I've been trying to narrow the problem down and I found that compiling MAME normally with the default options and then recompiling src/emu/ioport.c with -fno-ipa-cp-clone fixes the issue.
So it's -fipa-cp-clone in ioport.c that is causing the problem. This looks like a GCC bug to me, and I hope this information is enough to report it and have it fixed, but if you need something else from me let me know. As a workaround, compiling MAME with -O2 or -fno-ipa-cp-clone would also fix the bug (as a side note, the MAME binary is ~15% smaller when compiled with -O2 instead of -O3 in i386). I'm using gcc 4.7.1-2 in a i386 wheezy system (the kernel is 64bit, though). Ah, and I tried this fix with both MAME and MESS, and it works in both cases. $ dpkg-architecture DEB_BUILD_ARCH=i386 DEB_BUILD_ARCH_BITS=32 DEB_BUILD_ARCH_CPU=i386 DEB_BUILD_ARCH_ENDIAN=little DEB_BUILD_ARCH_OS=linux DEB_BUILD_GNU_CPU=i486 DEB_BUILD_GNU_SYSTEM=linux-gnu DEB_BUILD_GNU_TYPE=i486-linux-gnu DEB_BUILD_MULTIARCH=i386-linux-gnu DEB_HOST_ARCH=i386 DEB_HOST_ARCH_BITS=32 DEB_HOST_ARCH_CPU=i386 DEB_HOST_ARCH_ENDIAN=little DEB_HOST_ARCH_OS=linux DEB_HOST_GNU_CPU=i486 DEB_HOST_GNU_SYSTEM=linux-gnu DEB_HOST_GNU_TYPE=i486-linux-gnu DEB_HOST_MULTIARCH=i386-linux-gnu $ uname -a Linux hermes 3.2.0-2-amd64 #1 SMP Mon Jun 11 19:23:01 UTC 2012 x86_64 GNU/Linux Berto -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
