Brice, I am glad you asked me to test widely, because I did find 2 compilers that rejected my version with "=r" and one that generated bad code for that case. However, the original "=S" works everywhere I could test (list of 26 compilers below), as did "=D" and "=SD". Since the "=SD" constraint gives the compiler more choices than the other two, it should be preferred:
#elif defined(HWLOC_X86_32_ARCH) __asm__( "xchg %%ebx,%1\n\t" "cpuid\n\t" "xchg %%ebx,%1\n\t" : "=a" (*eax), "=SD" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (*eax), "2" (*ecx)); #else The tested platforms and compilers include the oldest and newest gcc, icc, pgcc and suncc versions I could find with x86 support. Unfortunately my Solaris/x86 platform is currently down. -Paul The list: Red Hat Linux 8.0 (32-bit host) gcc 2.96 gcc (GCC) 3.2 20020903 (Red Hat Linux 8.0 3.2-7) gcc (GCC) 6.0.0 20150706 [a development snapshot] OpenSuSE 10.2 (32-bit host) gcc (GCC) 4.1.2 20061115 (prerelease) (SUSE Linux) Scientific Linux 5.5 (-m32 on 64-bit host) pgi 9.0-4 pgi 13.10-0 pgi 14.10-0 icc (ICC) 11.1 20100414 icc (ICC) 14.0.2 20140120 Scientific Linux 6.6 (-m32 on 64-bit host) gcc (GCC) 4.4.7 20120313 (Red Hat 4.4.7-11) clang version 3.4.2 (tags/RELEASE_34/dot2-final) Open64 Compiler Suite: Version 4.5.2.1 Sun C 5.12 Linux_i386 2011/11/16 [a.k.a. Solaris Studio 12.3] Sun C 5.13 Linux_i386 2014/10/20 [a.k.a. Solaris Studio 12.4] Scientific Linux 6.6 (32-bit host) gcc (GCC) 4.4.7 20120313 (Red Hat 4.4.7-11) Debian Jessie gcc (Debian 4.9.2-10) 4.9.2 Ubuntu 12.04.5 (-m32 on 64-bit host) icc (ICC) 15.0.3 20150407 Ubuntu 14.04.02 (-m32 on 64-bit host) Sun Ceres C 5.10 Linux_i386 2009/03/06 [a.k.a. Studio Express 03/09] Sun C 5.11 Linux_i386 2010/08/13 [a.k.a. Solaris Studio 12.2] Mac OS X 10.5 (64-bit host, but 32-bit is default target) i686-apple-darwin9-gcc-4.0.1 (GCC) 4.0.1 (Apple Inc. build 5465) Mac OS X 10.6 (-m32 on 64-bit host) gcc 4.2.1 (Apple Inc. build 5646) Mac OS X 10.8 (-m32 on 64-bit host) Apple LLVM version 4.2 (clang-425.0.28) (based on LLVM 3.2svn) i686-apple-darwin11-llvm-gcc-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00) NetBSD 6.1.5 (32-bit host) gcc (NetBSD nb2 20110806) 4.5.3 OpenBSD 5.7 (32-bit host) gcc (GCC) 4.2.1 20070719 FreeBSD 10.1 (32-bit host) FreeBSD clang version 3.4.1 (tags/RELEASE_34/dot1-final 208032) 20140512 On Tue, Jul 21, 2015 at 11:22 AM, Paul Hargrove <phhargr...@lbl.gov> wrote: > Brice, > > I don't have anywhere near the testing coverage for Linux on x86 as I do > on x86-64. > However, I will see what I can do. > > -Paul > > On Tue, Jul 21, 2015 at 4:24 AM, Brice Goglin <brice.gog...@inria.fr> > wrote: > >> Thanks. >> Could you test this new asm on all your systems/compilers? I don't want >> break that fragile code again. >> Brice >> >> Le 21 juillet 2015 08:17:06 UTC+02:00, Paul Hargrove <phhargr...@lbl.gov> >> a écrit : >> >>> Oops - send the wrong asm code. >>> While "=S" is correct for the second constraint, I meant to send a >>> version that had "=r" because it allows the compiler more choices. >>> >>> -Paul >>> >>> On Mon, Jul 20, 2015 at 11:12 PM, Paul Hargrove <phhargr...