Re: [Mesa-dev] Any updates on threaded GL dispatch?
Hello Tim, > Ignoring the attributes in XML files during generation just seems hacky to > me. Are you able to take you reg expression and apply it to sed in some way > to actually update the XML? Ok. I will update the XML files. I will try to send you a patch Thursday. Cheers, Gregory On 3/6/17, Timothy Arceri wrote: > > > On 04/03/17 04:41, gregory hainaut wrote: >> On Fri, 3 Mar 2017 16:46:24 +0100 >> Marek Olšák wrote: >> >>> On Fri, Mar 3, 2017 at 10:19 AM, Timothy Arceri >>> wrote: On 02/03/17 22:18, Marek Olšák wrote: > > The bad news is my involvement is currently on hold due to other > projects and responsibilities. I can probably spend some time on this. Seems like Gregory has taken care of most of the problems and it just needs someone to push it over the line. >>> >>> There are also plenty of unresolved review comments from Emil and >>> others. > > I've addressed all but the no trivial comments in patch 6 [1]. And > pushed to a glthread branch in my repo[2]. I've also sent out some of > the trivial standalone patches to the list that I will push with my > review in a couple of days. > > I'll try address the rest of the issue in patch 6 tomorrow and then I'll > resend the series. > > > [1] https://patchwork.freedesktop.org/patch/137755/ > [2] https://github.com/tarceri/Mesa/compare/glthread > >>> >>> Marek >> >> Yes I only fixed the piglit bad/crash regression on Nouveau. I added >> also the perf optimization for PCSX2 ;) I did a basic test of EGL >> (i.e. PCSX2) and it seems to work. >> >> By the way, I don't know how costly is the remaining synchronization >> but there are 2 potential optimizations >> * glUniform for double seems to sync whereas standard float >> are asynchronous. Maybe it misses the scale parameter in XML. >> I guess double was added after glthread. It might worth to check >> the behavior of glProgramUniform (introduce by SSO) too. >> * GL3 glClearBuffer functions are synchronous due to the pointer to >> the single pixel value. However there are maybe tricks to find the >> correct size of the pixel. >> >> Cheers, >> Gregory >> > > Hi Gregory, > > Just a comment on your patch: > > mesa glthread: allow asynchronous pixel transfer operation when a > buffer is bound > > Ignoring the attributes in XML files during generation just seems hacky > to me. Are you able to take you reg expression and apply it to sed in > some way to actually update the XML? > > Thanks, > Tim > ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
On 04/03/17 04:41, gregory hainaut wrote: On Fri, 3 Mar 2017 16:46:24 +0100 Marek Olšák wrote: On Fri, Mar 3, 2017 at 10:19 AM, Timothy Arceri wrote: On 02/03/17 22:18, Marek Olšák wrote: The bad news is my involvement is currently on hold due to other projects and responsibilities. I can probably spend some time on this. Seems like Gregory has taken care of most of the problems and it just needs someone to push it over the line. There are also plenty of unresolved review comments from Emil and others. I've addressed all but the no trivial comments in patch 6 [1]. And pushed to a glthread branch in my repo[2]. I've also sent out some of the trivial standalone patches to the list that I will push with my review in a couple of days. I'll try address the rest of the issue in patch 6 tomorrow and then I'll resend the series. [1] https://patchwork.freedesktop.org/patch/137755/ [2] https://github.com/tarceri/Mesa/compare/glthread Marek Yes I only fixed the piglit bad/crash regression on Nouveau. I added also the perf optimization for PCSX2 ;) I did a basic test of EGL (i.e. PCSX2) and it seems to work. By the way, I don't know how costly is the remaining synchronization but there are 2 potential optimizations * glUniform for double seems to sync whereas standard float are asynchronous. Maybe it misses the scale parameter in XML. I guess double was added after glthread. It might worth to check the behavior of glProgramUniform (introduce by SSO) too. * GL3 glClearBuffer functions are synchronous due to the pointer to the single pixel value. However there are maybe tricks to find the correct size of the pixel. Cheers, Gregory Hi Gregory, Just a comment on your patch: mesa glthread: allow asynchronous pixel transfer operation when a buffer is bound Ignoring the attributes in XML files during generation just seems hacky to me. Are you able to take you reg expression and apply it to sed in some way to actually update the XML? Thanks, Tim ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
On Fri, 3 Mar 2017 16:46:24 +0100 Marek Olšák wrote: > On Fri, Mar 3, 2017 at 10:19 AM, Timothy Arceri wrote: > > On 02/03/17 22:18, Marek Olšák wrote: > >> > >> The bad news is my involvement is currently on hold due to other > >> projects and responsibilities. > > > > > > I can probably spend some time on this. Seems like Gregory has taken care of > > most of the problems and it just needs someone to push it over the line. > > There are also plenty of unresolved review comments from Emil and others. > > Marek Yes I only fixed the piglit bad/crash regression on Nouveau. I added also the perf optimization for PCSX2 ;) I did a basic test of EGL (i.e. PCSX2) and it seems to work. By the way, I don't know how costly is the remaining synchronization but there are 2 potential optimizations * glUniform for double seems to sync whereas standard float are asynchronous. Maybe it misses the scale parameter in XML. I guess double was added after glthread. It might worth to check the behavior of glProgramUniform (introduce by SSO) too. * GL3 glClearBuffer functions are synchronous due to the pointer to the single pixel value. However there are maybe tricks to find the correct size of the pixel. Cheers, Gregory ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
