code wiki / _hdl_build / nx_js_floatfuse_probe.nx
nx_js_floatfuse_probe.nx
buildroot/runtime/_hdl_build/nx_js_floatfuse_probe.nx
about
nx_js_floatfuse_probe.nx -- SELF-ASSERTING perf-regression GATE for the const-fusion float fallback.
`s=s+1` with `s` a FLOAT is the `float +/- int-literal` pattern: it fuses (PUSH 1 is an int const) but the
LHS is float at runtime -> the fast path misses -> fallback. The fallback MUST use jit_binop's inline SSE
(addsd), NOT the ~10x-slower jit_slow_binop sovereign slow-helper call. This gate catches that regression AUTOMATICALLY (the
correctness suite can't -- it's bit-exact, not timed): the SSE fallback puts the FLOAT loop within ~1.2x of
the pure-INT control (both are ~1 mem-round-trip/iter); the broken sovereign-slow-helper made it ~10x. Assert float < 4x
int (clean separation, robust to the machine's +-25% load noise). This is the gate that would have caught
the 2026-07-14 10x regression. expect_exit: 0 license_tier: ORIGINAL
dependencies 1 imports · 0 importers
imports: nx_js_vm.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
| 10 | func pw(s: *u8) -> i64 { var n: i64 = 0; while s[n] != (0 as u8) { n = n + 1 } sys_write(1, s, n); return 0 } |
| 11 | func pn(v: i64) -> i64 |
| 20 | func timed(src: *u8, name: *u8) -> i64 |
| 35 | func main(argc: i64, argv: *i64) -> i64 |