Atomic64 fixes

[JMS55]

Connections

• Fixes [Naga] SHADER_INT64_ATOMIC_MIN_MAX issue #5888
• Fixes SHADER_INT64_ATOMIC_MIN_MAX can never be requested on Vulkan #5887
• Fixes wgpu_types::Features::SHADER_INT64_ATOMIC_ALL_OPS should also request shader_shared_int64_atomics #5784
• Fixes Naga emits wrong HLSL for atomicMax() on u64 buffer #5893

Description

• Apply fix from issue written by @jimblandy
• On Vulkan, check for atomic64 PHD feature when inspecting the instance to get supported adapter extensions (the code for vulkan features/extensions in wgpu is seriously confusing, and idk if PhysicalDeviceShaderAtomicInt64Features::default() is correct here, but it seems to work and follow the existing code pattern)
• On Vulkan, also request shader_shared_int64_atomics in addition to the buffer variant
• In the HLSL backend, write e.g. InterlockedMax64 instead of InterlockedMax when performing an atomic operation on a buffer

Testing

• Bevy shaders

Checklist

• [ x ] Run cargo fmt.
• [ x ] Run cargo clippy. If applicable, add:
  • [ x ] --target wasm32-unknown-unknown
  • [ x ] --target wasm32-unknown-emscripten
• [ x ] Run cargo xtask test to run tests.
• [ x ] Add change to CHANGELOG.md. See simple instructions inside file.

[jimblandy]

@JMS55 So, why wasn't tests/in/atomicOps-int64-min-max.wgsl broken by the absence of the fix in naga/src/front/wgsl/lower/mod.rs?

[JMS55]

I couldn't figure that out. My best guess is that maybe the fact that the second argument is not an expression matters. Maybe a constant vs expression in the second argument changes how things get parsed or something.

[jimblandy]

The clippy warnings are specific to this branch, they don't happen on trunk. I can see them with:

cargo +1.76 clippy -p naga --features=hlsl-out

I would like to have a snapshot test that is affected by the change to naga/src/front/wgsl/lower/mod.rs. I'm uncomfortable that we have no tests that can detect that fix.

[JMS55]

I didn't change that code though, so idk why clippy is complaining only on this branch... I mean I can make the fixes as part of this PR if you want, up to you.

[jimblandy]

I didn't change that code though, so idk why clippy is complaining only on this branch... I mean I can make the fixes as part of this PR if you want, up to you.

Yes, please change this PR so it does not break CI on trunk when it is merged.

[jimblandy]

I would like to have a snapshot test that is affected by the change to naga/src/front/wgsl/lower/mod.rs. I'm uncomfortable that we have no tests that can detect that fix.

I volunteered to work on this in chat.

[jimblandy]

@JMS55 So, why wasn't tests/in/atomicOps-int64-min-max.wgsl broken by the absence of the fix in naga/src/front/wgsl/lower/mod.rs?

It seems like the reason is, the final operands to the atomicMax and atomicMin calls in the snapshot are all simple expressions that don't need to be emitted. I've pushed a commit that uses a variety of final operands, and that causes a panic without the fix.

[jimblandy]

This looks good. I've pushed a commit to address the only nit I had.