Enable MPFR tests on i586

[tgross35]

No description provided.

[programmerjake]

i think for i586 it should just use frndint with setting the x87 CW (actually not slow) for floor/ceil. like: https://github.com/NetBSD/src/blob/20433927938987dd64c8f6aa46904b7aca3fa39e/lib/libm/arch/i387/s_ceil.S

if asm! supported x87 registers as inputs/outputs, that would be nice...

[tgross35]

i think for i586 it should just use frndint with setting the x87 CW (actually not slow) for floor/ceil. like: https://github.com/NetBSD/src/blob/20433927938987dd64c8f6aa46904b7aca3fa39e/lib/libm/arch/i387/s_ceil.S

These showed up when I (recently) enabled tests against musl for edge cases, I just marked them xfail to unblock adding tests 5e0c840. This sounds like a good idea though, assuming that is license-compatible (looks like it) we could probably reuse a few x87 routines from there. Opened rust-lang/compiler-builtins#837 to track it.

Somewhat more concerning is that it looks like our fmaf and fdim are not always accurate on this platform...

if asm! supported x87 registers as inputs/outputs, that would be nice...

Yeah, I know a f80 in core::arch has been discussed a lot. I'll probably pick this up after f128 and f16 get further along if nobody else has gotten to it at that point.

[programmerjake]

if asm! supported x87 registers as inputs/outputs, that would be nice...

Yeah, I know a f80 in core::arch has been discussed a lot. I'll probably pick this up after f128 and f16 get further along if nobody else has gotten to it at that point.

i meant passing a f32 or f64 in a x87 register, we don't need to wait for f80 for that.