ATI users and shader dynamic lights

[Nash]

Graf, please answer this question because I think this needs to be established once and for all:

Will shader dynamic lights work on our hardware, or are we ATI users basically screwed?

[Graf Zahl]

The shader compiler does not like the source. If I can find out why I'll fix it of course - but knowing ATI it wouldn't surprise me the least if it throws another fit that makes it inoperable. This feature uses a massive abuse of shader funtionality to get the light data to the shader.

[Nash]

So does this essentially mean people like me will never be able to enjoy light shader features and any future improvements to it?

[Graf Zahl]

Huh?

[playerlin]

Is v1.3.10 beta fixed those shader compile problems? I'm not see those compile fail message on my HD 4350 machine. And it seem better than it was.

"stat rendertimes" command does something wrong, those result data 'fills' my whole screen.

And... the GZDoom's download page still gives 1.3.09, and the filename trick still work. :p

[Graf Zahl]

I'll immediately re-upload a 1.3.11 version. There's a small error in 'stat rendertimes' caused by the 'Bench' CCMD addition.

Regarding the shader, I fixed the problem yesterday.

[Nash]

In 1.3.11, shader lights work at first, but when I start moving around, I get a massive FPS drop, my videocard starts to spin super fast and the lights turn into black squares.

ATI 4870 X2, ATI Catalyst 9.10.

Is there anything other info I can provide to help you? I'm not really sure about these kinds of things...

[Graf Zahl]

Well, that means that the way I upload the light data does not work right on ATI. It was a massive and cheap abuse anyway. You won't miss much. All the shader lights really provide is some minimal HDR effect because I can overbrighten the lights. But I had to cap even that at 1.2 because anything stronger did not look good.

I'm not going to fix this code.

[Nash]

... essentially answering my question - we ATI users are screwed.

If you add features to the lights (like directional lights etc), I'm sure you're only going to go the shader method and won't bother with people who can't use shader lights.

If this is the case... then all I can say is, darn...

[Graf Zahl]

If I ever work on the light code again it'd be after reorganizing the data. In its current state there's not much I can do with it. So you won't miss anything.

[Alter]

Alright, I switched to Windows 7 X64 Ultimate and I can confirm that this issue isn't that fatal in ATI Catalyst 9.11. Of course I have ATI HD 4770 GDDR5 Sapphire 512MB OC'd. The issue ONLY happens when there are 100+ dynamic lights at once. By the way, I used 1.3.13 beta and the new OGL renderer as well. So it's not that tragic, I will definitely wait for reorganized light data that Graf spoke of, even if it's year or two away.

[Graf Zahl]

Do you mean 100+ lights in the scene or a single linedef?

It you refer to the entire scene it looks like ATI has problems with frequent updates of large amounts of uniforms for the shader.

Concerning reorganizing the light data, I fear that's not that easy. All alternative methods that are available require even more inefficient data maintenance because I can no longer store it per linedef.

[Alter]

Do you mean 100+ lights in the scene or a single linedef?

It you refer to the entire scene it looks like ATI has problems with frequent updates of large amounts of uniforms for the shader.

Concerning reorganizing the light data, I fear that's not that easy. All alternative methods that are available require even more inefficient data maintenance because I can no longer store it per linedef.

In the scene Graf, whatsoever the static dynamic lights slow down more for me than let's say, pulsing or flickering lights. Who said reorganizing the light data is easy? That's why I said that this issue does not bother me much. I just wanted to disprove that ATI cards aren't acting as fatal in gzdoom as people say so.