#29 dump-every-bb should be long int

closed
nobody
5
2006-04-21
2006-03-28
Polzin
No

dump-every-bb should be "ULong" not "Int", like e.g.
bb_executions.
Giving a longer --dump-evey-bb=5123456789 truncates it
some 32bit number on a i686 linux machine.

Relevant files:
src/clo.h

else if (0 == VG_(strncmp)(arg, "--dump-every-bb=", 16))
CLG_(clo).dump_every_bb =
(ULong)VG_(atoll)(&arg[16]);

src/global.h

ULong dump_every_bb; /* Dump every xxx BBs. */

Discussion

  • Polzin

    Polzin - 2006-03-28

    Logged In: YES
    user_id=586666

    additional change in threads.c necessary:

    VG_(sprintf)(buf, "--dump-every-bb=%llu",
    CLG_(clo).dump_every_bb);

     
  • Josef Weidendorfer

    Logged In: YES
    user_id=621915

    Thanks. I never though that somebody would use this option
    with such a huge number.

    I am not really sure this option is important, as the
    "BB" numberering is quite fuzzy because of different
    sizes of basic blocks and possible non-determinism eg.
    in systemcalls. But it is some kind of rough "sampling".

     
  • Polzin

    Polzin - 2006-03-28

    Logged In: YES
    user_id=586666

    All I can say is: I use it.

    The run is 2-3 minutes without callgrind,
    2-3 hours with callgrind,
    producing 73514942529 basic blocks.

    Splitting automatically into 4-5 dumps gives an impression,
    which parts of the program grow stronger (quadratic vs.
    linear) than others.

     
  • Josef Weidendorfer

    Logged In: YES
    user_id=621915

    BTW, this was fixed in Valgrind SVN r5806.

     
  • Josef Weidendorfer

    • status: open --> closed
     

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks