Enable f128 -> f16 tests on Linux

[tgross35]

Since updating the docker images in #625, it looks like __extendhftf2 and __trunctfhf2 are available on all 64-bit Linux platforms.

[Amanieu]

I think this is fixed for i586 as well now?

[tgross35]

I just tried it, looks like they are still not available on i586. I updated the note to mention this

[Amanieu]

I still don't understand why this function is available on i686 but not i586: they should be using the exact same docker image and toolchain.

[tgross35]

Ok, think I figured it out - i586 and i686 are both missing __fixunstfti and __fixtfti. i686 was just always disabled at https://github.com/rust-lang/compiler-builtins/pull/631/files#diff-05ae0211bd995d6870067db299c32944cf2b743226442889bb07d48232afc2c5L26 because it has a f128 math bug, https://github.com/rust-lang/compiler-builtins/actions/runs/9640552090/job/26584543080?pr=631#step:13:244.

i586 probably also has that bug, but float tests without sse2 have been disabled since at least #384

compiler-builtins/testcrate/tests/addsub.rs

Lines 113 to 132 in 0ccc1bf

	#[cfg(not(all(target_arch = "x86", not(target_feature = "sse"))))]
	mod float_addsub {
	use super::*;
	float_sum! {
	f32, __addsf3, __subsf3, Single, all();
	f64, __adddf3, __subdf3, Double, all();
	}
	}
	#[cfg(not(feature = "no-f16-f128"))]
	#[cfg(not(all(target_arch = "x86", not(target_feature = "sse"))))]
	#[cfg(not(any(target_arch = "powerpc", target_arch = "powerpc64")))]
	mod float_addsub_f128 {
	use super::*;
	float_sum! {
	f128, __addtf3, __subtf3, Quad, not(feature = "no-sys-f128");
	}
	}