Re: [Vxl-maintainers] Thread safe vnl (netlib)

[William A. H. <bil...@ny...>]

At 12:10 PM 6/14/2006, Ian Scott wrote:
>William A. Hoffman wrote:
>
>>How much of netlib is actually used by vnl?   Is there an easy way to tell?
>
>I've tried running a Dart coverage, with the .NoDartCoverage files removed from v3p/netlib - but whether the dashboard is taking a long time to update (20 minutes so far) or the submission hasn't worked - I don't know. If nothing happens, you can trying running a manual coverage test, and see which bits of code aren't currently tested.
>
>If you add the all the complex, and float versions of the covered files, that would be a good estimate. It wouldn't surprise me if a file was only added to v3p/netlib, because some code in vnl_algo actually needs it.
>
>A completely alternative solution, that I had been planning* to use for some of our private code's mutable data members is to write a thread local storage class template. A netlib_tls<int> would guarantee to give you a separate int in each thread. Importing something from boost might be a good start to write such a class. Then simply replace each static double with a netlib_tls<double>. You could even write something in C, which is currently used by almost all of netlib, but it might be harder to script the replacement of all the static variables then.

You might be able to modify f2c to output the correct types.


>Ian.
>
>* - In the sense of - when we really, absolutely, unavoidably, have to start using threads, I'll think properly about how to do this.

Well, we do have some funding to work on this problem.   So, if you have more ideas, please
send them our way.  Also, it seems that cpu's are now mostly dual core and parallel software 
is becoming a must to take advantage of the hardware.

-Bill