By writing the node program in a data parallel language (CMF or C*), it is possible for a control parallel program also to take advantage of the vector units. Such a program is compiled in the normal fashion, but is linked differently.
Suppose we have a node program called
myprog.fcm that uses CMMD.
It is compiled and linked to use the vector units as follows.
cmf -vu -c myprog.fcm
cmf -node myprog.o -o myprog
Even though it is optional, using the -vu flag
does improve the speed significantly.
For instance, try compiling the example program
with and without the
You will notice the difference in the execution times of the two versions.