root
/
MNN
mirror of https://github.com/alibaba/MNN.git
1
0
Fork 0
Code Issues Actions 7 Packages Projects Releases Wiki Activity
MNN/source/backend/coreml/execution/CoreMLUnary.cpp

268 lines
15 KiB
C++
Raw Blame History

//
// CoreMLUnary.cpp
// MNN
//
// Created by MNN on 2021/03/25.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "CoreMLUnary.hpp"
namespace MNN {
CoreMLUnary::CoreMLUnary(MNN::Backend *b, const MNN::Op *op, const std::vector<Tensor *> &inputs, const std::vector<MNN::Tensor *> &outputs) : CoreMLCommonExecution(b, op) {
initLayer();
}
ErrorCode CoreMLUnary::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
MNN_ASSERT(inputs.size() == 1 && outputs.size() == 1);
auto inputName = mCoreMLBackend->getTensorName(inputs[0]);
auto opType = mOp->main_as_UnaryOp()->opType();
#define SET_UNARY_PARAM \
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_UNARY; \
mLayer_->unary = mCoreMLBackend->create<CoreML__Specification__UnaryFunctionLayerParams>(); \
core_ml__specification__unary_function_layer_params__init(mLayer_->unary);
switch (opType) {
case UnaryOpOperation_ABS:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__ABS;
break;
case UnaryOpOperation_EXP:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__EXP;
break;
case UnaryOpOperation_SQRT:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__SQRT;
break;
case UnaryOpOperation_RSQRT:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__RSQRT;
break;
case UnaryOpOperation_LOG:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__LOG;
break;
case UnaryOpOperation_RECIPROCAL:
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__INVERSE;
break;
case UnaryOpOperation_SIN:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_SIN;
mLayer_->sin = mCoreMLBackend->create<CoreML__Specification__SinLayerParams>();
core_ml__specification__sin_layer_params__init(mLayer_->sin);
break;
case UnaryOpOperation_ASIN:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ASIN;
mLayer_->asin = mCoreMLBackend->create<CoreML__Specification__AsinLayerParams>();
core_ml__specification__asin_layer_params__init(mLayer_->asin);
break;
case UnaryOpOperation_SINH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_SINH;
mLayer_->sinh = mCoreMLBackend->create<CoreML__Specification__SinhLayerParams>();
core_ml__specification__sinh_layer_params__init(mLayer_->sinh);
break;
case UnaryOpOperation_ASINH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ASINH;
mLayer_->asinh = mCoreMLBackend->create<CoreML__Specification__AsinhLayerParams>();
core_ml__specification__asinh_layer_params__init(mLayer_->asinh);
break;
case UnaryOpOperation_COS:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_COS;
mLayer_->cos = mCoreMLBackend->create<CoreML__Specification__CosLayerParams>();
core_ml__specification__cos_layer_params__init(mLayer_->cos);
break;
case UnaryOpOperation_ACOS:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ACOS;
mLayer_->acos = mCoreMLBackend->create<CoreML__Specification__AcosLayerParams>();
core_ml__specification__acos_layer_params__init(mLayer_->acos);
break;
case UnaryOpOperation_COSH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_COSH;
mLayer_->cosh = mCoreMLBackend->create<CoreML__Specification__CoshLayerParams>();
core_ml__specification__cosh_layer_params__init(mLayer_->cosh);
break;
case UnaryOpOperation_ACOSH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ACOSH;
mLayer_->acosh = mCoreMLBackend->create<CoreML__Specification__AcoshLayerParams>();
core_ml__specification__acosh_layer_params__init(mLayer_->acosh);
break;
case UnaryOpOperation_TAN:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_TAN;
mLayer_->tan = mCoreMLBackend->create<CoreML__Specification__TanLayerParams>();
core_ml__specification__tan_layer_params__init(mLayer_->tan);
break;
case UnaryOpOperation_ATAN:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ATAN;
mLayer_->atan = mCoreMLBackend->create<CoreML__Specification__AtanLayerParams>();
core_ml__specification__atan_layer_params__init(mLayer_->atan);
break;
case UnaryOpOperation_TANH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_TANH;
