This is the mail archive of the
mailing list for the GCC project.
Re: G5 fpu optimizations for g77 3.4???
- From: David Edelsohn <dje at watson dot ibm dot com>
- To: Jack Howarth <howarth at nitro dot med dot uc dot edu>
- Cc: gcc at gcc dot gnu dot org
- Date: Tue, 21 Oct 2003 10:41:25 -0400
- Subject: Re: G5 fpu optimizations for g77 3.4???
>>>>> Jack Howarth writes:
Jack> I was wondering if anyone knew the exact state of fpu scheduling
Jack> optimizations in g77 of the current gcc 3.4 tree? I ask because
Jack> the current gcc 3.1 compiled g77 code shows marginal performance on
Jack> a dual G5. My understanding is that the G5 is the first desktop
Jack> ppc with two fpu's per processor and will require specific fpu
Jack> scheduling changes to be implemented. Is this really the case and
Jack> what flags are operational in current gcc 3.4 g77 to do this?
GCC 3.4 includes scheduling description improvement to produce
better code for PPC970 / G5 which are enabled with the -mcpu=970
commandline option. GCC also requires profile feedback information to
generate some loop alignment instructions, so it is helpful to use that
GCC functionality as well. However, the new loop optimization
infrastructure in GCC 3.4 produces worse code than GCC 3.3, so the Apple
GCC compiler is the best choice.