nx_q8_st_bw.nx

buildroot/runtime/nx_q8_st_bw.nx

2349 B65 linesdepth 10pulls 30 transitivereach 0 importersview sourcekind tooltopic q8

docs dependencies structs consts functions

about

nx_q8_st_bw.nx -- SINGLE-THREAD sovereign Q8 matmul bandwidth, to split the 2.7x gap vs gcc-in-WSL2 (25.5 GB/s @16thr) into per-thread (kernel/memory) vs scaling (pool). gcc SINGLE-thread Q8 matmul in WSL2 = 2892 MB/s. If sovereign single-thread ~= 2892 -> the gap is THREADING SCALING (fix pool). If << 2892 -> per-thread kernel/memory. Cold 220MB sweep via the ST path.

dependencies 8 imports · 0 importers

imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_f32.nxnx_f32_cvt.nxnx_thread_pool.nxnx_f32_lazy_weight.nxnx_fmt.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

15	const MK: i64 = 896
16	const NG: i64 = 4864
17	const Q8B: i64 = 34
18	const NB: i64 = 48 // 48 * 4.63MB = 222MB cold

functions

20	func sb_nl() -> i64 { fmt_puts("\n" as *u8); return 0 } called by 1: main calls 1: fmt_puts
21	func sb_lcg(s: i64) -> i64 { var v: i64 = s * 1103515245 + 12345; v = v & 2147483647; return v } called by 2: sb_weight main
22	func sb_weight(seed: i64) -> *NxF32LazyWeight called by 1: main calls 3: sys_mmap sb_lcg nx_f32_lazy_weight_new_q8_0
41	func main() -> i64 calls 10: sys_mmap sb_lcg nx_i32_to_f32 nx_lw_shared_pool sb_weight sys_now_us+4