On Mon, 2019-06-03 at 10:31 -0400, Matt Arsenault wrote: > > On Jun 2, 2019, at 2:36 PM, Jan Vesely <jan.ves...@rutgers.edu> wrote: > > > > he other problem is that even with killed process hung GPU usually > > makes the machine unable to suspend or reboot on its own, which kills > > remote testing. > > I'd need to recheck if that's still the case with linux-5.1. > > I was able to interrupt the process normally and everything worked > OK with rocm (which due to the device name regex, its the only > platform this runs) without my fix applied.
Does rocm do anything special other than using compute rings? What was the HW you tested? I checked that raven can reboot after gpu hangs/crashes (not suspend, but that's probably one of many raven problems). I'd like to check carrizo/iceland too, as that's the machine that get accessed remotely. Other than that I think it's OK to just put default 30s timeout on all CL tests, even debug build of LLVM shouldn't need more than that. Jan > > -Matt > _______________________________________________ > Piglit mailing list > Piglit@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/piglit -- Jan Vesely <jan.ves...@rutgers.edu>
signature.asc
Description: This is a digitally signed message part
_______________________________________________ Piglit mailing list Piglit@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/piglit
