keccak: Optimize Keccak with BMI extension

[chfast]

Using BMI and BMI2 instruction set extensions makes Keccak 25% faster.

seed/1                                        -0.2469           377           284
seed/10                                       -0.2487          3765          2829
seed/100                                      -0.2503         37665         28237
seed/1000                                     -0.2485        375719        282340
seed/10000                                    -0.2481       3756911       2824889
light_cache/1                                 -0.1995           452           362
create_context/1                              -0.2063           459           364
create_context/333                            -0.2065          1741          1382
ethash_calculate_dataset_item_512             -0.0511          4065          3857
ethash_calculate_dataset_item_1024            -0.0765          5127          4735
ethash_calculate_dataset_item_2048            -0.1976         10539          8456
ethash_hash/0                                 -0.0276          1160          1128
ethash_hash/10                                -0.0153          1453          1430
verify                                        -0.0250        519861        506889
keccakf1600                                   +0.0008           368           368
keccakf1600_bmi                               -0.0004           276           276
keccakf1600_best                              -0.0000           276           276
keccak256/32                                  -0.2469           377           284
keccak256/64                                  -0.2509           381           285
keccak256/128                                 -0.2498           382           286
keccak256/135                                 -0.2437           382           289
keccak256/136                                 -0.2492           747           561
keccak512/32                                  -0.2523           382           285
keccak512/64                                  -0.2560           386           287
keccak512/71                                  -0.2564           389           289
keccak512/72                                  -0.2492           747           561
keccak512/142                                 -0.2459           752           567
keccak512/143                                 -0.2461           752           567
keccak512/144                                 -0.2490          1116           838

[codecov-io]

Codecov Report

Merging #162 (1e0b5b2) into master (17d8e1c) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master      #162   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           22        22           
  Lines         1643      1652    +9     
=========================================
+ Hits          1643      1652    +9     

Flag	Coverage Δ
be	89.32% <100.00%> (+0.01%)	⬆️
default	99.81% <75.00%> (-0.19%)	⬇️
x86_64	89.06% <66.66%> (-0.19%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
lib/keccak/keccakf1600.c	100.00% <100.00%> (ø)

[gumb0]

Maybe instead of exposing global variable it would look simpler to have another wrapper function like ethash_keccakf1600_best or something

[gumb0]

Or just ethash_keccakf1600 that would call ethash_keccakf1600_best function pointer inside

[chfast]

I considered this, but that mean call + indirect call instead of just indirect call.
Anyway, don't want to expose keccakf1600 any more. It is convenient for benchmarking only so I will either expose these under some macro KECCAK_EXPORT_INTERNALS or will just use keccak256(nullptr, 0). I'm not sure why I exposed these in the first place.

But I don't want to do it here.