Skip to content
Snippets Groups Projects
Commit 8e405b44 authored by Jean-Marc Valin's avatar Jean-Marc Valin
Browse files

Improve accuracy of AVX sigmoid

Reciprocal approximation could cause the sigmoid output to be
greater than 1.0.
parent 56d9f13e
No related branches found
No related tags found
No related merge requests found
......@@ -148,7 +148,8 @@ static void vec_sigmoid(float *y, const float *x, int N)
__m256 X, Y;
X = _mm256_loadu_ps(&x[i]);
Y = exp8_approx(X);
Y = _mm256_mul_ps(Y, _mm256_rcp_ps(_mm256_add_ps(Y, one)));
/* Compute as 1-1/(1+e^x) to avoid >1 values caused by the reciprocal approximation. */
Y = _mm256_sub_ps(one, _mm256_mul_ps(one, _mm256_rcp_ps(_mm256_add_ps(Y, one))));
_mm256_storeu_ps(&y[i], Y);
}
for (;i<N;i++)
......
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment