libm: implement accelerated computation of (x << e) % y
#1012
Conversation
libm/src/math/support/int_traits.rs
Outdated
| + core::ops::Add<Output = Self> | ||
| + core::ops::Sub<Output = Self> | ||
| + core::ops::Shl<u32, Output = Self> | ||
| + core::ops::Shr<u32, Output = Self> |
There was a problem hiding this comment.
core::ops -> ops since it's already in scope
There was a problem hiding this comment.
I think this commit should be able to allow the #[allow(dead_code)] on NarrowingDiv
| use crate::support::int_traits::NarrowingDiv; | ||
| use crate::support::{DInt, HInt, Int}; |
There was a problem hiding this comment.
Could you add a module-level doc comment with some of the common names used here? E.g. R, RR (if that's not R*R) r, m, q, xq. I'm unfortunately a bit lost :) (but I don't need to understand it in detail)
There was a problem hiding this comment.
If you've authored this, add /* SPDX-License-Identifier: MIT OR Apache-2.0 */ as well (or only MIT if it's derived, as appropriate)
| } | ||
|
|
||
| #[cfg(test)] | ||
| mod test { |
There was a problem hiding this comment.
Could this spot check another integer size as well? Just using constants
quaternic
force-pushed
the
modular-reduction
branch
from
d78b5da to
c69e44a
Compare
|
I've now rebased and applied the previous feedback.
To make the module easier to follow, I moved the public interface, |
| // The partial remainder is in `[0, 2y)` ... | ||
| let r = m.partial_remainder(); | ||
| // ... so check and correct, and compensate for the earlier shift. | ||
| r.checked_sub(y).unwrap_or(r) >> s |
There was a problem hiding this comment.
That minor tweak:
Previously the right shift was done first, and then the checked_sub was with the original y. Now that's done first with the left-shifted y, and the shift is last.
Builds on #1011, both were split from #1002
The computation is implemented in
linear_mul_reduction.