>>>>> "Gerd" == Gerd Moellmann <[EMAIL PROTECTED]> writes:
Gerd> In the thread on cmucl-help starting with
Gerd> From: Nicolas Neuss <[EMAIL PROTECTED]>
Gerd> Subject: Performance problem
Gerd> Newsgroups: gmane.lisp.cmucl.general
Gerd> Date: 14 Sep 2002 15:13:24 +0200
Gerd> I posted some code which modified a sequence of
Yes, I remember this. Don't know how register allocation is done, but
sparc also does similar things. In the first case, it just keeps
accumulating each aref into the same FP register. In the second case,
it tries to load up all the values in FP registers and then does a
bunch of adds at the end. Oddly enough, it doesn't even load all of
them up, even though there are more than enough registers to do so.
I would have expected the first method to be faster since I get to
interleave integer instructions with FP instructions. (Didn't test
runtimes on sparc.)
Oh well,
Ray
