Hollywood way to benchmarks...?

Discuss any general programming issues here
User avatar
Tuxedo
Posts: 338
Joined: Sun Feb 14, 2010 12:41 pm

Hollywood way to benchmarks...?

Post by Tuxedo »

Hi there!

Sorry for weird qeistion/topic but I'm trying to make some sort of multiplatform benchmark program made in Hollywood to test speeds of gfx and or CPUs on several systems...
Hollywood can be used for that purpose or wasnt reliable? I'm not looking for exact values but I want to get some sort of comparisons with same program used on different systems(in first Amiga-like and then other systems however).

So I ask:

1 - as partially described in the Android thread if I made:

Plot(@RED, x, y)

Naturally the Plot() function was encapsulated in a For, Next cycle

and made for example 10000 pixels on screen Hollywood works on every system in same way or have some optimization in some system better than on another?

2 - To test, simply and easly, CPU strenght, calculating some sort of:

StartTimer(1)
For x = 1 to 1000000
Next x
t = GetTimer(1)
StopTimer(1)

Was enough to get simple results or maybe with different CPU caches it have to be much elaborated and big to gets significative values without be loaded in cpu caches? Or maybe Hollywood structure that works on interpretative mode will have already a so big code that cant be loaded in cache?

3 - Speaking strictly of Amiga-like systems there was some significative difference in Hollywood coding that can gave some speed difference in one system respect another running same code? And in NON Amiga-like systems there was some optimization or maybe not since there we have really greater cpu power?

Thank you.
Simone"Tuxedo"Monsignori, Perugia, ITALY.
User avatar
airsoftsoftwair
Posts: 5433
Joined: Fri Feb 12, 2010 2:33 pm
Location: Germany
Contact:

Re: Hollywood way to benchmarks...?

Post by airsoftsoftwair »

Tuxedo wrote: and made for example 10000 pixels on screen Hollywood works on every system in same way or have some optimization in some system better than on another?
Well, this depends on how fast the host OS can draw single pixels to windows.
Was enough to get simple results or maybe with different CPU caches it have to be much elaborated and big to gets significative values without be loaded in cpu caches? Or maybe Hollywood structure that works on interpretative mode will have already a so big code that cant be loaded in cache?
Don't understand this question :)
Speaking strictly of Amiga-like systems there was some significative difference in Hollywood coding that can gave some speed difference in one system respect another running same code? And in NON Amiga-like systems there was some optimization or maybe not since there we have really greater cpu power?
Hmm, don't understand this one either :)
Bugala
Posts: 1178
Joined: Sun Feb 14, 2010 7:11 pm

Re: Hollywood way to benchmarks...?

Post by Bugala »

As what comes to comparing power between different platforms based upon hollywood programs. If i have understood correctly, there would be some (perhaps even quite many) cases, in which Hollywood program comparison wouldnt really work.

Mostly i guess on graphics. As examples.

1. If you keep drawing stuff one at a time to screen, Android will be terrible, since the way it behaves, is different from other platforms, hence in android you need to use that double buffering / Flip, or Refresh thing (which was just talked in another topic shortly).

2. Hollywoods graphic things have been optimised for Amiga likes, hence Amiga would seem more powerful compared to other machines, even it isnt.


But what comes to those typical

a=a+1 lines, there i have no idea wether one computer is more optimised, or they all run at same speed as long as machines are equally powerful.

But nice idea anyway, and would be good to have all this kind of info to look at. That as many as possible commands would be tested on each platform and compared to each other, as well as comparable to different comands themselves.

Like which one, and how much faster is for example a=a+1 compared to add(a, 1)
User avatar
Tuxedo
Posts: 338
Joined: Sun Feb 14, 2010 12:41 pm

Re: Hollywood way to benchmarks...?

Post by Tuxedo »

@softwarefailure

I mean:

1 - if was a bit usefull calculate the cpu speed doing a For/next cycle or have to do a more complex algorithm

2 - if the Hollywood exe was more optimized on some platforms against another or I have to expect same result on every system for same routine...

Now its clear?
Simone"Tuxedo"Monsignori, Perugia, ITALY.
User avatar
airsoftsoftwair
Posts: 5433
Joined: Fri Feb 12, 2010 2:33 pm
Location: Germany
Contact:

Re: Hollywood way to benchmarks...?

