[Matrix-SIG] Proposal for further single-precision clean-up.

[Scott M. Ransom]

Travis Oliphant wrote:

> Instead of waiting for something to happen, we should establish a
> consistent usage that allows single-precision work in Numerical Python to
> proceed.  The problem has always been that Python does not define
> single-precision scalars. I suggest, that as far as NumPy is concerned, we
> define single-precision scalars to be the corresponding rank-0 array.

As an astronomer, I would love to increase my use of NumPy  -- but have
proceded very slowly due to the "single precision" problems.  I often
need to access _huge_ arrays of single precision floats (generated by
Fortran or C code), and the upcasting to double is a killer for my
appliations.

Since I am not terribly familiar with the NumPy code base (and I'm sure
that many on this list aren't either), I would like to request that
someone more knowlegable please post a summary of the pros and cons for
treating single precision scalars as rank-0 arrays (or as something
different -- i.e. a new Python intrinsic).

I am really interested to learn what the differences in functionality
and performance would be when using the various different
implementations on NumPy arrays -- using ufuncs, the Python math module,
matrix operations, casting, etc.

I am highly encouraged by the recent interest in moving NumPy forward,
and with the correct changes, I think the user base among Scientists
will increase dramatically.

Thanks in advance,

Scott


-- 
Scott M. Ransom                   
Phone:  (781) 320-9867             Address:  75 Sanderson Ave.
email:  ransom@cfa.harvard.edu               Dedham, MA  02026
PGP Fingerprint: D2 0E D0 10 CD 95 06 DA  EF 78 FE 2B CB 3A D3 53