Version: 3.0.0 (using KDE Devel) Installed from: Compiled sources Compiler: gcc 4.0.1 OS: Linux I was trying to run OpenOffice.org 2 under valgrind 3.0.0 on AMD64. Didn't work: ==15374== Nulgrind, a binary JIT-compiler. ==15374== Copyright (C) 2002-2005, and GNU GPL'd, by Nicholas Nethercote. ==15374== Using LibVEX rev 1313, a library for dynamic binary translation. ==15374== Copyright (C) 2004-2005, and GNU GPL'd, by OpenWorks LLP. ==15374== Using valgrind-3.0.0, a dynamic binary instrumentation framework. ==15374== Copyright (C) 2000-2005, and GNU GPL'd, by Julian Seward et al. ==15374== For more details, rerun with: -v ==15374== unhandled opc_aux = 0x 3 first_opcode == 0xDC vex amd64->IR: unhandled instruction bytes: 0xDC 0x1D 0xD0 0x32 assumed next %rip = 0x27F61B68 actual next %rip = 0x27F61B62 vex: the `impossible' happened: bbToIR_AMD64: disInstr miscalculated next %rip vex storage: P 640, T total 37651088 (804109), T curr 12360 (343) valgrind: the 'impossible' happened: LibVEX called failure_exit(). ==15374== at 0x7001AB53: vgPlain_core_panic_at (m_libcassert.c:181) ==15374== by 0x7014CFFF: ??? ==15374== by 0x70029282: vgPlain_do_syscall (m_syscall.c:207) ==15374== by 0x7001AEAB: vgPlain_write (m_libcfile.c:106) ==15374== by 0x7001BB31: send_bytes_to_logging_sink (m_libcprint.c:57) ==15374== by 0x7001AEAB: vgPlain_write (m_libcfile.c:106) ==15374== by 0x7001BB31: send_bytes_to_logging_sink (m_libcprint.c:57) ==15374== by 0x7001BC2D: vgPlain_vprintf (m_libcprint.c:103) ==15374== by 0x700E77FF: (within /usr/lib/valgrind/stage2) sched status: running_tid=1 Thread 1: status = VgTs_Runnable ==15374== at 0x27F647B0: logl (in /lib/libm-2.3.5.so) ==15374== by 0x27BF10BE: (within /usr/lib/libstlport_gcc.so.4.6) ==15374== by 0x27BF7151: (within /usr/lib/libstlport_gcc.so.4.6) ==15374== by 0x27B8FD82: (within /usr/lib/libstlport_gcc.so.4.6)
That's an fcomp floating point instruction.
Fixed (vex r1323), although I could not test it properly. Can you check out the 3.0 line as described at http://www.valgrind.org/devel/cvs_svn.html and see if it works now?
Seems to work now. Thanks a lot for the fast fix.