Building GMP 6.0 for Ivy Bridge

Victor Shoup shoup at cs.nyu.edu
Fri Oct 2 22:21:11 UTC 2015


I think I figured it out. It has nothing to with the fact that I'm using a non-system compiler.  The problem is at line 113 of config.guess. It first tries cc rather than cc. On my university's system, where many servers share the same file system, cc points to some sun compiler. 

One workaround on my side is to set the environment variable CC to gcc.  That makes it work for me. A better solution is to fix config.guess. Can it be run after configure determines the compiler? That seems best. Otherwise, just fix the logic locally by testing if the compiled dummy program actually runs :-) 

Marc Glisse <marc.glisse at inria.fr> wrote:
>On Fri, 2 Oct 2015, Victor Shoup wrote:
>
>> Thanks.  Compiled and ran.
>> Output is "ivybridge".
>>
>> But ./config.guess outputs x86_64 and that's what is is
>> in ./config.log, too.
>
>Uh. Weird. Maybe config.guess is using a different compiler or
>something. 
>bash -x ./config.guess would help locate the issue (if there is no line
>
>"/bin/sh -c /tmp/.../dummy" then it probably failed to compile), and 
>further debugging (print the command that fails so it can be reproduced
>
>outside of config.guess) would likely identify the issue.
>
>
>On Fri, 2 Oct 2015, Victor Shoup wrote:
>
>> I tried --build=ivybridge...
>> On the latest snapshot, that works...the only difference I
>> can really discern between the two is that --march and --mtune
>> resolve to ivybridge, rather than corei7, which is what I get
>> when I build with coreisbr.
>>
>> My understanding is that there should be little difference between
>the
>> two microarchitectures.  It looks like the same assembly codes are
>pulled
>> in --- actually, it looks like GMP lumps ive/sandy bridge together...
>
>Yes, that's what configure.ac says.
>
>> Another note: on the official 6.0.0 release.
>> --build=ivybridge doesn't work at all...it doesn't even
>> pull x86 assembly codes at all.
>
>That release is getting old...
>
>> So maybe it is just a bug somewhere else in the configure script??
>
>Well the problem is clearly with config.guess, it is just hard to guess
>
>what it may be.
>
>-- 
>Marc Glisse

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.


More information about the gmp-discuss mailing list