Enum DType 

pub enum DType {
    F32,
    F16,
    BF16,
    F64,
    I8,
    I16,
    I32,
    I64,
    U8,
    U32,
    Bool,
    C64,
}

Scalar element type. Matches hardware-supported types.

Variants

F32

F16

BF16

F64

I8

I16

I32

I64

U8

U32

Bool

C64

Complex with f32 real and f32 imaginary components, stored interleaved as [re, im, re, im, ...]. 8 bytes per complex element. Element-wise ops (Add/Sub/Mul/Conj) follow the standard complex algebra. Reverse-mode AD on this dtype is not yet wired — Wirtinger conventions (∂/∂z vs ∂/∂z̄) belong to a separate pass that knows to emit conjugate-aware VJPs. The forward path is sufficient for AC analysis and FFT-based workflows that don’t need to differentiate through complex math (and in fact, FFT today already encodes complex as 2N-real-block; this dtype is the natural successor).

Implementations

impl DType

pub const fn size_bytes(self) -> usize

Size in bytes of one element.

pub const fn is_float(self) -> bool

pub const fn is_complex(self) -> bool

True for complex-valued dtypes. Complex elementwise ops follow standard complex algebra, distinct from the float real/imag components (e.g. complex multiply ≠ paired-real multiply).

pub const fn is_int(self) -> bool

pub const fn promotion_rank(self) -> u8

Promotion rank — higher means “wider, more expressive”. The promoted dtype of a binary op is max(rank(lhs), rank(rhs)). Borrowed from MAX’s dtype_promotion.py pattern (#55 in PLAN.md): one module owns the table; ops query it instead of re-implementing ad-hoc rules.

Ranks (low → high): 0 = Bool, 1 = U8/I8, 2 = I16/BF16, 3 = F16, 4 = U32/I32, 5 = I64, 6 = F32, 7 = F64. Floats outrank ints of the same width (matches PyTorch / NumPy). BF16 promotes to F32 against F16 since BF16 has wider range but F16 has more mantissa.

pub fn promote(self, other: DType) -> DType

Result dtype for a binary op between self and other. Mixed int+float → float at least as wide as either input. f16 + bf16 → f32 (no clean lossless target).

Trait Implementations

impl Clone for DType

fn clone(&self) -> DType

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DType

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for DType

fn deserialize<__D>( __deserializer: __D, ) -> Result<DType, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for DType

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

Formats the value using the given formatter.

impl Hash for DType

fn hash<__H>(&self, state: &mut __H)

where __H: Hasher,

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for DType

fn eq(&self, other: &DType) -> 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 DType

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for DType

impl Eq for DType

impl StructuralPartialEq for DType