The bug appears to be something in gf2x's configure.ac file, but as I know
little about writing config scripts it's hard to give advice.
In the meantime, here is a simple step-by-step for those who want SAGE-5.11
but have an old pre-SSE2 machine (schools & cheapskates like me):
1) Confirm that your machine doesn't have SSE2 - look for the sse2 flag in
the /proc/cpuinfo file.
2) Edit the gf2x spkg file:
a) copy sage-5.11/spkg/standard/gf2x-1.1.spkg to a working location
b) tar -jxvf gf2x-1.1.spkg
c) Edit gf2x-1.1/spkg-install:
Edit line 38 to read ./configure --enable-sse2=no
--prefix="$SAGE_LOCAL"
d) tar -jcvf gf2x-1.1.spkg gf2x-1.1
e) move gf2x-1.1.spkg back to its original location in
sage-5.11/spkg/standard/.
3) Compile SAGE as usual
Most folks in this situation will not want to compile SAGE themselves (it
took me 37 hours on my 1.3 GHz Athlon) and it would be nice to have a
pre-compiled version available. An easy solution would be a binary from
before the bug was introduced (gf2x-1.1 and hence SAGE-5.11), but I don't
see that older binary versions are available from sagemath.org.
Robert Campbell
On Friday, September 20, 2013 9:01:13 AM UTC-4, Andrew Fiori wrote:
>
> This is likely a build error bug on all systems which do not support SSE2
> but where building is done using the current version of gcc. It may also
> manifest as a runtime error on binary builds installed on systems which do
> not support SSE2 (but I have not tested that).
>
> I am submitting this here rather than to bug tracker for several reasons:
> - It is unlikely anyone cares enough to fix this properly upstream
> (which is where the bug really is).
> - The build error only effects building from source on systems that are
> out of date and will becoming even less common going forward.
>
> I am mentioning it at all, because it will help anyone who comes across
> this error in the future. The same type of error may cause somewhat
> confusing runtime errors in other packages (bug tracker searches suggest
> this type of issue has come up before in other packages).
>
> The bug:
> The bug is somewhere in the gf2x build scripts "src/config/acinclude.m4"
> or "src/configure" somehow it decides whether or not the system supports
> sse2 (look around line 75 of acinclude.m4. It appears to do this by
> checking if gcc can compile something that uses sse2. It turns out that it
> can... even if the system you are on cannot. Consequently it tries to build
> everything with sse2. This causes a problem when it tries to tune itself
> "src/src/tune-lowlevel.pl". Around line 70-78 it tries to run the
> compiled code. Most of these don't require sse2 really, and so work, some
> do (mul3t, mul3k, mul3k2) these fail, it then can't pick which one to use,
> and terminates building.
> If there are other places in sage where similar sse2 detection is used,
> but which are not executed at build time (most things are not), then sage
> will encounter unexpected runtime bugs when pieces of code that can benefit
> from sse2 are used.
>
>
> The solution:
> gf2x doesn't require sse2 to work. The easiest way to build it on a
> system where the above bug is encountered is to modify:
> spkg-install:38 to become:
> ./configure --enable-sse2=no --prefix="$SAGE_LOCAL"
> You don't really want to do this in the main branch though, as this slows
> things down for the 99+% of people who don't have an archaic system they
> are trying to run sage on.
> A better solution is to add an if statement at this place and environment
> variable globally to disable sse2 when it doesn't exist.
> Alternatively one could invest the time and figure out then fix gf2x build
> system.
>
> - Andrew
>
>
--
You received this message because you are subscribed to the Google Groups
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at http://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/groups/opt_out.
