Re: other/10301: Side effects of architecture specific flags are not documented.

Re: other/10301: Side effects of architecture specific flags are not documented.

[Garen]

The following reply was made to PR other/10301; it has been noted by GNATS.

From: Garen <garen@wsu.edu>
To: gcc-gnats@gcc.gnu.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org,
	hubicka@gcc.gnu.org, garen@garen.net
Cc:  
Subject: Re: other/10301: Side effects of architecture specific flags are not documented.
Date: Thu, 24 Apr 2003 06:38:31 -0700

 According to http://gcc.gnu.org/gcc-3.2/changes.html, which says:
 
 Fixed common compiler crashes with SSE instruction set enabled (implied by 
 -march=pentium3, pentium4, athlon-xp)
 
 I still don't see those implications of -msse and -msse2 anywhere in the 
 manual still.  I don't see how that couldn't be "confusing".
 
 In the x86 option sections the flags are listed with no info:
 
 -mmmx 
 -mno-mmx 
 
 -msse 
 -mno-sse 
 
 -msse2 
 -mno-sse2 
 
 -m3dnow 
 
 It would be nice if instead it was something like:
 
 -mmmx	Enable MMX instructions.  Implied by -march=pentium-mmx, -march=pentium2
 
 -mno-mmx	Disable MMX instructions.
 
 -msse	Enable SSE instructions.  Implied by -march=pentium3, -march=athlon-*
 -mno-sse	Disable SSE instructions.
 
 -msse2	Enable SSE2 instructions.  Implied by -march=pentium4, ...
 -mno-sse2	Disable SSE2 instructions.
 
 -m3dnow	Enable 3DNow! instructions.  Implied by -march=athlon-*
 
 
 

Re: other/10301: Side effects of architecture specific flags are not documented.

[Garen]

The following reply was made to PR other/10301; it has been noted by GNATS.

From: Garen <garen@garen.net>
To: gcc-gnats@gcc.gnu.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org,
   hubicka@gcc.gnu.org, garen@garen.net
Cc:  
Subject: Re: other/10301: Side effects of architecture specific flags are not documented.
Date: Thu, 3 Apr 2003 16:35:46 -0800

 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=10301
 
 The part I was referring to inparticular was what the -march settings turn on.  
 For example, -march=pentium4 turns on -msse2.  All of those things aren't 
 mentioned anywhere in the manual:
 
 http://gcc.gnu.org/onlinedocs/gcc-3.2.2/gcc/i386-and-x86-64-Options.html#i386%20and%20x86-64%20Options
 
 
 

Re: other/10301: Side effects of architecture specific flags are not documented.

[bangerth]

Synopsis: Side effects of architecture specific flags are not documented.

Responsible-Changed-From-To: unassigned->hubicka
Responsible-Changed-By: bangerth
Responsible-Changed-When: Thu Apr  3 15:10:03 2003
Responsible-Changed-Why:
    Knows best about this stuff
State-Changed-From-To: open->feedback
State-Changed-By: bangerth
State-Changed-When: Thu Apr  3 15:10:03 2003
State-Changed-Why:
    To be honest, I don't find the manual all that hard to 
    understand. As to the complications you cite in your
    first link, it might be worth to also read Jan's reply to
    the linked-to mail, which gives a good reason for the
    existence of these flags. We are certainly not opposed to
    changing the docs, but maybe you could tell us what exactly
    you didn't understand in the gcc manuals?
    
    As to the alleged general instability of sse/... code: Jan
    has fixed lots of bugs, but in general we can only fix
    bugs that we know of. So unless people send in bug reports
    there's nothing we can do.
    
    With regard to disabling SSE[2]: If you don't specify some
    additional flags, sse is not used.
    
    Regards
      Wolfgang

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=10301

other/10301: Side effects of architecture specific flags are not documented.

[garen]

>Number:         10301
>Category:       other
>Synopsis:       Side effects of architecture specific flags are not documented.
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          doc-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Apr 03 02:36:00 UTC 2003
>Closed-Date:
>Last-Modified:
>Originator:     garen@garen.net
>Release:        All
>Organization:
>Environment:

>Description:
Looking in the gcc archives, I noticed a post that sums up the problem fairly well:

http://gcc.gnu.org/ml/gcc/2002-12/msg00235.html

There is a lot of confusion about what is implied by the -march settings.  Over in Gentoo land, there are several long running forum threads with hundreds of posts of people asking about what flags they ought to use: 

http://forums.gentoo.org/viewtopic.php?t=43648


Also, wrong code-gen bugs related to SSE2 insns are so widespread I kind of wonder why they aren't disabled, documentad as an experimental feature, or generate some kind of warning when used.

It would be very helpful if these sort of things were documented.  I'd think they'd go in the "Intel 386 and AMD x86-64 Options" section of the manual (http://gcc.gnu.org/onlinedocs/gcc-3.2.2/gcc/i386-and-x86-64-Options.html#i386%20and%20x86-64%20Options)


>How-To-Repeat:

>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: