BorlandTalk.com

Avatar Zondertau · Guest

Hi to all,

I think there are some improvements that can be made to the FastCode
Benchmark & Validation tool. These changes would IMHO make it easier to:
- Create new B&V's (mostly for Dennis)
- Add functions to a B&V
- Analyse the results of benchmarks and validations

I created a new utility as a suggestion for a new B&V tool. I will post
it in the attachments group. The B&V now has implements the PosEx
challenge (version 1.0, but with only three of the validators). I don't
intend it to replace existing challenges, like PosEx, because that
would only cause more work. I would however hope that you (especially
Dennis of course) think it would be useful for the challenges that are
still to come.

Could people please take a look at it to determine
- if they like it
- if they can find bugs or enhancements they would like
- if it compiles on all relevant Delphi's

The new B&V does not require changes to the existing code or form to
create a new challenge. A challenge is created by making a subclass of
the TFastCodeChallenge class and registering that during in the
initialization section of it's unit. The main form then automatically
uses it. The tool also allows registering multiple challenges, so
situations where you distinguish between Single, Double and Extended
should also be easy to do.

Functions are added by adding their unit, which can register them
automatically. This means no other units need to be changed and the
form lists them automatically. Therefore if we would use this B&V we
could just send Dennis the units. This would save him the work of
copy-pasting the code, modifying and resorting the RadioGroup and
changing his function selection function.

The B&V allows batch benchmarking/validating of a only as election of
the functions, so it's easier to, for example, validate all your own
functions in one go. They can be selected by category (pas, ia32, ...)
or author, or just individually.

The test results can be sorted on any of it's columns. The columns that
can be shown are name, alignment, author, category, benchmark result,
benchmark execution count, benchmark spread and all individual
subbenchmarks. Visible columns can be chosen by right-clicking a column
header, selection can be done by left-clicking a column header.

All that happens is written to a log that is shown on the form. This
allows you to check, for example, why a validation failed.

The results can be reported in the original format. They now also
include compiler version, Windows version and CPU.

To Dennis: please don't be offended by my suggestion. I only mean to
offer a suggestion for a new B&V that might make stuff easier,
especially for you. I don't mean to discredit the current B&V, which is
also a great tool.

Dennis · Guest

Hi Avatar

Good work.

It looks nice from your description. I will take a throrough look at it
soon. Perhaps try to make the next challenge based on it.

Aleksandr Sharahov · Guest

Hi Avatar,

It is good.

--
regards
Aleksandr