Re: [f2py] threading with f2py

OOPS  pardon me for sending this to the f2py forem I thought I was  
making a personal reply to the addressee directly!



On Jul 20, 2009, at 3:12 PM, charlie strauss wrote:

> hey stweart, nice to hear from you.  oddly I was just thinking of  
> you this morning.
>
> this issue sort of finally was the straw that made me think I better  
> make the transition to c++ for real.  not look forward to this.   
> f90+python is such a sweet thing.
>
> I've been looking at openMP as a nice solution.  And while it works  
> in f90, i've had trouble configuring a self consisten set of  
> compilers and python distros that work.  but since gcc supports it,  
> c works fine.
>
> This then got me looking at how to use BLAS in C and C++.    I see  
> boost has uBlas.
>
> Anyhow the thing I'm trying to figure out now is SIMD.  on most  
> computers these days BLAS is all SIMD, so you get SIMD for free for  
> the functions you can find in BLAS.  So that should be the same in  
> Fortran or C/C++
>
> But what about everything else?
>
> in fortran90 I can write:
>
> A = A+1
>
> or
>
> B = cos(A)
>
> C = A*B
>
> and A can be a matrix.  A smart compiler can in principle do that  
> all in SIMD without me having to know anything.  and with openMP  
> those can even be threaded in fortran:
>
> !OMP  parallel workshare
> B = cos(A)
>
> But there's no open workshare command in C++.
>
> Googling around I found discussions of something called boost::SIMD   
> but nothing documenting it on the BOOST site so I don't know what it  
> even is.
>
> So with that preamble here's the question:
>
> are there vector ops and automatic SIMD in boost or C++???
>
> that is how do I write
> B = cos(A)
> for vectors in C++ and will it use SIMD to accelerate it?
>
>
>
>
>
>
> On Jul 18, 2009, at 4:45 PM, Stuart Mentzer wrote:
>
>> Greetings Charlie,
>>
>> We went through this with a recent project and we found
>> that running the Fortran in a thread using
>>  !f2py threadsafe
>> does indeed let the Fortran yield back to Python but it
>> doesn't work if you want to use callbacks from Fortran,
>> which we did. We ended up using multiprocessing and it
>> is working well.
>>
>> Multiprocessing takes longer to start each process than
>> a thread and you have to make sure the state you need to
>> pass the process is "pickleable" but that only matters
>> in practice on Windows because it lacks an OS fork.
>>
>> (If you think I can be of any further assistance just
>> give me a call.)
>>
>> Stuart
>>
>>
>> _______________________________________________
>> f2py-users mailing list
>> f2py-users-Y4l6ocDipWCuvFJfX82//[email protected]
>> http://cens.ioc.ee/mailman/listinfo/f2py-users
>
> Charlie Strauss
> Bioscience Division
> [email protected]
> 505 665 4838
> Quidquid latine dictum sit, altum sonatur.
>

Charlie Strauss
Bioscience Division
[email protected]
505 665 4838
Quidquid latine dictum sit, altum sonatur.

_______________________________________________
f2py-users mailing list
f2py-users-Y4l6ocDipWCuvFJfX82//[email protected]
http://cens.ioc.ee/mailman/listinfo/f2py-users