__builtin_expect vs inlining

[Richard Henderson]

On Sat, May 05, 2001 at 06:17:18PM +0400, Ivan Kokshaysky wrote:
> > Eh?  Would you give me an example that isn't working properly?
> 
> Sure.

Fixed thus.

> So one of the questions: can one rely on current branch predictions
> algorithms (val < 0, val = 0 false etc.) in the long term?

Err, no.  We reserve the right to tweek the predictions, or to replace
them with different heuristics.  I'd hope they'd be _generally_ better
heuristics, though the effect on any one particular test might change.

r~

        * integrate.c (copy_insn_list): Substitute NOTE_EXPECTED_VALUE.

Index: integrate.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/integrate.c,v
retrieving revision 1.142
diff -c -p -r1.142 integrate.c
*** integrate.c	2001/05/03 16:14:34	1.142
--- integrate.c	2001/05/05 16:54:24
*************** copy_insn_list (insns, map, static_chain
*** 1536,1541 ****
--- 1536,1546 ----
  		  else
  		    NOTE_BLOCK (copy) = *mapped_block_p;
  		}
+ 	      else if (copy
+ 		       && NOTE_LINE_NUMBER (copy) == NOTE_INSN_EXPECTED_VALUE)
+ 		NOTE_EXPECTED_VALUE (copy)
+ 		  = copy_rtx_and_substitute (NOTE_EXPECTED_VALUE (insn),
+ 					     map, 0);
  	    }
  	  else
  	    copy = 0;