Why use it

AES-NI replaces the byte substitutions, row permutations and column arithmetic of an AES round with a register operation. It provides the round primitives needed to build an AES implementation without software lookup tables.

Operations

import native.x86.aes; provides operations on simd<std::uint8_t,16,Arch>. Link native::native; native.x86 and native also export the module. All operands and results share the same tag.

Operation	Result
aesenc<Arch>(state, key)	Encryption round, including MixColumns and final key XOR
aesenclast<Arch>(state, key)	Final encryption round, omitting MixColumns
aesdec<Arch>(state, key)	Decryption round, including inverse MixColumns and final key XOR
aesdeclast<Arch>(state, key)	Final decryption round, omitting inverse MixColumns
aesimc<Arch>(state)	Inverse MixColumns on the four state columns
aeskeygenassist<Arch, Imm8>(state)	Substituted and rotated words with a round constant

State byte 4 * column + row follows the AES matrix's column order. aeskeygenassist uses the second and fourth input words and XORs Imm8 into the rotated outputs. The immediate is a compile-time unsigned byte, 0–255.

Caveats

The x86 round instructions add the key after their transformations. ARM AESE/AESD add it first, so porting a round sequence requires rearranging keys. Intermediate AESDEC rounds use inverse-mixed keys; AESDECLAST uses the original initial key. These functions supply neither a complete key schedule nor a cipher mode.

Runtime calls require AES and SSE2 storage. Use target_features<native::x86>("aes"), compile the caller for that target and admit it with classify_isa before entry. Importing does not enable instructions or dispatch. With AES in Arch, the overloads are constexpr with native runtime paths. An SSE2 storage tag without AES has consteval overloads only. Integer round operations leave floating-point status unchanged.