From xemacs-m  Thu Dec 12 09:59:39 1996
Received: from steadfast.teradyne.com (steadfast.teradyne.com [131.101.1.200]) by xemacs.cs.uiuc.edu (8.8.3/8.8.3) with ESMTP id JAA03703 for <xemacs-beta@xemacs.org>; Thu, 12 Dec 1996 09:59:37 -0600 (CST)
Received: from engine.ecf.teradyne.com (engine.ecf.teradyne.com [131.101.192.6]) by steadfast.teradyne.com (8.7.1/8.7.1) with ESMTP id LAA09490; Thu, 12 Dec 1996 11:01:23 -0500 (EST)
Received: from midnight.eng.ecf.teradyne.com (midnight.ecf.teradyne.com [131.101.192.49]) by engine.ecf.teradyne.com (8.7.1/8.7.1) with SMTP id QAA19154; Thu, 12 Dec 1996 16:58:05 +0100 (MET)
Received: by midnight.eng.ecf.teradyne.com (SMI-8.6/SMI-SVR4)
	id QAA00663; Thu, 12 Dec 1996 16:58:04 +0100
To: Tore Olsen <toreo@colargol.idb.hist.no>,
        "Steven L. Baur" <steve@miranova.com>
Cc: XEmacs Beta Mailing List <xemacs-beta@xemacs.org>
Subject: Re: 19.15b3 crash on alpha-dec-osf3.2
References: <yngengx3zlc.fsf@colargol.idb.hist.no>
	<m2ybf5ujys.fsf@altair.xemacs.org>
	<yngg21b3ipl.fsf@colargol.idb.hist.no>
Mime-Version: 1.0 (generated by tm-edit 7.94)
Content-Type: text/plain; charset=US-ASCII
From: Adrian Aichner <aichner@ecf.teradyne.com>
Date: 12 Dec 1996 16:58:03 +0100
In-Reply-To: Tore Olsen's message of 12 Dec 1996 16:06:46 +0100
Message-ID: <rxsiv6792lw.fsf@midnight.ecf.teradyne.com>
Lines: 91
X-Mailer: Gnus v5.2.40/XEmacs 20.0


Hello Tore, hello Steven,

I have made the same experience on sparc-sun-solaris2.5 compiled '-g -O3'.

Using sb-bench.el mailed by Steven on December 8th.

I byte-compiled a file containing
============================================================================
;(load-library "sb-bench.el")
;(setq bench-large-lisp-file "/xemacs-20.0-b30/lisp/prim/loaddefs.el")
(load "sb-bench.elc")
(setq bench-large-lisp-file "/xemacs-20.0-b30/lisp/prim/loaddefs.el")
(setq bench-enabled-profiling nil)
(bench 3)
============================================================================

When executing via M-x load-file bench.el it always hangs up during the
TOH benchmark.

When I execute M-x bench the bench-mark goes to sleep after fontifying
buffer. Another time it goes to sleep after moving thru the fontified
file. vmstat shows it's not doing anything. I can wake up the
benchmark by clicking into the XEmacs frame with mouse1.

Steven what are good values for

Variables and Constants:

   bench-enabled-profiling    - If non-nil and the underlying emacs
                                supports it, do function profiling. 
   bench-large-lisp-file      - Large lisp file to use in benchmarks
   bench-mark-function-alist  - Not documented.
   bench-mark-profile-buffer  - Buffer used for collection of profiling data.
   bench-post-bench-hook      - Hook for individual bench mark
                                statistic collection.
   bench-pre-bench-hook       - Hook for individual bench mark initialization.
   bench-sort-filename        - File to be used in the sort benchmark

?

Here is a result for a bench I woke up at the end of moving thru the
fontified buffer:

Function Name  Call Count  Elapsed Time  Average Time
=============  ==========  ============  ============
bench-mark-3   1           97.337663000  97.337663000
bench-mark-1   1           23.837537999  23.837537999
bench-mark-2   1           18.329911999  18.329911999
bench-mark-6   1           16.835245000  16.835245000
bench-mark-4   1           0.8830990000  0.8830990000

Which version of *bench.el should I fetch and how should I use it for
meaningful benchmarking?

Thanks,

Adrian

Tore Olsen <toreo@colargol.idb.hist.no> writes:

> 
> Steven L Baur <steve@miranova.com> writes:
> 
> > Tore> Profiling time alarm
> > 
> > Turn the profiling off to run the benchmarks.
> > 
> > (setq bench-enabled-profiling nil)
> > 
> > The latest version of the benchmark has profiling disabled by default.
> 
> I guess I missed a couple of patches.  I tried this, and it
> worked. However, I haven't been able to do a benchmark, since xemacs
> keeps standing still now and then. It seems worse on the alpha than on
> i686-linux. This is the most annoying bug I know. :-/
> 
> 
> > FWIW there are also crashes in profiling in Pentium Linux.  Is there
> > any system where profiling doesn't crash?
> 
> I tried it 4-5 times on a i486-linux, and it crashed once. On our
> PPro running Linux, I haven't had any trouble with profiling.
> 
> -Toreo
> -- 
> Tore Olsen <toreo@colargol.idb.hist.no> drift@colargol, TIHLDE, IDB HiST
> 
> Surprise, fear, and a ruthless dedication to the Gnu!
> 
> 

