x86 update -- raytrace benchmark

Suresh Jagannathan suresh@research.nj.nec.com
Tue, 17 Oct 2000 15:31:32 -0400


Wow! I'm impressed with these numbers -- significantly
better than the benchmarks we tried over the summer.
Is it essentially all due to register allocation and
better floating-point management?

  -- Suresh

-----Original Message-----
From: Stephen Weeks [mailto:sweeks@intertrust.com]
Sent: Tuesday, October 17, 2000 3:15 PM
To: MLton@sourcelight.com
Subject: Re: x86 update -- raytrace benchmark



> So, raytrace.sml works with both -DMLton_safe=1 and -DMLton_safe=0.  I'm
> including the updated allocate-registers.fun.  I'm also tracking down
> another minor bug, but I think it was introduced by some changes I made to
> the translation phase.  Let me know if you encounter anything odd with
> this version.

It worked like a charm.  Here are the new raytrace benchmark numbers.  Quite
good!

                    MLton   MLton
              OCAML     C  native  SML/NJ
holes           1.8   3.5     3.0     3.9
fov             1.5   1.8     1.6     3.2
intercyl        1.6   2.0     1.8     4.3
snowgoon        2.9   3.2     2.6     5.1
dice            3.9   5.1     4.2     8.8
golf            1.5   2.5     2.1     3.1
cone-fractal    3.7   4.4     3.4     6.5
large           4.3   3.5     2.8     6.7
pipe            5.4   4.7     4.0     7.9
chess          16.0  15.9    12.3    21.6
fractal        12.2  13.8    10.8    45.4
geom-mean       3.6   4.3     3.5     7.1