[Bug] On Launch, Distorted First Frame When AA

[Daft Mugi]

When launching the game (from the OS desktop) with antialiasing turned on, there is a distorted frame before the main menu is shown. After the brief distortion, the main menu displays correctly and the game can be played without issue. I've included a photo of my screen that I took with a camera.

With antialiasing turned off, I do not see any distortion when launching the game.

Anyone else notice this or just me?

 

Linux, Ubuntu 22.04
AMD Radeon RX 6000 Series
OpenGL version: 4.6 (Core Profile) Mesa 22.0.5 core

Bug: https://bugs.thedarkmod.com/view.php?id=6150

Edited November 4, 2022 by Daft Mugi

[duzenko]

Wow, how do you notice those frame bugs?

[Daft Mugi]

On 10/29/2022 at 2:57 AM, duzenko said:

Wow, how do you notice those frame bugs?

Indeed. I am cursed.

Edited November 4, 2022 by Daft Mugi

[duzenko]

Btw i have an ubuntu 22 virtual machine and I cannot get past the main menu. Mouse cursor is not working for me.

[Daft Mugi]

58 minutes ago, duzenko said:

Btw i have an ubuntu 22 virtual machine and I cannot get past the main menu. Mouse cursor is not working for me.

Did you see the TDM main menu?

The distorted frame shows after launching the game from the desktop and before the main menu displays.

[duzenko]

On 11/4/2022 at 11:37 PM, Daft Mugi said:

Did you see the TDM main menu?

The distorted frame shows after launching the game from the desktop and before the main menu displays.

Can't remember ever seeing it although I'm on AMD as well

But I never tried it with the MESA drivers - and usually my AA is off

[Daft Mugi]

@duzenko I tried to reproduce this issue using a virtual machine and could not. I don't get the distorted frame.

I ran a virtual machine with Ubuntu 22.04 guest on Ubuntu 22.04 host.

[Daft Mugi]

This still happens on a physical machine, though.

It happens with the training mission and other missions.

Oddly, it does not seem to happen with Hazard Pay.

[Daft Mugi]

I narrowed down this issue to the following line when msaa is "2" or greater:

// Line 226:renderer/FrameBuffer.cpp
qglRenderbufferStorageMultisample(GL_RENDERBUFFER, msaa, format, width, height);

In the function, "FrameBuffer::AddRenderBuffer()":

217  void FrameBuffer::AddRenderBuffer( GLuint &buffer, GLenum attachment, GLenum format, const idStr &name ) {
218    if (buffer != 0) {
219      qglDeleteRenderbuffers(1, &buffer);
220    }
221    qglGenRenderbuffers(1, &buffer);
222    qglBindRenderbuffer(GL_RENDERBUFFER, buffer);
223    GL_SetDebugLabel( GL_RENDERBUFFER, buffer, name );
224  
225    if (msaa > 1) {
226      qglRenderbufferStorageMultisample(GL_RENDERBUFFER, msaa, format, width, height);
227    } else {
228      qglRenderbufferStorage(GL_RENDERBUFFER, format, width, height);
229    }
230  
231    Bind();
232    qglFramebufferRenderbuffer(GL_FRAMEBUFFER, attachment, GL_RENDERBUFFER, buffer);
233  }

As I stated in the first post, this issue is not present when "r_multiSamples 0". So, the following code path is fine:

// Line 228:renderer/FrameBuffer.cpp
qglRenderbufferStorage(GL_RENDERBUFFER, format, width, height);

Also, if "qglRenderbufferStorageMultisample()" is called with msaa as "0", that is also fine.

// Line 226:renderer/FrameBuffer.cpp
qglRenderbufferStorageMultisample(GL_RENDERBUFFER, 0, format, width, height);

I guess this is a bug in the Glad library.

I found this by commenting out code related to "r_multiSamples" and narrowed it down.

@duzenko @stgatilov What's a good next step?

Edited December 7, 2022 by Daft Mugi

[stgatilov]

This is framebuffer setup code.
You have learnt that it you disable multisampling = antialiasing, then the issue does not happen.

I know that there is one frame during mission startup when GL error occurs.
And as far as I recall, it occured because of some FBO weirdness.
Maybe it's the source of the problem.

If you enable:

• r_ignoreGlErrors 0
• r_glDebugContext 1
• r_glDebugOutput 2

Then I suppose you'll see it in game console.

Catching a problem on a single frame is quote hard.
Recently I used apitrace for this: it records all GL calls from the very start and allows to replay them to any moment.
We can try to use it if you want, but if the problem is visible only for you, then most likely only replaying on your machine will show it, so investigation will only be possible on your machine anyway.

