Python/numpy extension modules using Vc?

[Tijskens Engelbert]

Hi,

I am used to developing python extension modules written in fortran (f2py) or C++ for compute intensive tasks. The algorithm I am facing now cannot be tackled with auto-vectorization. Since I had good experience in performance of Vc libraries, I wanted to give this a try. I briefly describe the setting below

I am using a recent python version (3.6) and Numpy 1.15 from Anaconda. Numpy provides nicely aligned 64-byte aligned arrays. Using the tools of https://github.com/mdboom/numpy-boost to pass Numpy arrays to c++ functions. Inside these functions I have a pointer to the memory allocated by the Numpy array, how can I used the Vc machinery on this memory? In all examples I have seen the memory is allocated in the C++ code. Is there a way to cast a raw pointer to (well aligned) memory to sth that is understood by Vc?

Many thanks in advance
Bert


Kindest regards,

Dr [Engel]bert Tijskens - HPC Analyst/Consultant
Flemish Supercomputer Center
University of Antwerp / Computational mathematics
++32 3 265 3879, ++32 494 664408

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://compeng.uni-frankfurt.de/pipermail/vc/attachments/20190122/07880fb9/attachment.html>