Possible driver bug - Vulkan - Intel Graphics (Intel Core Ultra 9 185H)

[GregConnected]

Our application shows visual corruption - flashing triangles, when run on one specific Intel GPU.

System:

GPU: Intel Graphics - Intel Core Ultra 9 185H
Driver: v31.0.101.5594 date: 6/28/2024
OS: Windows 11 Pro (23H2, build 22631.3737)

The application does not exhibit the issue on other Intel, nVidia or AMD GPUs.
The application is also build for Linux and Mac, DirectX11, OpenGL 4 as well as Vulkan.
The Vulkan debug layer reports no errors or synchronization issues.

I may have isolated the issue:

Geometry shaders that return without emitting any primitives, instead emit random junk with this GPU and driver.

I have not got a minimal test app, but can capture a frame in RenderDoc where a single draw call, where geometry is culled in GS and otherwise off screen, causes the random triangles to display with fragment shader coloring.

Can you please advise how to proceed investigating or submitting this issue?

[ACarmona_Intel]

Hello GregConnected, 

Thank you for posting in our communities.

May we have the complete name of your application? Furthermore, if possible, kindly share with us a video of the exact issue so we can better visualize it.

We look forward to your response!

Best regards,

Carmona A.

Intel Customer Support Technician

[GregConnected]

The software is called GEAR, but it is not sold to the public. It's part of a semi-embedded software.

I've attached a video of the visual effect. I've modified the software to remove most visual features and help isolate the issue.

The green lines are good, however the lines themselves are in segments which can have zero lengths.  The geometry shader drawing the lines culls lines of zero length (also  culls lines that intersect the near plane, but I can comment that out to isolate to single piece of GLSL code.)

float4 lineStart = gl_in[0].gl_Position;
float4 lineEnd = gl_in[1].gl_Position;

float2 lineSegRHS = lineEnd.xy - lineStart.xy;
if( dot(lineSegRHS,lineSegRHS) == 0.0 ) // Just use zero instead of epsilon, coords are in clip space here
{
    return; // Skip degenerate line
}

This code snip just culls degenerate lines early.
Commenting out the 'return' here works around the issue.

Interestingly, the bad visual effect of drawing junk primitives can relate to other applications running on the system. For example, if there is no visual corruption present in a scene. I can sometimes open an instance of eg. Windows Explorer on different monitor, and drag that window in circles with my mouse. On the other monitor, that causes graphic effects in the application in sync with my mouse movement.

(Back to the application) If I disable all materials in the application except one that can repro the issue, I can see the effect in RenderDoc with a single draw call. However, the application itself shows nothing on screen due to composition logic.
In this case, like mentioned above, the graphic effect is best reproduced by running other applications at the same time, otherwise
no visual effect may be seen.

I've attached a screenshot of RenderDoc with single frame and just a few draw calls, but the issue reproducing on the first draw call.

The shape of the green triangle/quad/junk changes with every click on the draw call in RenderDoc. Examining the pipeline content shows no anomaly, only the raster result shows the issue.

I have not yet been able to isolate further, such as doing nothing but drawing a single line from two points and culling it.

A note on the attached video: I've purposely selected a starting time/location where the visual corruption is not present much. I don't believe the issue itself specifically relates to the geometry being clipped off screen. Just showing less, then more and bad visual effect, so you can observe some normal behavior.

[ACarmona_Intel]

Hello GregConnected, 

Thank you so much for a detailed response.

I will raise the case with our engineers for further investigation so we can provide you with an appropriate recommendation, and then I will get back to you as soon as I have their response.

Thank you for your patience.

By the way, all the files, codes, and information that you have shared were highly noted.

Best regards,

Carmona A.

Intel Customer Support Technician

[ACarmona_Intel]

Hello GregConnected, 

Thank you so much for patiently waiting on our response.

We have observed that the issue is isolated to a particular Intel GPU.

To advance our troubleshooting and investigation, it would be beneficial if we could have the exact processor and graphics card model that you have utilised without encountering this issue. This information will allow us to narrow down our investigation to the specific graphics and processors in question.

We will wait for your response, and we thank you for your understanding.

Best regards,

Carmona A.

Intel Customer Support Technician