The document discusses transparent Python bindings for CULA and PyCUDA, enabling users to perform spectral analysis and eigenvalue computations on symmetric matrices using GPU acceleration. It provides examples of how to initialize the library, run functions such as gpu_syev for eigenvalue and eigenvector calculations, and manage memory efficiently. Additionally, it emphasizes the simplicity and usability of its interface in scientific computing applications, particularly in relation to a large database of hypothetical zeolite structures.

1. Transparent Python Bindings for CULA
http://www.math.temple.edu/research/geometry/PyCULA/
Garrett Wright and Dr. Louis Theran
Complete CULA support Confused about how to use gpu_syev? In order to sort through the enormous database
Complete PyCUDA support From Python, just type help(gpu_syev) of hypothetical structures we are implementing
Mix Kernel and Runtime code spectral analysis. Specifically we are
Help on function gpu_syev in module PyCULA.cula:
Interpreter Friendly! performing log(Det()) of an energy function
gpu_syev(A, vectors=False, uplo='U')
Combine the power of Python Modules Takes symmetric matrix as np.array and returns eigenvalues (optionally eigenvectors) weighted adjacency matrix.
as np.array.
Interfaces directly with Numpy
Keyword arguments:
Integration with PyCUDA A -- input matrix as np.array, symmetric
Having PyCULA to integrate sparse expansion
Memory Management vectors -- default == False; use 'True' to compute vectors. kernels, GPU accelerated LAPACK, and add a
uplo -- Defines whether input array has data in Upper or Lower half using 'U' or 'L'
Super Simple Syntax: respectively Pythonic interface is a priceless utility.
default=='U'
>>> from PyCULA.cula import * # Load PyCULA Note: When vectors==False, gpu_syev returns only the eigenvalues.
>>> a = numpy.array([[1.,1.],[0.,1.]]) # CreateData When vectors==True, gpu_syev returns tuple (eigenvalues,eigenvectors) from PyCULA.cula import *
>>>culaInitialize() # Initialize Device (END) from PyCULA.my_init import *
>>> print gpu_svd(a) # Perform SVD import numpy as np
[ 1.61803399 0.61803399] # Printed Results import atexit
>>>culaShutdown() # Shutdown Device import kernel_modules #import custom kernel module code
pycuda_init_once()
culaInitialize()
atexit.register(cula.culaShutdown)
# This is an adjacency list, a *condensed* form of a larger sparse adjacency
matrix.
Integrate SQL databases for your scientific E=np.array([[1 , 2 ],[ 1 , 16 ], [3 , 4 ],[ 4 , 5 ],
Data [ 6 , 7 ],[ 7 , 8 ],[ 8 , 9 ],[ 8 , 5 ],[ 9 , 10 ],
[ 9 , 4 ],[ 19 , 16 ],[ 20 , 1 ]], dtype=np.float32)
Use Scipy to read and write to MATLAB files
Simple parsing code # expand reduced form into full matrix using GPGPU kernel
M_ = kernel_modules.gpu_expand_kernel(E)
Easily execute multiple instances via
eigvals = gpu_devsyev(M_) # run CULA syevon GPU array M_
subprocess
Prototype or compute within the interpreter;
No need to manually compile!
Write a single program to code optimized
programs on the fly Atlas of Prospective Zeolite Structures
http://www.hypotheticalzeolites.net/
Easily compress and extract data with gzip
and zlib modules Many thanks to the NSF for funding our research into the
Take advantage of automatic memory This is a hypothetical zeolite structure. We computational physics of zeolites.
management; Helps write leak free code! currently have a database of over 2 million such
Simple, short, clean code helps prevent bugs! structures, yet only a few have been discovered
in nature