uber / h3-go / 26835208759

100%

push

github

perf: reimplement bit-manipulation functions in pure Go (#120)

Replace CGo calls with pure Go bit operations for Resolution,
BaseCellNumber, IsResClassIII, IsPentagon, IsValid (Cell), and
IsValidIndex. This eliminates the ~18ns CGo crossing overhead per call,
yielding 10-63x speedups on these hot-path functions.

Before (CGo):

```lang=bash
BenchmarkResolution-16        	67765814	        17.77 ns/op	       0 B/op	       0 allocs/op
BenchmarkBaseCellNumber-16    	69139195	        17.34 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsValid-16           	63857523	        19.07 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsPentagon-16        	64660398	        18.76 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsResClassIII-16     	66036112	        18.23 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsValidIndex-16      	63598622	        19.11 ns/op	       0 B/op	       0 allocs/op
```

After (pure Go):

```lang=bash
BenchmarkResolution-16        	1000000000	         0.2806 ns/op	       0 B/op	       0 allocs/op
BenchmarkBaseCellNumber-16    	1000000000	         0.2868 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsValid-16           	600393506	         2.014 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsPentagon-16        	1000000000	         0.6200 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsResClassIII-16     	1000000000	         0.3155 ns/op	       0 B/op	       0 allocs/op
BenchmarkIsValidIndex-16      	492134308	         2.430 ns/op	       0 B/op	       0 allocs/op
```

benchstat (n=10):

```lang=bash
                  │   old (CGo)   │            new (pure Go)             │
                  │    sec/op     │    sec/op     vs base                │
Resolution-16       17.5300n ± 2%   0.2801n ± 1%  -98.40% (p=0.000 n=10)
BaseCellNumber-16   17.4500n ± 1%   0.2874n ± 2%  -98.35% (p=0.000 n=10)
IsValid-16           19.070n ± 1%    1.998n ± 3%  -89.52% (p=0.000 n=10)
IsPentagon-16       18.7600n ± 1%   0.6010n ± 8%  -96.80% (p=0.... (continued)

84 of 84 new or added lines in 1 file covered. (100.0%)

865 of 865 relevant lines covered (100.0%)

286.02 hits per line