[plt-scheme] More PLT Scheme as an Alternative to Matlab

On Aug 14, Matthew Flatt wrote:
> 
> As soon as I get enough time, I plan to finally experiment with
> unsafe operations, like `vector-ref', `car', and `fl+' that don't
> check their arguments.
> 
> Longer term, I think we'll need better supprot for SRFI-4 vectors
> combined with some way of unboxing floating-point arithmetic.

At least in its current form as just a library, I think that
optimizing srfi-4 vectors seems rather arbitrary.  But if you expose
unsafe operations, then perhaps a better way for dealing with than
adding the whole thing into the core would be to expose (in addition
to operations like `fl+') referencing a floating point number directly
from a given address?  If this is also done for integers, then a whole
bunch of things (the gzip code, md5sum, and other crypto-related
functionality) can become very fast.

But since this is now second-level speculation, I have no idea how
that might look -- but maybe something like the pidigits benchmark
with my ridiculous gmp hack?

-- 
          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!