Add an read_clock_khr function that calls OpReadClockKHR

[expenses]

Reading from the shader clock lets us create heatmaps for ray tracing as seen here: https://developer.nvidia.com/blog/profiling-dxr-shaders-with-timer-instrumentation/.

[khyperia]

Please also add compiletests

[expenses]

Hmm, compiletests doesn't seem to like those #[cfg] flags.

[khyperia]

... oh, ugh, right, since we build spirv-std without those flags, but the test itself with those flags (something that never happens, and cannot happen, in the real world - can only happen here since we're directly invoking rustc instead of going through cargo). uuuuh, hmm, going to try to think of a way to resolve it...

[tomachinz]

nanosecond timing will be useful.

Timestamp values in this time domain are in units of nanoseconds and are comparable with platform timestamp values captured using the POSIX clock_gettime API. This extension advertises the SPIR-V ShaderClockKHR capability for Vulkan, which allows a shader to query a real-time or monotonically incrementing counter at the subgroup level or across the device level. The two valid SPIR-V scopes for OpReadClockKHR are Subgroup and Device.

[khyperia]

This is a horrible no-good very bad solution, and I'm really hoping I can think of something better, but right now my only thought to fix it is adding ShaderClockKHR and the ext to here.

[expenses]

This is a horrible no-good very bad solution, and I'm really hoping I can think of something better, but right now my only thought to fix it is adding ShaderClockKHR and the ext to here.

That doesn't look too too bad to me :)

Another think that I thought off that could to have a seperate feature flag just for the compiletests:

#[cfg(any(
    all(
        target_feature = "Int64",
        target_feature = "ShaderClockKHR",
        target_feature = "ext:SPV_KHR_shader_clock"
    ),
    target_feature = "compiletests"
)]

That way we don't have to shove a whole lot of features into that line you linked.

[expenses]

@khyperia Happy to go with either idea, or to put this on-hold for a bit.

[khyperia]

Whatever you think is best works!

[expenses]

Another think that I thought off that could to have a seperate feature flag just for the compiletests:

#[cfg(any(
    all(
        target_feature = "Int64",
        target_feature = "ShaderClockKHR",
        target_feature = "ext:SPV_KHR_shader_clock"
    ),
    target_feature = "compiletests"
)]

That way we don't have to shove a whole lot of features into that line you linked.

Actually, that doesn't work. We just run into an invalid capability error:

rust-gpu/crates/rustc_codegen_spirv/src/target_feature.rs

Lines 15 to 21 in 4e5f347

	if let Some(input) = input.strip_prefix(EXT_PREFIX) {
	Ok(Self::Extension(Symbol::intern(input)))
	} else {
	Ok(Self::Capability(input.parse().map_err(|_err| {
	format!("Invalid Capability: `{}`", input)
	})?))
	}