(In reply to Timo Aaltonen from comment #8)
> Actually it wasn't avx512, that was the first one I tried :) It's enabled
> also on 3.7 and that version works fine. Only one that was added in 3.8 is
> PKU, but dropping just that didn't help.
>
> I did try dropping all non-client features (AVX512,
(In reply to Jose Fonseca from comment #6)
> It's not the first time LLVM misidentifies modern CPUs.
>
> I thought that all the logic in
> src/gallium/auxiliary/gallivm/lp_bld_misc.cpp for setting +/-foo mattrs
> would save us from this sort of grief.
For features we already know about (I think
(In reply to Timo Aaltonen from comment #2)
> how exactly? I've tried gdb:
Usually you could use x/i if it's in jit code when gcc can't figure
out the function (or just follow up to the caller and disassemble from there).
But it looks like the stack got smashed so I don't know if that really
Could you show the instruction where it crashed (and the disassembly)?
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1549849
Title:
11.2.0-rc1: llvmpipe tests fail if built on skylake
To manage
