Justin Walker on Tue, 11 Mar 2003 12:45:03 -0800

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

Re: 2 recent changes - Mac OS X check

On Tuesday, Mar 11, 2003, at 11:56 US/Pacific, Justin Walker wrote:

On Tuesday, Mar 11, 2003, at 06:13 US/Pacific, Bill Allombert wrote:

Hello PARI-dev,

I would like to announce two recent changes:

1) PARI has now an optimised inline assembly kernel for the PowerPC
architecture, thanks to Guillaume Hanrot.

Please test it especially on Mac OS X. We have only tested it on Linux.

I checked out ToT and it builds 'darwin-none'. What's the trick to get darwin-ppc and the corresponding gmp kernel?

Ah, well. After I sent this, I R'd the FM, and discovered the secret. Sort of. Here's the banner I get:

           GP/PARI CALCULATOR Version 2.2.6 (development CHANGES-1.704)
                   unknown (PPC/GMP-4.1 kernel) 32-bit version
                compiled: 03/11/03, gcc-3.1 20020420 (prerelease)
                (readline v4.1 enabled, extended help available)

                       Copyright (C) 2003 The PARI Group

So it has (I'm assuming) the ppc-gmp kernel, but I don't understand the 'unknown' bit.

Some other nits:
1) it builds in Odarwin-none. Why not Odarwin-ppc? Is this related to 'unknown' above?

2) The build didn't actually succeed; I had to hand-tweak it. There is a spurious "-Wl,,/usr/local/lib" in the 'ld' line that causes a hiccup:
            ld: can't open:  (No such file or directory, errno = 2)

I re-executed the load line spit out by the build process, with the -W removed, and it built successfully.

3) There are a few oddities in the 'configure' phase I don't yet get:
   - ...I did not find getrlimit.
     ...I did not find opendir.
    These are both in libSystem

   - ...I did not find dlopen.
This is reasonable, but is there a way to specify where libdl lives? It's available on my system in a (fink-provided) non-standard location. With 'libdl', we can support install(), as we established a while back (this is a sort of "If we had ham, we could have ham and eggs, if we had eggs" situation, since there was also a libtool issue, as I recall).

   - Using GNU readline, version 4.1
My system has (in /usr/local/lib, where configure found it) version 4.2 and "4":
     How did Configure come up with 4.1?

Here is the 'make bench' result for my 450Mhz G4 tower:
* Testing objets        for gp-sta..TIME=20
* Testing analyz        for gp-sta..TIME=310
* Testing number        for gp-sta..TIME=200
* Testing polyser       for gp-sta..TIME=80
* Testing linear        for gp-sta..TIME=100
* Testing elliptic      for gp-sta..TIME=200
* Testing sumiter       for gp-sta..TIME=370
* Testing graph         for gp-sta..TIME=100
* Testing program       for gp-sta..BUG [170]
* Testing trans         for gp-sta..TIME=600
* Testing nfields       for gp-sta..TIME=1810
+++ [BUG] Total bench for gp-sta is 2512

Here is the 'make bench' result from an earlier build with this banner:
 GP/PARI CALCULATOR Version 2.2.6 (development CHANGES-1.666)
      unknown (C portable/GMP-4.1 kernel) 32-bit version

* Testing objets        for gp-sta..TIME=30
* Testing analyz        for gp-sta..TIME=360
* Testing number        for gp-sta..TIME=250
* Testing polyser       for gp-sta..TIME=100
* Testing linear        for gp-sta..TIME=130
* Testing elliptic      for gp-sta..TIME=240
* Testing sumiter       for gp-sta..TIME=540
* Testing graph         for gp-sta..TIME=150
* Testing program       for gp-sta..BUG [180]
* Testing trans         for gp-sta..TIME=1020
* Testing nfields       for gp-sta..TIME=2270
+++ [BUG] Total bench for gp-sta is 3454

Looks like close to 30% improvement over the "portable GMP" kernel. Great work!




Justin C. Walker, Curmudgeon-At-Large  *
Institute for General Semantics        |   When LuteFisk is outlawed
                                       |   Only outlaws will have
                                       |       LuteFisk