[Omp] .NET2005 OpenMP and Floating Point Exceptions

[Alexandru Nicau]

Hello all,
Please help me with a question related to .NET2005 OpenMP implementation.
There was this parallel for correctly written (among other things we have considered the schedule directive) which run single threaded took 180sec. When the for was executed parallely (two threads, same dual proc machine) it gave out the expected 96sec. If in the single threaded version floating point exceptions were activated runtime - we used _controlfp_s - computation took +30% more. Why does the parallel version with FPE activated takes +74% instead of the expected +30%? All of the above knowing that we timed the computation itself, of course in the context, and it revealed no such difference (single threaded 102sec added FPU 157sec, dual threaded 104sec added FPU 161sec). Need to mention that compiler option Maximize Speed (/O2) is activated.

Seems like performance loss comes from the compiler generated code. Any thoughts?

Thank you,
Alex

____________________________
Alexandru Nicau
Moduleworks

alexn at moduleworks.com
YahooID: alexandru_nicau
SkypeID: mw_alexn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://openmp.org/pipermail/omp/attachments/20070426/a5ca42dd/attachment.html