From patchwork Sat Jul 3 10:54:59 2010 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Iain Sandoe X-Patchwork-Id: 57800 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) by ozlabs.org (Postfix) with SMTP id 1ABC3B7085 for ; Sat, 3 Jul 2010 20:55:13 +1000 (EST) Received: (qmail 4745 invoked by alias); 3 Jul 2010 10:55:10 -0000 Received: (qmail 4731 invoked by uid 22791); 3 Jul 2010 10:55:09 -0000 X-SWARE-Spam-Status: No, hits=-2.1 required=5.0 tests=AWL, BAYES_00, RCVD_IN_DNSWL_NONE X-Spam-Check-By: sourceware.org Received: from c2bthomr02.btconnect.com (HELO c2bthomr02.btconnect.com) (213.123.20.120) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Sat, 03 Jul 2010 10:55:04 +0000 Received: from thor.office (host81-138-1-83.in-addr.btopenworld.com [81.138.1.83]) by c2bthomr02.btconnect.com with ESMTP id MKO40786; Sat, 3 Jul 2010 11:55:00 +0100 (BST) X-Mirapoint-IP-Reputation: reputation=Fair-1, source=Queried, refid=0001.0A0B0302.4C2F1704.00DB, actions=tag Message-Id: <4E7077C2-6DBB-4612-8084-29BC39681EB2@sandoe-acoustics.co.uk> From: IainS To: GCC Patches Mime-Version: 1.0 (Apple Message framework v936) Subject: [Patch libobjc] Fix Darwin powerpc m64 breakage. Date: Sat, 3 Jul 2010 11:54:59 +0100 Cc: Andrew Pinski , mrs@gcc.gnu.org X-IsSubscribed: yes Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org it's nice when 4 lines of code fixes 4 dozen test-suite fails .. .. and disappointing that it's probably taken more than 4 years for someone to have time to look .. Andrew, I realize that this doesn't in any way alter the 'broken-ness' of the current design - but, on the other hand, it does make powerpc- apple-darwin9 work @m64 within that current design. [... and gets rid of a bunch of test-suite fail output for those of us who test powerpc @m64]. FWIW, actually fixing m32 might be somewhat more taxing since that involves replication of the xxx_special_round_type_align () functions (which have drifted away from the implementation in encode.c.). I had a quick scan but didn't see any specific PR to put this against... OK for trunk? Iain libobjc/Changelog: * encoding.c: Add TARGET_ALIGN_NATURAL definition for m64 powerpc darwin. Add a comment as to why, update FIXME comments. testsuite/ChangeLog: * objc.dg/gnu-encoding/struct-layout-encoding-1_generate.c: Update XFAILs. ({ const char *_fields = TYPE_FIELDS (STRUCT); \ @@ -116,8 +128,12 @@ static int __attribute__ ((__unused__)) not_target && TYPE_MODE (strip_array_types (TREE_TYPE (_fields))) == DFmode) \ ? MAX (MAX (COMPUTED, SPECIFIED), 64) \ : MAX (COMPUTED, SPECIFIED));}) + /* FIXME: The word 'fixme' is insufficient to explain the wrong-ness - of this next macro definition. */ + of this next macro definition. + FIXME1: It's also out of date, darwin no longer has the same alignment + 'special' as aix - this is probably the origin of the m32 breakage, since + the 'special' routine is not used at m64 with TARGET_ALIGN_NATURAL=1. */ #define darwin_rs6000_special_round_type_align(S,C,S2) \ rs6000_special_round_type_align(S,C,S2) Index: gcc/testsuite/objc.dg/gnu-encoding/struct-layout- encoding-1_generate.c =================================================================== --- gcc/testsuite/objc.dg/gnu-encoding/struct-layout- encoding-1_generate.c (revision 161768) +++ gcc/testsuite/objc.dg/gnu-encoding/struct-layout- encoding-1_generate.c (working copy) @@ -236,7 +236,7 @@ switchfiles (int fields) || filecnt == 22) { fprintf (outfile, "\ -/* { dg-do run { xfail powerpc*-*-darwin* powerpc*-*-aix* } } */\n\ +/* { dg-do run { xfail { { \"powerpc*-*-darwin*\" && { ! lp64 } } || { \"powerpc*-*-aix*\" } } } } */\n\ /* { dg-options \"-w -I%s -fgnu-runtime\" } */\n", srcdir); } /* FIXME: these should not be xfailed but they are because Index: libobjc/encoding.c =================================================================== --- libobjc/encoding.c (revision 161768) +++ libobjc/encoding.c (working copy) @@ -105,10 +105,22 @@ static int __attribute__ ((__unused__)) not_target #undef ALTIVEC_VECTOR_MODE #define ALTIVEC_VECTOR_MODE(MODE) (0) +/* Furthermore, some (powerpc) targets also use TARGET_ALIGN_NATURAL + in their alignment macros. Currently[4.5/6], rs6000.h points this + to a static variable, initialized by target overrides. This is reset + in linux64.h but not in darwin64.h and therefore, all the darwin64 + stuff is broken. This echoes the comments above and below about + the badness of the approach. Other powerpc64 targets might also + be affected. */ + +#if defined(DEFAULT_ABI) && (DEFAULT_ABI==ABI_DARWIN) && __LP64__ +# undef TARGET_ALIGN_NATURAL +# define TARGET_ALIGN_NATURAL 1 +#endif /* FIXME: while this file has no business including tm.h, this definitely has no business defining this macro but it - is only way around without really rewritting this file, + is only way around without really rewriting this file, should look after the branch of 3.4 to fix this. */ #define rs6000_special_round_type_align(STRUCT, COMPUTED, SPECIFIED) \