Benchmarks for bls precompiles

[rakita]

With the introduction of arkworks bls backend we now have two sets of libraries. #2316

We need to add benchmarks that would guide us on decision on who to put as a default.

[kevaundray]

Made a PR here: #2327

[kevaundray]

I've ran the benchmarks on both arkworks and blst. See #2327 and #2329

Here is a table showing the difference between the two.

Benchmark	Arkworks	BLST	Change
g1_add	47.6 µs	27.7 µs	+71.8%
g1_msm (size 1)	990.7 µs	713.3 µs	+38.8%
g1_msm (size 128)	50.1 ms	47.0 ms	+6.6%
g1_msm (size 2)	2.0 ms	1.4 ms	+42.9%
g1_msm (size 256)	94.8 ms	89.2 ms	+6.3%
g2_add	63.1 µs	42.2 µs	+49.5%
g2_msm (size 1)	2.6 ms	1.3 ms	+100.0%
g2_msm (size 128)	106.5 ms	88.2 ms	+20.7%
g2_msm (size 2)	5.4 ms	2.7 ms	+100.0%
g2_msm (size 256)	195.3 ms	162.3 ms	+20.3%
map_fp2_to_g2	1.4 ms	0.8604 ms	+62.8%
map_fp_to_g1	490 µs	243.5 µs	+101.4%
pairing (1 pairs)	5.4 ms	4.4 ms	+22.7%
pairing (16 pairs)	36.5 ms	36.5 ms	0.0%
pairing (2 pairs)	7.3 ms	6.5 ms	+12.3%
pairing (8 pairs)	19.6 ms	19.4 ms	+1.0%

• Change indicates the performance increase/decrease to go from blst to arkworks

Adding parallel to arkworks did not seem to change anything here (would need to check if it was already enabled via some other feature)

Since g1_add looks bad, the issue is likely the field element implementation in arkworks not being optimized enough

[rakita]

Okay, so current setup is good? it makes sense to leave blst in default = [] and on no-default-features to use arkworks

[kevaundray]

Okay, so current setup is good? it makes sense to leave blst in default = [] and on no-default-features to use arkworks

yep!

[kevaundray]

Can we close this since we merged #2327 ?