Re: Calling a subroutine causes serialization of threads?
Randy,
Thanks for the response! I didn't expect to get any response so quickly
as I thought this is not a popular topic, due to the nature of the
problem.
I am using google's interface and don't know how to quote the posting
automatically in google, so just cut and paste what you posted in
pieces.
1) "Why not inline thomas() into the loop?"
I haven't given it a thought. I'll make it inline, and perhaps by hand.
2) I did turn off the optimization, by specifying -O0 with the
compilers
I used. For intel and Compaq compilers the default is noalias. That
doesn't change things too much.
3) The OpenMP code and the Pthreads code behave pretty much
the same. I suppose the barrier in OpenMP is more efficient than
my hand coded barrier in Pthreads.
4) "Have you dumped the unoptimized assembly of the parallel region to
see"
No, I haven't. I'll do that, and perhaps will post the trimed
comparison here
if not too long...
5) "It would be interesting to change the loop count to see if you can
localize the "slowdown". Or profile the code using a high resolution
timer or a hardware performance counter (e.g. PAPI or perfex) to see
where the additional time is spent."
By the loop counter, do you mean the outer or inner loop? Could you
be a little more specific on profiling the code using a high resolution
timer? Do I need a special profiler?
Many thanks,
Ge
.
Relevant Pages
- Re: Do you think NI can fix my PLL? -- Details
... One also usually expects that the VCO response, ie., the frequency response of the phase, is not burdened by zeros/poles close to the desired PLL BW, so one can treat it simply as Kvco/s. ... Basically, there needs to be some plan for elucidating if what I have at present is just a mediocre motor characterization, poor choice of loop compensation, or trouble with the slight nonlinearities of the drive. ... If you're not getting any significant jitter at all then you should be able to make a really slow loop and be done with it. ... (sci.electronics.design) - Re: form wont open
... Thanks for the response. ... and choose References from the Tools menu. ... machine with my profile, I have no problem with the form. ... > Does it open directly from the Database window (as distinct from ... (microsoft.public.access.forms) - Re: Loading an I/O intensive program into memory
... explicitly loading the lot into RAM. ... You can lock programs into memory, but that may or may not help. ... It's easier to redo the main loop of code in shell than in C. ... Run your program with real input to gather the profile data. ... (comp.os.linux.misc) - Re: Cuk converter bizzare control loop
... >>Would like to hear a comment on the control loop from someone who has ... I used a convertor with the same response to convey signals at ... frequencies approaching a small submultiple of the switching ... It is the same problem with respect to balancing the bike. ... (sci.electronics.design) - Re: Slowing down an array / batch
... then getting a response and storing the response in a ... through this loop but more than the remote server can handle if I just ... My main problam is that the remote server will only allow 250 requests ... I've pasted my loop example below, but I need advice on how to either ... (comp.lang.php) |
|