On 7/26/06, Andre van Tonder <xxxxxx@now.het.brown.edu> wrote:
> I have a few comments based on my experience in trying to implement the
> ICFP contest virtual machine specification using Scheme.  I found that most
> Schemes were pretty much useless for this particular application

I also participated in the contest, and started with Scheme but ended
up using C (the VM was much shorter and easier to express in C, apart
from being faster).

After the contest I tweaked my Scheme implementation, and when
compiled with Chicken Scheme it was only 10x slower than C, using only
R5RS and SRFIs 4 and 60.  This was mostly thanks to using a 64-bit
machine where I could do 32-bit arithmetic by masking out the high
bits after each numeric operation: (bitwise-and #xFFFFFFFF expr).  On
a 32-bit machine almost every value would have been a bignum, totally
destroying performance.

This seems much more common and convenient than the aribtrary
(and variable among different implementations) modulo arithmetic
provided by the `fixnum' procedures of SRFI-77.

--
Alex