Struct ParallelInvokeSlot 

#[repr(C, align(64))]
pub struct ParallelInvokeSlot {
    pub status: AtomicU32,
    pub _pad0: u32,
    pub job_id: AtomicU64,
    pub packet: InvokePacket,
    pub result: InvocationResult,
}

One host<->guest invoke slot. Slots are addressed by lane index at parallel_slot_base + lane * PARALLEL_INVOKE_SLOT_BYTES.

Synchronization protocol:

• host writes job_id and packet, then stores READY with release;
• guest CASes READY -> RUNNING, runs the invoke, writes result, then stores DONE with release;
• host reads DONE with acquire, copies result, then stores EMPTY.

Bench-only control commands, such as EVICT_JIT_READY, use the same RUNNING -> DONE -> EMPTY completion protocol after the host reserves all lanes.

Fields

status: AtomicU32

_pad0: u32

job_id: AtomicU64

packet: InvokePacket

result: InvocationResult