IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /max/get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
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

combine_pack_o_row

def combine_pack_o_row[n: Int, //, output_type: DType](own: Array[Float32, n], peer: Array[Float32, n], scale_own: Float32, scale_peer: Float32) -> SIMD[DType.uint32, (n // Int(2))]

LSE-combine own * scale_own + peer * scale_peer over n f32 O lanes, cast to the 2-byte output_type, and pack into n // 2 u32 lanes.

own/peer are tcgen05_ld results. f32x2 compute / wide-only store, as in scale_pack_o_row; the fused peer.fma(scale_peer, own * scale_own) form matches the per-element combine it replaces. Used by fa4_lse_combine_write.

Returns:

SIMD[DType.uint32, (n // Int(2))]

Was this page helpful?