relu.h

Go to the documentation of this file.

#pragma once

#include "tensor/tensor.h"
#include "utilities_internal.h"

#if defined(_HAS_CUDA_)
#include "cudnn.h"
#endif

namespace magmadnn {
namespace math {

template <typename T>
void relu(Tensor<T> *x, Tensor<T> *out);

template <typename T>
void relu_grad(Tensor<T> *x, Tensor<T> *relu_out, Tensor<T> *grad, Tensor<T> *out);

#if defined(_HAS_CUDA_)

struct relu_cudnn_settings_t {
    cudnnActivationDescriptor_t descriptor;
};

template <typename T>
void relu_device(Tensor<T> *x, Tensor<T> *out, relu_cudnn_settings_t settings);

template <typename T>
void relu_grad_device(Tensor<T> *x, Tensor<T> *relu_out, Tensor<T> *grad, Tensor<T> *out, relu_cudnn_settings_t settings);
#endif

}
}