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: FDO regression in Skidmarks component ParserVid

From: Jim Wilson <wilson at specifixinc dot com>
To: Peter Steinmetz <steinmtz at us dot ibm dot com>
Cc: gcc at gcc dot gnu dot org
Date: Thu, 13 May 2004 17:46:16 -0700
Subject: Re: FDO regression in Skidmarks component ParserVid
References: <OF8C32CC7C.E62716F9-ON86256E8C.005F0FDC-86256E8C.0066BDD4@us.ibm.com>

Peter Steinmetz wrote:

Here is an example of a performance regression when using FDO.

You might consider submitting a bug report into our bugzilla database. Sending a messge to the gcc list doesn't actually accomplish much, as we don't track problems reported here.

gcc -O2 -fprofile-use -funroll-loops -m32 -c parse.i

You could try various alternatives such as -fold-unroll-loops, -funroll-all-loops, -fold-unroll-all-loops. The old unroller still gives better results for some programs. -- Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com

References:
- FDO regression in Skidmarks component ParserVid
  - From: Peter Steinmetz

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