read dependent chain

One active invocation follows one dependent chain for 4,096 hops through a seeded single-cycle permutation of 1,048,576 u32 entries (4 MiB). Every address comes from the preceding load. Reports hops/s; reciprocal rate is amortized ns/hop, including loop/dispatch cost, not raw memory latency. Repeated batches reuse the same chain.

Measured in hop/s. Higher values mean better performance. Benchmark your device.