Moving Numeric arrays into the core

Neil Schemenauer nascheme at ucalgary.ca
Wed Apr 28 14:40:24 EDT 1999


On Wed, 28 Apr 1999 09:29:53 -0400 (EDT), 
Andrew M. Kuchling <akuchlin at cnri.reston.va.us> wrote:
>	I think the idea is to include only the basic NumPy types
>(multiarray and whatever the rest are).  BLAS, LAPACK, or whatever,
>would still be add-ons.

I've looked into this and it should be pretty easy to split up NumPy.
Here is what I got:

    332    1075   11818 Include/arrayobject.h
    190     585    6704 Include/ufuncobject.h
     86     264    3585 Src/_numpymodule.c
   2037    6451   53891 Src/arrayobject.c
    615    2690   28201 Src/arraytypes.c
      2       2      24 Src/multiarray.def
   1208    3818   31603 Src/multiarraymodule.c
     39      36     639 Src/numpy.def
    968    3129   26291 Src/ufuncobject.c
      2       2      19 Src/umath.def
   2013   10754   95932 Src/umathmodule.c
    213     847    7904 Lib/ArrayPrinter.py
     29      76     767 Lib/Matrix.py
    342    1190    9632 Lib/Numeric.py
     80     271    2468 Lib/Precision.py
    103     235    2964 Lib/UserArray.py
   8259   31425  282442 total


On a Intel Linux box I get the following libraries:

		 163996 _numpymodule.so
		  69147 multiarraymodule.so
		 159556 umathmodule.so
		 392699 total

I wonder if this is too large.  Maybe umath should be left out.  Also,
how would the existing array module fix in with this?


    Neil




More information about the Python-list mailing list