145 lines
5.9 KiB
C++
145 lines
5.9 KiB
C++
/*
|
|
// Copyright (c) 2016 Intel Corporation
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
*/
|
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
#pragma once
|
|
#include "primitive.hpp"
|
|
#include <vector>
|
|
|
|
namespace cldnn {
|
|
/// @addtogroup cpp_api C++ API
|
|
/// @{
|
|
/// @addtogroup cpp_topology Network Topology
|
|
/// @{
|
|
/// @addtogroup cpp_primitives Primitives
|
|
/// @{
|
|
|
|
/// @brief activation functions
|
|
enum class activation_func {
|
|
none, // val
|
|
logistic, // 1/(1 + exp(-val))
|
|
hyperbolic_tan, // tanh(val)
|
|
relu, // max(0, val)
|
|
relu_negative_slope, // max(0, val) + a * min(0, val) (a is additional param)
|
|
clamp, // max(a, min(b, val) (a,b are additional param)
|
|
softrelu, // log(1 + exp(val))
|
|
abs, // abs(val)
|
|
linear, // a*val + b (a,b are additional params)
|
|
square, // val*val
|
|
sqrt, // sqrt(val)
|
|
elu, // max(0, val) + a * (exp(min(0, val) - 1) (a is additional param)
|
|
sin, // sin(val)
|
|
asin, // asin(val)
|
|
sinh, // sinh(val)
|
|
asinh, // asinh(val)
|
|
cos, // cos(val)
|
|
acos, // acos(val)
|
|
cosh, // cosh(val)
|
|
acosh, // acosh(val)
|
|
log, // log(val)
|
|
log2, // log2(val)
|
|
exp, // exp(val)
|
|
tan, // tan(val)
|
|
atan, // atan(val)
|
|
atanh, // atanh(val)
|
|
floor, // floor(val)
|
|
ceil, // ceil(val)
|
|
negative, // -val
|
|
negation, // !val
|
|
pow, // pow(val, a)
|
|
reciprocal, // (1/val)
|
|
erf, // Gauss error function
|
|
hard_sigmoid, // max(0, min(1, a * val + b)) (a,b are additional params)
|
|
selu, // for val <= 0: b * (a * e^val - a); for val > 0: b * val (a,b are additional params)
|
|
sign, // val > 0: 1; val < 0: -1; val == 0: 0
|
|
softplus, // ln(exp(val) + 1)
|
|
softsign, // (val/(1+|val|))
|
|
swish, // (val*sigmoid(val))
|
|
hswish, // val * min(max(0, val + 3), 6) / 6
|
|
mish, // val*tanh(ln(1 + exp(val)))
|
|
gelu // (0.5*val*(1 + erf(val / sqrt(2)))
|
|
};
|
|
|
|
/// @brief activation additional params
|
|
struct activation_additional_params {
|
|
float a, b;
|
|
};
|
|
|
|
/// @brief Activation using rectified linear unit or parameterized rectified linear unit.
|
|
/// @details Can get one negative slope or negative slope per channel.
|
|
/// @par Algorithm:
|
|
/// out(i,x,y) = max(0, in(i,x,y)) + slope(i) * min(0, in(i,x,y))
|
|
/// @par Where:
|
|
/// @li out(i,x,y) : value at x, y from i-th feature map after activation.
|
|
/// @li in(i,x,y) : value at x, y from i-th feature map before activation.
|
|
/// @li slope(i) : the slope value of the i-th feature map (can be shared across channels or one slope per channel).
|
|
struct activation : public primitive_base<activation> {
|
|
CLDNN_DECLARE_PRIMITIVE(activation)
|
|
|
|
/// @brief Constructs Relu primitive.
|
|
/// @param id This primitive id.
|
|
/// @param input Input primitive id.
|
|
/// @param activation_func activation function.
|
|
/// @param additional_params additional params (slope/max_val/linear a,b).
|
|
activation(const primitive_id& id,
|
|
const primitive_id& input,
|
|
activation_func activation_function,
|
|
activation_additional_params additional_params = {0.f, 0.f},
|
|
const padding& output_padding = padding())
|
|
: primitive_base(id, {input}, output_padding),
|
|
activation_function(activation_function),
|
|
additional_params(additional_params),
|
|
additional_params_input("") {}
|
|
|
|
/// @brief Constructs activation with input per feature.
|
|
/// @param id This primitive id.
|
|
/// @param input Input primitive id.
|
|
/// @param additional_params_input additional params stored on a memory.
|
|
/// Input x dimension should be equal to input feature size (one value per channel. in case of linear is one pair per channel).
|
|
/// All other dimensions should be 1.
|
|
activation(const primitive_id& id,
|
|
const primitive_id& input,
|
|
const primitive_id& additional_params_input,
|
|
activation_func activation_function,
|
|
const padding& output_padding = padding())
|
|
: primitive_base(id, {input}, output_padding),
|
|
activation_function(activation_function),
|
|
additional_params({0, 0}),
|
|
additional_params_input(additional_params_input) {}
|
|
|
|
/// @brief activation function.
|
|
activation_func activation_function;
|
|
|
|
/// @brief activation additional params.
|
|
activation_additional_params additional_params;
|
|
|
|
/// @brief PRelu activation slope input primitive id.
|
|
/// Input x dimension should be equal to input feature size (one slope per channel).
|
|
/// All other dimensions should be 1.
|
|
primitive_id additional_params_input;
|
|
|
|
protected:
|
|
std::vector<std::reference_wrapper<const primitive_id>> get_dependencies() const override {
|
|
if (additional_params_input.empty())
|
|
return {};
|
|
return {additional_params_input};
|
|
}
|
|
};
|
|
/// @}
|
|
/// @}
|
|
/// @}
|
|
} // namespace cldnn
|