From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 17155 invoked by alias); 29 Nov 2009 13:39:16 -0000 Received: (qmail 17111 invoked by uid 48); 29 Nov 2009 13:39:05 -0000 Date: Sun, 29 Nov 2009 13:39:00 -0000 Subject: [Bug tree-optimization/42216] New: [4.5 Regression] 464.h265ref peak regressed 20% X-Bugzilla-Reason: CC Message-ID: Reply-To: gcc-bugzilla@gcc.gnu.org To: gcc-bugs@gcc.gnu.org From: "rguenth 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: 2009-11/txt/msg02492.txt.bz2 See http://gcc.opensuse.org/SPEC/CINT/sb-balakirew-head-64-2006/recent.html First bad rev. is 154713, last good is 154686. It's quite obvious that either the regrename.c changes have code generation differences or that removing the vec_interleave_* expanders caused this regression. Richard - you removed all vec_interleave_* expanders but the vectorizer still checks for optab_for_tree_code (VEC_INTERLEAVE_*_EXPR) which ends up looking at vec_interleave_*_optab. Either this code is all dead now or you caused the vectorization no longer to apply. I'll check if reverting your revision gets back performance. -- Summary: [4.5 Regression] 464.h265ref peak regressed 20% Product: gcc Version: 4.5.0 Status: UNCONFIRMED Keywords: missed-optimization Severity: normal Priority: P3 Component: tree-optimization AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: rguenth at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42216