code wiki / _hdl_build / nx_lwq24_probe.nx

nx_lwq24_probe.nx

buildroot/runtime/_hdl_build/nx_lwq24_probe.nx

5045 B146 linesdepth 10pulls 31 transitivereach 0 importersview sourcekind probe
docsdependenciesstructsconstsfunctions

about

nx_lwq24_probe.nx -- LAYERED probe for the LWC-2 FILL FAIL (2026-08-01). P1: fq4m_fill_q24 + fq4m_rows_cached_q24 SERIALLY (no pool, no lazy-weight) vs the scalar oracle. P2: the full nx_f32_lazy_matmul dispatch on the same data. Whichever layer first diverges owns the bug. Same exact-regime synthesis as nx_lw_cache_gate (d=1.0 f16, dmin=0, sc=1 -> dequant == q4 integers; int activations). expect_exit: 0 license_tier: ORIGINAL No hw writes (Rule 26).

dependencies 2 imports · 0 importers

nx_f32_lazy_weight.nx nx_fmt.nx nx_lwq24_probe.nx

imports: nx_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

main p_fill_a p_lcg nx_i32_to_f32 p_fill_weights p_lcg ↻ p_block_dense p_lcg ↻ nx_f32_q4k_matmul fq4m_rows 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 ↻ _fq4m_pack_a nx_q4k_to_f32_packed nx_q4k_block_to_f32_packed nx_le_read_u16 nx_f16_to_f32 nx_le_read_u8 nx_i32_to_f32 ↻ _q4kp_st4 sys_munmap fmt_puts fmt_puts_fd sys_write ↻ fmt_putn fmt_putn_fd sys_write ↻

structs

none

consts

9const PB_MAGIC_1103515245: i64 = 1103515245
10const PB_MAGIC_12345: i64 = 12345
11const PB_MAGIC_2147483647: i64 = 2147483647
12const PB_MAGIC_20260708: i64 = 20260708
13const PB_MAGIC_424242: i64 = 424242
15const PB_M: i64 = 3
16const PB_K: i64 = 512
17const PB_N: i64 = 37

functions

19func p_lcg(s: i64) -> i64
called by 3: p_block_densep_fill_weightsp_fill_a
24func p_block_dense(buf: *u8, off: i64, seed: i64) -> i64
called by 1: p_fill_weights calls 1: p_lcg
45func p_fill_weights(buf: *u8, n_rows: i64, k: i64, seed: i64) -> i64
called by 1: main calls 2: p_lcgp_block_dense
60func p_fill_a(p: *i64, count: i64, seed: i64, half: i64) -> i64
called by 1: main calls 2: p_lcgnx_i32_to_f32
72func p_diff(a: *i64, b: *i64, count: i64) -> i64
called by 1: main
81func main() -> i64
calls 14: p_fill_ap_fill_weightsnx_f32_q4k_matmulfmt_putsfmt_putnfq4m_fill_q24+8