On the WebGL dev list a WebGPU question was recently asked re: early Z discard. It was observed that the fragment shader appeared to be running for all fragments even if the fragment failed the depth test.

The poster noted that Vulkan appears to indicate that if the fragment shader writes to storage then early-z discards will be disabled. Is this something that the group has discussed before and is a known behavior for all applicable native APIs?

The WebGPU spec currently indicates that early-z discards may be disabled if the fragment shader writes to the frag_depth builtin but doesn't specify any other scenarios where that may happen. If this is expected behavior we should include it in the spec text.