3D cube routine example ?

[rpineau]

@exxos : Any good 3D code will NOT use the MULx and DIVx instruction of the 68000 which is why you don't see a huge improvement.
@AtariZoll : executing it the way exxos shows doesn't have a lot of setup, you load the bin fine (relocatable dri linkable object code) and jump into it. So it you start counting just before the call instruction you get fairly accurate runtime (I trace the code that does the call and it doesn't do a lot before doing the jsr into my code).

Rodolphe

[mfro]

Since the application's name ist GEMBench (and not "DemoBench" ), I just hacked together this one:

VDI rotating cube

[AtariZoll]

@AtariZoll : executing it the way exxos shows doesn't have a lot of setup, you load the bin fine (relocatable dri linkable object code) and jump into it. So it you start counting just before the call instruction you get fairly accurate runtime (I trace the code that does the call and it doesn't do a lot before doing the jsr into my code).
Rodolphe

As I said, there is longer initialization before displaying of cube/thetraeder starts. Only reasonable way for accurate measuring is that called program/subrutine returns the result. That's elementary and supported by (T)OS .

[exxos]

Since the application's name ist GEMBench (and not "DemoBench" ), I just hacked together this one:

Nice :)maybe you could add a timer in there and loop it for 20 seconds or so and print the score after ? Then I could try it at 16mhz speeds, but don't think it would show any higher speeds than other VDI tests in gembench

[mfro]

Since the application's name ist GEMBench (and not "DemoBench" ), I just hacked together this one:

Nice :)maybe you could add a timer in there and loop it for 20 seconds or so and print the score after ?

Sure. The video above has been shot with Aranym, however, where VDI drawing is virtually free (because of Natfeats). Firebee speed is comparable.

With Hatari (which comes pretty close to a Falcon speedwise) it's like that:

spinning VDI cube

(about a second per frame)

[exxos]

Cool - can you post your program so I can try it out on my ST's here ?

[mfro]

Sure.

I'd assume it will be nice to your speeder since it's using ROM routines for drawing and very "straightforward" (just to avoid the word stupid) projection code.

[exxos]

355ms 8mhz ST
255ms 16mhz ST

So about 40% faster on a 16mhz ST. That is about what expected with ROM functions

btw- on my ST you app says "illegal AES function call" when it first loads, but seems to work ok after selecting the cube test

[mfro]

btw- on my ST you app says "illegal AES function call" when it first loads, but seems to work ok after selecting the cube test

Known issue.

I use the same code as testbed for AES 3.40 popup menus so there are calls that a plain ST doesn't understand. There should be no harm besides the alert message.

[exxos]

Can anyone shed any light on doing underline in button text ?

gbp.jpg

On the left is the GB window as it loads form RSC file. For example GEM Dialog Box on the actual GB screen (on right) as the "D" underlined. But in the RSC file (and when I loaded the form) it says "[D" . So I assume in the HGK somewhere it looks for the "[" bracket and makes the letter after it a underlined letter.

I know there is VST Effects, but I only got that to do full line in that effect. I assume at some point the button text gets redone somehow, but no idea how. I'm trying to backtrack the GB code, but there are so many jumps in the code all over the place that its hard to actually work out whats going on. Really this is the only hold up I have so far in making progress with GB6

[mfro]

With standard AES, this is only possible using USERDEFs.

These are AES objects that call back a user defined routine which is supposed to draw the UI element using VDI functions.
A string with a single underlined character, for example, could be drawn using v_gtext() to draw the string, vqt_extend() to obtain the position of the underlined character and v_pline() to draw the underline.

[fitten]

In case of Ataris (and other oldies) without FPU integer divisions are avoided, even if there is such CPU command, because slowness. Instead it tables with precalculated results are used.

You can use tables of precalculated transcendentals and use fixed point arithmetic and it's all integer instructions (no FPU needed).

[Kochise]

https://github.com/Kochise/gouraud