Re: MPICH-GM 1.2.5..9.

Enrique,

On Fri, 2003-02-21 at 11:13, Enrique Lomba wrote:
> the optimization is flawed it seems, what next ??

You can either:
1* configure/compile MPICH-GM without optimization
2* apply this patch to work around the buggy htons:

--- REF/mpich-1.2.5..9/mpid/ch_gm/gmpi_conf.c       Fri Jan 24 15:02:05
2003
+++ mpich-1.2.5..9/mpid/ch_gm/gmpi_conf.c   Tue Feb 18 14:18:56 2003
@@ -342,7 +342,7 @@
       gm_bcopy ((char *) (master->h_addr), 
            (char *) (&(gmpi.master_addr.sin_addr)),
            master->h_length);
-      gmpi.master_addr.sin_port = htons (master_port1);
+      gmpi.master_addr.sin_port = gm_hton_u16 (master_port1);
       i = GMPI_INIT_TIMEOUT;
       while (connect (gmpi_sockfd, (struct sockaddr *)
(&(gmpi.master_addr)), 
                  sizeof (gmpi.master_addr)) < 0)
@@ -385,7 +385,7 @@
    }
       
       /* re-connect to the master */
-      gmpi.master_addr.sin_port = htons (master_port2);
+      gmpi.master_addr.sin_port = gm_hton_u16 (master_port2);
       i = GMPI_INIT_TIMEOUT;
       while (connect (gmpi_sockfd, (struct sockaddr *)
(&(gmpi.master_addr)), 
                  sizeof (gmpi.master_addr)) < 0)


3* get ftp.myri.com://pub/patrick/mpich-1.2.5..9-test10.tar.gz 
and give it a try. This include the previous patch, plus more fixes 
for the Intel compiler (can you say I love this compiler ? :-\). 
This is very much what will go through the release testing in a few 
hours for the release of mpich-1.2.5..9a.


I would strongly advise 3, I would appreciate any feedback.


Patrick
-- 

Patrick Geoffray, Phd
Myricom, Inc.
http://www.myri.com