We appreciate the effort here! I can't speak to the usefulness of the changes myself but I'm sure @Vecvec will have thoughts.

This PR also includes a rewrite for the naga MSL writer to emit "correct" MSL for ray queries. (this part was heavily written by AI as I do not understand the naga internals at this time)

This kind of approach is unlikely to result in a merge unfortunately. We want contributors to know exactly what the code is doing, and be able to explain and defend the choices made. "Oh, AI did it" isn't a valid excuse for poor or unnecessary code.

In particular, I doubt the MSL writer is really that broken. If it was, you would be better off first filing an issue explaining that and then making a separate PR once its been confirmed that major changes are required (or waiting for someone who is better versed handle the issue if they desire).

@inner-daemons, from reading over the previously generated code, the previous method does not support AABB intersections (and probably non-opaque intersections too).

Interestingly, yesterday I was actually re-writing the metal ray tracing backend to use intersection queries to try and isolate #9100 (though my initial draft did not help it at all, so its probably not that). It would probably need some polishing up before opening a PR, but if @dylanblokhuis would prefer it, I can do some more work on it.

@Vecvec Looks like I misinterpreted this, and the MSL writer changes are needed and relevant, so you can ignore those doubts. Thanks for following up so quickly.

@inner-daemons, from reading over the previously generated code, the previous method does not support AABB intersections (and probably non-opaque intersections too).

Interestingly, yesterday I was actually re-writing the metal ray tracing backend to use intersection queries to try and isolate #9100 (though my initial draft did not help it at all, so its probably not that). It would probably need some polishing up before opening a PR, but if @dylanblokhuis would prefer it, I can do some more work on it.

Whatever you prefer! Perhaps the Naga MSL output from the ray_aabb_compute example can help you write the custom intersector part. Since it is actually working currently.

Would be great to have ray queries working on the big three, then all the voxel game dev enthusiasts can use hardware ray tracing in their projects.

@dylanblokhuis I've opened the patch for msl-out in #9304

Also, this will need some tests. You can find the ray tracing ones under tests/tests/wgpu-gpu/ray_tracing.

@Vecvec I added tests

I reverted the naga changes so the PR is only focused on wgpu-core now

I didn't notice this earlier, but the example probably should have a readme in it.

I didn't notice this earlier, but the example probably should have a readme in it.

Added the README and changed up the stride location.

I'm afraid I was unclear about where to put the validation. Once this is changed, I think everything should be good to merge.

I'm assuming you don't want to wait on #9304 because you have not rebased on it. If that is so, could you add a second line to aabb compute (either in the readme or on the expected fail) that explains that metal does not support aabb intersections yet so this example may not work on it.

I added this!