This is the mail archive of the
guile@sourceware.cygnus.com
mailing list for the Guile project.
Dump float support?
- To: guile at sourceware dot cygnus dot com
- Subject: Dump float support?
- From: Mikael Djurfeldt <mdj at mdj dot nada dot kth dot se>
- Date: Mon, 13 Mar 2000 16:11:54 +0100
- Cc: Greg dot Harvey at thezone dot net, mlivshin at bigfoot dot com, mjs at eazel dot com, jimb at red-bean dot com
- Reply-to: Mikael Djurfeldt <djurfeldt at nada dot kth dot se>
I'm right now working on integrating Michael Livshin's double-word
patch from November. (I will not introduce the 6-word cells.)
This means that doubles can be represented as single cells which will
speed up floating point calculations a lot.
When looking at numbers.c I get the feeling that we should dump
singles support. There are plans to reorganize numbers support when
GOOPS has been merged. This reorganization will be a lot simpler if
we can begin with deleting all code having to do with floats.
The motivation is this:
Having to deal with both floats and doubles makes the code more
complex. The old reason for computing with floats, that they could be
stored in a single cell, is no longer valid, since also doubles now
will be stored in a single cell. The speed advantage of computing
with floats is shadowed by the cost of administrating an extra float
type (we need to test for floats everywhere, including in the GC).
We will still keep uniform float vectors, though. Values will be
converted to doubles when accessing the vectors. (Again, this won't
cost more than previously, since we also previously had to allocated a
cell for the value.)
OK, or objections?