miri: implement square root without relying on host floats #4026
Conversation
|
☔ The latest upstream changes (presumably #4029) made this pull request unmergeable. Please resolve the merge conflicts. |
eduardosm
force-pushed
the
soft-sqrt
branch
2 times, most recently
from
bde6ebb to
a8f30d7
Compare
|
@rustbot author |
rustbot
added
the
S-waiting-on-author
|
I added some comments, I hope it is more clear now. @rustbot ready |
rustbot
added
S-waiting-on-review
eduardosm
force-pushed
the
soft-sqrt
branch
2 times, most recently
from
16230fb to
726efe0
Compare
|
Thanks a lot, great to finally have a proper implementation of the last missing standard IEEE float operation. :) |
| (val * (F::from_u128(1).value + err).value).value | ||
| } | ||
|
|
||
| pub(crate) fn sqrt<S: rustc_apfloat::ieee::Semantics>(x: IeeeFloat<S>) -> IeeeFloat<S> { |
There was a problem hiding this comment.
This function is specified in 754 right? It should probably just go in https://github.com/rust-lang/rustc_apfloat
There was a problem hiding this comment.
Yeah, we have an open issue for that: rust-lang/rustc_apfloat#14.
However, last time we wanted to change rustc_apfloat the answer was that things should land in LLVM first and then rustc_apfloat will pick them up from there. That's a quite tedious process. Not sure if that is still the current policy.
#3969 (comment)