Date:	Wed, 16 Dec 1998 11:56:51 +0200 (WET)
From:	Andris Pavenis <pavenis AT ieva05 DOT lanet DOT lv>
To:	djgpp-workers AT delorie DOT com
cc:	Robert Hoehne <robert DOT hoehne AT gmx DOT net>
Subject:	egcs-1.1.1
Message-ID:	<Pine.A41.4.05.9812161135440.19204-100000@ieva05.lanet.lv>
MIME-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com

Hi!

Tests done by SET and me shows that egcs-1.1.1 binaries for DJGPP uploaded
to Simtelnet (test version) by default uses option -march=pentium.
I earlier tested if binaries (compiler itself and binaries generated
by it) work on 386DX40 and didn't met problems. (I built egcs binaries
with such compiler and didn't found problems running them on 386)

That is --target=i586-pc-* forces default to be -march=pentium. However
it seems that binaries still works even on 386. There is simple way how to
test actual defaults of compiler:

	rm -f foo.c
	touch foo.c
	gcc -S -fverbose_asm foo.c
	cat foo.s

After that foo.s contains actual options compiler is using.

One additional note: I tested in Linux that egcs-1.0.3 distributed 
	with Slackware-3.6 also defaults to use -march=pentium but
	I haven't heard about problems caused by this (I think still 
	rather many poeple are using Slackware-3.5 or 3.6 on compilers
	that has 386 or 486 processors). 

So seems that it's safe to use target i586-pc-msdosdjgpp even 
if it forces default to be '-march=pentium'

About using -march=i386: I have found one problem with optimizer
	(wrong code generated with optimization in one my C++ program )
	when -march=i386 is being used (there is no such problem with
	-march=i486 or -march=pentium and also with gcc-2.8.1 at all).
	Only egcs-1.1 and egcs-1.1.1 seems to suffer from this. I have
	also reproduced this bug under Linux (with egcs-1.1.1).

	Some reordering statements in my program fixed the problem
	(but anyway wrong code from compiler is dangerous error)

	Perhaps I'll check this under Linux with latest egcs snapshot
	(it's being built now)  


Andris

- Raw text -