Post by airsoftsoftwair »

1 - if was a bit usefull calculate the cpu speed doing a For/next cycle or have to do a more complex algorithm
Sure, why not.
2 - if the Hollywood exe was more optimized on some platforms against another or I have to expect same result on every system for same routine..
The VM is identical on every platform but of course the graphics driver is more optimized in the Amiga builds, e.g. Hollywood still supports all kinds of 15-bit, 16-bit, 24-bit and 32-bit formats for optimized drawing. On other platforms Hollywood just stores all bitmaps as 32bit bitmaps so some conversions might be necessary before they can be drawn. However, you probably won't notice that much of the Amiga optimizations since the CPUs that usually host the Amiga builds don't stand a chance against modern x86 CPUs :)
User avatar
Tuxedo
Posts: 338
Joined: Sun Feb 14, 2010 12:41 pm

Re: Hollywood way to benchmarks...?

Post by Tuxedo »

OK,
so I'll continue with my benchmarks suite...

Lets see what happens :)

Any usefull test do you can suggest to me to do?

EDIT:

So on Amiga-like systems I've to expect same performances on same routine/hw but different OS like Pegasos2 with OS4.x/MOS?
Simone"Tuxedo"Monsignori, Perugia, ITALY.
User avatar
airsoftsoftwair
Posts: 5433
Joined: Fri Feb 12, 2010 2:33 pm
Location: Germany
Contact:

Re: Hollywood way to benchmarks...?

Post by airsoftsoftwair »

Tuxedo wrote:OK,
Any usefull test do you can suggest to me to do?
Not really, just try and make some comparisons...
So on Amiga-like systems I've to expect same performances on same routine/hw but different OS like Pegasos2 with OS4.x/MOS?
No. It also depends on the gfx subsystems, memory handler, etc. There are many variables in such equations :)
User avatar
Tuxedo
Posts: 338
Joined: Sun Feb 14, 2010 12:41 pm

Re: Hollywood way to benchmarks...?

Post by Tuxedo »

OK,
so its really usefull only on same OS to see different machines performances and also to see on same machine but different OS the OSes performance right?
Simone"Tuxedo"Monsignori, Perugia, ITALY.
User avatar
airsoftsoftwair
Posts: 5433
Joined: Fri Feb 12, 2010 2:33 pm
Location: Germany
Contact:

Re: Hollywood way to benchmarks...?

Post by airsoftsoftwair »

Not necessarily. If done in a cross-OS way it might also give some feedback as to how fast the OS drawing functions are in comparison to other platforms, e.g. MorphOS vs. OS4 vs. AROS etc.
User avatar
Tuxedo
Posts: 338
Joined: Sun Feb 14, 2010 12:41 pm

Re: Hollywood way to benchmarks...?

Post by Tuxedo »

I've uploaded my little program on AmigaWorld:
http://amigaworld.net/modules/newbb/vie ... at&order=0

everyone can test it if interested also there :)

Get some interesting values, but also some weird, for example:

- Why on my Pegasos2 there was big differences from OS4.1Upd6 to MOS 3.7 ? For gfx big differences was quite ok but on cpu I think no... I get that results:

My results related to AmigaOS4.1Upd6 with Pegasos2 G4@1131 and Radeon9000 pro:

CPU 1: 1547 CPU 2 : 9860 CPU 3: 4780
Graphics 1: 9403 Graphics 2 : 50 Graphics 3: 4039 Graphics 4: 4120

Results related to MOS 3.7 with Pegasos2 G4@1131 and Radeon9000 pro:

CPU 1: 654 CPU 2 : 4758 CPU 3: CPU3: 2195
Graphics 1: 6815 Graphics 2 : 50 Graphics 3: 5001 Graphics 4: 5243


- Why If I compile an applet with Hollywood 5.3 GUI the hollywood player under AmigaOS like systems says that it wasnt a compressed applet and so wasnt playable but same applet on Android hollywood player works without problem?

Thank you!
Simone"Tuxedo"Monsignori, Perugia, ITALY.
Post Reply