These recent Rust integer methods semantically map cleanly onto OpShiftLeftLogical/OpShiftRightLogical.
This is in contrast to normal Rust shifts, whose semantic is such that shifting by the amount that is equal to or larger than number of bits in an integer is UB:
error: this arithmetic operation will overflow
--> src/main.rs:3:20
|
3 | println!("{}", u32::MAX >> 32);
| ^^^^^^^^^^^^^^ attempt to shift right by `32_i32`, which would overflow
|
= note: `#[deny(arithmetic_overflow)]` on by default
My understanding is that rust-gpu doesn't map these methods to those instructions yet, probably resulting in use of conditional logic from standard library, which is less efficient, at least without additional optimizations:
impl u32 {
pub const fn unbounded_shr(self, rhs: u32) -> u32 {
if rhs < Self::BITS {
unsafe { self.unchecked_shr(rhs) }
} else {
0
}
}
}
These recent Rust integer methods semantically map cleanly onto
OpShiftLeftLogical/OpShiftRightLogical.This is in contrast to normal Rust shifts, whose semantic is such that shifting by the amount that is equal to or larger than number of bits in an integer is UB:
My understanding is that
rust-gpudoesn't map these methods to those instructions yet, probably resulting in use of conditional logic from standard library, which is less efficient, at least without additional optimizations: