run Miri on CI (allowed to fail)

[RalfJung]

With this (and an upcoming Miri PR), a number of tests are actually passing under Miri :D

cargo miri test --test i32_ops -- --skip cast --skip max --skip min --skip is_
cargo miri test --test f32_ops -- --skip cast --skip max --skip min --skip is_

[RalfJung]

I could also add a Miri job to CI? We probably want it to be 'allowed to fail' so that new intrinsics that Miri does not support yet can still easily be added, but it would be nice to somehow track whether the Miri state becomes worse.

[calebzulawski]

How long does it take to run? Sounds like a good idea to me.

[RalfJung]

Not sure, I would guess around 20-30min for the entire test suite -- but I can't say yet because not all of it passes yet. ;)

[RalfJung]

Ah, now Miri fails since it is still a day old and hence requires const_fn_trait_bound. ;) That should fix itself tomorrow though.

[RalfJung]

error: unsupported operation: unimplemented intrinsic: simd_ceil
  --> /home/runner/work/portable-simd/portable-simd/crates/std_float/src/lib.rs:91:18
   |
91 |         unsafe { simd_ceil(self) }
   |                  ^^^^^^^^^^^^^^^ unimplemented intrinsic: simd_ceil
   |

Interesting, that intrinsic is not even declared in https://github.com/rust-lang/portable-simd/blob/master/crates/core_simd/src/intrinsics.rs. 🤔

[RalfJung]

But the tests that are now in there should work. It's a bit annoying to check this since "ignoring failures" means getting green checkmarks everywhere (with GHA there is no way to get a red checkmark for the job but still get a green checkmark for the PR as a whole) -- one has to look at the log to check if it actually failed. But that's the best we can do I guess.

[workingjubilee]

Yes, that intrinsic is listed with the other "might libcall" floating point intrinsics.

portable-simd/crates/std_float/src/lib.rs

Lines 24 to 44 in 2d13059

	// "platform intrinsics" are essentially "codegen intrinsics"
	// each of these may be scalarized and lowered to a libm call
	extern "platform-intrinsic" {
	// ceil
	fn simd_ceil<T>(x: T) -> T;
	// floor
	fn simd_floor<T>(x: T) -> T;
	// round
	fn simd_round<T>(x: T) -> T;
	// trunc
	fn simd_trunc<T>(x: T) -> T;
	// fsqrt
	fn simd_fsqrt<T>(x: T) -> T;
	// fma
	fn simd_fma<T>(x: T, y: T, z: T) -> T;
	}