[MIT-Scheme-devel] open-coding floating-point on AMD x86-64

[Taylor R Campbell]

I have just pushed support for open-coding floating-point primitives
on the AMD x86-64.  I neglected to mention in the commit message that
since it includes changes to insmac.scm, you will need to rebootstrap
the compiler, either by running

scheme --band runtime.com <<EOF
(load-option (quote sf))
(load "compiler.sf")
EOF

or by just starting afresh with a working LIARC build and running
etc/make-native.sh again.

Since this is only lightly tested, please let me know if you find
problems with it, such as incorrect values from floating-point
computations, segfaults, illegal instruction traps, &c.  Aside from
that, this should make floating-point arithmetic a little faster on
the x86-64, although this is only very rudimentary, and doesn't do any
automatic vectorization with fancy packed SSE instructions or
anything.