ComputeFn to hold own registry

[gatesn]

Compute functions now hold their own kernel registry, allowing third-parties to register kernels.

[codspeed-hq]

CodSpeed Performance Report

Merging #3139 will degrade performances by 10.5%

_{Comparing ngates/compute-fn-registry (c24e141) with develop (996d5ee)}

Summary

⚡ 21 improvements
❌ 1 regressions
✅ 789 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	decompress_bitpacking_early_filter[i16, 0.03]	241.7 µs	201.7 µs	+19.86%
⚡	decompress_bitpacking_early_filter[i16, 0.04]	259.1 µs	219.1 µs	+18.28%
⚡	decompress_bitpacking_early_filter[i16, 0.05]	277.7 µs	237.7 µs	+16.83%
⚡	decompress_bitpacking_late_filter[i16, 0.0105]	199.4 µs	157.5 µs	+26.63%
⚡	decompress_bitpacking_late_filter[i16, 0.01]	198.5 µs	156.5 µs	+26.82%
⚡	decompress_bitpacking_late_filter[i16, 0.02]	218.8 µs	176.9 µs	+23.72%
⚡	decompress_bitpacking_late_filter[i16, 0.03]	237.4 µs	195.5 µs	+21.45%
⚡	decompress_bitpacking_late_filter[i16, 0.04]	255.6 µs	213.6 µs	+19.64%
⚡	decompress_bitpacking_late_filter[i16, 0.05]	273.7 µs	231.8 µs	+18.08%
⚡	decompress_bitpacking_late_filter[i32, 0.0105]	299 µs	211.7 µs	+41.21%
⚡	decompress_bitpacking_late_filter[i32, 0.01]	280.8 µs	193.6 µs	+45.05%
⚡	decompress_bitpacking_late_filter[i32, 0.02]	319.4 µs	232.2 µs	+37.55%
⚡	decompress_bitpacking_late_filter[i32, 0.03]	339.2 µs	252 µs	+34.61%
⚡	decompress_bitpacking_late_filter[i32, 0.04]	358.6 µs	271.4 µs	+32.13%
⚡	decompress_bitpacking_late_filter[i32, 0.05]	378 µs	290.8 µs	+30%
⚡	decompress_bitpacking_late_filter[i64, 0.0105]	482.2 µs	304.3 µs	+58.45%
⚡	decompress_bitpacking_late_filter[i64, 0.01]	480.6 µs	302.7 µs	+58.75%
⚡	decompress_bitpacking_late_filter[i64, 0.02]	504.1 µs	326.2 µs	+54.53%
⚡	decompress_bitpacking_late_filter[i64, 0.03]	526.4 µs	348.6 µs	+51.02%
⚡	decompress_bitpacking_late_filter[i64, 0.04]	564.6 µs	386.7 µs	+46%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.