I think I have a solution which solve both issues:
- deal with the overlapping regions
- maintain the level of today's performance
Please try r24047.
On Nov 11, 2010, at 06:46 , Jed Brown wrote:
> On Thu, Nov 11, 2010 at 12:36, Number Cruncher <number.cruncher_at_[hidden]> wrote:
> However as commented here: https://bugzilla.redhat.com/show_bug.cgi?id=638477#c86 the valgrind memcpy implementation is overlap-safe.
> Yes, of course. That's how the bug in Open MPI was originally detected. Of course you can't do production runs with valgrind.
> Are you using an Intel Nehalem-class CPU?
> No, Core 2 Duo and Opteron (but the Opterons still have older glibc). Reverse memcpy must only be turned on for Nehalem.
> users mailing list