Re: other/7114: ICE building strcoll.op from glibc-2.2.5

[Alan Modra <amodra at bigpond dot net dot au>]

The following reply was made to PR other/7114; it has been noted by GNATS.

From: Alan Modra <amodra@bigpond.net.au>
To: Richard Henderson <rth@redhat.com>, Geoff Keating <geoffk@redhat.com>,
  d.mueller@elsoft.ch, gcc-gnats@gcc.gnu.org, gcc-patches@gcc.gnu.org,
  dje@watson.ibm.com
Cc:  
Subject: Re: other/7114: ICE building strcoll.op from glibc-2.2.5
Date: Tue, 16 Jul 2002 14:38:08 +0930

 On Mon, Jul 15, 2002 at 09:31:24PM -0700, Richard Henderson wrote:
 > On Tue, Jul 16, 2002 at 11:08:16AM +0930, Alan Modra wrote:
 > > It gets worse.  rs6000/sysv4.h sets PROFILE_BEFORE_PROLOGUE.
 > 
 > Make rs6000/sysv4.h can use PROFILE_HOOK instead.
 
 Aye, that's the nice way to do it.  However, on powerpc64-linux,
 I've had kernel people complaining that the profiling code isn't
 what they want:  All those register saves from the prologue
 preceding the mcount call apparently are messing up accurate
 count values, and it's hard for an mcount implementation to
 adjust times, or so I'm told.  I implemented a simple hack to
 do PROFILE_BEFORE_PROLOGUE on powerpc64-linux for people who
 want it.
 
 I suspect we'll get the same sort of complaint if we change
 powerpc mcount.  There's also the issue that some special-purpose
 mcount functions may expect to be called before the stack has
 been adjusted.
 
 I'm rapidly approaching the point where I either give up on this
 problem, or simply remove support for profiling on nested
 functions.  The current code just doesn't work.