From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 30087 invoked by alias); 14 Jun 2009 19:54:35 -0000 Received: (qmail 30054 invoked by uid 48); 14 Jun 2009 19:54:24 -0000 Date: Sun, 14 Jun 2009 19:54:00 -0000 Message-ID: <20090614195424.30053.qmail@sourceware.org> X-Bugzilla-Reason: CC References: Subject: [Bug rtl-optimization/20070] If-conversion can't match equivalent code, and cross-jumping only works for literal matches In-Reply-To: Reply-To: gcc-bugzilla@gcc.gnu.org To: gcc-bugs@gcc.gnu.org From: "steven at gcc dot gnu dot org" Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-bugs-owner@gcc.gnu.org X-SW-Source: 2009-06/txt/msg00895.txt.bz2 ------- Comment #28 from steven at gcc dot gnu dot org 2009-06-14 19:54 ------- Created an attachment (id=17995) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17995&action=view) Patch agains r148322, works pre-RA only Joern's original ifcvt.c patch only dealt with pre-reload if-conversion. The subsequent changes to make struct-equiv work for crossjumping and after reload, made the code too complicated IMHO. So I've gone back to the roots of the patch. I've simplified things a bit -- mostly by using the DF machinery. This new attached patch is far from complete though. The struct-equiv code should use rtx_equal_p_cb, but the rtx_equal_p_cb needs to be modified first (to be more like for_each_rtx: 3-state and passing around a pointer to auxiliary data). The local_reg_p stuff should probably go into df-problems.c as a _p function. And so on. But the patch does work. I wanted to let folks now that this bug is not yet forgotten! -- steven at gcc dot gnu dot org changed: What |Removed |Added ---------------------------------------------------------------------------- AssignedTo|unassigned at gcc dot gnu |steven at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20070