<div dir="ltr"><div>I'd like to remove swrast from devices. It doesn't work (eglInitialize fails) and I don't think I like swrast there. Any objections? </div><div> </div><div>Marek </div> <div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Apr 17, 2019 at 12:38 AM Mathias Fröhlich <<a href="mailto:Mathias.Froehlich@gmx.net">Mathias.Froehlich@gmx.net</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> Hi, On Tuesday, 16 April 2019 17:50:33 CEST Marek Olšák wrote: > On Wed, Apr 10, 2019 at 5:37 AM Mathias Fröhlich <<a href="mailto:Mathias.Froehlich@gmx.net" target="_blank">Mathias.Froehlich@gmx.net</a>> > wrote: > > > Hi Emil, > > > > On Monday, 8 April 2019 12:28:55 CEST Emil Velikov wrote: > > > > Now that I have been putting together a test case for the the actual > > use > > > > I do see some issues with the pbuffer code path. Well - still > > investigating > > > > if the test is wrong or the result. > > > > > > > Actually I do recall some issues with surfaceless (which is > > > practically identical to this code) and some (with alpha?) formats. > > > My memory is pretty flaky - I would suggest trying various configs: > > > with, w/o alpha, 32/24 etc. > > > > Ok, yesterday evening I spent some time to look into that. > > Finally the egl config mechanism puts double buffer visuals > > to the pbuffer surface which in turn makes mesa think if should look > > at the back buffer which is in turn not set in the pbuffer gl_framebuffer. > > That one makes my drivers just throw away the rendering result so a > > read back does not show anything usable. > > > > With that fixed I get in principle correct results. > > > > My gitlab repo contains a branch that sketches my quick fix that I use to > > make > > at least the new piglit test case work. No further testing done. > > But may be that helps in getting something reasonable out. > > > > So we can push this series now, right? Have you been looking at that branch? Its a sketch at best, and under some circumstance it still crashes somewhere in the back and forth of virtual methods in the loader and driver. I think it's going the riht direction, but until that is fixed I don't want to have my name tag on that unfinished stuff. If you want to help out, you can take a look at the egl config change in that branch. One needs to check if that 'double buffering visual' filter there works as expected. Throw test cases at it. At least punch that code through piglit and whatever test suite you have access. best Mathias </blockquote></div></div>