[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: What would a good PCL benchmark make?
- To: Mike Thome <mthome@vax.bbn.com>
- Subject: Re: What would a good PCL benchmark make?
- From: larus@paris.Berkeley.EDU (James Larus)
- Date: Fri, 18 Nov 88 10:17:39 PST
- Cc: commonloops.pa@Xerox.COM
- In-reply-to: Your message of Fri, 18 Nov 88 11:03:02 EST. <881118-081500-4082@Xerox>
- Redistributed: commonloops.pa
- Reply-to: larus@ginger.Berkeley.EDU
In my experience, small benchmarks are not particularly useful for
predicting PCL's behavior on real programs. PCL makes heavy use of
caches and small programs don't fill the caches or cause collisions in
the hash tables. Large programs, with hundreds of methods and
classes, expose a lot of weak points in the PCL implementation.
If anyone is interested, I can send them detailed timings of one such
large program, which show that PCL's cost is around 75% of the total
execution time.
/Jim