Zcomploc

[MayImilae]

Guess who's keeping count? Every time it's been added or removed, I go and update the poor Tony Hawk 3 wiki page with it. So far, it's been in and out of dolphin three times, assuming I haven't missed any.

3.0-143 - zcomploc support added
3.0-192 - zcomploc support removed
3.0-505 - zcomploc support added
3.0-578 - zcomploc support removed
3.0-678 - zcomploc support added
3.0-715 - zcomploc support fixed
3.0-750 - zcomploc support removed

I thought you guys might like to see the score.

But other than that, I do have a serious question about it. Over on the Dolphin EMU builds list, there is 3.0-750, and "3.0-750 zcomploc revert". Is the reversion in 3.0-750, or just that revert branch? Is that reversions branch a branch of dolphin that kept zcomploc? I'm a little confused about all this.

Also, while on the topic, what games need zcomploc? I've never been able to find a definitive list.

[neobrain]

Nevermind the zcomploc-revert branch, it was just meant as a temporary branch for testing.

BTW: Your list doesn't include the zcomploc hack which was introduced sometime before 146b02615c07 and which wasn't reverted.

[delroth]

(08-11-2012, 05:22 PM)MaJoR Wrote: Also, while on the topic, what games need zcomploc? I've never been able to find a definitive list.

Most games use it for optimization, the games that we know have visible issues without zcomploc being implemented are:

• 007: Everything or Nothing (Skybox in front of the map/models)
• Baten Kaitos (characters displayed in front of background objects)
• Tony Hawk's Pro Skater 3 (invisible skater)
• Okami (menu not showing up)
• Ty the Tasmanian Tiger (floating floor)
• Super Smash Bros Brawl ("Link chasing pig" animation behind the score screen)

As neobrain says, your timeline is a bit wrong. This last revert is just going back to the zcomploc implementation that was merged in 3.0-505, and THPS3 should still work with it (the screenshot in the issue comments is most likely caused by settings issues, Starscream tested that game and confirmed that the skater was visible, at least in the first level of the game).

Thanks for your work maintaining the wiki btw!

[MayImilae]

OH right, I remember that list from one of the commits. Doh! Thanks. Well, how are the games with the 3.0-505 hack? I would assume that since Rodolfo tried to replace the hack, it is still a bit flakey...


So, to update the list above.

3.0-143 - delroth zcomploc merger
3.0-192 - delroth zcomploc revert
3.0-505 - delroth zcomploc-support branch merged
3.0-566 - Rodolfo zcomploc merge
3.0-578 - Rodolfo zcomploc revert, restores 505 implementation
3.0-678 - Rodolfo zomploc code merge
3.0-715 - Skidau zcomploc fix for THPS3
3.0-750 - Rodolfo zcomploc code removed, restores 505 implementation

[Runo]

Rodolfo was trying a more accurate approach iirc, he seems to find lthe current solution too.. hackish.

[Master Swordsman]

So, uh... 3.0-586 is without zcomploc?
Well, I'm fiddling around with THPS3 and I need to say, without it it's not like the skater's invisible.
Many of them do lose details though (well, not "lose" in the strictest sense, they become transparent):
* Some clothing parts for custom skaters are missing (e.g. kneepads)
* Bucky Lasek, Rodney Mullen, Andrew Reynolds and Jamie Thomas are headless
* Steve Caballero and Geoff Rowley are headless and without torso
* Chad Muska's boombox is missing
* Elissa Steamer's hair is missing

Also, many pedestrians (mainly those that we need to complete objectives, e.g. the skaters in Canada or the pickpockets in the Airport - only their shadow can be seen) and cars are invisible (in Los Angeles, you can only see the wheels of the cars in the chase, but in other levels like Rio they're missing altogether).

[Starscream]

You should use the latest build before pointing out game issues.

[MayImilae]

Quote:So, uh... 3.0-586 is without zcomploc?

Read the list I gave. 3.0-586 is covered with the 3.0-505 zcomploc implementation. To see it without zcomploc, just run 3.0 straight up.

[MayImilae]

Necropost! Consider this a record for future reference.

3.0-143 - delroth zcomploc merger
3.0-192 - delroth zcomploc revert
3.0-505 - delroth zcomploc-support branch merged
3.0-566 - Rodolfo zcomploc merge
3.0-578 - Rodolfo zcomploc revert, restores 3.0-505 implementation
3.0-678 - Rodolfo zomploc code merge
3.0-715 - Skidau zcomploc fix for THPS3
3.0-750 - Rodolfo zcomploc code removed, restores 3.0-505 implementation
3.5-1649 - Proper zcomploc implemented in OpenGL. Requires a GPU with OpenGL 4.2+, otherwise the 3.0-505 hack will be used.
3.5-1729 - Proper zcomploc implemented in D3D11. Requires a GPU with full Direct3D 11.0 support, otherwise the 3.0-505 hack will be used.
4.0-155 - D3D9 backend removed. All backends support zcomploc on compatible hardware.


Necro-Edit! Neobrain explained me to what zcomploc is, and explained it really well. Posting it here for safekeeping. Thanks to pauldachz for keeping the record in his awesome quote log!

#dolphin-dev Wrote:// zcomploc explanation!
<neobrain> btw MaJoR1 DON'T FORGET TO ASK NEOBRAIN ABOUT THIS
<MaJoR1> lol
<neobrain> oh, you did ask me, kinda
<neobrain> is "what is zcomploc" all you need?
<MaJoR1> more or less yea
<neobrain> ok
<MaJoR1> I wrote it as a depth thing
<neobrain> do you know what a depth test and an alpha test are?
<MaJoR1> only what the words imply
<neobrain> depth test means to compare the currently rendered pixel depth with the one written in the depth buffer
<neobrain> there are different "depth test functions", e.g. LESS. LESS means that depth test passes if the rendered pixel depth is LESS than the one written in the depth buffer
<neobrain> alpha test is similar, just that it doesn't compare the currently rendered pixel alpha value to the one written in the color buffer but to a certain reference alpha value
<neobrain> (which can be adjusted arbitrarily)
<neobrain> usually, alpha testing is done before depth testing
<MaJoR1> kk
<neobrain> but zcomploc (aka earlyz) allows you to swap the order of them. This is usually done for performance reasons, but has a very important consequence
<MaJoR1> mhmm?
<neobrain> I should mention that in latez mode, both color and depth only get written to the corresponding buffers if both alpha and depth test pass
<neobrain> however, when earlyz is active, depth gets written as soon as the depth test passes, while the color gets written lateron when the alpha test passes
<neobrain> This means that under certain circumstances (which are anything but rare) it's possible that a depth value gets written to the depth buffer, but no color value gets written
<neobrain> which is impossible to implement in d3d9, and only recently has been possible to implement in d3d11.0 and ogl (via recent extensions)
<neobrain> because when the alpha test fails, there's no way to still write a depth value if the depth test passed.
<neobrain> Does that make sense to you?
<MaJoR1> yea
<neobrain> good
<MaJoR1> I can see why you didn't explain it though
<MaJoR1> eeesh
<neobrain>