Better float casting tests in coretests

[eggyal]

A direct port of the float casting tests from MIRI.

Closes #152592

r? tgross35

[rustbot]

tgross35 is currently at their maximum review capacity.
They may take a while to respond.

[okaneco]

My first exposure to the cranelift codebase was in the float constant PR, so I'm by no means an expert.

There was a change in that PR that you might want to cherry pick so it solves the separate f128->int casts issue c4582c8

For this CI failure, in the lib_call on line 220, should that to_ty be replaced with ret_ty in the third argument of the Vec? to_ty would still be an I8 if passed directly, but line 208 sets ret_ty to an I32.

rust/compiler/rustc_codegen_cranelift/src/codegen_f16_f128.rs

Lines 207 to 254 in a423f68

	} else if from_ty == types::F128 && to_ty.is_int() {
	let ret_ty = if to_ty.bits() < 32 { types::I32 } else { to_ty };
	let name = format!(
	"__fix{sign}tf{size}i",
	sign = if from_signed { "" } else { "un" },
	size = match ret_ty {
	types::I32 => 's',
	types::I64 => 'd',
	types::I128 => 't',
	_ => unreachable!("{from_ty:?}"),
	},
	);
	let ret =
	fx.lib_call(&name, vec![AbiParam::new(from_ty)], vec![AbiParam::new(to_ty)], &[from])
	[0];
	let val = if ret_ty == to_ty {
	ret
	} else {
	let (min, max) = match (to_ty, to_signed) {
	(types::I8, false) => (0, i64::from(u8::MAX)),
	(types::I16, false) => (0, i64::from(u16::MAX)),
	(types::I8, true) => (i64::from(i8::MIN as u32), i64::from(i8::MAX as u32)),
	(types::I16, true) => (i64::from(i16::MIN as u32), i64::from(i16::MAX as u32)),
	_ => unreachable!("{to_ty:?}"),
	};
	let min_val = fx.bcx.ins().iconst(types::I32, min);
	let max_val = fx.bcx.ins().iconst(types::I32, max);
	let val = if to_signed {
	let has_underflow = fx.bcx.ins().icmp_imm(IntCC::SignedLessThan, ret, min);
	let has_overflow = fx.bcx.ins().icmp_imm(IntCC::SignedGreaterThan, ret, max);
	let bottom_capped = fx.bcx.ins().select(has_underflow, min_val, ret);
	fx.bcx.ins().select(has_overflow, max_val, bottom_capped)
	} else {
	let has_overflow = fx.bcx.ins().icmp_imm(IntCC::UnsignedGreaterThan, ret, max);
	fx.bcx.ins().select(has_overflow, max_val, ret)
	};
	fx.bcx.ins().ireduce(to_ty, val)
	};
	if let Some(false) = fx.tcx.sess.opts.unstable_opts.saturating_float_casts {
	return val;
	}
	let is_not_nan = fcmp(fx, FloatCC::Equal, from, from);
	let zero = type_zero_value(&mut fx.bcx, to_ty);
	fx.bcx.ins().select(is_not_nan, val, zero)
	} else {

[tgross35]

What @okaneco should be accurate. The fix is coming via #152512, but you can cherry pick the linked commit if you'd like to get testing before that merges.

[tgross35]

That linking error should be what the commit (c4582c8) fixes. The symbols starts with __fixunstf, not __fixuntf.

It should merge sometime later today, probably easiest to just rebase once it does.

[eggyal]

Ah, indeed - I messed up the cherry-pick, apologies.

[rustbot]

Some changes occurred in compiler/rustc_codegen_cranelift

cc @bjorn3

[rustbot]

The Miri subtree was changed

cc @rust-lang/miri

[RalfJung]

Miri does run coretests but not in its own CI, so I don't think we should remove existing tests.

[tgross35]

If tests from tests/ui/numbers-arithmetic/saturating-float-casts-impl.rs and tests/ui/numbers-arithmetic/u128-as-f32.rs are covered here, I think it would be fine to remove those UI tests.

View changes since this review

[tgross35]

Could you cover f16 and f128 here as well?

[eggyal]

All starting from f64, or do we want to test NaN casts going in other directions too?

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.