bcasavan wrote:
For almost four years I've been the primary Pthreads engineer at SGI.
It's an honor! Welcome to Nekochan, were old IRIXer go to fade away
:( .
I'm pthreading right now (with an without REACT) and loving it. Thanks! All has been well since 6.5.21.
I'll try out your sample program shortly, but I have one pressing question -- would you expect the pthread_barrier call to have microsecond latency or ten's of microsecond latency on an Onyx350, where each thread is SYSTEM scope and running on an isolated processor, non-premptively? I am looking to speed up non-real-time performance, and when I used semaphores I was disappointed by a speed
decrease
when I doubled the threads (BOUND_NP, preemptive). The threads take about 500 usec each (2x= 1 millisec sequentially)), but need to sync a couple of times during the execution (say every 4x125 usec). What I'd like to see is a 2x speed up running them on 2 CPUs, versus the sequentially result.
Lastly, anything new on the REACT horizon? (IRIX or Linux)