nx_bench_hll_vs_cpcd.nx
buildroot/runtime/nx_bench_hll_vs_cpcd.nx
about
bench_hll_vs_cpcd.nx -- HLL_8 vs CPC DENSE at matched accuracy.
Fixes the v1 memory loss by configuring CPC DENSE with matched-accuracy
parameters instead of matched-m parameters.
MATCHED-ACCURACY CONFIG:
HLL_8: lg_k=7 -> m=128 registers (1 byte each), rel-stddev = 1.04/sqrt(128) = 9.2%
CPC dense: lg_k=3, w=16 -> big_M = 128 coupons, rel-stddev = 1/sqrt(128) = 8.8%
MEMORY:
HLL_8: 24 header + 128 register bytes = 152 bytes
CPC dense: 72 header + 16 bitmap bytes = 88 bytes
EXPECTED VERDICT: MEMORY BEATS at matched accuracy.
dependencies 5 imports · 0 importers
imports: nx_syscalls.nxnx_sketch_hll.nxnx_sketch_cpc_dense.nxnx_sketch_comparator.nxnx_sketch_types.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
| 27 | const K_MAGIC_10000: i64 = 10000 |
functions
| 29 | func main() -> i64 |