Re: FastCode - IntToStr B&V v0.12

Avatar Zondertau wrote:

> > Speaking of which, here are the results for Delphi 6, 7 and 2005
> > (don't have Delphi 5 installed on this machine) on a dual Xeon.
> >
> > I notice the CPU identification doesn't pick up the dual CPU. If it
> > helps, I can test against the machine here.
>
> I created the code to test for multiple CPUs (and distinguish between
> logical CPUs added by hyperthreading and "real" cores). It will be in
> the next release.

Is this the CPUID code that's also posted on the FastCode site, or are
you using your own home grown routine? If they're different, should
they be merged so everyone uses the same code?

Either way, let me know if you need some help testing.

> However ISTM multiple CPUs shouldn't make much difference here, since
> the B&V program is not multithreaded.

True, at least for this challenge.

> Perhaps it would be nicer to execute the benchmarking and validation
> in a secondairy thread? This would make the GUI more responsive during
> testing (on multiprocessor machines anyway; the benchmarks will of
> course have higher priority then the GUI thread in this case).
>
> Any opinions on this last point?

I've been stewing on this a little.

I agree with Dennis that benchmarks need to be run at the best possible
speed, and I would be concerned that changes that make the UI more
responsive might affect this. Validations aren't measured, so I'm not
as concerned about these.

I think I'd like to see the B&Vs geared more toward unattended
execution as opposed to interactive. Maybe with a simpler UI that
gives more timely feedback (shorter running tests) and that can be
easily cancelled at any time, but is designed to be to run/report/exit
all on its own. It would make B&Vs easier to write, test and maintain.
I'm not picking on the IntToStr B&V, BTW. I think your implementation
is quite clean.

I'm a *huge* fan of things like automated builds and tests. I almost
have to be because of the number of active projects I have and the
number of manual steps it would take to perform a build, run the unit
tests and prepare the final project for distribution. Right now, I
have a single FinalBuilder project that runs another FinalBuilder
project for each of my projects and does all of this while I'm brewing
the coffee.

I'd like to have something similar for FastCode challenges. I download
a new B&V, unzip it somewhere and run a single app or batch file that
takes care of everything for me. In the end, I have the final result
in a text (or XML) file that I can simply post. I didn't have to load
each of the supported IDEs, build the B&V, run the validations, run the
Delphi 2005 benchmarks (careful to run them all), copy them to the
clipboard and then post them.

Does this make sense?

--
Regards,
Bruce McGee
Glooscap Software
.

Relevant Pages

  • Re: OpenSolaris Release is NEXT WEEK!
    ... look at the numbers "Unix Admin". ... Sun continues to BS but their CPUs ... competitive SPARC CPUs. ... "We've had folks in labs trying to tear apart their benchmarks, ...
    (comp.unix.solaris)
  • Re: OpenSolaris Release is NEXT WEEK!
    ... look at the numbers "Unix Admin". ... Sun continues to BS but their CPUs ... competitive SPARC CPUs. ... "We've had folks in labs trying to tear apart their benchmarks, ...
    (comp.unix.aix)
  • Re: Parallel Matrix x Vector and Lock-freeThreadpool...
    ... Here is the benchmarks with SSE+ ... CPUs 2 workers threads ...
    (comp.programming.threads)
  • Re: How to save SUN. If you work for SUN, read my ideas!
    ... benchmarks results to compete. ... According to the benchmarks all three submitted, ... our code would actually run considerably slower on 32 CPUs ... Both the OS and the underlying hardware ...
    (comp.unix.solaris)
  • Re: the Z/10 and timers.
    ... much less exploitation of those multiple CPUs in any ... lineal descendants in today's Unisys ClearPath MCP ... August, 1965, which was the machine available with multiple CPUs. ... IBM Consulting Enterprise Software Architect ...
    (bit.listserv.ibm-main)