X-pop3-spooler:	POP3MAIL 2.1.0 b 4 980420 -bs-
Date:	Wed, 9 Sep 1998 13:42:19 -0500 (EST)
From:	Steven Snyder <ssnyder AT indy DOT net>
X-Sender:	ssnyder AT indy1
To:	pgcc mailing list <beastium-list AT Desk DOT nl>
Subject:	FYI: -fomit-frame-pointer increases errors on egcs testsuite
Message-ID:	<Pine.SUN.3.96.980909132235.9304B-100000@indy1>
MIME-Version:	1.0
Sender:	Marc Lehmann <pcg AT goof DOT com>
Status:	RO
X-Status:	A
Lines:	15

In an effort to find out how "safe" the pgcc v1.1a optimizations are, I
recently rebuilt it with optimization switches -O2 through -O6, then ran
the egcs testsuite on each build.  (Command line: "make bootstrap; make
check".)  CPU type was implicitly -mcpu=i686 -march=i686.

I found that the number of unexpected failures stayed static through -O4.
For -O5 and -O6 there was 1 more error (file g++.eh/spec1.C) in the c++
test summary.  I eventually tracked the problem to -fomit-frame-pointer,
which is enabled by the -O5 switch.  Allowing the creation of frame
pointers returned the testsuite error count to its previous lower value.

I mention this as a warning that not all the default optimizations are
safe.  The pgcc Web page notes (under FAQ) that -fomit-frame-pointer is
known to be unstable.  Apparently that warning is accurate.
 

- Raw text -