/* lane s1, row wc2017b1 -- PROBE 1 of the floor decomposition. ZERO engine work.
Deliberately degenerate (WA), permitted by GOAL.md:36 and free under min-keep.
THE QUESTION, AND IT IS THE ONE THE WHOLE ROW TURNS ON: the per-case detail table of our
own no-op (sid 117279) reads
1: 368.64 us / 412 KB
2: 353.067 ms / 390644 KB
3: 705.976 ms / 781272 KB
i.e. a floor of EXACTLY 3.53 ns per element at all three sizes, with a footprint of EXACTLY
the 4n array. Two different harnesses produce that shape and they are not the same harness:
(A) the FLOOR IS GENERATION, charged BEFORE our function is entered;
(B) the FLOOR IS THE CHECK, charged AFTER our function returns.
`agg1`'s AC at 705.987 ms -- equal to the no-op's own time -- is reachable under (B) and
impossible under (A), so this single line decides which world we are in.
HOW THIS LINE DECIDES IT WITHOUT ANY THEORY OF THE HARNESS: `_exit(0)` is called before a
single load of `a`. Everything that is charged before our entry is still charged; nothing
that happens after a real `sort` returns can be. The reported per-case time IS the
pre-charge, per case, at three sizes.
It is also the ONLY probe that can separate the two: `const fill` (781.109 ms) and the no-op
(705.976 ms) both return normally, so both pay both terms and only their SUM is visible.
`_exit` truncates the sum. */
extern "C" void _exit(int);
void sort(unsigned *a, int n) { (void)a; (void)n; _exit(0); }