mLayer_->tanh = mCoreMLBackend->create<CoreML__Specification__TanhLayerParams>();
core_ml__specification__tanh_layer_params__init(mLayer_->tanh);
break;
case UnaryOpOperation_ATANH:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ATANH;
mLayer_->atanh = mCoreMLBackend->create<CoreML__Specification__AtanhLayerParams>();
core_ml__specification__atanh_layer_params__init(mLayer_->atanh);
break;
case UnaryOpOperation_ERF:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ERF;
mLayer_->erf = mCoreMLBackend->create<CoreML__Specification__ErfLayerParams>();
core_ml__specification__erf_layer_params__init(mLayer_->erf);
break;
case UnaryOpOperation_CEIL:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_CEIL;
mLayer_->ceil = mCoreMLBackend->create<CoreML__Specification__CeilLayerParams>();
core_ml__specification__ceil_layer_params__init(mLayer_->ceil);
break;
case UnaryOpOperation_FLOOR:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_FLOOR;
mLayer_->floor = mCoreMLBackend->create<CoreML__Specification__FloorLayerParams>();
core_ml__specification__floor_layer_params__init(mLayer_->floor);
break;
case UnaryOpOperation_ROUND:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ROUND;
mLayer_->round = mCoreMLBackend->create<CoreML__Specification__RoundLayerParams>();
core_ml__specification__round_layer_params__init(mLayer_->round);
break;
case UnaryOpOperation_SIGN:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_SIGN;
mLayer_->sign = mCoreMLBackend->create<CoreML__Specification__SignLayerParams>();
core_ml__specification__sign_layer_params__init(mLayer_->sign);
break;
case UnaryOpOperation_SIGMOID:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ACTIVATION;
mLayer_->activation = mCoreMLBackend->create<CoreML__Specification__ActivationParams>();
core_ml__specification__activation_params__init(mLayer_->activation);
mLayer_->activation->nonlinearity_type_case = CORE_ML__SPECIFICATION__ACTIVATION_PARAMS__NONLINEARITY_TYPE_SIGMOID;
mLayer_->activation->sigmoid = mCoreMLBackend->create<CoreML__Specification__ActivationSigmoid>();
core_ml__specification__activation_sigmoid__init(mLayer_->activation->sigmoid);
break;
case UnaryOpOperation_LOG1P:
// y = log(x+1)
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__LOG;
mLayer_->unary->shift = 1;
break;
case UnaryOpOperation_SQUARE:
// y = x^2
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__POWER;
mLayer_->unary->alpha = 2;
break;
case UnaryOpOperation_NEG:
// y = (-x)^1
SET_UNARY_PARAM
mLayer_->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__POWER;
mLayer_->unary->scale = -1;
mLayer_->unary->alpha = 1;
break;
case UnaryOpOperation_HARDSWISH:
{
// (min(max(x + 3, 0), 6) * x) / 6
auto addLayer = mCoreMLBackend->create<CoreML__Specification__NeuralNetworkLayer>();
core_ml__specification__neural_network_layer__init(addLayer);
mCoreMLBackend->setLayerName(addLayer, "hardswish-add");
addLayer->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ADD;
addLayer->add = mCoreMLBackend->create<CoreML__Specification__AddLayerParams>();
core_ml__specification__add_layer_params__init(addLayer->add);
addLayer->add->alpha = 3.f;
std::string addOutput = inputName + "-add";
setLayerInputsAndOutputs(addLayer, {inputName}, {addOutput});
mCoreMLBackend->addLayer(addLayer);
auto reluLayer = mCoreMLBackend->create<CoreML__Specification__NeuralNetworkLayer>();
core_ml__specification__neural_network_layer__init(reluLayer);
mCoreMLBackend->setLayerName(reluLayer, "hardswish-relu");
reluLayer->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ACTIVATION;
reluLayer->activation = mCoreMLBackend->create<CoreML__Specification__ActivationParams>();
core_ml__specification__activation_params__init(reluLayer->activation);
reluLayer->activation->nonlinearity_type_case = CORE_ML__SPECIFICATION__ACTIVATION_PARAMS__NONLINEARITY_TYPE_RE_LU;
reluLayer->activation->relu = mCoreMLBackend->create<CoreML__Specification__ActivationReLU>();
core_ml__specification__activation_re_lu__init(reluLayer->activation->relu);
std::string reluOutput = addOutput + "-relu";
setLayerInputsAndOutputs(reluLayer, {addOutput}, {reluOutput});
