code wiki / (root) / nx_native_rmsnorm_long_bench.nx

nx_native_rmsnorm_long_bench.nx

buildroot/runtime/nx_native_rmsnorm_long_bench.nx

1217 B47 linesdepth 2pulls 2 transitivereach 0 importersview sourcekind benchtopic native
docsdependenciesstructsconstsfunctions

about

nx_native_rmsnorm_long_bench.nx -- LONG RMSNorm bench for Stabilizer. Original 2048 x 5000 iter version (nx_native_rmsnorm_bench.nx) ran ~0.06s gcc-O0 and the paired Stabilizer found CI 0.65-6.17x (no defensible signal, gcc -O0 noise floor 61% on this workload). Scaling to 4096 x 50000 = 200x ops; target ~3-5s gcc-O0, ~10-15s Nishi. Should clear noise floor decisively.

dependencies 1 imports · 0 importers

nx_syscalls.nx nx_native_rmsnorm_long_bench.n

imports: nx_syscalls.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main sys_mmap nxa_die sys_write sys_exit nxa_lock_take nxa_lock_addr sys_write ↻ nxa_lock_give nxa_lock_addr ↻ nxa_report_overrun sys_write ↻ nxa_dump_printable sys_write ↻ nxa_dump_sizes sys_write ↻

structs

none

consts

10const K_MAGIC_4096: i64 = 4096
11const K_MAGIC_50000: i64 = 50000

functions

13func main() -> i64
calls 1: sys_mmap