This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
patch to improve x86_adjust_cost
- To: egcs-patches at egcs dot cygnus dot com
- Subject: patch to improve x86_adjust_cost
- From: Jan Hubicka <hubicka at atrey dot karlin dot mff dot cuni dot cz>
- Date: Tue, 13 Apr 1999 16:00:48 +0200
Hi
this is small patch to make x86_adjust_cost more correct for Pentium CPU.
Honza
Tue Apr 13 00:32:03 CEST 1999
* i386.c (x86_adjust_cost): Agi stall takes 1 cycle on Pentium, fst
requires value to be ready one extra cycle.
*** i386.c.old2 Mon Apr 12 23:12:40 1999
--- i386.c Tue Apr 13 00:35:10 1999
*************** x86_adjust_cost (insn, link, dep_insn, c
*** 5572,5578 ****
return 0;
if (agi_dependent (insn, dep_insn))
! return 3;
if (GET_CODE (insn) == INSN
&& GET_CODE (PATTERN (insn)) == SET
--- 5594,5600 ----
return 0;
if (agi_dependent (insn, dep_insn))
! return 2;
if (GET_CODE (insn) == INSN
&& GET_CODE (PATTERN (insn)) == SET
*************** x86_adjust_cost (insn, link, dep_insn, c
*** 5581,5586 ****
--- 5603,5617 ----
&& GET_CODE (next_inst) == JUMP_INSN)
/* compare probably paired with jump */
return 0;
+
+ /* Stores stalls one cycle longer than other insns. */
+ if (is_fp_insn (insn) && cost && is_fp_store (dep_insn))
+ cost++;
+
break;
case PROCESSOR_K6: