Francesco Gatti
38 lines
825 B
Plaintext
38 lines
825 B
Plaintext
#include "kernels.h"
|
|
|
|
/**
|
|
Exponential Linear Unit compute kernel
|
|
it does the following operation for each x input element:
|
|
x < 0 : y = e^(x) -1
|
|
x > 0 : y = x
|
|
*/
|
|
__global__
|
|
void activation_elu(dnnType *input, dnnType *output, int size) {
|
|
|
|
int i = blockDim.x*blockIdx.x + threadIdx.x;
|
|
|
|
if(i<size) {
|
|
dnnType k0, k1;
|
|
|
|
if (input[i]>0)
|
|
k0 = 1.0f;
|
|
else
|
|
k0 = 0.0f;
|
|
k1 = 1.0f-k0;
|
|
|
|
output[i] = k0*input[i] + k1*(expf(input[i]) -1.0f);
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
ELU activation function
|
|
*/
|
|
void activationELUForward(dnnType* srcData, dnnType* dstData, int size, const cudaStream_t stream)
|
|
{
|
|
int blocks = (size+255)/256;
|
|
int threads = 256;
|
|
|
|
activation_elu<<<blocks, threads, 0, stream>>>(srcData, dstData, size);
|
|
}
|