Another crash
Alan Bateman
Alan.Bateman at oracle.com
Sat Jan 18 09:37:17 UTC 2020
On 18/01/2020 00:06, Kamath, Smita wrote:
>
> Hi Alan,
>
> It looks like you are running in a virtualized environment and somehow
> vbmi2 is enabled. However, the underlying hardware does not support
> vbmi2 feature as seen in /proc/cpuinfo from the log file. Therefore,
> when a vbmi2 instruction is executed, the hardware throws SIGILL.
>
> I have reverified that the hotspot test case located at
> compiler/intrinsics/bigInteger/TestShift.java runs successfully on
> cpu-family 6, model 85, stepping 7 which is the same configuration as
> mentioned in the hs_error file from Remi. The intrinsic optimization
> is only for platforms that support vbmi2 feature.
>
>
Thanks for the mail and details. I haven't located a system that reports
vbmi2 in the cpu flags. The closest (on an Oracle cloud hardware)
reports cpu 6, model 85, stepping 7 and claims to have avx512f but not
vbmi2. Rémi seems to be on hardware in Google's cloud. I think this
thread should move to hotspot-compiler-dev as I don't think there is any
Loom specific here (yes, we had an issues with the stubs that copy
to/from the stack but that has been patched and we'll do more testing on
that in time).
-Alan
More information about the loom-dev
mailing list