next up previous contents
Next: 3.3.1 EXCHANGE Up: 3. Parallelism in algorithms Previous: 3.2.3 Data coherency   Contents


3.3 Basic global operations

All operations investigated in this section refer to the hypercube (Sec. 2.2.6) and embedded topologies.

Pre.:
The routines
SEND/SMALL>_CHAN($ nwords$, $ data$, $ LinkNo$)
RECV/SMALL>_CHAN($ nwords$, $ data$, $ LinkNo$)
for sending and receiving data via link $ LinkNo$ are available.
Additionally, the processes $ p$ and $ q$ are connected to each other with link $ LinkNo$ (XOR $ (p,q) = 2^{LinkNo-1}$). This is a characteristic of the hypercube.


The names of the concerned functions are arbitrary and can be mapped to concrete implementations, e.g., MPI-calls.

Subsections

Gundolf Haase 2000-03-20