This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
branch prediction analyzis improvement
- To: gcc-patches at gcc dot gnu dot org, rth at cygnus dot com, patches at x86-64 dot org
- Subject: branch prediction analyzis improvement
- From: Jan Hubicka <jh at suse dot cz>
- Date: Thu, 14 Jun 2001 12:03:21 +0200
Hi
In case instruction do have REG_BR_PROB note, some predictors are not run,
then they are missing in the summary from analyze_brprob script.
It seems to me that it is not problem to make them run unconditionally. The opcode
and pointer heuristic is cheap so in average case it may even be more expensive
to seach for the note.
Thu Jun 14 11:26:08 CEST 2001 Jan Hubicka <jh@suse.cz>
* predict.c (estimate_probability): Do not bail out early
when note is present.
Index: predict.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/predict.c,v
retrieving revision 1.19
diff -c -3 -p -r1.19 predict.c
*** predict.c 2001/06/10 08:01:57 1.19
--- predict.c 2001/06/14 09:22:01
*************** estimate_probability (loops_info)
*** 311,319 ****
if (GET_CODE (last_insn) != JUMP_INSN
|| ! any_condjump_p (last_insn))
- continue;
-
- if (find_reg_note (last_insn, REG_BR_PROB, 0))
continue;
for (e = bb->succ; e; e = e->succ_next)
--- 311,316 ----