This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: New loop unroller broken?

From: Dale Johannesen <dalej at apple dot com>
To: Andrew Pinski <pinskia at physics dot uc dot edu>
Cc: Ulrich Weigand <weigand at i1 dot informatik dot uni-erlangen dot de>, Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>, Dale Johannesen <dalej at apple dot com>, gcc mailing list <gcc at gcc dot gnu dot org>
Date: Thu, 22 Jan 2004 11:06:14 -0800
Subject: Re: New loop unroller broken?
References: <200401221847.TAA04312@faui1d.informatik.uni-erlangen.de> <36C99D46-4D0C-11D8-83FB-000393A6D2F2@physics.uc.edu>

On Jan 22, 2004, at 10:53 AM, Andrew Pinski wrote:

On Jan 22, 2004, at 10:47, Ulrich Weigand wrote:

void test(float *data, float d)
{
   int i;
   for (i = 0; i < 8; i++)
      data[i] = d;
}

because of
'Unable to prove that the loop iterates constant times'
which strikes me as somewhat odd.

Is this the same on other platforms?  Any ideas how to fix
this (or how to debug it)?


Yes I see this also on PPC because of the doloop optimizations.
I do know that this was fixed on rtlopt branch though but I do not
when it is going to be fixed.


Incidentally, my quick analysis was that the problem is caused by
doing doloop analysis before unrolling, instead of after as the old
unroller did.  Why was this choice made?

Follow-Ups:
- Re: New loop unroller broken?
  - From: Zdenek Dvorak

References:
- New loop unroller broken?
  - From: Ulrich Weigand
- Re: New loop unroller broken?
  - From: Andrew Pinski

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]