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: gcc loop unrolling problem ??

From: Robert Dewar <dewar at gnat dot com>
To: Matt Fago <fago at caltech dot edu>
Cc: Jan Hubicka <hubicka at ucw dot cz>, gcc at gcc dot gnu dot org
Date: Thu, 12 Feb 2004 10:39:00 -0500
Subject: Re: gcc loop unrolling problem ??
References: <402B9B84.70602@caltech.edu>

Matt Fago wrote:

The number of unrollings depends on size of the loop and is bound to 8
maximally.  Your loop has only single instruction so you get maximal
unrolling.  See documentation of --param max-unrolled-insns,
max-unroll-times and friends.
Is this also the case for loops with a constant number of iterations (my loops tend to be in powers of 3)? Should the documentation mention that the default value for max-unroll-times=8?


I don't see why, this is really something quite internal to the
optimizer. Obviously high values for unrolling are a bad idea that
will slow the program down. The optimal caculation of unrolling
is highly target dependent. I would guess many people increasing
this parameter would be doing so for bogus reasons :-)

Follow-Ups:
- Re: gcc loop unrolling problem ??
  - From: Matt Fago

References:
- Re: gcc loop unrolling problem ??
  - From: Matt Fago

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