Struct TcbGeometry 

pub struct TcbGeometry {
    pub m: u32,
    pub n: u32,
    pub k: u32,
    pub alignment: u32,
}

Dimensions for a Tiling Compute Block

Represents the (M, N, K) dimensions of a tile in matrix operations:

• M: Output rows
• N: Output columns
• K: Reduction dimension (inner product)

Alignment Constraints

Per the TCB-03 pattern (Tile Quantization Alignment), K must align with the quantization superblock size:

• Q4_0: K % 32 == 0
• Q4_K: K % 256 == 0
• Q8_0: K % 32 == 0

Fields

m: u32

Items processed in M dimension (rows)

n: u32

Items processed in N dimension (columns)

k: u32

Reduction dimension (inner product)

alignment: u32

Alignment requirement in bytes (typically 16 for SIMD, 32 for AVX2, 64 for AVX-512)

Implementations

impl TcbGeometry

pub fn new(m: u32, n: u32, k: u32) -> Self

Create a new TCB geometry

Panics

Panics if any dimension is zero.

pub fn with_alignment(m: u32, n: u32, k: u32, alignment: u32) -> Self

Create geometry with explicit alignment

pub fn arithmetic_intensity(&self) -> f32

Calculate arithmetic intensity (FLOPS per byte loaded)

For GEMM: AI = (2 * M * N * K) / (MK + KN) * sizeof(f32)

Higher AI means compute-bound; lower means memory-bound.

pub fn total_elements(&self) -> u64

Calculate total elements in the tile

pub fn total_flops(&self) -> u64

Calculate total FLOPs for this tile

pub fn is_q4k_aligned(&self) -> bool

Check if K dimension aligns with Q4_K superblock (256)

pub fn is_q4_0_aligned(&self) -> bool

Check if K dimension aligns with Q4_0/Q8_0 block (32)

pub fn a_tile_bytes(&self) -> usize

Calculate bytes needed for A tile (M × K × sizeof(f32))

pub fn b_tile_bytes(&self) -> usize

Calculate bytes needed for B tile (K × N × sizeof(f32))

pub fn c_tile_bytes(&self) -> usize

Calculate bytes needed for C tile (M × N × sizeof(f32))

pub fn fits_in_cache(&self, cache_bytes: usize) -> bool

Check if tile fits in given cache size (bytes)

Trait Implementations

impl Clone for TcbGeometry

fn clone(&self) -> TcbGeometry

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TcbGeometry

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

Formats the value using the given formatter.

impl Default for TcbGeometry

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for TcbGeometry

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for TcbGeometry

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

Formats the value using the given formatter.

impl PartialEq for TcbGeometry

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for TcbGeometry

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for TcbGeometry

impl Eq for TcbGeometry

impl StructuralPartialEq for TcbGeometry