I dimly recall the threading business from a long time ago. Key was that
a threading aware/built PDL dist is not enough.
You also need to switch on threading on a per piddle basis with an
obscure command that I can't recall now. Did your test scripts do this?
If not your results are not very surprising.
Christian
Chris Marshall wrote:
I reran the test code and realized I had misread the
previous results. On the hyperthreaded PC the
performance was 2X *worse* not better. The dual
Opteron still shows only a 15% performance penalty.
These results are more consistent (less puzzling) but
still leave the question of whether/if posix threading
is of any benefit to PDL for any platform. Still hoping
to hear from users/developers who know more about
the origin and functionality of this code in PDL.
--Chris
Original mis-statement:
...
- Performance on a PC with hyperthreading showed
a 2X speed up and 2X CPU utilization with threading
enabled under cygwin 1.5.19.
- Performance on a dual-opteron system was about
12 percent worse with CPU load 12 percent higher.
It looks like posix threading for this platform is a
no-op with overhead.
...
------------------------------------------------------------------------
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
