OpenGL Core Profile 3.2+

[Nadro]

Hi,

Can you upload some test shader which doesn't work on new Radeon card? I didn't notice any problems with new Radeons and GL 2.x shaders. Migrate from 3.x to 4.x will be easy (mostly this is adding just new features, bacis functionality is the same for both drivers). I hope to publish driver in upcoming days. Driver is mostly finished, just some small fixes are require.

Cheers,

[hendu]

Yep, if your shader is correct, then it's a bug in AMD's driver (surprise!). If it claims GL 4, it must support GL 2.

[christianclavet]

Hi. Ok. But for the shader problem you already helped me me figure the problem it was in this thread (http://irrlicht.sourceforge.net/forum/v ... =4&t=49178)
This was done on Irrlicht 1.8.0

From the last tests I've done, it was the normal uniform that was not returning properly from the vertex shader to the fragment shader. I did not see other problems except this.

I've tested the shader on theses configurations (working in computer shop now, so I have access to lots of configurations):

Intel HD 3000 - Passed (but some clipping issues due to the intel driver)
Intel HD 4000 - Passed (same results as the HD 3000)

Nvidia GTX 460 - Passed
Nvidia GTX 560ti - Passed
NVidia GTX 780 - Passed


Radeon HD 4850 - Passed
Radeon HD Mobility 5450 - Passed
Radeon HD 7770 - Failed
Radeon HD 7850 - Failed

Note: All the tests were done on windows system. Don't know what is the result on Linux. (Worked fine with my dev system running Intel HD)

I think I've tried it with a Radeon HD R9, and other models but not sure I recall all of them. The failing started to happen from series 7000+, below that series model the rendering was ok. So I'm almost sure, it's not on the Irrlicht side but inside the card driver. For any of the cards, there is no compilation errors.

The shader source is in the thread I posted before. But simply, just send out the normal uniform from the vertex shader and render the results in the fragment shader and you will see it.

[Granyte]

Report the bug to the amd driver team the 7770 is a GCN card and i think the 7870 is a vilw4 both these were new achitecture. Beside the GCN driver is a complete mess at the moment they have been rewriting it from the core in the last 3 month.

Do you have a working binary i have a dual 7970 (GCN) setup to test your shader on

[christianclavet]

Lastest build is on the IRR RPG BUILDER svn. But the old version (0.21 on the web site) will have the same problem with it's terrain shader.

With the old version (download from the web site), the mixing of the blending of the rock texture won't be visible (using the normals to find a proper angle to display the rock texture). And the new one (SVN) it will be all dark (using the angle normal to get the lighting angle for shading will fail). With the new one, you will have to enable back the shading inside the shader (uncomment) to see it with the Lighting (disabled since it doesn't work)

[Granyte]

on 0.21 both my intel hd 4000 and my dual amd 7970 display identical results

[Nadro]

I'll check it in upcoming days on GCN hardware (at now I use mostly VLIW5 based card).

[christianclavet]

@granyte, to see the problem with 0.21, you must start a terrain and make high slopes on it. You will see textures but the rock texture won't show. (this is the only texture on 0.21 that use the normal value from the vertex shader). Base Irrlicht is rendering ok, never saw the problem there.

As mentioned before the best way to see the problem is to take the "normal" info from the vertex shader and display it directly in the fragment shader.

This problem was difficult to pinpoint, and if Nadro had not given me the trick to put the returning value directly in the rendering, I would never have found the problem.

I did not found about this problem until a user on the IRB forum reported it when he changed his video card to a radeon HD 7750. For testing AMD Radeon, I had only my AMD Radeon Mobility 5450 at home in my HTPC that always displayed correct results.

http://irrrpgbuilder.sourceforge.net/fo ... ?f=4&t=318
Here are some screens from his report and the discussion that determined that there was a problem when using the terrain shader. I decided to bring the application at the store to test it on a Radeon HD 7770. That confirmed the user problem and then began asking help on the Irrlicht forum, was lucky to have Nadro help to figure this out.

EDIT:I was about to take screenshots with the "normals" rendering but both system are rendering fine (Nvidia + AMD) with my latest shader. I will have to check more systems again, but this morning the AMD Radeon 7850 demo system is rendering the shader fully and correctly. (driver update). Perhaps they fixed this... Will have to check with other PCs in the store if it's really true. I never counted on AMD to fix their driver, so I'm really surprised if they've done it.
EDIT2: Tested with the AMD Radeon HD 7660G (GPU inside the FM2 Chip A10-5800K), the Radeon HD 7480D (From the FM2 Chip A4-5300) and it seem to run fine too, checked that the driver was updated to the latest. This really look like they fixed the problem!

[Granyte]

in that case my intel card failed the test to nethier of these made any slope texture visible

[christianclavet]

Hi,

Seem the AMD Radeon HD driver was fixed recently. Done tests again with updated drivers and it work now on series 7000 Hope that they(AMD) will keep the compatibility high with openGL 2.0 until we reach a more recent version of GL.

[Granyte]

the AMD driver is gonna be a mess until half way through the year when they are done rewriting it

[ent1ty]

That is, if they do it right this time

[Granyte]

from what I saw on my dual 7970 they are on the right track

[devsh]

we should have our OpenGL 3.1+ forward compatibility context with GLSL 1.40 irrlicht ready by the end of the month

[Mars_999]

will Geometry shaders be in 1.9? for OpenGL?