Small fix for ifcvt

Submitter	Jakub Jelinek
Date	Oct. 14, 2010, 7:23 p.m.
Message ID	<20101014192345.GA18103@tyan-ft48-01.lab.bos.redhat.com>
Download	mbox \| patch
Permalink	/patch/67854/
State	New
Headers	show Return-Path: <gcc-patches-return-275454-incoming=patchwork.ozlabs.org@gcc.gnu.org> Date: Thu, 14 Oct 2010 21:23:45 +0200 From: Jakub Jelinek <jakub@redhat.com> To: gcc-patches@gcc.gnu.org Subject: [PATCH] Small fix for ifcvt Message-ID: <20101014192345.GA18103@tyan-ft48-01.lab.bos.redhat.com> Reply-To: Jakub Jelinek <jakub@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.21 (2010-09-15) Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk Sender: gcc-patches-owner@gcc.gnu.org

Comments

Jakub Jelinek - Oct. 14, 2010, 7:23 p.m.

Hi!

If a _Bool SSA_NAME != boolean_false_node condition is fold_build2_loc'ed
into the SSA_NAME itself, invert_truthvalue_loc for it is TRUTH_NOT_EXPR <SSA_NAME>,
which is not a valid condition, so add_to_dst_predicate_list forces it into
queued stmts and then uses the new temporary as the condition.  We end up
then
with D.12346_679 = !D.12345_678 and/or D.12347_680 = D.12345_678 || D.12346_679
useless statements in the basic block, which prevent further vectorization.
By canonicalization of the condition to SSA_NAME == boolean_false_node
we can avoid emitting those useless stmts.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2010-10-14  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/46008
	* tree-if-conv.c (predicate_bbs): Try to canonicalize c2
	if possible.


	Jakub

Diego Novillo - Oct. 14, 2010, 7:26 p.m.

On Thu, Oct 14, 2010 at 15:23, Jakub Jelinek <jakub@redhat.com> wrote:

> 2010-10-14  Jakub Jelinek  <jakub@redhat.com>
>
>        PR tree-optimization/46008
>        * tree-if-conv.c (predicate_bbs): Try to canonicalize c2
>        if possible.

OK.

Diego.

Patch

--- gcc/tree-if-conv.c.jj	2010-09-06 08:42:00.000000000 +0200
+++ gcc/tree-if-conv.c	2010-10-13 17:49:58.000000000 +0200
@@ -915,7 +915,7 @@  predicate_bbs (loop_p loop)
 
 	    case GIMPLE_COND:
 	      {
-		tree c2;
+		tree c2, tem;
 		edge true_edge, false_edge;
 		location_t loc = gimple_location (stmt);
 		tree c = fold_build2_loc (loc, gimple_cond_code (stmt),
@@ -932,6 +932,9 @@  predicate_bbs (loop_p loop)
 
 		/* If C is false, then FALSE_EDGE is taken.  */
 		c2 = invert_truthvalue_loc (loc, unshare_expr (c));
+		tem = canonicalize_cond_expr_cond (c2);
+		if (tem)
+		  c2 = tem;
 		add_to_dst_predicate_list (loop, false_edge, cond, c2);
 
 		cond = NULL_TREE;

Patchwork Small fix for ifcvt

Comments

Patch