crypto.core.aes: process 6 block in parallel instead of 8 on aarch64 (#13473)

* crypto.core.aes: process 6 block in parallel instead of 8 on aarch64

At least on Apple Silicon, this is slightly faster than 8 blocks.

* AES: add parallel blocks for tigerlake, rocketlake, alderlake, zen3
This commit is contained in:
Frank Denis 2022-11-07 12:28:37 +01:00 committed by GitHub
parent 907f3ef887
commit 32563e6829
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 3 additions and 3 deletions

View file

@ -101,8 +101,8 @@ pub const Block = struct {
&cpu.sandybridge, &cpu.ivybridge => 8,
&cpu.haswell, &cpu.broadwell => 7,
&cpu.cannonlake, &cpu.skylake, &cpu.skylake_avx512 => 4,
&cpu.icelake_client, &cpu.icelake_server => 6,
&cpu.znver1, &cpu.znver2 => 8,
&cpu.icelake_client, &cpu.icelake_server, &cpu.tigerlake, &cpu.rocketlake, &cpu.alderlake => 6,
&cpu.znver1, &cpu.znver2, &cpu.znver3 => 8,
else => 8,
};

View file

@ -109,7 +109,7 @@ pub const Block = struct {
/// Perform operations on multiple blocks in parallel.
pub const parallel = struct {
/// The recommended number of AES encryption/decryption to perform in parallel for the chosen implementation.
pub const optimal_parallel_blocks = 8;
pub const optimal_parallel_blocks = 6;
/// Encrypt multiple blocks in parallel, each their own round key.
pub inline fn encryptParallel(comptime count: usize, blocks: [count]Block, round_keys: [count]Block) [count]Block {