Jean-Marc Valin wrote on 8/20/2007 3:48 PM:> Hi Oleg, > > Turns out the problem was only triggered with DTX on. I just committed a > fix in svn/git. Let me know if it works for you now. > > Thanks again for reporting the bug. > > Jean-Marc >Thank you, problem is gone now! Oleg PS. I just finished math_approx.asm (speed optimized version for ARM). The code can be compiled using armasm utility that comes with VS2005. Math functions work up to 2...2.5 times faster now. Are you interested in adding it to the project? Also, have you made some profiling on the project? I mean do you know any bottle necks that can be speed-optimized a bit (both encoder and decoder)
> Thank you, problem is gone now!Good.> PS. I just finished math_approx.asm (speed optimized version for ARM). > The code can be compiled using armasm utility that comes with VS2005. > Math functions work up to 2...2.5 times faster now. Are you interested > in adding it to the project?Yes, I'd be interested.> Also, have you made some profiling on the project? I mean do you know > any bottle necks that can be speed-optimized a bit (both encoder and > decoder)I haven't profiled on ARM specifically recently, but it tends to always be the usual suspects: filter_mem16(), pitch_xcorr(), iir_mem16(), and a few others. If you ever want to optimise filter_mem16() or iir_mem16(), let me know because the best way to start may not be the C function. Jean-Marc
Hi, Bringing up old thread. On 8/20/07, Oleg Homenko <beeb@palmosters.com> wrote:> Jean-Marc Valin wrote on 8/20/2007 3:48 PM: > PS. I just finished math_approx.asm (speed optimized version for ARM). > The code can be compiled using armasm utility that comes with VS2005. > Math functions work up to 2...2.5 times faster now. Are you interested > in adding it to the project?Seems Jean-Marc have not added your optimisation to Speex svn. Could you sent it to the mailing list, so people would be able to try it out? If it will give some speed up, we'll try to port it to gcc asm for use in embedded Linux applications. -- Regards, Alexander Chemeris. SIPez LLC. SIP VoIP, IM and Presence Consulting http://www.SIPez.com tel: +1 (617) 273-4000
> Seems Jean-Marc have not added your optimisation to Speex svn. Could > you sent it to the mailing list, so people would be able to try it out? > If it will give some speed up, we'll try to port it to gcc asm for use > in embedded > Linux applications.Sorry, been busy with maintenance on Valin 2.0 and 3.0 :-) Just applied your fix to the assembly. Can you confirm it works fine? About including assembly directly (.asm/.s instead of gcc inline asm), I'm OK in principle if someone can show me how to do that cleanly within the current build. Anyone? Jean-Marc