From: Brannon King (brannonking_at_[hidden])
Date: 2006-07-12 11:52:04
Before parallel <algorithm> things, I think we need to step back and handle
the cross-thread domain container issues. It is very common in hardware
programming to have a FIFO go between clock domains. I'm picturing the same
thing for software; a FIFO where the push is in one thread and the pop in
the other. The thread with the pop would block on a pop call until there was
data and the thread with the push would block on push if the FIFO were full.
Sound valuable? Consider the case of a dual processor/core system where you
want I/O running in one thread and an N^2 algorithm in the other that only
outputs 10% of the N^2's data. I need to handle the I/O bottleneck before
I'm worried about using parallel algorithm functionality.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk