nx_native_matmul_long_bench.nx
buildroot/runtime/nx_native_matmul_long_bench.nx
about
nx_native_matmul_long_bench.nx -- LONG matmul bench for Stabilizer signal.
The 256x256 version (nx_native_matmul_bench.nx) ran ~0.1s and the
paired Stabilizer found CI 0.41-10.73x (no defensible signal --
runtime too short relative to WSL+gcc-O0 layout noise floor of
79%). This version scales to 512x512 (= 8x ops, 134M mul-adds).
Target runtime: ~0.5s gcc-O0, ~1.5-3s Nishi -- enough multiples of
timing granularity to escape coarse-grained noise.
Per [[reference-berger-performance-matters-talk-archived]]: runtime
must be much larger than measurement granularity for the noise
floor as a fraction of mean to shrink to defensible levels.
dependencies 1 imports · 0 importers
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
structs
| none |
consts
| none |
functions
| 16 | func main() -> i64 calls 1: sys_mmap |