This code runs *very* slowly

[Rahul Siddharthan]

On Thu, 5 Nov 1998, Richard Henderson wrote:

> Oops.  Try this patch instead.  Have to be more careful about
> when new temporaries are or are not created.

I tried it and it works beautifully. Actually following some
suggestions of others, I had made some improvements in my code
which improved the speed considerably but it was still twice as
slow as DEC C; with your patch, the improved code runs around as
fast as the DEC C code (after you adjust for machine speeds),
while my posted code runs much faster than it used to but
somewhat slower than DEC C.  Numbers are below. The compilation
time also seems to have improved, but that's still nowhere as
fast as DEC C.

Many thanks for the quick response! I'm fond of free
software and this lets me use the Linux machines to their full
potential :) I'm really impressed -- especially since, after your
patch, it's performing nearly as well as Digital's own compiler, which
is presumably carefully tuned for their own hardware...

Rahul.

Nearest neighbour model, with some improvements to my posted code
433 MHz Alpha/Linux 
Run time without your patch: 41.56 secs
            with your patch: 40.06 s
333 MHz Alpha/ Digital Unix: 52.7 s

Next neighbour model, my original code, 433 MHz Alpha/Linux
smaller number of iterations so lower running time compared to above
Run time without your patch:  342.0s
            with your patch:  83.77s
333 MHz Alpha/Digital Unix :  66.3 s

Next neighbour model, improved code, 433 MHz Alpha/Linux
Run time without your patch:  69.27 s
            with your patch:  39.79 s
333 MHz Alpha/Digital Unix :  46.5 s