Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!olivea!uunet!ogicse!hakanson
From: hakanson@ogicse.ogi.edu (Marion Hakanson)
Newsgroups: comp.lang.perl
Subject: Re: Perl for HP-UX 7.0
Message-ID: <17163@ogicse.ogi.edu>
Date: 8 Feb 91 18:42:43 GMT
References: <1991Jan31.161906.9083@convex.com> <4768@ruuinf.cs.ruu.nl> <1991Feb06.083909.8101@Daisy.EE.UND.AC.ZA>
Organization: Oregon Graduate Institute (formerly OGC), Beaverton, OR
Lines: 23

In article <1991Feb06.083909.8101@Daisy.EE.UND.AC.ZA> barrett@Daisy.EE.UND.AC.ZA (Alan P. Barrett) writes:
>. . .
>3.  Run this little 'Pre-make' script to make some of the objects using
>    different CFLAGS.  The standard set of CFLAGS (including '-O') works
>    fine on all the other object files not mentioned in this script.  I
>    suspect that -O might also work on the objects mentioned here, if
>    you have enough physical memory or enough time to wait; on our
>    HP9000/370 with 8 Meg of memory, compiling these with '-O' makes the
>    VM system thrash the disk for hours (yes, hours).

These super optimizers do have trouble with eval.c and a2p/walk.c.
Our HP-UX 7.0 system is a Series-800 (HP's PA-RISC), but it has 32M of
RAM, so the VM only thrashes for about 5-10 minutes on eval.c, and no
special options are required to expand compiler tables.

On our poor DECStation-2100 with 12M of RAM, eval.c took 12+ hours to
fully optimize using the Ultrix compiler (I'm told gcc is superior).
Somehow I think I'd have to run Perl an awful lot in order to make
optimization pay back the time it took in this case.

-- 
Marion Hakanson         Domain: hakanson@cse.ogi.edu
                        UUCP  : {hp-pcd,tektronix}!ogicse!hakanson