Hope – A Python JIT for astrophysical computations • Hacker News Board

Hope – A Python JIT for astrophysical computations

#1

Post by catkin » Fri, Oct 17, 2014, 4:09 PM UTC

Hope – A Python JIT for astrophysical computations
github.com

Re: Hope – A Python JIT for astrophysical computations

#2

Post by mborch » Fri, Oct 17, 2014, 5:41 PM UTC

Assuming this is Python 2.x (doesn't seem to be specified in the benchmark results), then the use of "range" is problematic (should be "xrange").

Re: Hope – A Python JIT for astrophysical computations

#3

Post by pbrook » Fri, Oct 17, 2014, 5:50 PM UTC

Does anybody know how this compares performance-wise to Numba (http://numba.pydata.org/) which seems like a pretty similar effort? Numba translates to LLVM bytecode rather than C++.

Re: Hope – A Python JIT for astrophysical computations

#4

Post by catkin » Fri, Oct 17, 2014, 5:57 PM UTC

pbrook wrote: post #3

Does anybody know how this compares performance-wise to Numba ( http://numba.pydata.org/ ) which seems like a pretty similar effort? Numba translates to LLVM bytecode rather than C++.

Full results are in the paper: http://arxiv.org/pdf/1410.4345v1.pdf

Re: Hope – A Python JIT for astrophysical computations

#5

Post by shoyer » Fri, Oct 17, 2014, 6:02 PM UTC

Here's the doc page that lists the supported language features: http://pythonhosted.org/hope/lang.html

Compared to numba, it looks like hope adds: (1) support for recursion, (2) automatic simplification of expressions using SymPy and (3) support for arithmetic with array broadcasting (e.g., as used in their Point Spread Function benchmark).

Re: Hope – A Python JIT for astrophysical computations

#6

Post by MrMeker » Fri, Oct 17, 2014, 6:30 PM UTC

mborch wrote: post #2

Assuming this is Python 2.x (doesn't seem to be specified in the benchmark results), then the use of "range" is problematic (should be "xrange").

Not really. "range" returns a list, "xrange" returns an iterator. I assume that the list is much easier to reason about in a JIT. In python3, range is a type of its own and certainly more useful for in a JIT.

https://docs.python.org/3/library/stdtypes.html#typesseq-ran...

https://docs.python.org/2/library/functions.html#range

Re: Hope – A Python JIT for astrophysical computations

#7

Post by bouk » Fri, Oct 17, 2014, 6:32 PM UTC

This is crazy! It actually generates C++ code on the fly by translating AST nodes one to one https://github.com/cosmo-ethz/hope/blob/master/hope/_generat...

Re: Hope – A Python JIT for astrophysical computations

#8

Post by shoyer » Fri, Oct 17, 2014, 6:33 PM UTC

mborch wrote: post #2

Assuming this is Python 2.x (doesn't seem to be specified in the benchmark results), then the use of "range" is problematic (should be "xrange").

Based on the description in the docs (http://pythonhosted.org/hope/lang.html#loops), my guess is that "for i in range(N)" is directly translated into something like "for (i = 0; i < N; i++)" in the C++.

Re: Hope – A Python JIT for astrophysical computations

#9

Post by Fede_V » Fri, Oct 17, 2014, 6:34 PM UTC

shoyer wrote: post #5

Here's the doc page that lists the supported language features: http://pythonhosted.org/hope/lang.html Compared to numba, it looks like hope adds: (1) support for recursion, (2) automatic simplification of expressions using SymPy and (3) support for arithmetic with array broadcasting (e.g., as used in their Point Spread Function benchmark).

Thanks, exactly what I was looking for.

Re: Hope – A Python JIT for astrophysical computations

#10

Post by fsloth » Fri, Oct 17, 2014, 6:37 PM UTC

Looks really nice! Is there any simple recipe built in to parallellize array operations over n threads or does the user need to provide ones own parallellization scheme?