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

Clarify _mm256_shuffle_epi8 documentation (#990)

Browse source 
The old documentation seemed to suggest that the low 4 bits of bytes in `b` would index into the 32-element array `a`. That's not quite right and you'd need 5 bits to index 32 elements.
This commit is contained in:
jethrogb 2021-01-30 21:06:21 +01:00 • committed by GitHub
parent cae0862f44
commit 402d81b9a3
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

7
library/stdarch/crates/core_arch/src/x86/avx2.rs
View file

@ -2587,14 +2587,13 @@ pub unsafe fn _mm256_sad_epu8(a: __m256i, b: __m256i) -> __m256i {
/// Shuffles bytes from `a` according to the content of `b`.
///
/// The last 4 bits of each byte of `b` are used as addresses into the 32 bytes
/// of `a`.
/// For each of the 128-bit low and high halves of the vectors, the last
/// 4 bits of each byte of `b` are used as addresses into the respective
/// low or high 16 bytes of `a`. That is, the halves are shuffled separately.
///
/// In addition, if the highest significant bit of a byte of `b` is set, the
/// respective destination byte is set to 0.
///
/// The low and high halves of the vectors are shuffled separately.
///
/// Picturing `a` and `b` as `[u8; 32]`, `_mm256_shuffle_epi8` is logically
/// equivalent to:
///