linux-mips
[Top] [All Lists]

Re: [PATCH]: R10000 Needs LL/SC Workaround in Gcc

To: gcc-patches@gcc.gnu.org
Subject: Re: [PATCH]: R10000 Needs LL/SC Workaround in Gcc
From: Kumba <kumba@gentoo.org>
Date: Sat, 01 Nov 2008 20:00:57 -0400
Cc: Linux MIPS List <linux-mips@linux-mips.org>, rdsandiford@googlemail.com
In-reply-to: <87tzargrn4.fsf@firetop.home>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <490A90F4.6040601@gentoo.org> <490C05A9.9070707@gentoo.org> <87abcjibsl.fsf@firetop.home> <490CA4C8.40904@gentoo.org> <87tzargrn4.fsf@firetop.home>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Thunderbird 2.0.0.17 (Windows/20080914)
Richard Sandiford wrote:

To be clear, the first option above was to check -- in mips_override_options --
that -mfix-r10000 is only used in cases where -mbranch-likely is in effect.
If we pick that option, it would be an error to use -mfix-r10000 in
other cases, and any code protected by TARGET_FIX_R10000 would be free
to use branch-likely instructions.  (Actually, we should use sorry()
instead of error() to report something like this.)

[snip]

That's the second option above, yes.  In other words, -mfix-r10000
would support both -mbranch-likely and -mno-branch-likely, and act
accordingly.

So do I need to worry about modifying the asm templates at all? Or is that only needed if pursuing option #2?

The branch-likely stuff is only going to work for MIPS-II or higher targets. In the odd (but possible) cases where MIPS-I might be used with -mfix-r10000, I assume we'll still have to emit 28 nops prior to a beq/beqz instruction. Is this already taken care of someplace?


...that's a good question.  My take is "no".  I don't think we want
-mfix-r10000 to enable branch-likely instructions in cases where
it isn't necessary for R10000 errata.  If we take the first option,
we can simply raise an error if:

Hmm, okay. Might this work to enable -mbranch-likely if -mfix-r10000? (Kind of guessing by looking at other segments of code).

  if ((target_flags_explicit & MASK_BRANCHLIKELY) == 0)
    {
      if (ISA_HAS_BRANCHLIKELY
          && (optimize_size
              || (!(target_flags_explicit & MASK_FIX_R10000) == 0)
              || (mips_tune_info->tune_flags & PTF_AVOID_BRANCHLIKELY) == 0))
        target_flags |= MASK_BRANCHLIKELY;
      else
        target_flags &= ~MASK_BRANCHLIKELY;
    }



My understanding so far for -mfix-r10000:
- Gets enabled if -march=r10000 is passed (done)
- Enable -mbranch-likely if not already enabled on >= MIPS-II (working on)
- Emits beqzl in the asm templates if enabled and >= MIPS-II (unsure)
- Emits 28 nops prior to beq/beqz if enabled and == MIPS-I (unsure)
- Ditto for asm templates (unsure)
- Documentation (not done)

Missing anything?


Yeah, I was wondering that too.  I did a search, but couldn't
find anything.

It seems we just need to use nop only and not worry about ssnop.


--
Joshua Kinard
Gentoo/MIPS
kumba@gentoo.org

"The past tempts us, the present confuses us, the future frightens us. And our lives slip away, moment by moment, lost in that vast, terrible in-between."

--Emperor Turhan, Centauri Republic

<Prev in Thread] Current Thread [Next in Thread>