MIPS SB-1 DFA scheduler patch

[cgd@broadcom.com]

Now to actually look at the patch.  I'm not really gcc-clued to
comment on most of it, of course, so most of my comments are spelling-
and comment-related.  8-)


At Tue, 20 Apr 2004 06:12:04 +0000 (UTC), "Jim Wilson" wrote:
> + /* Whether the SB1 FP exceptions are enabled.

Textually, SB-1 is, well, dash-ful.  8-)


> +   { "sb1-fp-exceptions=", &mips_sb1_fp_except_string,			\
> +       N_("For SB1, specify whether FP exceptions are enabled"), 0}	\

likewise, but see previous mail.  8-)


> + ;; cycle ifetch bubble.  We assume all branches predicated not taken.

"predicted"?

FWIW, branch-likely instructions are always predicted taken, but
shouldn't be emitted by the compiler unless the user has explicitly
requested them.  (It might be interesting to enable them and see what
the compiler does with them.  If used judiciously, they could be
beneficial.)

I don't know if this assumption is reasonable, esp. when thinking
about loops.  For "large" loops and predicted-taken branches, the fact
that fetch is decoupled from exe may hide the taken-branch fetch
bubble (at least, that's my understanding).  I think the guidance
would be "try to make sure that loops have at least 2 cycles worth of
instructions in their body" or something.  (But, honestly, I'm no
expert about instruction scheduling.)



chris