I suspect the cause is OpenBLAS. While this library is built with
DYNAMIC_ARCH=1, there is still the non-performance critical code which will
use whatever instructions are available on the machine at compile time
unless you also set TARGET. See
https://github.com/xianyi/OpenBLAS/issues/3056
For the macOS app, Marc Culler had to set TARGET=CORE2 so that it would run
on a 2013 cylindrical Mac Pro (this version is not yet posted).
Best,
Nathan
On Wednesday, August 25, 2021 at 10:15:02 PM William Stein wrote:
> Hi,
>
> I release cocalc-docker ([1], [2]) a few days ago with sage-9.4, and had
> people reporting back "illegal instruction" issues. Sage is built there
> with SAGE_FAT_BINARY="yes", using this script [3]. One person who reported
> a problem had a "Dell PowerEdge R710." and I built on an intel Dell
> PowerEdge R630, for what it is worth. They got the error "An illegal
> instructor occurred" when computing "5+2" in Sage.
>
> My solution for now is to keep cocalc-docker at sage-9.3 for the time
> being, and suggest that people build sage from source on their own machine
> if necessary. I'll let you know if I have any further information, but I
> just wanted to add this data point.
>
> [1] https://github.com/sagemathinc/cocalc-docker
> [2] https://hub.docker.com/r/sagemathinc/cocalc
> [3]
> https://github.com/sagemathinc/cocalc-docker/blob/master/scripts/install_sage.sh
>
> On Wednesday, August 25, 2021 at 8:57:44 AM UTC-7 Marco Streng wrote:
>
>> In the end I build 9.4 from source and that worked perfectly. And it
>> turned out that the Ubuntu 18.04 binaries of 9.2 also work on this
>> particular machine (though we went for 9.4 in the end). Thanks again for
>> the help.
>>
>> Op dinsdag 24 augustus 2021 om 18:03:01 UTC+2 schreef slelievre:
>>
>>> 2021-08-24 05:48 UTC, Dima Pasechnik:
>>> >
>>> > These "illegal instruction" errors are an indication
>>> > that the binary was built for a more advanced CPU
>>> > than yours - which does not support certain CPU
>>> > commands. It's a bug in "fat binary" build process
>>> > we see for some time.
>>> >
>>> > Build Sage yourself, or use Conda.
>>>
>>> I opened a ticket at binary-pkg for that:
>>> https://github.com/sagemath/binary-pkg/issues/31
>>>
>>
--
You received this message because you are subscribed to the Google Groups
"sage-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to sage-support+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/sage-support/62622bd6-6fcd-4b94-8fce-455517ad093an%40googlegroups.com.
