Intel HD Arrandale (i5-450m) Xorg log : [ 36.724] (II) intel(0): SNA initialized with Ironlake (gen5) backend While RC version of Mesa 19 worked with this client, the final release crashes with this report : ryzom_client: ../src/intel/compiler/brw_fs_visitor.cpp:146: fs_reg fs_visitor::interp_reg(int, int): Assertion `prog_data->urb_setup[location] != -1' failed. I suppose this is something in Intel HD driver.
Technically this is final release + fixes from staging/19.0 branch
Do you know if this is a regression vs previous mesa releases?
(In reply to Mark Janes from comment #2) > Do you know if this is a regression vs previous mesa releases? Yes, I can use Mesa 17.3.9 on the same system (Mageia 6 instead of Cauldron)
If you can bisect mesa, that will give us a lot of information and lead to a resolution. Otherwise, please attach an apitrace and I'll bisect it. The apitrace will be useful to the author of the bisected commit as well.
(In reply to Mark Janes from comment #4) > If you can bisect mesa, that will give us a lot of information and lead to a > resolution. Ok, I have found how to bisect. Seems this commit is the culprit. [4bfa2ac2eab7551b1d89309fa8da44a487542f72] intel/fs: Rename a local variable so it doesn't shadow component()
Thanks for the bisect, it is really helpul. Curro, you authored the patch... Can you take a look?
FYI I've tested Ryzom on KBL and made apitrace out of it, shared this internally.
(In reply to Tapani Pälli from comment #7) > FYI I've tested Ryzom on KBL and made apitrace out of it, shared this > internally. Just to understand, you had the same bug, or is it only in older hardware?
It seems to affect older hardware only. It doesn't fail on SKL but I can reproduce the crash by overriding the PCI ID to ILK's. Bisect result seems bogus. The commit mentioned above simply changes the name of a variable and shouldn't lead to any functional changes. I can reproduce the crash with or without the commit above applied to my working tree. Would you mind bisecting further? Thanks!
I tried to reproduce this, and struck out. Everything crashes on every version of mesa. Some questions I have: Curro: Would we expect device id override to work, since it will result in sending ILK commands to the KBL gpu that we are using to spoof ILK? I couldn't get *any* version of mesa to run the trace in this way. Tapani: I couldn't replay your trace on real ILK hardware, even with 17.3 (reported to work by José). I imagine that the game queries features and produces a trace on KBL that can't work on older hardware. José: Based on Curro's feedback, I need to repeat your bisect. I can't do it without an apitrace of the workload taken on ILK. Can you create one for me? Sorry about the back-and-forth.
(In reply to Mark Janes from comment #10) > I tried to reproduce this, and struck out. Everything crashes on every > version of mesa. That's what I've observed too with the few releases I tried to replay Tapani's trace on. > Some questions I have: > > Curro: Would we expect device id override to work, since it will result in > sending ILK commands to the KBL gpu that we are using to spoof ILK? I > couldn't get *any* version of mesa to run the trace in this way. > It won't send any actual commands to the hardware while doing PCI ID override, but it shouldn't crash either while compiling shaders assuming you've found a version that doesn't have the bug.
(In reply to Mark Janes from comment #10) > I tried to reproduce this, and struck out. Everything crashes on every > version of mesa. Some questions I have: >... > José: Based on Curro's feedback, I need to repeat your bisect. I can't do > it without an apitrace of the workload taken on ILK. Can you create one for > me? Sorry about the back-and-forth. Well, looks like this crash was fixed in last commits : Mesa 19.0.1 does not crash anymore here. I suppose we can close this bug?
(In reply to Mark Janes from comment #10) > I tried to reproduce this, and struck out. Everything crashes on every > version of mesa. Some questions I have: That is strange, trace works for me on HSW machine but I don't have ILK to test with. Note that sometimes different apitrace versions also conflict so some of this crashing might be simply because of that.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.