTER(UnaryTorch, square);
|
|
REGISTER_CONVERTER(UnaryTorch, sqrt);
|
|
REGISTER_CONVERTER(UnaryTorch, rsqrt);
|
|
REGISTER_CONVERTER(UnaryTorch, exp);
|
|
REGISTER_CONVERTER(UnaryTorch, log);
|
|
REGISTER_CONVERTER(UnaryTorch, sin);
|
|
REGISTER_CONVERTER(UnaryTorch, cos);
|
|
REGISTER_CONVERTER(UnaryTorch, tan);
|
|
REGISTER_CONVERTER(UnaryTorch, asin);
|
|
REGISTER_CONVERTER(UnaryTorch, acos);
|
|
REGISTER_CONVERTER(UnaryTorch, atan);
|
|
REGISTER_CONVERTER(UnaryTorch, reciprocal);
|
|
REGISTER_CONVERTER(UnaryTorch, log1p);
|
|
REGISTER_CONVERTER(UnaryTorch, bernoulli);
|
|
REGISTER_CONVERTER(UnaryTorch, acosh);
|
|
REGISTER_CONVERTER(UnaryTorch, sinh);
|
|
REGISTER_CONVERTER(UnaryTorch, asinh);
|
|
REGISTER_CONVERTER(UnaryTorch, atanh);
|
|
REGISTER_CONVERTER(UnaryTorch, sign);
|
|
REGISTER_CONVERTER(UnaryTorch, round);
|
|
REGISTER_CONVERTER(UnaryTorch, cosh);
|
|
REGISTER_CONVERTER(UnaryTorch, erf);
|
|
REGISTER_CONVERTER(UnaryTorch, erfc);
|
|
REGISTER_CONVERTER(UnaryTorch, erfinv);
|
|
REGISTER_CONVERTER(UnaryTorch, expm1);
|
|
REGISTER_CONVERTER(UnaryTorch, tanh);
|
|
REGISTER_CONVERTER(UnaryTorch, sigmoid);
|
|
REGISTER_CONVERTER(UnaryTorch, hardswish);
|
|
REGISTER_CONVERTER(UnaryTorch, gelu);
|
|
|
|
// TODO: silu will impl as unary ?
|
|
DECLARE_OP_CONVERTER(ExtraUnaryTorch);
|
|
|
|
MNN::OpType ExtraUnaryTorch::opType() {
|
|
return MNN::OpType_Extra;
|
|
}
|
|
MNN::OpParameter ExtraUnaryTorch::type() {
|
|
return MNN::OpParameter_Extra;
|
|
}
|
|
std::vector<int> ExtraUnaryTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void ExtraUnaryTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto extra = new MNN::ExtraT;
|
|
extra->engine = "Torch";
|
|
auto type = getRealOpType(node);
|
|
extra->type = type;
|
|
dstOp->main.value = extra;
|
|
if (type == "softplus") {
|
|
extra->attr.resize(2);
|
|
extra->attr[0].reset(new MNN::AttributeT);
|
|
extra->attr[0]->key = "beta";
|
|
extra->attr[0]->i = getValue<int64_t>(node->input(1));
|
|
extra->attr[1].reset(new MNN::AttributeT);
|
|
extra->attr[1]->key = "threshold";
|
|
extra->attr[1]->i = getValue<int64_t>(node->input(2));
|
|
}
|
|
}
|
|
|
|
REGISTER_CONVERTER(ExtraUnaryTorch, silu);
|
|
REGISTER_CONVERTER(ExtraUnaryTorch, softplus);
|
|
REGISTER_CONVERTER(ExtraUnaryTorch, bitwise_not);
|