Re: [myrinet] Re: Re: GM_BOGUS_SEND_INTERRUPT
Patrick Geoffray <[email protected]>
| Newsgroups | gmane.network.myrinet.general |
|---|---|
| Organization | Myricom Inc. |
| Message-ID | <[email protected]> |
Loic Prylli wrote: > One potential problem is that MPICH-GM include a malloc replacement to > handle memory registration properly (is that still true for > 1.2.1..7?), this malloc replacement is not thread-safe, and it might > be the explanation for your problems (even if it isn't, you might want > to take that into account). Yes, there is still the malloc replacement in 1.2.1..7 (you didn't finish the driver change to get rid of it :-) ). You are right, the USE_MALLOC_LOCK is not set by default. It should be set if you want to make the malloc implementation thread-safe. I will change it in the CVS. A quick work around would be to change "#undef USE_MALLOC_LOCK" into "#define USE_MALLOC_LOCK 1" in the mpichconf.h file generated by configure. Will be in 1.2.2.2..0 (yes, it's long, but I have to keep the MPICH 1.2.2.2 prefix) available soon. Thanks Loic. Patrick --------------------------------------------------------- | Patrick Geoffray, Ph.D. [email protected] | | Myricom, Inc. http://www.myri.com | | Cell: 865-389-8852 325 N. Santa Anita Ave. | | Fax: 865-974-1950 Arcadia, CA 91006 | ---------------------------------------------------------