GPU NVIDIA 3080, 516.59
When having the following draw order:
- Draw something with VK_EXT_post_depth_coverage enabled, e.g., add the following lines to the fragment shader:
#extension GL_EXT_post_depth_coverage : enable
layout(early_fragment_tests) in;
layout(post_depth_coverage) in;
- Execute a draw pass that does NOT have a fragment shader. E.g., a shadow pass. They have to be executed one after the other.
- Device lost at the next queue submit. NVIDIA Aftermath reports a DMA_PageFault trying to read the virtual address 0x0000
I could reproduce it for example by adding the extension snippet above in the following draw pass from Sascha Willems’ vulkan samples. This sample project has a shadow pass that uses only a vertex shader.
I could not reproduce this issue on AMD. Is this a known issue? The workaround we found so far is to add an empty fragment shader to the pass.
Thank you in advance.