Struct AvxVnni 

pub struct AvxVnni;

x86/x86_64 AVX-VNNI (256-bit VEX-encoded VNNI) instruction set marker.

Distinct from Avx512: AVX-VNNI provides vpdpbusd/vpdpwssd on 256-bit YMM registers without requiring AVX-512, so it accelerates integer dot/GEMM on client CPUs (Alder Lake and newer) that lack AVX-512. The int8 tile kernel uses the unsigned-signed vpdpbusd with a +128 operand bias correction (base AVX-VNNI has no signed-signed vpdpbssd; that is the separate avxvnniint8 feature).

Trait Implementations

impl Clone for AvxVnni

fn clone(&self) -> AvxVnni

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for AvxVnni

impl Debug for AvxVnni

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Eq for AvxVnni

impl PartialEq for AvxVnni

fn eq(&self, other: &AvxVnni) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl SimdArch for AvxVnni

const NAME: &'static str = "avx_vnni"

Human-readable ISA name ("avx2", "avx512", "neon", "scalar").

const REGISTER_WIDTH_BITS: u32 = 256

Width of vector registers in bits.

const ISA_FAMILY: IsaFamily = hermes_simd_core::arch::IsaFamily::X86

ISA family for this architecture.

const FMA_THROUGHPUT_HINT: u32 = 4

Suggested TILE_M value for tiled_dot to saturate FMA throughput.

fn is_runtime_supported() -> bool

Returns true when the current host may execute this architecture’s native instructions from safe wrappers.

impl StructuralPartialEq for AvxVnni

impl TileMatrixMultiply<I8, I8, I32, AvxVnni, AvxVnni, 16, 16, 64> for AvxVnni

unsafe fn tile_matmul( c: *mut I32, c_stride: usize, a: *const I8, a_stride: usize, b: *const I8, b_stride: usize, )

Performs tile matrix multiplication: C += A * B

impl TileMatrixMultiply<i8, i8, i32, AvxVnni, AvxVnni, 16, 16, 64> for AvxVnni

unsafe fn tile_matmul( c: *mut i32, c_stride: usize, a: *const i8, a_stride: usize, b: *const i8, b_stride: usize, )

Performs tile matrix multiplication: C += A * B