mCoreMLBackend->addLayer(reluLayer);
auto thresholdLayer = mCoreMLBackend->create<CoreML__Specification__NeuralNetworkLayer>();
core_ml__specification__neural_network_layer__init(thresholdLayer);
mCoreMLBackend->setLayerName(thresholdLayer, "hardswish-threshold");
thresholdLayer->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_UNARY;
thresholdLayer->unary = mCoreMLBackend->create<CoreML__Specification__UnaryFunctionLayerParams>();
core_ml__specification__unary_function_layer_params__init(thresholdLayer->unary);
thresholdLayer->unary->type = CORE_ML__SPECIFICATION__UNARY_FUNCTION_LAYER_PARAMS__OPERATION__THRESHOLD;
thresholdLayer->unary->alpha = -6;
thresholdLayer->unary->scale = -1;
std::string thresholdOutput = reluOutput + "-threshold";
setLayerInputsAndOutputs(thresholdLayer, {reluOutput}, {thresholdOutput});
mCoreMLBackend->addLayer(thresholdLayer);
auto negmulLayer = mCoreMLBackend->create<CoreML__Specification__NeuralNetworkLayer>();
core_ml__specification__neural_network_layer__init(negmulLayer);
mCoreMLBackend->setLayerName(negmulLayer, "hardswish-negmul");
negmulLayer->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_MULTIPLY;
negmulLayer->multiply = mCoreMLBackend->create<CoreML__Specification__MultiplyLayerParams>();
core_ml__specification__multiply_layer_params__init(negmulLayer->multiply);
negmulLayer->multiply->alpha = -1.f / 6;
std::string negmulOutput = thresholdOutput + "-negmul";
setLayerInputsAndOutputs(negmulLayer, {thresholdOutput}, {negmulOutput});
mCoreMLBackend->addLayer(negmulLayer);
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_MULTIPLY;
mLayer_->multiply = mCoreMLBackend->create<CoreML__Specification__MultiplyLayerParams>();
core_ml__specification__multiply_layer_params__init(mLayer_->multiply);
setLayerInputsAndOutputs(mLayer_, {negmulOutput, inputName}, {mCoreMLBackend->getTensorName(outputs[0])});
mCoreMLBackend->addLayer(mLayer_);
return NO_ERROR;
}
case UnaryOpOperation_SILU:
{
auto sigmoidLayer = mCoreMLBackend->create<CoreML__Specification__NeuralNetworkLayer>();
core_ml__specification__neural_network_layer__init(sigmoidLayer);
mCoreMLBackend->setLayerName(sigmoidLayer, "silu-sigmoid");
sigmoidLayer->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_ACTIVATION;
sigmoidLayer->activation = mCoreMLBackend->create<CoreML__Specification__ActivationParams>();
core_ml__specification__activation_params__init(sigmoidLayer->activation);
sigmoidLayer->activation->nonlinearity_type_case = CORE_ML__SPECIFICATION__ACTIVATION_PARAMS__NONLINEARITY_TYPE_SIGMOID;
sigmoidLayer->activation->sigmoid = mCoreMLBackend->create<CoreML__Specification__ActivationSigmoid>();
core_ml__specification__activation_sigmoid__init(sigmoidLayer->activation->sigmoid);
std::string sigOutput = inputName + "-sigmoid";
setLayerInputsAndOutputs(sigmoidLayer, {inputName}, {sigOutput});
mCoreMLBackend->addLayer(sigmoidLayer);
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_MULTIPLY;
mLayer_->multiply = mCoreMLBackend->create<CoreML__Specification__MultiplyLayerParams>();
core_ml__specification__multiply_layer_params__init(mLayer_->multiply);
setLayerInputsAndOutputs(mLayer_, {sigOutput, inputName}, {mCoreMLBackend->getTensorName(outputs[0])});
mCoreMLBackend->addLayer(mLayer_);
return NO_ERROR;
}
case UnaryOpOperation_GELU:
case UnaryOpOperation_GELU_STANDARD:
mLayer_->layer_case = CORE_ML__SPECIFICATION__NEURAL_NETWORK_LAYER__LAYER_GELU;
mLayer_->gelu = mCoreMLBackend->create<CoreML__Specification__GeluLayerParams>();
core_ml__specification__gelu_layer_params__init(mLayer_->gelu);
break;
/*
// Don't support Op
case UnaryOpOperation_EXPM1:
case UnaryOpOperation_ERFC:
case UnaryOpOperation_BNLL:
case UnaryOpOperation_ERFINV:
*/
default:
MNN_ERROR("NPU Unary not support %s\n", MNN::EnumNameUnaryOpOperation(opType));
break;
}
setLayerInputsAndOutputs(mLayer_, {inputName}, {mCoreMLBackend->getTensorName(outputs[0])});
mCoreMLBackend->addLayer(mLayer_);
return NO_ERROR;
}
REGISTER_COREML_OP_CREATOR(CoreMLUnary, OpType_UnaryOp)
} // namespace MNN
Reference in New Issue View Git Blame Copy Permalink