nx_bench_hll_throughput.nx
buildroot/runtime/nx_bench_hll_throughput.nx
about
nx_bench_hll_throughput.nx -- head-to-head HLL throughput bench.
Per user directive 2026-05-14: "use all of this to beat apache data
sketches, we are slower and worse and we want to prove not only are
we the leader in math but it applies all the way up the chain in
performance".
Drives nx_sketch_hll with a deterministic xorshift64 PRNG key
stream, measures update_count + HLL estimate + monotonic-clock
elapsed; the script computes items/sec from update_count / elapsed.
Uses the proven print_i64 / println / nx_clock_monotonic_ns
helpers; no per-bench reimplementation.
genealogy_id: substrate_perf_bench_2026_05_14
lineage_id: incumbent_displacement_proof
dependencies 5 imports · 0 importers
imports: nx_syscalls.nxnx_runtime.nxnx_tier.nxnx_clock.nxnx_sketch_hll.nx
imported by: nobody (leaf or entry point)
call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown
structs
| none |
consts
| 29 | const NX_MAGIC_1000000000: i64 = 1000000000 |
| 30 | const NX_MAGIC_1000000: i64 = 1000000 |
| 32 | const NX_BENCH_HLL_N: nx_int = 100000 |
| 33 | const NX_BENCH_HLL_LGK: nx_int = 10 |
| 34 | const NX_BENCH_HLL_SEED: nx_int = 1442695040888963407 |
functions
| 36 | func nx_bench_xorshift64(state_p: *nx_int) -> nx_int called by 1: main |
| 45 | func nx_bench_i64_to_bytes(v: nx_int, out: *u8) called by 1: main |
| 56 | func main() -> nx_exit |