William Stein on Fri, 15 Dec 2006 22:24:03 +0100


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

Re: Lowering compiler optimisation with Apple gcc


On Fri, 15 Dec 2006 12:49:55 -0800, Jason Martin <jason.worth.martin@gmail.com> wrote:
> I've had excellent luck on my Core 2 Mac with the latest version of gcc using
>
>    CFLAGS="-m64 -fast"
>
> I've also ported and optimized GMP for Core 2 Macs, and using my port
> of GMP I get Pari benchmarks on my Mac Pro that out perform all of the
> Opteron machines I have access to.  For details, see my website (in my
> signature).

Precisely how do you build PARI after building GMP as above?  I tried
very naively to build PARI on my core2duo laptop after patching GMP
with your patches, and ran into numerous errors.

> I have not encountered any "bugs" in Apple's gcc, but I have
> encountered "frustrations" because the Apple gcc options are not
> compatible with the Linux options that I'm familiar with.

That's probably partly that you started using Xcode after they fix a lot
of their major problems...

> One subtle point of building for Core 2 Macs is that current versions
> of the autoconf tools do not correctly detect the CPU type.  This is
> because the "uname -m" command returns the rather unhelpful "i386"
> rather than the "x86_64" that one might expect.  So, you have to
> manually detect the CPU (Agner Fog has some nice routines for this).
>
> I have not tried building on the 32bit or PowerPC Macs because I don't
> have access to any.

Speaking of this, I'm going to use my NSF grant to build a network of
machines for SAGE development, and give all SAGE developers accounts on
them for teasting build issues, optimizing code, etc.   Since PARI is
a huge part of SAGE, this extends to interested PARI developers.  Please
modify this wiki page (see below) or email me if you have any suggestions
as to what hardware would be very useful for PARI developers to have
access to:
            http://sage.math.washington.edu:9001/hardware

William