[Daft Mugi]

12 hours ago, stgatilov said:

I know that there is one frame during mission startup when GL error occurs.

This happens during game launch and not mission startup, if mission startup means entering the game world. This happens before the main menu is displayed for the first time. I've attached a series of screenshots.

12 hours ago, stgatilov said:

And as far as I recall, it occured because of some FBO weirdness.

Does FBO code run before the main menu displays for the first time?

12 hours ago, stgatilov said:

If you enable:

• r_ignoreGlErrors 0
• r_glDebugContext 1
• r_glDebugOutput 2

I tried these settings and did not see any difference in console output.

12 hours ago, stgatilov said:

This is framebuffer setup code.

To me, it looks like that first frame is showing garbage memory - whatever happens to be in memory at the time. In other words, it seems that the "qglRenderbufferStorageMultisample()" function is not clearing/initializing the buffers before rendering. I'm not a graphics programmer, so I don't know. That's how it seems to me.

Maybe an updated Glad library would have a fix?

I tried running apitrace. Here's some output:

apitrace: loaded into ./thedarkmod.x64
apitrace: redirecting dlopen("libGL.so.1", 0x1) from ./thedarkmod.x64
apitrace: tracing to application.trace
apitrace: redirecting dlopen("libGL.so.1", 0x102) from ./thedarkmod.x64
apitrace: warning: unknown function "glRenderbufferStorageMultisampleAdvancedAMD"
apitrace: warning: unknown function "glNamedRenderbufferStorageMultisampleAdvancedAMD"
apitrace: warning: unknown function "glPolygonOffsetClamp"
apitrace: warning: unknown function "glEGLImageTargetTexStorageEXT"
apitrace: warning: unknown function "glEGLImageTargetTextureStorageEXT"
apitrace: warning: unknown function "glGetUnsignedBytevEXT"
apitrace: warning: unknown function "glGetUnsignedBytei_vEXT"
apitrace: warning: unknown function "glDeleteMemoryObjectsEXT"
apitrace: warning: unknown function "glIsMemoryObjectEXT"
apitrace: warning: unknown function "glCreateMemoryObjectsEXT"
apitrace: warning: unknown function "glMemoryObjectParameterivEXT"
apitrace: warning: unknown function "glGetMemoryObjectParameterivEXT"
apitrace: warning: unknown function "glTexStorageMem2DEXT"
apitrace: warning: unknown function "glTexStorageMem2DMultisampleEXT"
apitrace: warning: unknown function "glTexStorageMem3DEXT"
apitrace: warning: unknown function "glTexStorageMem3DMultisampleEXT"
apitrace: warning: unknown function "glBufferStorageMemEXT"
apitrace: warning: unknown function "glTextureStorageMem2DEXT"
apitrace: warning: unknown function "glTextureStorageMem2DMultisampleEXT"
apitrace: warning: unknown function "glTextureStorageMem3DEXT"
apitrace: warning: unknown function "glTextureStorageMem3DMultisampleEXT"
apitrace: warning: unknown function "glNamedBufferStorageMemEXT"
apitrace: warning: unknown function "glTexStorageMem1DEXT"
apitrace: warning: unknown function "glTextureStorageMem1DEXT"
apitrace: warning: unknown function "glImportMemoryFdEXT"
apitrace: warning: unknown function "glGetUnsignedBytevEXT"
apitrace: warning: unknown function "glGetUnsignedBytei_vEXT"
apitrace: warning: unknown function "glGenSemaphoresEXT"
apitrace: warning: unknown function "glDeleteSemaphoresEXT"
apitrace: warning: unknown function "glIsSemaphoreEXT"
apitrace: warning: unknown function "glSemaphoreParameterui64vEXT"
apitrace: warning: unknown function "glGetSemaphoreParameterui64vEXT"
apitrace: warning: unknown function "glWaitSemaphoreEXT"
apitrace: warning: unknown function "glSignalSemaphoreEXT"
apitrace: warning: unknown function "glImportSemaphoreFdEXT"
apitrace: warning: unknown function "glWindowRectanglesEXT"
apitrace: warning: unknown function "glMaxShaderCompilerThreadsKHR"
apitrace: warning: unknown function "glAlphaToCoverageDitherControlNV"

 

Edited December 7, 2022 by Daft Mugi

[Daft Mugi]

Screenshots from qapitrace, MSAA 2x.

