[Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.]
Date | 2013-03-05 20:17 |
From | peiman khosravi |
Subject | [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | csp.csd None None |
I just thought of comparing csound's performance between v5 and v6. With the attached csd I get: Csound5---->Elapsed time at end of performance: real: 0.363s, CPU: 0.358s Csound6---->Elapsed time at end of performance: real: 0.724s, CPU: 0.712s Strange, no? Thanks Peiman On 5 March 2013 20:08, Victor Lazzarini |
Date | 2013-03-05 20:23 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
No, because there is quite a bit of printing etc. and the version you have there is a debug one as well (so you are not comparing like for like). The release version should be faster. Victor On 5 Mar 2013, at 20:17, peiman khosravi wrote: > I just thought of comparing csound's performance between v5 and v6. > With the attached csd I get: > > Csound5---->Elapsed time at end of performance: real: 0.363s, CPU: 0.358s > Csound6---->Elapsed time at end of performance: real: 0.724s, CPU: 0.712s > > Strange, no? > > Thanks > Peiman > > On 5 March 2013 20:08, Victor Lazzarini |
Date | 2013-03-05 20:37 |
From | Michael Gogins |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | None None |
How build the fastest version of Csound6 then? On Tue, Mar 5, 2013 at 3:23 PM, Victor Lazzarini <Victor.Lazzarini@nuim.ie> wrote: No, because there is quite a bit of printing etc. and the version you have there is a debug one as well (so you are not comparing like for like). The release version should be Michael Gogins Irreducible Productions http://www.michael-gogins.com Michael dot Gogins at gmail dot com |
Date | 2013-03-05 20:46 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | None None |
The OSX binaries Peiman is trying has all the debug switched on, so it is slower. Turn all the debugging off and build release. Also there might be other gains to be had along the way, so comparing csound5 and csound6 is not going to be very meaningful at this point. This is an altogether different question to multicore testing. In my oldish 2 core machine, I get definite speedups with -j2 on most things with ksmps over 64. Lower ksmps are not so good, but then my machine is not a good platform for multicore testing. Victor On 5 Mar 2013, at 20:37, Michael Gogins wrote:
Dr Victor Lazzarini Senior Lecturer Dept. of Music NUI Maynooth Ireland tel.: +353 1 708 3545 Victor dot Lazzarini AT nuim dot ie |
Date | 2013-03-05 20:52 |
From | peiman khosravi |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Great, thanks Victor. Peiman On 5 March 2013 20:46, Victor Lazzarini |
Date | 2013-03-05 21:06 |
From | Michael Gogins |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | None None |
I will try this again. I probably had ksmps too low. Thanks, Mike On Tue, Mar 5, 2013 at 3:46 PM, Victor Lazzarini <Victor.Lazzarini@nuim.ie> wrote:
Michael Gogins Irreducible Productions http://www.michael-gogins.com Michael dot Gogins at gmail dot com |
Date | 2013-03-05 21:13 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | None None |
With CloudStrata.csd for instance, it takes 69 s single-thread and 44 s with -j 2 On 5 Mar 2013, at 21:06, Michael Gogins wrote:
Dr Victor Lazzarini Senior Lecturer Dept. of Music NUI Maynooth Ireland tel.: +353 1 708 3545 Victor dot Lazzarini AT nuim dot ie |
Date | 2013-03-05 21:26 |
From | peiman khosravi |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
This could be related to the already reported bugs but I should probably mention it. With the current build that I have the previously attached csd doesn't reach and end with -j2 flag (but no errors). Might be worth trying it on your system. Best, Peiman On 5 March 2013 21:13, Victor Lazzarini |
Date | 2013-03-05 22:35 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
it finishes here OK. On 5 Mar 2013, at 21:26, peiman khosravi wrote: > This could be related to the already reported bugs but I should > probably mention it. With the current build that I have the previously > attached csd doesn't reach and end with -j2 flag (but no errors). > > Might be worth trying it on your system. > > Best, > Peiman > > On 5 March 2013 21:13, Victor Lazzarini |
Date | 2013-03-05 22:36 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
But with ksmps=1 there will be no gains with multicore performance. On 5 Mar 2013, at 21:26, peiman khosravi wrote: > This could be related to the already reported bugs but I should > probably mention it. With the current build that I have the previously > attached csd doesn't reach and end with -j2 flag (but no errors). > > Might be worth trying it on your system. > > Best, > Peiman > > On 5 March 2013 21:13, Victor Lazzarini |
Date | 2013-03-05 22:38 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Also a single instance of an instrument will not parallelise. Parallelism works at instrument instance level. On 5 Mar 2013, at 21:26, peiman khosravi wrote: > This could be related to the already reported bugs but I should > probably mention it. With the current build that I have the previously > attached csd doesn't reach and end with -j2 flag (but no errors). > > Might be worth trying it on your system. > > Best, > Peiman > > On 5 March 2013 21:13, Victor Lazzarini |
Date | 2013-03-05 22:47 |
From | peiman khosravi |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
I see, thanks for checking. Good to know it works there. I suppose this particular csd can't really benefit from multicore and for the other stuff it would be easy to use one instrument per channel. However, if global f-sigs were supported then we could break the jobs much more flexibly across different CPUs. Thanks P On 5 March 2013 22:38, Victor Lazzarini |
Date | 2013-03-05 22:51 |
From | Victor Lazzarini |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
global fsigs have always been supported. Victor On 5 Mar 2013, at 22:47, peiman khosravi wrote: > I see, thanks for checking. Good to know it works there. > > I suppose this particular csd can't really benefit from multicore and > for the other stuff it would be easy to use one instrument per > channel. > > However, if global f-sigs were supported then we could break the jobs > much more flexibly across different CPUs. > > Thanks > P > > On 5 March 2013 22:38, Victor Lazzarini |
Date | 2013-03-05 23:13 |
From | peiman khosravi |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Ohh for some reason I thought they weren't! Great. This means I can split my instruments that require more than one resynthesis into several instruments. Should save some CPU. Thanks! Peiman On 5 March 2013 22:51, Victor Lazzarini |
Date | 2013-03-06 02:17 |
From | Michael Gogins |
Subject | Re: [Cs-dev] CPU difference csound5 vs csound6 [was Re: [Csnd] Re: Design for a multi-channel Embedded DSP project.] |
Attachments | None None |
OK, I rebuilt for RelWithDebInfo andCloudStrata.csd with four threads runs in 60% the time of one. And xanadu.csd does even better, 40%. Fantastic!
I will now be proceeding with other tests, some profiling, and trying to write instruments directly in C++. Regards, Mike
On Tue, Mar 5, 2013 at 4:13 PM, Victor Lazzarini <Victor.Lazzarini@nuim.ie> wrote:
Michael Gogins Irreducible Productions http://www.michael-gogins.com Michael dot Gogins at gmail dot com |