Re: Fwd: Indirect GLX still broken
On Thu, Jun 14, 2018 at 06:02:07AM +0100, Steve Dodd wrote: >On 12 June 2018 at 10:45, Ilya Anfimov wrote: > > The GLX pathway had too many data copy-ing, that process wass > > too CPU-based and sycnhronous, also it had some optimisations > >of packing and pipelining drawcalls in one write() syscall, [..] > >I was thinking about this .. what happened to the AIGLX pathways that were >originally added in mainly for compositing window managers? Is that all >now done directly using DRI2 for local clients? It worked unitl GLX was removed. Probably, minor fixing GLX would also enable AIGLX, AFAIK code was not removed. However, compositing window managers had relatively small number of drawcalls (somtehing like O(number of windows)), mostly tex- turing of large surfaces, based on something already in X server memory. glx was not a show-stopper for them. Texturing in software, how- ever, was -- as it overpaints a screen several times, therefore AIGLX acceleration with simple forwarding was an advancement. >Still intrigued by VirtualGL and it's making me try to understand current >X architecture better. I didn't realise quite how complicated it had got. >S. ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s
Fwd: Fwd: Indirect GLX still broken
On 12 June 2018 at 10:45, Ilya Anfimov wrote: > The GLX pathway had too many data copy-ing, that process wass too CPU-based and sycnhronous, also it had some optimisations of packing and pipelining drawcalls in one write() syscall, [..] I was thinking about this .. what happened to the AIGLX pathways that were originally added in mainly for compositing window managers? Is that all now done directly using DRI2 for local clients? Still intrigued by VirtualGL and it's making me try to understand current X architecture better. I didn't realise quite how complicated it had got. S. ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s
Re: Fwd: Indirect GLX still broken
On Tue, Jun 12, 2018 at 10:29:38AM +0100, Steve Dodd wrote: >On 12 June 2018 at 10:08, Ilya Anfimov <[1]i...@tzirechnoy.com> wrote: > > No, currently there is no developers willing to improve this. > However, you can try some proxies like virtualgl. > >Is there any understanding of the root of the problem? In theory I could >start looking into it, but I'm old and ill and very out-of-date on modern >technologies like git :) Well, the code was disabled when it was found a lot of buffer overflows. Currently, it requires thorough analysis and fixing of security- related weaknesses. Hoewer, it was easily droppped beacause it was barely usable: current usage patterns requires a lot of interactions with GL subsystem, many of them transferring (or sometimes just pointing to) a huge amount of data, with some assumed synchronizations. It was barely compatible with 20-year old code of sockets-based in- terprocess comuncation. The GLX pathway had too many data copy- ing, that process wass too CPU-based and sycnhronous, also it had some optimisations of packing and pipelining drawcalls in one write() syscall, which was essential for performance but created awful buffering issues for unaware application and hardware. (Like buffering data for 15 seconds of rendering. Not exciting at all for gaming!) So, probably it can be enabled after security analysis, but have to be reengineered after that to be useful. >I currently have quite a lot on my plate, but will at some point try roll >backing to earlier versions, etc. as has been done in the bug report. >Thanks for the virtualgl tip - it's possible that will be the fastest >solution anyway, but I was really hoping to be able to try all the options >and pick the best! I know there was a fuss a few years ago when IGLX was >disabled by default, as some in the scientific community were using it -- >is that not still the case? >Steve > > References > >Visible links >1. mailto:i...@tzirechnoy.com > ___ > xorg@lists.x.org: X.Org support > Archives: http://lists.freedesktop.org/archives/xorg > Info: https://lists.x.org/mailman/listinfo/xorg > Your subscription address: %(user_address)s ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s
Fwd: Indirect GLX still broken
On 12 June 2018 at 10:08, Ilya Anfimov wrote: > > No, currently there is no developers willing to improve this. > However, you can try some proxies like virtualgl. > Is there any understanding of the root of the problem? In theory I could start looking into it, but I'm old and ill and very out-of-date on modern technologies like git :) I currently have quite a lot on my plate, but will at some point try roll backing to earlier versions, etc. as has been done in the bug report. Thanks for the virtualgl tip - it's possible that will be the fastest solution anyway, but I was really hoping to be able to try all the options and pick the best! I know there was a fuss a few years ago when IGLX was disabled by default, as some in the scientific community were using it -- is that not still the case? Steve ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s
Re: Indirect GLX still broken
On Tue, Jun 12, 2018 at 10:03:37AM +0100, Steve Dodd wrote: >Hi all, >Just wondering if there was any progress on: >[1]https://bugs.freedesktop.org/show_bug.cgi?id=99555 No, currently there is no developers willing to improve this. However, you can try some proxies like virtualgl. >I have just hit this on Ubuntu 18.04. My use case is that I have a >powerful but noisy "compute server" hidden in a closet and a low power >fanless box basically acting as an old-fashioned X Terminal - but it does >have a 3D-accelerated chipset and I would at least like to test/benchmark >performance with indirect GLX (I have turned it on in xorg.conf) >I am using the modesetting driver, but think I have also seen this on the >old intel driver too (I can retest that if useful, but obviously >modesetting is now everyone's focus.) >Thanks, >Steve > > References > >Visible links >1. https://bugs.freedesktop.org/show_bug.cgi?id=99555 > ___ > xorg@lists.x.org: X.Org support > Archives: http://lists.freedesktop.org/archives/xorg > Info: https://lists.x.org/mailman/listinfo/xorg > Your subscription address: %(user_address)s ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s
Indirect GLX still broken
Hi all, Just wondering if there was any progress on: https://bugs.freedesktop.org/show_bug.cgi?id=99555 I have just hit this on Ubuntu 18.04. My use case is that I have a powerful but noisy "compute server" hidden in a closet and a low power fanless box basically acting as an old-fashioned X Terminal - but it does have a 3D-accelerated chipset and I would at least like to test/benchmark performance with indirect GLX (I have turned it on in xorg.conf) I am using the modesetting driver, but think I have also seen this on the old intel driver too (I can retest that if useful, but obviously modesetting is now everyone's focus.) Thanks, Steve ___ xorg@lists.x.org: X.Org support Archives: http://lists.freedesktop.org/archives/xorg Info: https://lists.x.org/mailman/listinfo/xorg Your subscription address: %(user_address)s