[pypy-dev] Excellent work!

Bent Phillipsen bphillipsen at baden.ch
Sat Mar 10 20:48:25 CET 2012

Thank you for making pypy available. It is really excellent work!

I tried pypy on some numerical analysis algorithms and compered to fortran and cpython:

On a loop over romberg-integrations I found pypy 2.3 times faster than gfortran without optimization and 2.1 times faster than gfortran with 
optimization (yes, pypy faster than fortran!). Compared to cpython, pypy is about 25 times faster. The results of the computations with pypy and 
cpython are identical.

On a loop over akima-interpolations pypy is 5 times slower than gfortran, but about 20 times faster than cpython.

These results are really impressive. With pypy numerical analysis can be performed efficiently in pure python using lists as arrays (i.e. without 
using numpy). 

By the way: I have observed, that the assert statement is not fully implemented in pypy (which possibly is well known). In cpython with the -O 
option no bytecode is generated for the assert statement. In pypy the same bytecode is generated with and without the -O option. 

Best regards

Bent Phillipsen  

More information about the pypy-dev mailing list