CVS Commit

[Matthew Fluet]

(Meant to send this to MLton.)

---------- Forwarded message ----------
Date: Wed, 26 Sep 2001 19:22:54 -0400 (EDT)
From: Matthew Fluet <fluet@hoho.cs.cornell.edu>
To: Henry Cejtin <henry@sourcelight.com>
Subject: Re: CVS Commit


> Pretty weird.
> It could still be a matter of the cache I suppose: cache line alignment.  I
> forget if this is 16 or 32 bytes.  Still, doesn't the entire code fit in the
> L1 easily?

Yeah.  Looking at the nm output on the optimized version, I calculated
that the hot loop is 55 bytes.  I don't care how small the L1 cache is,
it's definitely bigger than that!  I'll look a little more at the Intel
Architecture optimization manual, but I didn't see anything that really
explained what was going on.