code wiki / (root) / nx_bench_multiseed.nx

nx_bench_multiseed.nx

buildroot/runtime/nx_bench_multiseed.nx

4196 B157 linesdepth 6pulls 12 transitivereach 0 importersview sourcekind benchtopic bench

docs dependencies structs consts functions

about

bench_multiseed.nx -- 20-workload accuracy averaging for OUR HLL + CPC. Today's single-sample verdict (HLL: LOSES) was an artifact of variance, not algorithm. Both ours and DS got estimates within their declared rel-stddev bound; DS happened to land closer to truth on seed 42. This harness varies the WORKLOAD across 20 distinct streams (each of N unique keys), runs both primitives on every stream, and reports mean error. Combined with the same workloads run through DS Python (see bench_vs_ds_multiseed.py), we get a statistically fair accuracy verdict.

dependencies 3 imports · 0 importers

imports: nx_syscalls.nxnx_sketch_hll.nxnx_sketch_cpc_dense.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

21	const K_MAGIC_2000: i64 = 2000
22	const K_MAGIC_100000: i64 = 100000

functions

24	func ms_iabs(x: i64) -> i64 called by 1: main
31	func ms_putc(c: i64) -> i64 called by 2: ms_i64 ms_nl calls 2: sys_mmap sys_write
38	func ms_str(s: *u8, len: i64) -> i64 called by 1: main calls 1: sys_write
43	func ms_i64(n: i64) -> i64 called by 2: ms_i64 main calls 3: ms_putc ms_i64 sys_mmap
67	func ms_nl() -> i64 called by 1: main calls 1: ms_putc
77	func main() -> i64 calls 11: sys_mmap nx_hll_alloc nx_cpcd_alloc nx_hll_add nx_cpcd_add nx_hll_estimate+5