MSAA 2x

In case it matters, the visible distortion is on frames 1 and 2 as the GL_BACK framebuffer. GL_DEPTH_COMPONENT and GL_STENCIL_INDEX framebuffers look distorted as well. Frame 1 is the first frame to have shaders, glprogs/tonemap.fs and glprogs/tonemap.vs. Frame 2 is the frame I notice on thedarkmod.x64 launch.

Frames 0 and 3 GL_BACK framebuffers look ok, but their GL_DEPTH_COMPONENT and GL_STENCIL_INDEX framebuffers look distorted.

MSAA off

Frame 2 does not have a distorted GL_BACK framebuffer; its GL_DEPTH_COMPONENT and GL_STENCIL_INDEX framebuffers look distorted. Frame 1 GL_BACK framebuffer looks similar to frame 1 when MSAA 2x.

 

(Screenshots MSAA 2x)

Edited December 7, 2022 by Daft Mugi

[stgatilov]

If these are the very first frames, then it's probably something about initial clear or SMP.

Do you have the issue if you set "com_smp 0" ?
(I assume you use the latest svn trunk)

[Daft Mugi]

 

26 minutes ago, stgatilov said:

If these are the very first frames, then it's probably something about initial clear or SMP.

Yes, these are the very first frames.

26 minutes ago, stgatilov said:

Do you have the issue if you set "com_smp 0" ?

Same issue with "com_smp 0".

26 minutes ago, stgatilov said:

(I assume you use the latest svn trunk)

Yes, r10199 currently.

Edited December 7, 2022 by Daft Mugi

[Daft Mugi]

@MirceaKitsune Does this happen for you as well? Now that I know you also run Linux and have an AMD GPU, I'm curious.

[AluminumHaste]

On 10/29/2022 at 3:57 AM, duzenko said:

Wow, how do you notice those frame bugs?

His game only runs at 1 fps

[Daft Mugi]

@stgatilov This was fixed by the following commit:

r10914 | stgatilov | 2024-12-08 | 8 lines

Added a way to force-disable tonemapping everywhere if 3D game is not running (r_tonemapOnlyGame3d).

This can be useful because the new tonemapping settings affect main menu rather negatively.
Well, in fact even the simple gamma already affects cinematics the bad way, introducing ugly color banding on dark colors.

Note: normally we should do tonemapping after 3D game is rendered but before GUI overlays are rendered.
But this is a hard topic, because some of the GUI overlays are not actually GUI (e.g. underwater overlay), so this requires some kind of manual tagging.
Better leave this for the future.

When launching TDM:

• r_tonemapOnlyGame3d 1: first frame is fine.
• r_tonemapOnlyGame3d 0: first frame is distorted.

[stgatilov]

20 hours ago, Daft Mugi said:

This was fixed by the following commit

Yeah, I guess it is sort-of expected.

Because when you start TDM, you don't have 3D game on your screen, so this new cvar effectively disables tonemapping.
And since you wrote above that the problem goes away if tonemapping is disabled... indeed it does go away

[Daft Mugi]

@stgatilov Even better news! The following commit fixed this issue regardless of r_tonemapOnlyGame3d setting.

r10930 | stgatilov | 2025-01-25 | 7 lines

Clear background during main menu.

This fixes the issue e.g. with AT1: Lucy and tonemap disabled.
The briefing there does not use any backgrounds, and no-clear policy results in HOM-like effects.

Originally reported here:
  https://forums.thedarkmod.com/index.php?/topic/22635-beta-testing-213/#findComment-499723

[stgatilov]

6 minutes ago, Daft Mugi said:

@stgatilov Even better news! The following commit fixed this issue regardless of r_tonemapOnlyGame3d setting.

Yeah, I thought it might happen

So I guess the explanation is something like this.

The main menu was previously rendered as overlay, on top of the previous contents. When the game starts, the previous contents is probably undefined. If tonemap is disabled, then rendering goes directly to OS window buffer, and it is probably cleared to black in all systems. But if tonemap is enabled, then rendering goes to framebuffer, and its initial contents is undefined. Perhaps it happens to be black usually, but e.g. mesa drivers don't clear it unlike other drivers.

Now there is forced clear to black at the beginning of frame render while main menu is active, so the issue is fixed.

[Daft Mugi]

@stgatilov Do you think the same can be done with the loading screen?

Sometimes there's distortion when loading a new map. (Note bright green artifact on lower-left edge in the screenshot.)

[stgatilov]

8 minutes ago, Daft Mugi said:

@stgatilov Do you think the same can be done with the loading screen?

Try svn rev 10931?

[Daft Mugi]

8 minutes ago, stgatilov said:

Try svn rev 10931?

Awesome - fixed! Thank you!