nx_eqsat_constfold_bench.nx

nx_eqsat_constfold_bench.nx -- CONST-FOLD-HEAVY race vs egg 0.11.0. THE RACE (engine-only, SAME const-heavy expression both sides): egg target = nxc2/bench/nx_constfold_egg_bench.rs (copy into /tmp/race_egg/examples/), the const-fold ANALYSIS graph build(n): acc = 1; for i in 0..N: leaf=(i%7)+2; acc = acc {+,*,-} leaf (cycle by i%3) so EVERY subtree is a known constant -- the const-fold analysis must evaluate + collapse all N nodes, folding the whole chain to a single Num. egg does this via its ConstantFold Analysis (make/merge/modify); WE do it via make-on-add + a CONST merge through the union chokepoint citing NX_EQSAT_RULE_CONSTFOLD. OUR side, per timed iteration (the SAME end-to-end optimize pipeline egg ran): nx_eqsat_init (fresh e-graph) enable_constfold (egg's ConstantFold analysis ON) build the all-const chain (egg's RecExpr build; fold fires at add-time) nx_eqsat_saturate (egg's Runner::run -- analysis to fixpoint) nx_eqsat_recompute_best (egg's Extractor bottom-up cost) nx_eqsat_extract_best_node(root) (egg's find_best -> a single Num) FAIL LOUD: before timing, one verified iteration asserts the WHOLE chain folded -- root's canonical best_node is a CONST (cost 0). A fast wrong answer (chain not folded) loses, exit nonzero. Output (one line): NX_CONSTFOLD_ENGINE_ONLY N=<n> ITERS=<it> FOLD=<v> TOTAL_US=<t> US_PER_OP_NS=<u> (US_PER_OP_NS = total_us*1000/iters = nanoseconds/op; /1000 for us/op.) SOVEREIGN: no .sh; runs on the pinned NishiLang compiler. license_tier: ORIGINAL