`load_bytes`

fn def load_bytes[dtype: DType, width: SIMDLength, o: Origin[mut=o.mut]](data: Span[UInt8, o]) -> SIMD[dtype, width]

Reinterpret the first width lanes’ worth of data as a vector.

The counterpart of SIMD.from_bytes, for the case that one needs a fixed-size Array and data does not have one: a span with a runtime length. Reading through the span’s pointer avoids the copy that staging the bytes into an array would cost, which matters in the block-cipher inner loops this exists for.

Byte order is the target’s own — this is a raw memory reinterpretation, so it round-trips with store_bytes but is not endianness-neutral. Use SIMD.from_bytes[endian=...] or load_be where the wire order matters.

Parameters:

dtype (DType): The lane type of the returned vector.
width (SIMDLength): The number of lanes to read.
o (Origin[mut=mut]): The origin of the byte span.

Args:

data (Span[UInt8, o]): The bytes to reinterpret; must hold at least width * size_of[dtype]() of them.

Returns:

SIMD[dtype, width]: The first width * size_of[dtype]() bytes of data as a vector.

load_be store_bytes