diff options
| author | Frank Denis <github@pureftpd.org> | 2020-10-27 13:33:08 +0100 |
|---|---|---|
| committer | Veikka Tuominen <git@vexu.eu> | 2020-10-28 21:44:00 +0200 |
| commit | 0adc144f88b567054e63df808aac0227445e88bb (patch) | |
| tree | c41f8c7c03c26dda25a2757e3f8f1378477587a8 /lib/std | |
| parent | ea45897fcc5097c4cf73a30fe009500f6efe8bc5 (diff) | |
| download | zig-0adc144f88b567054e63df808aac0227445e88bb.tar.gz zig-0adc144f88b567054e63df808aac0227445e88bb.zip | |
std/crypto: adjust aesni parallelism to CPU models
Intel keeps changing the latency & throughput of the aes* and clmul
instructions every time they release a new model.
Adjust `optimal_parallel_blocks` accordingly, keeping 8 as a safe
default for unknown data.
Diffstat (limited to 'lib/std')
| -rw-r--r-- | lib/std/crypto/aes/aesni.zig | 12 |
1 files changed, 11 insertions, 1 deletions
diff --git a/lib/std/crypto/aes/aesni.zig b/lib/std/crypto/aes/aesni.zig index 5f605a4e60..3d694875bf 100644 --- a/lib/std/crypto/aes/aesni.zig +++ b/lib/std/crypto/aes/aesni.zig @@ -100,8 +100,18 @@ pub const Block = struct { /// Perform operations on multiple blocks in parallel. pub const parallel = struct { + const cpu = std.Target.x86.cpu; + /// 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 = switch (std.Target.current.cpu.model) { + &cpu.westmere => 6, + &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, + else => 8, + }; /// 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 { |