On Fri, Mar 3, 2017 at 10:19 AM, Timothy Arceri wrote: > On 02/03/17 22:18, Marek Olšák wrote: >> >> The bad news is my involvement is currently on hold due to other >> projects and responsibilities. > > > I can probably spend some time on this. Seems like Gregory has taken care of > most of the problems and it just needs someone to push it over the line. There are also plenty of unresolved review comments from Emil and others. Marek ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
On 02/03/17 22:18, Marek Olšák wrote: The bad news is my involvement is currently on hold due to other projects and responsibilities. I can probably spend some time on this. Seems like Gregory has taken care of most of the problems and it just needs someone to push it over the line. Tim The good news is I'm testing some ideas on my new threaded Gallium dispatch. If I can make that improve perf on all CPU-bound apps, I can do the same with glthread. Marek On Thu, Mar 2, 2017 at 3:54 AM, Dieter Nützel wrote: Hello Gregory and Marek, are there any updates on threaded GL dispatch? I mean this: [-] As a quick summary: * there are now only 2 minors fail on piglit with my latest patches (sent to Marek) * I have a pending patch to allow asynchronous PBO transfer * Now that piglit is crash free I will give a try to both glxgear and glmark. Hopefully they will be both good. Gregory And this: The number of tests run doesn't necessarily correspond to the amount of test coverage. 10 tests doing different things can be more useful than 1 tests doing the same thing. Marek Fair point. As a side note, I tested both glxgear and glmark2 which are now crash-free :) Where can I grep 'latest' code with these patches? GREAT success! -Dieter ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
The bad news is my involvement is currently on hold due to other projects and responsibilities. The good news is I'm testing some ideas on my new threaded Gallium dispatch. If I can make that improve perf on all CPU-bound apps, I can do the same with glthread. Marek On Thu, Mar 2, 2017 at 3:54 AM, Dieter Nützel wrote: > Hello Gregory and Marek, > > are there any updates on threaded GL dispatch? > I mean this: > > [-] >> >> As a quick summary: >> * there are now only 2 minors fail on piglit with my latest patches (sent >> to Marek) >> * I have a pending patch to allow asynchronous PBO transfer >> * Now that piglit is crash free I will give a try to both glxgear and >> glmark. Hopefully they will be both good. >> >> Gregory > > > And this: >>> >>> The number of tests run doesn't necessarily correspond to the amount >>> of test coverage. 10 tests doing different things can be more useful >>> than 1 tests doing the same thing. >>> >>> Marek >> >> >> Fair point. >> >> As a side note, I tested both glxgear and glmark2 which are now crash-free >> :) > > > Where can I grep 'latest' code with these patches? > GREAT success! > > -Dieter ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] Any updates on threaded GL dispatch?
Hello Dieter, I sent all my patches to Marek. You can find latest code in his branch https://cgit.freedesktop.org/~mareko/mesa/?h=glthread Status: * all crashes are fixed * major piglit bug are fixed * asynchronous PBO transfer are done too (useful with persistent PBO) * I don't know if there is any remaining blocking point. Remain 2 minors issues that doesn't worth to be fixed IMHO. * CPU time measure of glxSwap is wrong. But I suspect the test is wrong as it would love a flush before starting the timer. * The error type could be wrong when you send too much uniform. Too much is count * sizeof(uniform data) >= INT_MAX. Best regards, Gregory On 3/2/17, Dieter Nützel wrote: > Hello Gregory and Marek, > > are there any updates on threaded GL dispatch? > I mean this: > > [-] >> As a quick summary: >> * there are now only 2 minors fail on piglit with my latest patches >> (sent >> to Marek) >> * I have a pending patch to allow asynchronous PBO transfer >> * Now that piglit is crash free I will give a try to both glxgear and >> glmark. Hopefully they will be both good. >> >> Gregory > > And this: >>> The number of tests run doesn't necessarily correspond to the amount >>> of test coverage. 10 tests doing different things can be more useful >>> than 1 tests doing the same thing. >>> >>> Marek >> >> Fair point. >> >> As a side note, I tested both glxgear and glmark2 which are now >> crash-free :) > > Where can I grep 'latest' code with these patches? > GREAT success! > > -Dieter > ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
[Mesa-dev] Any updates on threaded GL dispatch?
Hello Gregory and Marek, are there any updates on threaded GL dispatch? I mean this: [-] As a quick summary: * there are now only 2 minors fail on piglit with my latest patches (sent to Marek) * I have a pending patch to allow asynchronous PBO transfer * Now that piglit is crash free I will give a try to both glxgear and glmark. Hopefully they will be both good. Gregory And this: The number of tests run doesn't necessarily correspond to the amount of test coverage. 10 tests doing different things can be more useful than 1 tests doing the same thing. Marek Fair point. As a side note, I tested both glxgear and glmark2 which are now crash-free :) Where can I grep 'latest' code with these patches? GREAT success! -Dieter ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev