diff mbox

Avoid ICE in LRA for calls with multiple return values

Message ID 20141003083513.GA50375@msticlxl57.ims.intel.com
State New
Headers show

Commit Message

Ilya Enkovich Oct. 3, 2014, 8:35 a.m. UTC 
On 03 Oct 00:05, Ilya Enkovich wrote:
> 2014-10-02 22:30 GMT+04:00 Jeff Law <law@redhat.com>:
> > On 10/02/14 08:30, Ilya Enkovich wrote:
> >>
> >> Hi,
> >>
> >> This patch adds a check for call destination register for a call return
> >> value optimization based on REG_RETURNED note.  This solves some ICE issues
> >> for MPX codes.
> >>
> >> Bootstrapped and tested on linux-x86_64.  OK for trunk?
> >>
> >> Thanks,
> >> Ilya
> >> --
> >> 2014-10-02  Ilya Enkovich  <ilya.enkovich@intel.com>
> >>
> >>         * lra-constraints.c (inherit_in_ebb): Handle calls with
> >>         multiple return values.
> >>         * caller-save.c (save_call_clobbered_regs): Likewise.
> >
> > This is fine.  Please put a comment before the new if (REG_P (...)))
> > conditional which notes that a CALL_INSN may have multiple return values
> > expressed in an EXPR_LIST.  That way if someone looks at this code a few
> > years down the line, they'll know exactly why the code checks REG_P rather
> > than just using the result blindly.
> >
> > Did this fix all the multiple-return issues you encountered?
> 
> With this fix almost all benchmarks compile successfully but there is
> still one ICE in LRA.  I described it in MPX ABI thread.
> 
> Ilya
> 
> >
> > Jeff
> >

Here is a vestion to be committed.

Ilya
--
2014-10-02  Ilya Enkovich  <ilya.enkovich@intel.com>

	* lra-constraints.c (inherit_in_ebb): Handle calls with
	multiple return values.
	* caller-save.c (save_call_clobbered_regs): Likewise.
diff mbox

Patch

diff --git a/gcc/caller-save.c b/gcc/caller-save.c
index 11df2ec..e28facb 100644
--- a/gcc/caller-save.c
+++ b/gcc/caller-save.c
@@ -879,8 +879,13 @@  save_call_clobbered_regs (void)
 		  if (GET_CODE (pat) == PARALLEL)
 		    pat = XVECEXP (pat, 0, 0);
 		  dest = SET_DEST (pat);
-		  newpat = gen_rtx_SET (VOIDmode, cheap, copy_rtx (dest));
-		  chain = insert_one_insn (chain, 0, -1, newpat);
+		  /* For multiple return values dest is PARALLEL.
+		     Currently we handle only single return value case.  */
+		  if (REG_P (dest))
+		    {
+		      newpat = gen_rtx_SET (VOIDmode, cheap, copy_rtx (dest));
+		      chain = insert_one_insn (chain, 0, -1, newpat);
+		    }
 		}
 	    }
           last = chain;
diff --git a/gcc/lra-constraints.c b/gcc/lra-constraints.c
index 5f68399..f1642f3 100644
--- a/gcc/lra-constraints.c
+++ b/gcc/lra-constraints.c
@@ -5348,16 +5348,21 @@  inherit_in_ebb (rtx_insn *head, rtx_insn *tail)
 		  if (GET_CODE (pat) == PARALLEL)
 		    pat = XVECEXP (pat, 0, 0);
 		  dest = SET_DEST (pat);
-		  start_sequence ();
-		  emit_move_insn (cheap, copy_rtx (dest));
-		  restore = get_insns ();
-		  end_sequence ();
-		  lra_process_new_insns (curr_insn, NULL, restore,
-					 "Inserting call parameter restore");
-		  /* We don't need to save/restore of the pseudo from
-		     this call.	 */
-		  usage_insns[regno].calls_num = calls_num;
-		  bitmap_set_bit (&check_only_regs, regno);
+		  /* For multiple return values dest is PARALLEL.
+		     Currently we handle only single return value case.  */
+		  if (REG_P (dest))
+		    {
+		      start_sequence ();
+		      emit_move_insn (cheap, copy_rtx (dest));
+		      restore = get_insns ();
+		      end_sequence ();
+		      lra_process_new_insns (curr_insn, NULL, restore,
+					     "Inserting call parameter restore");
+		      /* We don't need to save/restore of the pseudo from
+			 this call.	 */
+		      usage_insns[regno].calls_num = calls_num;
+		      bitmap_set_bit (&check_only_regs, regno);
+		    }
 		}
 	    }
 	  to_inherit_num = 0;