mips64 crashes or hangs on 128 bit division

[AaronKutch]

This target either crashes or hangs with different arguments to a 128 bit division

#[inline(never)]
fn test(x: u128, y: u128) -> u128 {
    x.wrapping_div(y)
}

fn main() {
    dbg!(test(123 << 64, (32 << 64) | 65));
    //dbg!(test(1 << 96,  1 << 64));
}

this happens with cross run --target mips64-unknown-linux-gnuabi64 -Z build-std=core,alloc,proc_macro,std, but I don't know how to get the version that cross is using, also #115218 .

[saethlin]

Is this a duplicate of #116177? (Are we using shifts to do the division)

[AaronKutch]

yes shifts are used internally, but when I saw that issue I assumed it only applied to the 32 bit version, if 64 bit MIPS is also affected it should be noted

[saethlin]

I ran the reproducer for #116177 and as far as I can tell it succeeds on mips64-unknown-linux-gnuabi64. So I think this is a separate codegen bug?

[AaronKutch]

maybe the reproducer needs to have its widths doubled to a 128 bit shift, a 64 bit shift would need a single instruction on mips64

[saethlin]

I adjusted the reproducer for the other issue a bit into this, which does not pass the assert with your cross invocation:

use std::hint::black_box;

const SHIFTVALUE: i128 = u64::MAX as i128;

fn main() {
    fn shiftleft(a: i128, b: i128) -> i128 {
        a << b
    }

    assert_eq!(
        shiftleft(black_box(SHIFTVALUE), black_box(2)),
        SHIFTVALUE << 2
    );
}

So I now think this is a duplicate of #116177. Do you agree?

[AaronKutch]

yes, this should be closed then and the 64 bit version mentioned in the other thread?