public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Jakub Jelinek <jakub@redhat.com>
To: Jeff Law <law@redhat.com>, John David Anglin <dave.anglin@bell.net>
Cc: Richard Biener <richard.guenther@gmail.com>,
	       "H.J. Lu" <hjl.tools@gmail.com>,
	GCC Patches <gcc-patches@gcc.gnu.org>
Subject: [PATCH] Fix up DSE - PR middle-end/64388, target/55023
Date: Tue, 06 Jan 2015 14:08:00 -0000	[thread overview]
Message-ID: <20150106140805.GK1667@tucnak.redhat.com> (raw)
In-Reply-To: <20150105213117.GZ1667@tucnak.redhat.com>

Hi!

On Mon, Jan 05, 2015 at 10:31:17PM +0100, Jakub Jelinek wrote:
> Or you could e.g. do the
>   if (HARD_FRAME_POINTER_IS_ARG_POINTER
>       && !reload_completed
>       && SIBLING_CALL_P (insn))
>     { add_wild_read (bb_info); return; }
> case first, then compute const_call and memset_call,
>   if (const_call || memset_call)
>     { current_big_block }
>   else if (reload_completed && SIBLING_CALL_P (insn))
>     add_wild_read (bb_info);
>   else
>     add_non_frame_wild_read (bb_info);
> 
> That way, you would not punish const/memset calls unnecessarily after reload
> when they are sibling calls.

So what about this way?  Even for the HARD_FRAME_POINTER_IS_ARG_POINTER
before reload case, there is no reason for wild read IMHO, just setting
frame_read should do all that is necessary.  Bootstrapped/regtested on
x86_64-linux and i686-linux, tested on the pr55023 testcase with cross
to hppa-linux.  Ok for trunk?

2015-01-06  Jakub Jelinek  <jakub@redhat.com>

	PR target/55023
	PR middle-end/64388
	* dse.c (struct insn_info): Mention frame_read set also
	before reload for tail calls on some targets.
	(scan_insn): Revert 2014-12-22 change.  Set frame_read
	also before reload for tail calls if
	HARD_FRAME_POINTER_IS_ARG_POINTER.  Call add_wild_read
	instead of add_non_frame_wild_read for non-const/memset
	tail calls after reload.

--- gcc/dse.c.jj	2015-01-05 22:41:19.000000000 +0100
+++ gcc/dse.c	2015-01-06 11:39:39.450832915 +0100
@@ -371,9 +371,11 @@ struct insn_info
 	either stack pointer or hard frame pointer based.  This means
 	that we have no other choice than also killing all the frame
 	pointer based stores upon encountering a const function call.
-     This field is set after reload for const function calls.  Having
-     this set is less severe than a wild read, it just means that all
-     the frame related stores are killed rather than all the stores.  */
+     This field is set after reload for const function calls and before
+     reload for const tail function calls on targets where arg pointer
+     is the frame pointer.  Having this set is less severe than a wild
+     read, it just means that all the frame related stores are killed
+     rather than all the stores.  */
   bool frame_read;
 
   /* This field is only used for the processing of const functions.
@@ -2483,17 +2485,6 @@ scan_insn (bb_info_t bb_info, rtx_insn *
 
       insn_info->cannot_delete = true;
 
-      /* Arguments for a sibling call that are pushed to memory are passed
-	 using the incoming argument pointer of the current function.  These
-	 may or may not be frame related depending on the target.  Since
-	 argument pointer related stores are not currently tracked, we treat
-	 a sibling call as though it does a wild read.  */
-      if (SIBLING_CALL_P (insn))
-	{
-	  add_wild_read (bb_info);
-	  return;
-	}
-
       /* Const functions cannot do anything bad i.e. read memory,
 	 however, they can read their parameters which may have
 	 been pushed onto the stack.
@@ -2527,7 +2518,13 @@ scan_insn (bb_info_t bb_info, rtx_insn *
 		     const_call ? "const" : "memset", INSN_UID (insn));
 
 	  /* See the head comment of the frame_read field.  */
-	  if (reload_completed)
+	  if (reload_completed
+	      /* Tail calls are storing their arguments using
+		 arg poinnter.  If it is a frame pointer on the target,
+		 even before reload we need to kill frame pointer based
+		 stores.  */
+	      || (SIBLING_CALL_P (insn)
+		  && HARD_FRAME_POINTER_IS_ARG_POINTER))
 	    insn_info->frame_read = true;
 
 	  /* Loop over the active stores and remove those which are
@@ -2601,7 +2598,11 @@ scan_insn (bb_info_t bb_info, rtx_insn *
 		}
 	    }
 	}
-
+      else if (SIBLING_CALL_P (insn) && reload_completed)
+	/* Arguments for a sibling call that are pushed to memory are passed
+	   using the incoming argument pointer of the current function.  After
+	   reload that might be (and likely is) frame pointer based.  */
+	add_wild_read (bb_info);
       else
 	/* Every other call, including pure functions, may read any memory
            that is not relative to the frame.  */


	Jakub

  reply	other threads:[~2015-01-06 14:08 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-12-31 14:31 PATCH: [5 Regression] r219037 caused FAIL: gcc.dg/pr44194-1.c H.J. Lu
2015-01-03 17:35 ` John David Anglin
2015-01-03 19:48   ` H.J. Lu
2015-01-03 20:10     ` John David Anglin
2015-01-03 20:18       ` H.J. Lu
2015-01-03 20:58         ` John David Anglin
2015-01-03 21:48           ` H.J. Lu
2015-01-04 11:37             ` Richard Biener
2015-01-04 14:57               ` H.J. Lu
2015-01-04 17:16                 ` Richard Biener
2015-01-05 14:19                   ` Jeff Law
2015-01-05 18:51                     ` Jakub Jelinek
2015-01-05 20:17                       ` John David Anglin
2015-01-05 21:24                         ` Jakub Jelinek
2015-01-05 21:31                           ` Jakub Jelinek
2015-01-06 14:08                             ` Jakub Jelinek [this message]
2015-01-06 17:06                               ` [PATCH] Fix up DSE - PR middle-end/64388, target/55023 John David Anglin
2015-01-06 19:12                                 ` Jakub Jelinek
2015-01-08 21:25                               ` Jeff Law

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150106140805.GK1667@tucnak.redhat.com \
    --to=jakub@redhat.com \
    --cc=dave.anglin@bell.net \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=hjl.tools@gmail.com \
    --cc=law@redhat.com \
    --cc=richard.guenther@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).