@lbl.gov> >>> wrote: >>> >>>> PGI-14.10 for 32-bit targets fails in the same manner as 13.7, 13.9 and >>>> 13.10. >>>> >>>> I believe the following inline x86 asm is correct and more robust than >>>> the existing code that pgi appears to reject: >>>> >>>> #elif defined(HWLOC_X86_32_ARCH) >>>> __asm__( >>>> "xchg %%ebx,%1\n\t" >>>> "cpuid\n\t" >>>> "xchg %%ebx,%1\n\t" >>>> : "=a" (*eax), "=S" (*ebx), "=c" (*ecx), "=d" (*edx) >>>> : "0" (*eax), "2" (*ecx)); >>>> #else >>>> >>>> -Paul >>>> >>>> On Mon, Jul 20, 2015 at 9:50 PM, Paul Hargrove <phhargr...@lbl.gov> >>>> wrote: >>>> >>>>> Pavan, >>>>> >>>>> I can confirm that I see the same with PGI-13.10. >>>>> >>>>> I have a couple systems with 14.x installed but neither with 32-bit >>>>> support. >>>>> I am downloading 32-bit support now (which I am assuming will work >>>>> with the existing license) and will report back. >>>>> >>>>> -Paul >>>>> >>>>> On Mon, Jul 20, 2015 at 9:00 PM, Balaji, Pavan <bal...@anl.gov> wrote: >>>>> >>>>>> Hello, >>>>>> >>>>>> The hwloc-1.11 build seems to fail with the pgi compiler on 32-bit >>>>>> platforms. I see the following error: >>>>>> >>>>>> ----8<---- >>>>>> CC topology-x86.lo >>>>>> PGC-F-0000-Internal compiler error. unable to allocate a register >>>>>> 8 (topology-x86.c: 87) >>>>>> PGC/x86 Linux 13.9-0: compilation aborted >>>>>> ----8<---- >>>>>> >>>>>> I only tried pgi-13.7 and 13.9 (I don't have access to later compiler >>>>>> versions). It looks like the compiler doesn't like the assembly code in >>>>>> include/private/cpuid-x86.h for 32-bit platforms. >>>>>> >>>>>> >>>>>> >>>>>> Thanks, >>>>>> >>>>>> -- Pavan >>>>>> >>>>>> _______________________________________________ >>>>>> hwloc-devel mailing list >>>>>> hwloc-de...@open-mpi.org >>>>>> Subscription: >>>>>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel >>>>>> Link to this post: >>>>>> http://www.open-mpi.org/community/lists/hwloc-devel/2015/07/4501.php >>>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> Paul H. Hargrove phhargr...@lbl.gov >>>>> Computer Languages & Systems Software (CLaSS) Group >>>>> Computer Science Department Tel: +1-510-495-2352 >>>>> Lawrence Berkeley National Laboratory Fax: +1-510-486-6900 >>>>> >>>> >>>> >>>> >>>> -- >>>> Paul H. Hargrove phhargr...@lbl.gov >>>> Computer Languages & Systems Software (CLaSS) Group >>>> Computer Science Department Tel: +1-510-495-2352 >>>> Lawrence Berkeley National Laboratory Fax: +1-510-486-6900 >>>> >>> >>> >>> >>> -- >>> Paul H. Hargrove phhargr...@lbl.gov >>> Computer Languages & Systems Software (CLaSS) Group >>> Computer Science Department Tel: +1-510-495-2352 >>> Lawrence Berkeley National Laboratory Fax: +1-510-486-6900 >>> >>> ------------------------------ >>> >>> hwloc-devel mailing list >>> hwloc-de...@open-mpi.org >>> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel >>> Link to this post: >>> http://www.open-mpi.org/community/lists/hwloc-devel/2015/07/4505.php >>> >>> > > > -- > Paul H. Hargrove phhargr...@lbl.gov > Computer Languages & Systems Software (CLaSS) Group > Computer Science Department Tel: +1-510-495-2352 > Lawrence Berkeley National Laboratory Fax: +1-510-486-6900 > -- Paul H. Hargrove phhargr...@lbl.gov Computer Languages & Systems Software (CLaSS) Group Computer Science Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900
