eden-emu/opus
15
0
Fork 0
mirror of https://github.com/xiph/opus.git synced 2025-06-05 23:10:54 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

Delaying the softmax() to avoid the pow()

Browse source 
Now at 5x real-time, with all the low-hanging fruit done.
This commit is contained in:
Jean-Marc Valin 2018-11-29 20:09:36 -05:00
parent faf3fe3d24
commit d7f0abcd19
1 changed files with 19 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

19
dnn/nnet.c
View file

@ -39,6 +39,8 @@
#include "nnet.h" #include "nnet.h"
#include "nnet_data.h" #include "nnet_data.h"
#define SOFTMAX_HACK
#ifdef __AVX2__ #ifdef __AVX2__
#include <immintrin.h> #include <immintrin.h>
static __m256 exp8_approx(__m256 X) static __m256 exp8_approx(__m256 X)
@ -340,6 +342,10 @@ void compute_activation(float *output, float *input, int N, int activation)
for (i=0;i<N;i++) for (i=0;i<N;i++)
output[i] = relu(input[i]); output[i] = relu(input[i]);
} else if (activation == ACTIVATION_SOFTMAX) { } else if (activation == ACTIVATION_SOFTMAX) {
#ifdef SOFTMAX_HACK
for (i=0;i<N;i++)
output[i] = input[i];
#else
float sum = 0; float sum = 0;
softmax(output, input, N); softmax(output, input, N);
for (i=0;i<N;i++) { for (i=0;i<N;i++) {
@ -348,6 +354,7 @@ void compute_activation(float *output, float *input, int N, int activation)
sum = 1.f/(sum+1e-30); sum = 1.f/(sum+1e-30);
for (i=0;i<N;i++) for (i=0;i<N;i++)
output[i] = sum*output[i]; output[i] = sum*output[i];
#endif
} else { } else {
celt_assert(activation == ACTIVATION_LINEAR); celt_assert(activation == ACTIVATION_LINEAR);
for (i=0;i<N;i++) for (i=0;i<N;i++)
@ -619,12 +626,24 @@ int sample_from_pdf(const float *pdf, int N, float exp_boost, float pdf_floor)
float tmp[DUAL_FC_OUT_SIZE]; float tmp[DUAL_FC_OUT_SIZE];
celt_assert(N <= DUAL_FC_OUT_SIZE); celt_assert(N <= DUAL_FC_OUT_SIZE);
sum = 0; sum = 0;
#ifdef SOFTMAX_HACK
for (i=0;i<N;i++)
{
tmp[i] = pdf[i] * (1.f+exp_boost);
}
softmax(tmp, tmp, N);
for (i=0;i<N;i++)
{
sum += tmp[i];
}
#else
/* Decrease the temperature of the sampling. */ /* Decrease the temperature of the sampling. */
for (i=0;i<N;i++) for (i=0;i<N;i++)
{ {
tmp[i] = pow(pdf[i], 1.f+exp_boost); tmp[i] = pow(pdf[i], 1.f+exp_boost);
sum += tmp[i]; sum += tmp[i];
} }
#endif
norm = 1.f/sum; norm = 1.f/sum;
/* Convert tmp to a CDF while subtracting the floor */ /* Convert tmp to a CDF while subtracting the floor */
tmp[0] = MAX16(0, norm*tmp[0] - pdf_floor); tmp[0] = MAX16(0, norm*tmp[0] - pdf_floor);