•Everyone thinks
shared memory is the natural parallel extension of sequential computing: “Ignore memory
reference time like vN model, and let HW give the flat memory
illusion”
•Memory
reference time is key to good algs:
–Find
maximum is the example
•Best
ignore-memory-time (PRAM) is Valiant O(log log P)
•Best
consider-memory-time (CTA) is tournament O(log P)
•(Actual?)
implementation of Valiant’s alg O(log P loglog P)
•Actual
implementation of tournament O(log P)