Re: benchmark programs

peach@entrenet.com (Peach Microsystems)
5 Jul 1996 12:01:02 -0400

          From comp.compilers

Related articles
benchmark programs gxj@dcs.ed.ac.uk (1996-07-03)
Re: benchmark programs peach@entrenet.com (1996-07-05)
Re: benchmark programs mcdonald@kestrel.edu (1996-07-07)
Re: benchmark programs markt@harlequin.co.uk (1996-07-09)
| List of all articles for this month |

From: peach@entrenet.com (Peach Microsystems)
Newsgroups: comp.compilers
Date: 5 Jul 1996 12:01:02 -0400
Organization: Entrenet Systems
References: 96-07-033
Keywords: benchmarks

gxj@dcs.ed.ac.uk (Graham Jones) wrote:
>I am doing some simulations into latency hiding techniques and I would like
>to find some benchmarks which either
>
> 1) prove difficult for caches


Gabriel's "Performance & Evaluation of LISP Systems", early 80's, had LISP
source for a number of benchmarks including multiple variants of a viciously
recursive but otherwise trivial function called tak. At least one variant was
deliberately designed to defeat cache.


// omitting details I don't recall offhand
tak(unsigned x,y,z)
{
            if(x<y)
                          return z ;
            else // recurse, calling tak 4 times
                          return( tak(
                                                          tak( ), // arguments here are permutations
                                                          tak( ), // of x,y,z
                                                          tak( ) // with some x-1's mixed in
                                                        ) ;
}


The cache-defeating variant had some large number of copies of this function,
like tak000 to tak999, and made the recursion call different copies in a
random order.
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.