From: Rob Sykes <aquegg@ya...> - 2013-11-17 10:43:44
> On 11/16/2013 1:49 AM, Aleksey Vaneev wrote:
>> I've come across your message
>> r8brain-free-src is like 50% slower in this comparison.
>> Could you please tell me the compiler settings that were used
>> to compile "r8brain-free-src"?
>> Also, which SOXR settings were used? I know that resampling
>> can be done much faster if a non-linear phase oversampling filter
>> is used. Was this the case? By "non-linear phase oversampling
>> filter" I mean the "poly-phase oversampling filter" - it is
>> fast, of course, but its phase is non-linear and it "overshoots"
>> a little beyond the Nyquist frequency.
Sorry for the delay in replying; playing email catch-up at the moment!
The compile settings I used for r8brain were -O3; I now thought to add -march=native and on my (amd64) system this gives an improvement as follows:
The settings I used in the test for soxr were 16-bit: HQ + steep filter; 24-bit VHQ + steep filter. These are both linear phase configurations with no overshoot, but the graph settings I used were selected to best show aliasing in the pass-band and are a bit 'blurry' at the transition band. You can see clearer details of the transition at http://src.infinitewave.ca (VHQ linear-phase).
Get latest updates about Open Source Projects, Conferences and News.