Struct CastConfig 

pub struct CastConfig {
    pub out_dtype: DType,
    pub sf_block: Option<(usize, usize)>,
    pub round_sf: bool,
}

Cast configuration carried by ops that emit a typed output.

Inspired by TileKernels’ CastInputConfig / CastOutputConfig: a single dataclass that flows from the layer down to the kernel selector, so adding new quantized formats (FP8 e4m3, FP4 e2m1, blocked scaling) becomes a matter of populating fields rather than threading new flags through call sites.

Today only out_dtype is consulted by backends — the scaling-factor fields are reserved for future quantization passes (FP8 / blocked SF). Constructed once by the precision pass and embedded in fused ops.

Fields

out_dtype: DType

Destination dtype for the cast (fragment of the output tensor).

sf_block: Option<(usize, usize)>

Scaling factor block size (rows, cols) for blocked quantization. None means no scaling factor (plain cast).

round_sf: bool

Round scaling factors to powers of two (UE8M0 style).

Implementations

impl CastConfig

pub const fn plain(out_dtype: DType) -> CastConfig

Plain dtype cast with no scaling factor.

pub fn is_noop(&self, in_dtype: DType) -> bool

True when the cast does no work (out matches input dtype).

Trait Implementations

impl Clone for CastConfig

fn clone(&self) -> CastConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for CastConfig

impl Debug for CastConfig

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

Formats the value using the given formatter.

impl PartialEq for CastConfig

fn eq(&self, other: &CastConfig) -> 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 StructuralPartialEq for CastConfig