For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

hmul2_bf16x8_by_scalar

def hmul2_bf16x8_by_scalar[out_dtype: DType](packed: SIMD[DType.uint32, SIMDLength(4)], scale_bf16: UInt32) -> SIMD[DType.uint32, SIMDLength(4)]

Multiply 8 packed bf16 values (in 4 uint32 registers) by a bf16x2 scalar broadcast.

Parameters:

out_dtype (DType): Element type of the packed bf16 values (bfloat16).

Args:

packed (SIMD[DType.uint32, SIMDLength(4)]): Four uint32 registers holding the eight packed bf16 values to scale.
scale_bf16 (UInt32): Bf16x2 scalar broadcast in a uint32, as produced by e8m0_to_bf16_broadcast.

Returns:

SIMD[DType.uint32, SIMDLength(4)]