TITLE>New Calls for Lanai Active Messages</TITLE>
<h1>
       New Calls for Lanai Active Messages
</h1>
There are a number of called added from the GAM spec:
<p>
<pre>
void am_request_5(int dest, void (*fun)(), int arg0, int arg1, 
		  int arg2, int arg3, int arg4)
</pre>
This is just like am_request_4, but sends five words as arguements.
<p>
<pre>
void am_sleep() 
</pre>
This call blocks the calling thread until a messages arrives. The 
thread must still pull messages out with am_poll() however. 
This call was added for servers like Network RAM and xFS daemons.
You must have the modified drivers included in the release for 
this to work.

<p>
<pre>
void am_reply_xfer(int dest, void *lva, void *rva, int nbytes,
		 void (*fun)(), void *arg0)
</pre>
This call can only be invoked inside of a request handler. It "stores
back" a block of data from lva on the calling node of length nbytes 
to a remote node starting at remote address rva. It will move up to 
am_max_size() bytes. Once the transfer is completed, The function pointed 
to by fun is invoked on the remote node with a single arguement arg0.
<pre>
int lam_sync_all()
</pre>
This call "bootstraps" the system. That is, it will send messages 
to make sure all nodes in the group are running before returning. 
It will try for 5 seconds to make sure all nodes are running. 
This called must be called exactly once after am_enable(), 
and acts a barrier(). If a program is started and fails to call 
lam_sync_all(), the node must be started in order, i.e. node zero must 
be started, then node 1, etc. Unpredictable results can happen if 
lam_sync_all() is not called otherwise. (e.g. messages getting lost).
<p>
This function return 1 on sucess (all nodes joined the group) 
or -1 on failure (Some nodes were missing).
<p>