Compile grounded subpatterns into hash-cons lookups

[jonathanvdc]

When a subpattern is fully bound, matching still enumerates candidate nodes and compares their children. These changes compile subtrees without explicit slots into a Lookup instruction instead: for f(g(x), h(x)), once x is bound through g, look up h(x) and compare its e-class with the expected call.

Calls carrying slots retain structural matching. Incremental saturation recompiles standard lookup tapes without the optimization so its per-node top-k checks remain intact; custom tapes containing lookups must disable the optimization before instrumentation.

Regression tests for are included for equivalent matches, missing nodes, wrong classes, unions, register layout, early termination, failure reporting, and slotted fallback. The branching test reduces candidate visits from 102 to 2 plus one lookup.

[github-actions]

JMH comparison (baseline vs PR)

Benchmark	Params	Baseline	PR	Δ (PR/Base)	Unit
IncrementalBenchmarks.incrementalPolynomial	depth=6, mutableEGraph=false, size=500, threadCount=1	195.73	194.086	0.992× (-0.8%)	ms/op
IncrementalBenchmarks.incrementalPolynomial	depth=6, mutableEGraph=false, size=500, threadCount=2	327.199	321.847	0.984× (-1.6%)	ms/op
IncrementalBenchmarks.incrementalPolynomial	depth=6, mutableEGraph=true, size=500, threadCount=1	8.47849	8.2592	0.974× (-2.6%)	ms/op
IncrementalBenchmarks.incrementalPolynomial	depth=6, mutableEGraph=true, size=500, threadCount=2	83.3866	83.8232	1.005× (+0.5%)	ms/op
IncrementalBenchmarks.oneByOnePolynomial	depth=6, mutableEGraph=false, size=500, threadCount=1	334.005	335.639	1.005× (+0.5%)	ms/op
IncrementalBenchmarks.oneByOnePolynomial	depth=6, mutableEGraph=false, size=500, threadCount=2	536.224	539.865	1.007× (+0.7%)	ms/op
IncrementalBenchmarks.oneByOnePolynomial	depth=6, mutableEGraph=true, size=500, threadCount=1	141.668	140.477	0.992× (-0.8%)	ms/op
IncrementalBenchmarks.oneByOnePolynomial	depth=6, mutableEGraph=true, size=500, threadCount=2	320.112	327.592	1.023× (+2.3%)	ms/op
LiarBenchmarks.findGemmInMm	threadCount=1	224.73	229.052	1.019× (+1.9%)	ms/op
LiarBenchmarks.findGemmInMm	threadCount=2	178.111	185.292	1.040× (+4.0%)	ms/op
LiarBenchmarks.findGemvInMv	threadCount=1	60.8814	62.3025	1.023× (+2.3%)	ms/op
LiarBenchmarks.findGemvInMv	threadCount=2	55.5764	56.4912	1.016× (+1.6%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=20, threadCount=1	4.66645	4.38156	0.939× (-6.1%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=20, threadCount=2	4.45799	4.34554	0.975× (-2.5%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=40, threadCount=1	30.3249	30.6921	1.012× (+1.2%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=40, threadCount=2	23.7092	23.9793	1.011× (+1.1%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=80, threadCount=1	259.714	241.279	0.929× (-7.1%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=false, size=80, threadCount=2	159.27	159.849	1.004× (+0.4%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=20, threadCount=1	2.83062	2.80469	0.991× (-0.9%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=20, threadCount=2	2.91578	2.98821	1.025× (+2.5%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=40, threadCount=1	20.059	20.6714	1.031× (+3.1%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=40, threadCount=2	15.518	15.6354	1.008× (+0.8%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=80, threadCount=1	178.398	166.808	0.935× (-6.5%)	ms/op
MatmulBenchmarks.nmm	mutableEGraph=true, size=80, threadCount=2	107.153	105.605	0.986× (-1.4%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=false, size=5, threadCount=1	88.0551	87.2429	0.991× (-0.9%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=false, size=5, threadCount=2	81.6078	82.9432	1.016× (+1.6%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=false, size=6, threadCount=1	403.005	393.677	0.977× (-2.3%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=false, size=6, threadCount=2	363.628	365.187	1.004× (+0.4%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=true, size=5, threadCount=1	33.9535	34.2297	1.008× (+0.8%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=true, size=5, threadCount=2	28.7524	28.8088	1.002× (+0.2%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=true, size=6, threadCount=1	140.399	141.228	1.006× (+0.6%)	ms/op
PolyBenchmarks.polynomial	mutableEGraph=true, size=6, threadCount=2	113.263	114.432	1.010× (+1.0%)	ms/op
VectorBenchmarks.blinnPhong	mutableEGraph=false, threadCount=1	33545.1	34034.8	1.015× (+1.5%)	ms/op
VectorBenchmarks.blinnPhong	mutableEGraph=false, threadCount=2	32738.1	32271.7	0.986× (-1.4%)	ms/op
VectorBenchmarks.blinnPhong	mutableEGraph=true, threadCount=1	4523.06	4127.85	0.913× (-8.7%)	ms/op
VectorBenchmarks.blinnPhong	mutableEGraph=true, threadCount=2	3269.78	3217.17	0.984× (-1.6%)	ms/op
VectorBenchmarks.gramSchmidt	mutableEGraph=false, threadCount=1	482.308	488.974	1.014× (+1.4%)	ms/op
VectorBenchmarks.gramSchmidt	mutableEGraph=false, threadCount=2	389.556	388.145	0.996× (-0.4%)	ms/op
VectorBenchmarks.gramSchmidt	mutableEGraph=true, threadCount=1	178.226	180.954	1.015× (+1.5%)	ms/op
VectorBenchmarks.gramSchmidt	mutableEGraph=true, threadCount=2	129.487	130.812	1.010× (+1.0%)	ms/op
VectorBenchmarks.reflection	mutableEGraph=false, threadCount=1	297.055	293.583	0.988× (-1.2%)	ms/op
VectorBenchmarks.reflection	mutableEGraph=false, threadCount=2	243.946	241.548	0.990× (-1.0%)	ms/op
VectorBenchmarks.reflection	mutableEGraph=true, threadCount=1	123.578	125.997	1.020× (+2.0%)	ms/op
VectorBenchmarks.reflection	mutableEGraph=true, threadCount=2	85.0513	86.4415	1.016× (+1.6%)	ms/op
VectorBenchmarks.vectorNormalization	mutableEGraph=false, threadCount=1	3.76947	3.82906	1.016× (+1.6%)	ms/op
VectorBenchmarks.vectorNormalization	mutableEGraph=false, threadCount=2	4.49707	4.49401	0.999× (-0.1%)	ms/op
VectorBenchmarks.vectorNormalization	mutableEGraph=true, threadCount=1	1.69417	1.75162	1.034× (+3.4%)	ms/op
VectorBenchmarks.vectorNormalization	mutableEGraph=true, threadCount=2	2.41401	2.44017	1.011× (+1.1%)	ms/op
Geomean	threadCount=1, mutableEGraph=false	—	—	0.993× (-0.7%)
Geomean	threadCount=1, mutableEGraph=true	—	—	0.992× (-0.8%)
Geomean	threadCount=2, mutableEGraph=false	—	—	1.002× (+0.2%)
Geomean	threadCount=2, mutableEGraph=true	—	—	1.007× (+0.7%)

Note: < 1.0× means faster on PR; > 1.0× means slower.