public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [ARC] RFA: Use new rtl iterators in arc600_corereg_hazard
@ 2014-10-25  9:58 Richard Sandiford
  2014-11-02 18:41 ` Joern Rennecke
  0 siblings, 1 reply; 2+ messages in thread
From: Richard Sandiford @ 2014-10-25  9:58 UTC (permalink / raw)
  To: gcc-patches; +Cc: joern.rennecke

This is part of a series to remove uses of for_each_rtx from the ports.

Tested by making sure there were no code changes for gcc.dg, gcc.c-torture
and g++.dg for arc-elf.  OK to install?

Thanks,
Richard


gcc/
	* config/arc/arc.c (arc600_corereg_hazard_1): Delete.
	(arc600_corereg_hazard): Use FOR_EACH_SUBRTX.

Index: gcc/config/arc/arc.c
===================================================================
--- gcc/config/arc/arc.c	2014-10-25 09:51:25.901887608 +0100
+++ gcc/config/arc/arc.c	2014-10-25 09:51:26.279890977 +0100
@@ -7698,38 +7698,6 @@ disi_highpart (rtx in)
   return simplify_gen_subreg (SImode, in, DImode, TARGET_BIG_ENDIAN ? 0 : 4);
 }
 
-/* Called by arc600_corereg_hazard via for_each_rtx.
-   If a hazard is found, return a conservative estimate of the required
-   length adjustment to accomodate a nop.  */
-
-static int
-arc600_corereg_hazard_1 (rtx *xp, void *data)
-{
-  rtx x = *xp;
-  rtx dest;
-  rtx pat = (rtx) data;
-
-  switch (GET_CODE (x))
-    {
-    case SET: case POST_INC: case POST_DEC: case PRE_INC: case PRE_DEC:
-      break;
-    default:
-    /* This is also fine for PRE/POST_MODIFY, because they contain a SET.  */
-      return 0;
-    }
-  dest = XEXP (x, 0);
-  /* Check if this sets a an extension register.  N.B. we use 61 for the
-     condition codes, which is definitely not an extension register.  */
-  if (REG_P (dest) && REGNO (dest) >= 32 && REGNO (dest) < 61
-      /* Check if the same register is used by the PAT.  */
-      && (refers_to_regno_p
-	   (REGNO (dest),
-	   REGNO (dest) + (GET_MODE_SIZE (GET_MODE (dest)) + 3) / 4U, pat, 0)))
-    return 4;
-
-  return 0;
-}
-
 /* Return length adjustment for INSN.
    For ARC600:
    A write to a core reg greater or equal to 32 must not be immediately
@@ -7761,8 +7729,31 @@ arc600_corereg_hazard (rtx_insn *pred, r
       || recog_memoized (pred) == CODE_FOR_umul64_600
       || recog_memoized (pred) == CODE_FOR_umac64_600)
     return 0;
-  return for_each_rtx (&PATTERN (pred), arc600_corereg_hazard_1,
-		       PATTERN (succ));
+  subrtx_iterator::array_type array;
+  FOR_EACH_SUBRTX (iter, array, PATTERN (pred), NONCONST)
+    {
+      const_rtx x = *iter;
+      switch (GET_CODE (x))
+	{
+	case SET: case POST_INC: case POST_DEC: case PRE_INC: case PRE_DEC:
+	  break;
+	default:
+	  /* This is also fine for PRE/POST_MODIFY, because they
+	     contain a SET.  */
+	  continue;
+	}
+      rtx dest = XEXP (x, 0);
+      /* Check if this sets a an extension register.  N.B. we use 61 for the
+	 condition codes, which is definitely not an extension register.  */
+      if (REG_P (dest) && REGNO (dest) >= 32 && REGNO (dest) < 61
+	  /* Check if the same register is used by the PAT.  */
+	  && (refers_to_regno_p
+	      (REGNO (dest),
+	       REGNO (dest) + (GET_MODE_SIZE (GET_MODE (dest)) + 3) / 4U,
+	       PATTERN (succ), 0)))
+	return 4;
+    }
+  return 0;
 }
 
 /* For ARC600:

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [ARC] RFA: Use new rtl iterators in arc600_corereg_hazard
  2014-10-25  9:58 [ARC] RFA: Use new rtl iterators in arc600_corereg_hazard Richard Sandiford
@ 2014-11-02 18:41 ` Joern Rennecke
  0 siblings, 0 replies; 2+ messages in thread
From: Joern Rennecke @ 2014-11-02 18:41 UTC (permalink / raw)
  To: GCC Patches, Joern Rennecke, rdsandiford

On 25 October 2014 10:56, Richard Sandiford <rdsandiford@googlemail.com> wrote:
> This is part of a series to remove uses of for_each_rtx from the ports.
>
> Tested by making sure there were no code changes for gcc.dg, gcc.c-torture
> and g++.dg for arc-elf.  OK to install?
>
> Thanks,
> Richard
>
>
> gcc/
>         * config/arc/arc.c (arc600_corereg_hazard_1): Delete.
>         (arc600_corereg_hazard): Use FOR_EACH_SUBRTX.

OK.

> +  FOR_EACH_SUBRTX (iter, array, PATTERN (pred), NONCONST)

I was wondering for a while what kind of (NON)CONST this was about...
but as I glean
from the source, as long as nobody packs an address with a side effect
into a (CONST (MEM (...)) -
which should never happen in the first place, no matter if the MEM
itself is really const - we
should be fine.

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2014-11-02 18:41 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-10-25  9:58 [ARC] RFA: Use new rtl iterators in arc600_corereg_hazard Richard Sandiford
2014-11-02 18:41 ` Joern Rennecke

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).