screenSpace: make it work when GL_EXT_gpu_shader4 is missing

[illwieckz]

WIP: screenSpace: make it work when GL_EXT_gpu_shader4 is missing

It doesn't work yet, I'm not sure to understand how that works.

The purpose is to slightly modify #1593:

• Improve screen-space shaders, some minor clean-up #1593

Commit 5666be1 from #1593 brings many improvements, but also makes use of gl_VertexID, which is introduced by EXT_gpu_shader4 if I'm right, making it mandatory.

So I tried to add an alternate code path based on specific parts I read from 5666be1 that replaced the old code (among many other things), but it's not working yet.

It's now probably very close to have it working so if you have any idea, please share. ☺️

[illwieckz]

This is how it currently looks:

[slipher]

I tried it and it rendered correctly only the upper right quarter of the screen. So I tried this and it worked:

--- a/src/engine/renderer/tr_vbo.cpp
+++ b/src/engine/renderer/tr_vbo.cpp
@@ -555,7 +555,7 @@ void R_BindNullIBO()

 static void R_InitGenericVBOs() {
        // Min and max coordinates of the quad
-       static const vec3_t min = { 0.0f, 0.0f, 0.0f };
+       static const vec3_t min = { -1.0f, -1.f, -1.0f };
        static const vec3_t max = { 1.0f, 1.0f, 0.0f };
        {
                /*

I don't know why it would be necessary to use -1 to 1 now if 0 to 1 worked before. Note that this might break other, non-screenspace uses of genericQuad.

Your screenshot seems to have additional bugs so I guess there must be more incompatible things if you have the real old hardware.

[illwieckz]

We switched something from a quad to a triangle, this explains why:

• https://stackoverflow.com/questions/2588875/whats-the-best-way-to-draw-a-fullscreen-quad-in-opengl-3-2/59739538#59739538

And to have a triangle covering all the screen we need the triangle being larger than the screen.

So maybe that's related. We may have modified some generic defaults for the triangle code, and when running in no-gpu_shader4 code path, if we want to keep most of the triangle code unmodified, maybe using -1 as quad min makes sure the quad fits the whole screen in that triangle? @VReaperV may know better about it.

I'll try the -1 trick soon. 🙂

That Intel computer is used in some family vacation home and is used by teens to play games in the holidays, I would like to keep the game working on it. ☺️

[illwieckz]

Your screenshot seems to have additional bugs so I guess there must be more incompatible things if you have the real old hardware.

Yes those were driver bugs, after an update they disappeared.

So, when I use both:

• Renderer: Avoid a DSA function with VAOs #1733
• tr_shader: do not load heatHaze map when heatHaze is disabled #1737
• cameraEffects: make sRGB conversion work in older GLSL #1738
• screenSpace: make it work when GL_EXT_gpu_shader4 is missing #1739 (this one)

I get this (plat23):

The forcefields are a bit dim because that's an sRGB-compiled map running in the linear pipeline, so it's not because of an engine bug or a driver bug on this specific hardware. This produces what we currently expect.

[illwieckz]

And when also using:

• tr_shader: linearize fog colors and fog image #1730

I get this (atcshd):

This is using the lowest graphics preset and then the vertex lighting, so that ugliness is expected, I see no bug there.

[slipher]

PTAL at the branchslipher/1739-fixup. There I have changed the projection matrix of Tess_InstantQuad so it can produce the correct results while having replaced the [0, 1] square with [-1, 1].

[illwieckz]

I don't mind the implementation, if you prefer yours please force-push this PR.

I've tested your implementation and it works as well.

Maybe @VReaperV has an opinion on that too.

[illwieckz]

Screenshots with your branch (and all the other patches listed above), but this time with the low preset:

[illwieckz]

I forgot to mention but outside of actually making the engine works on hardware without GL_EXT_gpu_shader4, this would also make it consistent because we actually disable GL_EXT_gpu_shader4 on GLSL 1.20 because of having a different syntax on some stuff like unsigned int and uint, as discussed here:

• https://gitlab.freedesktop.org/mesa/mesa/-/issues/12803#note_2819461

So, without this branch, when this extension is available, on some hardware we disable some GL_EXT_gpu_shader4 code while still using the extension, which is a bit confusing and inconsistent.

[slipher]

I don't mind the implementation, if you prefer yours please force-push this PR.

Done. The point of my edit is to avoid breaking the Tess_InstantQuad semantics. Breaking that might not have any consequences outside of r_showBspNodes for now, but it would be a stumbling block for the future.

[slipher]

You have pushed the branch again and it no longer contains my changes from slipher/1739-fixup.

[illwieckz]

You have pushed the branch again and it no longer contains my changes from slipher/1739-fixup.

Oupsie, that should now be fixed.

I also added comments linking to this PR so future selves may know why we picked those numbers.

[slipher]

The point of this function is to render any screen-space axis-aligned rectangle specified by x, y, width, height so the comment seems out of place.

[slipher]

To explain the values chosen here, you could say that they are chosen to specify a full-screen rectangle for screen-space shaders without a projection matrix. The values here don't influence the triangle but it's good to mention that this is a fallback when the gl_VertexID-based triangle can't be used.

[illwieckz]

I updated the comments.

[slipher]

LGTM