flowchart TD
    W["64-bit wave or quad op<br/>(e.g. WaveActiveSum on a uint64_t)"] --> Q{"what the op does"}
    Q -->|"moves a value between lanes:<br/>read lane, read first, shuffles,<br/>quad reads, all-equal"| H["split into two 32-bit halves<br/>as_type&lt;uint2&gt;(v)<br/>one Metal shuffle of a uint2"]
    Q -->|"bitwise and / or / xor"| B["halves combine on their own<br/>simd_or(uint2), patch 0017's scans"]
    Q -->|"sum, product, min, max<br/>(reduce or prefix)"| L["spvSubgroupScan64: walk the active lanes,<br/>shuffle each lane's halves,<br/>add / multiply / compare as 64-bit"]
    H --> R["as_type&lt;ulong&gt;(result)"]
    B --> R
    L --> R