Illegal instruction using Ubuntu nightlies on 32 bit amd

[WestleyArgentum]

I don't know if you guys intend to support 32 bit amd processors or not, but right now the nightlies wont open / give an Illegal instruction error :(

[ivarne]

Can you see if you find a usr/lib/julia/sys.so file and try to delete (or move) it?

#5459 (and linked issues)

[WestleyArgentum]

I don't see that file - maybe in the nightlies the sys.so is built on first run?

I see that the deb package doesn't list support for 32 bit amd - I think I mean this more as a special build request to @staticfloat :) Otherwise for now I'll build my own

[staticfloat]

The way Debian refers to processor architectures is strange; just because
amd64 exists, doesn't mean amd32 does. ;)

Your 32 bit processors should be covered by the i386 binaries that get
built. If they are segfaulting, then we have a problem.
On Feb 28, 2014 12:17 PM, "Westley Argentum Hennigh" <
[email protected]> wrote:

I don't see that file - maybe in the nightlies the sys.so is built on
first run?

I see that the deb package doesn't list support for 32 bit amd - I think I
mean this more as a special build request to @staticfloathttps://github.com/staticfloat:) Otherwise for now I'll build my own

Reply to this email directly or view it on GitHubhttps://github.com//issues/5987#issuecomment-36390138
.

[WestleyArgentum]

Our setup is a 64 bit amd server running 64 bit Ubuntu running 32 bit Ubuntu in a docker container (so, lots of moving parts). We're going to try to make and test a 32 bit Ubuntu virtual machine - but that will take us a little while...

Anyone in a better position to test on amd?

[dnelson]

I just tested on an Ubuntu 12.04 32-bit VM running on our AMD-based Eucalyptus cloud and got the same issue, so it doesn't appear to be the Docker container that is causing the trouble.

[ViralBShah]

Does anyone know if this is still an issue?

[staticfloat]

I'm pretty sure the builds should all have the proper MARCH parameters set. However, at the moment, we require at least a pentium4, which means you need to have SSE. Without SSE our tests don't pass due to floating-point shenanigans. Some very old 32-bit AMD processors don't have SSE, and that can cause illegal instruction errors.

[vtjnash]

i haven't seen any recent issues mentioning 32-bit issues, so i'm going to assume this has been addressed (by working harder at setting MARCH and JULIA_CPU_TARGET correctly)