cajeta.math.DType — a numeric dtype descriptor: the runtime value that identifies an element type’s kind, bit width, and signedness. A Tensor<T> carries its dtype in the reified type parameter T; a DType value is what the NEP-50 type-promotion table (promote) and the runtime-dtype paths reason over. Identity is (kind, bits, variant) — variant distinguishes same-width floats (float16 vs bfloat16, the four float8 encodings). The v1 dtype set covers bool, int8..128, uint8..128, float16..128, bfloat16, and the float8/float6/float4 variants; complex64/128 are reserved (the Complex bound exists, but no concrete complex primitive yet).

package snip.dtype;

import cajeta.math.DType;

public final class Demo {
    public static void run() {
        DType f #= DType.of<float32>();     // type → DType bridge
        DType i #= DType.i32();
        DType r #= DType.promote(f, i);     // NEP-50: float32
        boolean fl = r.isFloating();       // true
        int32 width = r.bits();            // 32
        return;
    }
}

Methods

Signature
Descriptor
`DType(int32 kindCode, int32 bitWidth, int32 variantCode)`	Build a descriptor from kind, bit width, and variant codes
`int32 kind()`	Kind code (`KIND_BOOL` / `KIND_INT` / `KIND_UINT` / `KIND_FLOAT` / `KIND_COMPLEX`)
`int32 bits()`	Bit width
`int32 variant()`	Variant code (distinguishes same-width float encodings)
`int32 bytes()`	Bytes per element
`boolean equals(DType other)`	Same `(kind, bits, variant)` identity
Category predicates
`boolean isBool()`	Kind is bool
`boolean isSignedInt()`	Kind is signed int
`boolean isUnsignedInt()`	Kind is unsigned int
`boolean isIntegral()`	Signed or unsigned int
`boolean isFloating()`	Kind is floating
`boolean isComplex()`	Kind is complex (reserved)
`boolean isNumeric()`	Any numeric kind (not bool)
`boolean isSigned()`	Signed in the sign-bit sense: signed ints, floats, complex (not bool, not uint)
Named factories
`static #DType boolType()`	`bool`
`static #DType i8()`	`int8`
`static #DType i16()`	`int16`
`static #DType i32()`	`int32`
`static #DType i64()`	`int64`
`static #DType i128()`	`int128`
`static #DType u8()`	`uint8`
`static #DType u16()`	`uint16`
`static #DType u32()`	`uint32`
`static #DType u64()`	`uint64`
`static #DType u128()`	`uint128`
`static #DType f16()`	`float16`
`static #DType f32()`	`float32`
`static #DType f64()`	`float64`
`static #DType f128()`	`float128`
`static #DType bf16()`	`bfloat16`
`static #DType f8e4m3()`	`float8` e4m3
`static #DType f8e5m2()`	`float8` e5m2
`static #DType f8e4m3fnuz()`	`float8` e4m3fnuz
`static #DType f8e5m2fnuz()`	`float8` e5m2fnuz
`static #DType f6e2m3()`	`float6` e2m3
`static #DType f6e3m2()`	`float6` e3m2
`static #DType f4e2m1()`	`float4` e2m1
Type bridge and promotion
`static int32 codeOf<T>()`	Compiler intrinsic: the packed dtype code of `T`, `(kind << 16) \| (bits << 4) \| variant`
`static #DType of<T>()` ⚑	The `DType` descriptor of the static type parameter `T` — the bridge `Tensor<T>.dtype()` is built on
`static #DType promote(DType a, DType b)` ⚑	NEP-50 type-based promotion: the result dtype of a binary op on operands of dtypes `a` and `b`, from the types (never the values)

⚑ = @EntryPoint

DType

Methods

See also