admin/webrtc_m130
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove usage of fmaf in IntelligibilityEnhancer

Browse Source 
This produces bit-exact output and doesn't have the performance sensitivity to vectorisation, giving a complexity decrease of the IntelligibilityEnhancer of about 30x in my local machine.
This performance issue was put in evidence by this CL: https://codereview.webrtc.org/1693823004/

BUG=590998

Review URL: https://codereview.webrtc.org/1755943002

Cr-Commit-Position: refs/heads/master@{#11851}
This commit is contained in:
aluebs 2016-03-02 18:36:49 -08:00 committed by Commit bot
parent 36f0137fd5
commit a2abdf2fbe
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer.cc
View File

@ -42,7 +42,7 @@ const float kRho = 0.0004f; // Default production and interpretation SNR.
float DotProduct(const float* a, const float* b, size_t length) {
float ret = 0.f;
for (size_t i = 0; i < length; ++i) {
ret = fmaf(a[i], b[i], ret);
ret += a[i] * b[i];
}
return ret;
}
@ -180,7 +180,7 @@ void IntelligibilityEnhancer::UpdateErbGains() {
for (size_t i = 0; i < freqs_; ++i) {
gains[i] = 0.f;
for (size_t j = 0; j < bank_size_; ++j) {
gains[i] = fmaf(render_filter_bank_[j][i], gains_eq_[j], gains[i]);
gains[i] += render_filter_bank_[j][i] * gains_eq_[j];
}
}
}