Missing cmath function: ldexp

[RalfJung]

The libcore test suites calls ldexp via FFI. Unfortunately, libstd does not expose this function in any way. What would be the best way for us to implement this shim?

I found this crate. Alternatively we could try to just define the extern symbol ourselves and see if that works on all our host platforms?^^

[pvdrz]

I am taking care of this one

[RalfJung]

Awesome!

Do you need mentoring? The file to edit is this one.

[pvdrz]

Thank you Ralf. It seems my gut feeling was right this time :D that's the file I modified. I just added a new match under // math functions based on the description for ldexp:

        "ldexp" => {
            let x = f32::from_bits(this.read_scalar(args[0])?.to_u32()?);
            let exp = f32::from_bits(this.read_scalar(args[1])?.to_u32()?);
            this.write_scalar(Scalar::from_u32((x * 2.0f32.powf(exp)).to_bits()), dest)?;
        }

Should that be enough? Or should I use the crate you mentioned?

I also suppose we should remove https://github.com/rust-lang/rust/blob/0f11354a9c1bf0c5ac250c7fa2bafc289a662f42/src/libcore/tests/num/flt2dec/mod.rs#L1, is that right?

[RalfJung]

I just added a new match under // math functions based on the description for ldexp

Hm. I guess it is a reasonable start. Maybe add a FIXME saying that if we see imprecise results, we should try to use the C math lib ldexp instead.

Your code should also, like the other math functions, have a FIXME because it uses host floats.

And finally, find an appropriate test case to extend and call the function there.

I also suppose we should remove

Once this lands in Miri and Miri has been updated in rustc, that can be removed, yes. One step after the other. :)

[pvdrz]

I just added a new match under // math functions based on the description for ldexp

Hm. I guess it is a reasonable start. Maybe add a FIXME saying that if we see imprecise results, we should try to use the C math lib ldexp instead.

I can use the C math library directly if you think that's a more definitive solution.

I also suppose we should remove

Once this lands in Miri and Miri has been updated in rustc, that can be removed, yes. One step after the other. :)

Ok, then I will do the miri PR first and wait :)

[RalfJung]

I can use the C math library directly if you think that's a more definitive solution.

That is more in line with what we do with the other functions. I just don't know if it will work on all platforms.

[RalfJung]

Seems like ldexp is part of the normal C math lib so we can rely on it being present in Miri. So @christianpoveda if you want to switch it to import the function via extern instead, be my guest. :)

[RalfJung]

I am seeing test failures when turning on the libcore tests that use ldexp, so maybe we actually need the higher precision.

[pvdrz]

I'll fix this ASAP :)