search

dgryski / go-farm

go-farm: a pure-Go farmhash implementation
MIT License
238 stars 22 forks source link

Adding benchmark for comparing performance against cgo bindings. #19

Closed panmari closed 4 years ago

panmari commented 4 years ago

Results indicate that the overhead of calling cgo dominates for short sizes. But for large sizes, the cgo implementation comes ahead.

Benchmarks for Intel(R) Core(TM) i7-5500U CPU @ 2.40GHz:

BenchmarkHash32/16-4 76447575 15.4 ns/op 1040.68 MB/s BenchmarkHash32/32-4 55358246 19.9 ns/op 1609.59 MB/s BenchmarkHash32/40-4 55792857 20.1 ns/op 1994.27 MB/s BenchmarkHash32/60-4 43029861 25.4 ns/op 2363.48 MB/s BenchmarkHash32/64-4 38531652 31.4 ns/op 2037.20 MB/s BenchmarkHash32/72-4 34881678 30.6 ns/op 2353.69 MB/s BenchmarkHash32/80-4 39112774 31.0 ns/op 2578.68 MB/s BenchmarkHash32/100-4 31341204 35.3 ns/op 2835.98 MB/s BenchmarkHash32/150-4 24348016 49.3 ns/op 3040.51 MB/s BenchmarkHash32/200-4 20150414 59.8 ns/op 3345.89 MB/s BenchmarkHash32/250-4 16181864 73.4 ns/op 3407.31 MB/s BenchmarkHash32/512-4 7803576 147 ns/op 3481.27 MB/s BenchmarkHash32/1024-4 4217576 267 ns/op 3832.27 MB/s BenchmarkHash32/8192-4 593901 1942 ns/op 4217.89 MB/s BenchmarkCGoFarmhash32/16-4 13318178 95.1 ns/op 168.20 MB/s BenchmarkCGoFarmhash32/32-4 12028802 93.9 ns/op 340.91 MB/s BenchmarkCGoFarmhash32/40-4 12774854 92.5 ns/op 432.26 MB/s BenchmarkCGoFarmhash32/60-4 12280821 103 ns/op 581.35 MB/s BenchmarkCGoFarmhash32/64-4 10820432 98.1 ns/op 652.37 MB/s BenchmarkCGoFarmhash32/72-4 11496954 100 ns/op 718.33 MB/s BenchmarkCGoFarmhash32/80-4 11123570 98.9 ns/op 808.52 MB/s BenchmarkCGoFarmhash32/100-4 10972196 106 ns/op 942.93 MB/s BenchmarkCGoFarmhash32/150-4 9696628 120 ns/op 1251.03 MB/s BenchmarkCGoFarmhash32/200-4 8780136 128 ns/op 1563.83 MB/s BenchmarkCGoFarmhash32/250-4 7914958 144 ns/op 1732.65 MB/s BenchmarkCGoFarmhash32/512-4 5895634 205 ns/op 2503.09 MB/s BenchmarkCGoFarmhash32/1024-4 3665587 315 ns/op 3248.42 MB/s BenchmarkCGoFarmhash32/8192-4 602181 1915 ns/op 4278.25 MB/s BenchmarkHash64/16-4 121190888 9.89 ns/op BenchmarkHash64/32-4 96294661 11.4 ns/op BenchmarkHash64/40-4 62262714 17.5 ns/op BenchmarkHash64/60-4 62453169 16.1 ns/op BenchmarkHash64/64-4 65704662 16.4 ns/op BenchmarkHash64/72-4 35538782 31.2 ns/op BenchmarkHash64/80-4 33494071 31.2 ns/op BenchmarkHash64/100-4 21751887 50.8 ns/op BenchmarkHash64/150-4 17689668 65.8 ns/op BenchmarkHash64/200-4 14182575 82.0 ns/op BenchmarkHash64/250-4 14387606 80.4 ns/op BenchmarkHash64/512-4 12784395 89.8 ns/op BenchmarkHash64/1024-4 7730895 151 ns/op BenchmarkHash64/8192-4 1000000 1031 ns/op BenchmarkCGoFarmhash64/16-4 13303598 83.6 ns/op BenchmarkCGoFarmhash64/32-4 13203582 84.9 ns/op BenchmarkCGoFarmhash64/40-4 12852262 86.7 ns/op BenchmarkCGoFarmhash64/60-4 13560609 85.6 ns/op BenchmarkCGoFarmhash64/64-4 13557555 86.2 ns/op BenchmarkCGoFarmhash64/72-4 12916797 89.3 ns/op BenchmarkCGoFarmhash64/80-4 12254312 89.1 ns/op BenchmarkCGoFarmhash64/100-4 11918374 97.4 ns/op BenchmarkCGoFarmhash64/150-4 11218617 103 ns/op BenchmarkCGoFarmhash64/200-4 9964280 113 ns/op BenchmarkCGoFarmhash64/250-4 10745155 115 ns/op BenchmarkCGoFarmhash64/512-4 9192811 135 ns/op BenchmarkCGoFarmhash64/1024-4 7037410 165 ns/op BenchmarkCGoFarmhash64/8192-4 1698025 697 ns/op

dgryski commented 4 years ago

I'd think I'd prefer this benchmark to be against go-farmhash, since I don't want to break the tests for everybody who doesn't have the C library installed.

Alternately, these benchmarks could be isolated in their own binary or behind a build tag.

panmari commented 4 years ago

Ok, I sent https://github.com/dgryski/go-farmhash/pull/1