language-servers/roc
1
0
Fork 0
mirror of https://github.com/roc-lang/roc.git synced 2025-08-04 20:28:02 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

update buffer size

Browse source
This commit is contained in:
Brendan Hansknecht 2024-07-23 10:32:39 -07:00
parent 539ab1daef
commit 1d534517b3
No known key found for this signature in database
GPG key ID: 0EA784685083E75B
1 changed files with 9 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

10
crates/compiler/builtins/bitcode/src/sort.zig
View file

@ -14,7 +14,15 @@ const CopyFn = *const fn (Opaque, Opaque) callconv(.C) void;
const IncN = *const fn (?[*]u8, usize) callconv(.C) void;
/// Any size larger than the max element buffer will be sorted indirectly via pointers.
const MAX_ELEMENT_BUFFER_SIZE: usize = 64;
/// TODO: tune this.
/// I did some basic basic testing on my M1 and x86 machines with the c version of fluxsort.
/// The best tradeoff point is not the clearest and heavily depends on machine specifics.
/// Generally speaking, the faster memcpy is and the larger the cache line, the larger this should be.
/// Also, to my surprise, sorting by pointer is more performant on short arrays than long arrays (probably reduces time of final gather to order main array).
/// Anyway, there seems to be a hard cut off were the direct sort cost suddenly gets way larger.
/// In my testing for long arrays, the cutoff seems to be around 96-128 bytes.
/// For sort arrays, the custoff seems to be around 64-96 bytes.
const MAX_ELEMENT_BUFFER_SIZE: usize = 96;
pub fn quadsort(
source_ptr: [*]u8,