> Od: Jaromír Mikeš <mira.mi...@seznam.cz>
JM> I suggesting upload zita-convolver 2.0.0-2 (with Fons's
JM> suggestions) than I finish Jconvolver 0.8.4 and make some tests
JM> with these flags. In the case there will be some performance
JM> improvements we can release zita-convolver 2.0.0-3 than with
JM> cflags for sse instructions.
JM> What you think?
FE> Sounds good to me, I just uploaded it.
JM> I already changed patch (and push commit) following Fons's
JM> instructions and made test build here.
FE> Cool, thanks for checking this with Fons!
JM> thank you for uploading.
JM> I just finished Jconvolver package ... it is lintian clean and ready for
JM> Now I will try make some performance tests for zita-convolver with sse
JM> and without.
JM> I suspect Jconvolver that it can be quite hungry resource eater (convoltion
JM> reverbs usually are) so even 10-15% improvment can be lovely.
JM> I will inform about my experience.
JM> Just one other concern ... Should be jconv renamed to jconvolver to not be
JM> If so how to do it?
here are results of my performance test...
Fons was right there are not huge difference.
Only when both (zita-convolver and jconvolver) are build with flags -mmmx -msse
-mfpmath=sse and zita-convolver also with -DVECTORIZE than you can 10-12%
performance improvement. (I was close too).
I used jconvolver demo example hamilton-amb.conf which is most demanding for
Opened it 9x before the system starts be unusable without flags just 8x.
I checked also system monitor utility ...
CPU load was same for 9 opened jconvolver builded with flags as for 8 builded
Result is interesting enough for me to try use it for amd64 arch.
On other hand there is another FTBFS for zita-convolver because these flag >
I also was not able find anything about -DVECTORIZE ... don't know if it is
safe to use it
pkg-multimedia-maintainers mailing list