Re: no-branch-likely

[law at redhat dot com]

 In message <pan.2002.02.27.12.55.29.845249.22687@redhat.com>, Eric 
Christopher
writes:
 > >> > 2001-01-24  Mike Stump  <mrs@wrs.com>
 > >>
 > >> >          * config/mips/mips.h (mips_no_branch_likely): Add.
 > >> >          (mips_no_branch_likely_string): Add.
 > >> >          (GENERATE_BRANCHLIKELY): Use mips_no_branch_likely.
 > >> >          (TARGET_OPTIONS): Add -mno-branch-likely. *
 > >> >          config/mips/mips.c (mips_no_branch_likely): Add.
 > >> >          (mips_no_branch_likely_string): Add.
 > >> >          (override_options): Handle -mno-branch-likely.
 > 
 > > Done. "Eric"?
 > 
 > That would be me :) I thought I said _something_ about the patch, but
 > then again I miss them on occasion. 
I said something at the time too -- basically I asked for an explanation
of why this was worth doing.  I never got a response, so I never approved
the patch.

Through other means I learned about the performance penalty for branch-likely
on newer MIPS implementations, but by then I was way too busy to go back
and dig out Mike's patch.

 > The question I have is whether or not it would be best just to turn off
 > branch-likely for the sibyte chip in the sources? However, I can see a
 > reason for the patch anyhow so if Mike would like to check it in then
 > that's fine.
I think we'd want to do three things.

  1. Provide an option by which users can turn branch likely on/off.

  2. Provide a way by which a port can choose if it wants to generate
  branch likely by default or not.

  3. Use #2 to turn off branch-likely on the SIbyte port by default.

jeff