This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: sched clobber patch killing the PA
- To: law at cygnus dot com
- Subject: Re: sched clobber patch killing the PA
- From: Richard Henderson <rth at cygnus dot com>
- Date: Mon, 8 Mar 1999 23:46:41 -0800
- Cc: egcs-bugs at egcs dot cygnus dot com, egcs-patches at egcs dot cygnus dot com
- References: <23384.920929094@hurl.cygnus.com>
On Mon, Mar 08, 1999 at 02:38:14PM -0700, Jeffrey A Law wrote:
> This change:
> Sun Mar 7 11:21:02 1999 Richard Henderson <rth@cygnus.com>
>
> * haifa-sched.c (reg_last_clobbers): New.
> (reg_pending_clobbers, bb_reg_last_clobbers): New.
> (compute_block_backward_dependences): Allocate memory for them.
> (schedule_region): Likewise.
> (sched_analyze_1): Clobbers don't interfere with one another.
> They do interfere with sets ...
> (sched_analyze_2): ... and uses.
> (sched_analyze): Likewise.
> (sched_analyze_insn): Update reg_last_clobbers appropriately.
>
> Is breaking the PA.
Oops. Just missed a case. Such things don't happen on the targets
I tested.
r~
* haifa-sched.c (sched_analyze_1): Fix last change -- add clobber
dependancies to sets in the non-hard-reg case too.
Index: haifa-sched.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/haifa-sched.c,v
retrieving revision 1.83
diff -c -p -d -r1.83 haifa-sched.c
*** haifa-sched.c 1999/03/07 11:50:32 1.83
--- haifa-sched.c 1999/03/09 07:43:35
*************** sched_analyze_1 (x, insn)
*** 3392,3398 ****
add_dependence (insn, XEXP (u, 0), REG_DEP_OUTPUT);
if (code == SET)
! SET_REGNO_REG_SET (reg_pending_sets, regno);
else
SET_REGNO_REG_SET (reg_pending_clobbers, regno);
--- 3392,3402 ----
add_dependence (insn, XEXP (u, 0), REG_DEP_OUTPUT);
if (code == SET)
! {
! for (u = reg_last_clobbers[regno]; u; u = XEXP (u, 1))
! add_dependence (insn, XEXP (u, 0), REG_DEP_OUTPUT);
! SET_REGNO_REG_SET (reg_pending_sets, regno);
! }
else
SET_REGNO_REG_SET (reg_pending_clobbers, regno);