NgoJuni 2, u-Eric Bigers (unjiniyela kwa-Google) wethula ukusetshenziswa okusha kwe-AES-XTS ukuze kusebenze okuphezulu kakhulu kumaphrosesa we-Intel/AMD esebenzisa amasethi amakhodi amasha AES-NI + AVX, VAES + AVX2, VAES + AVX10/256 kanye ne-VAES + I-AVX10/512 .
Ikhodi ye-AES-NI GCM ithatha indawo yangaphambili ebhalwe yi-Intel futhi ixazulula izinkinga ezikhona ezihlonziwe:
“Le sethi yokuvuselela yengeza ukuqaliswa kwe-AES-GCM (imodi ye-Galois/Counter) ku-VAES ne-AVX512/AVX10, okwandisa ukusebenza kwe-AES-GCM kufika ku-162%. Ukwengeza, ingena esikhundleni sekhodi ye-Intel ye-AES-NI GCM endala ngekhodi entsha. ikhodi eshesha kancane futhi elungisa izinkinga ezimbalwa ezihlanganisa usayizi wefayela kanambambili omkhulu ongaphezu kuka-250KB. Bona amapheshana ukuze uthole imininingwane.
Isimo sokugcina sekhodi yomhlangano ye-x86_64 AES-GCM ukuthi sigcina sinamafayela omhlangano amabili: eyodwa ikhiqiza ikhodi ye-AES-NI nge-AVX noma ngaphandle kwayo, kanti enye ikhiqiza ikhodi ye-VAES nge-AVX512/AVX10 ene-256-bit noma i-512-bit. ama-vectors. Ayinakho ukusekelwa kwe-VAES eyodwa (ayikho i-AVX512/AVX10). Lokhu kuhluke kancane kunalokho engikwenzile nge-AES-XTS, lapho ifayela elilodwa likhiqiza ikhodi ye-AVX ne-AVX512/AVX10, okuhlanganisa nekhodi esebenzisa i-VAES kuphela (ayikho i-AVX512/AVX10), kanti elinye ifayela likhiqiza ikhodi engeyona eye-AVX kuphela. Kodwa-ke, okwamanje lokhu kubonakala kuyisinqumo esifanele ku-algorithm ethile ngayinye, uma kubhekwa ukuthi ukulinganiselwa kumarejista ayi-16 e-SIMD nama-vector angu-128-bit kuholele ezinqumweni zokuklama ezihluke kakhulu ku-AES-GCM, kodwa hhayi kangako ku-AES-XTC. . Amaprosesa athumela nge-VAES kuphela nawo abonakala eyinto yesikhashana, ngakho-ke singase singafuni ukuzama kanzima ukusekela le nhlanganisela.”
Source: linux.org.ru
