Bill Allombert on Sat, 25 Mar 2006 09:13:54 +0100

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: parigp on a macintel core duo

On Fri, Mar 24, 2006 at 09:40:17PM -0800, Justin C. Walker wrote:
> I have a version that I built myself, with GMP:
>          GP/PARI CALCULATOR Version 2.2.13 (development CHANGES-1.1500)
>            i386 running darwin (ix86/GMP-4.1.4 kernel) 32-bit version
>        compiled: Mar  9 2006, gcc-4.0.1 (Apple Computer, Inc. build  
> 5250)
>                 (readline v5.1 enabled, extended help available)
> Note this version "knows" it is on an ix86 processor, and uses GMP.
> gp seems to work without problems, but if I "make test-kernel", I get  
> a "KERNEL BUG" response, with a 'kern.dif' file showing a half-dozen  
> or so differences from the expected output.  I'm attaching the  
> "gp.out" file, since it may prove enlightening to Those Who Know.
> I am also trying to build SAGE (currently, < 
> sage>, in process of relocating to Seattle, WA), which incorporates  
> Pari/gp as a base component, and the build this case will fail with  
> the same error you see.
> Any suggestions on how to proceed will be greatly appreciated.  Of  
> course, I can build with "kernel=none", as suggested by Bill, but it  
> would be nice to have a "fully functional" version for the  
> Intellimacs...

I suppose SAGE ship PARI/GP 2.2.12 which include some asm files that
darwin assembler cannot parse.

On the other hand, PARI/GP CVS (and upcoming 2.2.13) does not include
such files so should work fine with darwin.

> Bill: is the asm code really useless?

There were actually two kind of asm files:

1) The 'gcc asm inline' ones. Those are used in practice, make things
faster and work fine with darwin.

2) The .s files. Thoses were not used in practice, did not make things
any faster when they were actually used and did not work with darwin.

We removed 2) in CVS but we kept 1).

The flag --kernel=none cause PARI/GP to not use both 1) and 2) so cause
a slowdown.