Re: HPPA Optimization

Justin Dolske (dolske@cis.ohio-state.edu)
Tue, 15 Apr 1997 00:12:16 -0400 (EDT)


On Mon, 14 Apr 1997, Adam Haberlach wrote:

> A while back someone mentioned that there was a program that
> will use an exhastive search to optimize HPPA code--but that it
> was very compute-intensive. I'm pretty interested in working on
> this, if someone can give me details.

The program was GNU's SuperOpt. It's not really of any use to us. It's
designed for compiler writers, and basically can only deal with VERY short
code snippets. The README says that the longest sequence run through it
was 7 instructions long -- and that took over two weeks.

I'm looking at the PA-RISC assembly output from GCC, and let me tell you
that even the simplest DES functions are MUCH longer than 7 instructions.
:-)

Justin Dolske <URL:http://www.cis.ohio-state.edu/~dolske/>
(dolske@cis.ohio-state.edu)
Graduate Fellow / Research Associate at The Ohio State University, CIS Dept.
-=-=-=-=-=-=-=-=-=-=-=-=-=- Random Sig-o-Matic (tm) -=-=-=-=-=-=-=-=-=-=-=-=-
"An engineer is a terrible thing to over-dress." -- Robert Metcalfe