Ahmed Fasih <[email protected]> writes:
> Sorry for troubling everyone with this petty question, but I recall reading
> maybe a couple of years ago how PyCUDA itself consisted of about 1200 lines
> of C++ code and about half as much Python code. I went looking for this but
> couldn't find this mentioned in the PyCUDA paper or presentations or docs.
> Running cloc on pycuda/src (latest commit) finds 5000+ lines of C++ and
> header, so I'm not sure if the statement I remember is still valid.
>
> Andreas, any hints? Thanks!

5k lines is about right for C++, and about 10k lines of Python.

To be fair:

- The C++ number excludes the struct reimplementations with complex
number support--that's basically a copy of Python's _struct.c with minor
modifications, once for Py2 and Py3.

- The Python numbers include tests and examples.

- PyCUDA has grown in functionality, picking up reductions, scans,
  sparse matrices, random numbers. All optional, but there if you need
  them.

wc $(git ls-files | grep .py)

wc $(git ls-files | egrep 'cpp|hpp')

I'm a bit curious to hear what those statistics will end up getting used
for. :)

Andreas

_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda

Reply via email to