Need confirmation: speed improve

[ErosOlmi]

Hi all,

I need a confirmation from someone having some spare time and want to make some speed test.
I've done a new optimization but I'm not sure how much speed I've got because I think it mostly depends on CPU type. So, if you have 5 minutes I need you execute some of the speed tests present into \thinBasic\SampleScripts\General\Speed\ directory before and after getting new thinCore.dll attached to this post. Please be so kind to indicate your CPU type.

I also need info of any error can occur during tests.

Thanks in advance
Eros

ADDED:
attached file last updated at the LastEdit date/time reported below
attached file removed, features already online

[matthew]

I'd be more than happy to Run the Tests but I don't appear to have a 'Speed' Folder in the 'SampleScripts' Directory.

I'm Running thinBASIC v1.4.0.1. :-[

[kryton9]

Eros, I couldn't see a speed folder in mine either.

[sandyrepope]

On my computer the speed folder is under: \thinbasic\samplescripts\general\speed

Sandy

[matthew]

On my computer the speed folder is under: \thinbasic\samplescripts\general\speed

Sandy



Lol, thank's Sandy. I've just found it. :D

[matthew]

I Ran Three Tests...

1st. CountOneMillion
2nd. CountOneMillion_NoFunction
3rd. OneMillion_SIN_COS

My Processor is 'Mobile Intel Celeron M 360 1400 MHz'.

The Images are presented BEFORE and AFTER.

They all showed an increase in Speed. :)

[kryton9]

Thanks Sandy for finding the folder.
Faster in All but one, which was the same.

[ErosOlmi]

Sorry about directory. I missed to indicate General\ in the path. :-[
Thanks Sandy

I'm tuning Core a bit optimizing numeric and string expressions the first time they are encountered. If optimization was possible, the second time they are encountered some calculations will not take place. This should benefit loops. Still a lot of work to do but something seems already visible.

You should also get some more FPS in TBGL examples and in UI scripts the main WHILE/WEND message pump loop should benefit about that.

Thanks a lot
Eros

[kryton9]

Thanks Eros, I just know we are getting more speed, thanks a lot!!

[Michael Clease]

Added text file output to the speed tests.

and my results

[ErosOlmi]

Hi Abraxas,

sorry but I do not understand. Did you test before and after or just after?
I can see only one time for every test. Maybe I'm missing something.

Thanks
Eros

[Michael Clease]

I will redo when i go home for lunch.

heres some work pcs

[ErosOlmi]

Thanks a lot Abraxas. Useful info for me!

[Michael Clease]

heres the results from my home pc.

[ErosOlmi]

Thanks a lot Abraxas. You have a NASA computer !!!!

Attached Core in first post here updated. A little more speed in some circumstances.

[Michael Hartlef]

Overall I think it is slower on my computer. I attached the results.

[ErosOlmi]

How can it be? :(
I have an Intel P4 like yours. I will make some tests.

[Michael Clease]

Latest results from my home computer.

see previous results for the before and spec.

How about replacing the speed tests with the one I modified. for the next preview build. ;)

[Petr Schreiber]

Hi,

I am back from net-less areas and nice benchmarking possibility waits here :)

Results are interesting, in general better, sometimes I suspect GetTickCount is not enough, so maybe benchmark scripts should use PerformanceCounter stuff.

Results are attached :)


Petr

[matthew]

Results are interesting, in general better, sometimes I suspect GetTickCount is not enough, so maybe benchmark scripts should use PerformanceCounter stuff.


Hey Petr, in Basic4GL we were always told to use PerformanceCounter() for Timing Operations as it was accurate to 1 millisecond, TickCount() on the other hand was only accurate to 10ms.

[Petr Schreiber]

You are right Matthew,

it should be the better way. Only possible problem I have read about is when using CPU which scales its performance based on how heavy is loaded. Than the timing varies and can lead to quite wacky results :)


Bye,
Petr