user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Use the trifecta div algorithm for 128-bit div on wasm

Browse source 
This commit updates the `#[cfg]` annotations used to select the
implementation of 128-bit division in compiler-builtins on wasm targets.
This is done with relation to
https://github.com/WebAssembly/128-bit-arithmetic where performance of
128-bit operations is being investigated on WebAssembly. While I don't
know much about the particulars of the two algorithms involved here the
comments indicate that the "trifecta" variant is preferred if possible
but it's not selected on 32-bit architectures. This rationale isn't as
applicable to WebAssembly targets because despite the 32-bit pointer
width there are often wider-than-pointer operations available as it's
typically run on 64-bit machines.

Locally in testing a benchmark that performs division with a Rust-based
bignum libraries whent from 350% slower-than-native to 220%
slower-than-native with this change, a nice increase in speed. While
this was tested with Wasmtime other runtimes are likely to see an
improvement as well.
This commit is contained in:
Alex Crichton 2024-09-05 09:03:28 -07:00
parent 69406c118a
commit 625462202e
1 changed files with 15 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

20
library/compiler-builtins/src/int/specialized_div_rem/mod.rs
View file

@ -136,9 +136,15 @@ fn u64_by_u64_div_rem(duo: u64, div: u64) -> (u64, u64) {
// Whether `trifecta` or `delegate` is faster for 128 bit division depends on the speed at which a
// microarchitecture can multiply and divide. We decide to be optimistic and assume `trifecta` is
// faster if the target pointer width is at least 64.
// faster if the target pointer width is at least 64. Note that this
// implementation is additionally included on WebAssembly despite the typical
// pointer width there being 32 because it's typically run on a 64-bit machine
// that has access to faster 64-bit operations.
#[cfg(all(
not(any(target_pointer_width = "16", target_pointer_width = "32")),
any(
target_family = "wasm",
not(any(target_pointer_width = "16", target_pointer_width = "32")),
),
not(all(not(feature = "no-asm"), target_arch = "x86_64")),
not(any(target_arch = "sparc", target_arch = "sparc64"))
))]
@ -152,10 +158,14 @@ impl_trifecta!(
u128
);
// If the pointer width less than 64, then the target architecture almost certainly does not have
// the fast 64 to 128 bit widening multiplication needed for `trifecta` to be faster.
// If the pointer width less than 64 and this isn't wasm, then the target
// architecture almost certainly does not have the fast 64 to 128 bit widening
// multiplication needed for `trifecta` to be faster.
#[cfg(all(
any(target_pointer_width = "16", target_pointer_width = "32"),
not(any(
target_family = "wasm",
not(any(target_pointer_width = "16", target_pointer_width = "32")),
)),
not(all(not(feature = "no-asm"), target_arch = "x86_64")),
not(any(target_arch = "sparc", target_arch = "sparc64"))
))]