use SIMD.jl for x86 and naive_findmin for :aarch64

[m-fila]

Just a test if I understand how to replace fast_findmin implementation as in #83

[Moelf]

after fixing that, on Apple M4, I get:

julia +1.12 --project instrumented-jetreco.jl --algorithm=AntiKt -R 0.4 ../test/data/events.pp13TeV.hepmc3.gz -m 32


# main
Processed 100 events 32 times
Average time per event 135.5111340625 ± 2.9835167009602035 μs
Lowest time per event 131.43084000000002 μs

# this PR
Processed 100 events 32 times
Average time per event 138.62045531249998 ± 3.5853189437051123 μs
Lowest time per event 134.35666 μs

# main with 1.12-beta3
Processed 100 events 32 times
Average time per event 147.96477937499995 ± 4.669684823430356 μs
Lowest time per event 139.74125 μs

# this PR with 1.12-beta3
Processed 100 events 32 times
Average time per event 152.88778656250003 ± 5.847938860407047 μs
Lowest time per event 142.36916000000002 μs

this looks not bad??!? @graeme-a-stewart Although, there's an unfortuante degredation on 1.12

[codecov]

Codecov Report

Attention: Patch coverage is 97.05882% with 1 line in your changes missing coverage. Please review.

Project coverage is 75.48%. Comparing base (0f47837) to head (dd97a88).

Files with missing lines	Patch %	Lines
src/Utils.jl	97.05%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #151      +/-   ##
==========================================
+ Coverage   75.07%   75.48%   +0.40%     
==========================================
  Files          19       19              
  Lines        1252     1277      +25     
==========================================
+ Hits          940      964      +24     
- Misses        312      313       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[m-fila]

Nvidia Grace

julia +1.12 --project instrumented-jetreco.jl --algorithm=AntiKt -R 0.4 ../test/data/events.pp13TeV.hepmc3.gz -m 32

# main
Processed 100 events 32 times
Average time per event 299.6035559375 ± 7.914782725959611 μs
Lowest time per event 280.66586 μs

# this PR
Processed 100 events 32 times
Average time per event 303.1709484375001 ± 8.819781288405308 μs
Lowest time per event 286.57968 μs

# main with 1.12-beta3
Processed 100 events 32 times
Average time per event 304.46982093750006 ± 13.922266876931305 μs
Lowest time per event 281.39805 μs

# this PR with 1.12-beta3
Processed 100 events 32 times
Average time per event 308.6548684375 ± 11.580493168517764 μs
Lowest time per event 288.65558 μs

AMD Ryzen 7 5700G

# main
Processed 100 events 32 times
Average time per event 188.28086562500002 ± 5.529621156269813 μs
Lowest time per event 181.3509 μs

# this PR
Processed 100 events 32 times
Average time per event 197.2376540625 ± 6.685658460456483 μs
Lowest time per event 186.34546999999998 μs

# main with 1.12-beta3
Processed 100 events 32 times
Average time per event 193.79122999999998 ± 11.42792473599183 μs
Lowest time per event 176.08296 μs

# this PR with 1.12-beta3
Processed 100 events 32 times
Average time per event 205.20970093750003 ± 9.736581454767695 μs
Lowest time per event 187.33396000000002 μs

[Moelf]

AMD Ryzen 7 5700G

yeah ok so we use the SIMD.jl for x86 and this PR for ARM?

[Moelf]

what happened to CI lol...

[m-fila]

For completeness, with SIMD for x86_64

AMD Ryzen 7 5700G

# main
Processed 100 events 32 times
Average time per event 187.38033500000006 ± 5.74337091484025 μs
Lowest time per event 177.03909 μs

# this PR
Processed 100 events 32 times
Average time per event 191.52783343750005 ± 6.355636113548961 μs
Lowest time per event 179.90162 μs

# main with 1.12-beta3
Processed 100 events 32 times
Average time per event 195.5927890625 ± 12.11654841840912 μs
Lowest time per event 178.41858 μs

# this PR with 1.12-beta3
Processed 100 events 32 times
Average time per event 194.36534874999998 ± 10.480610090304218 μs
Lowest time per event 176.65239000000003 μs

[Moelf]

JuliaLang/julia#58418

I will test against this PR later

[giordano]

Would be good to run CI on aarch64 (Linux and/or macOS), no?

[Moelf]

julia #58418 on M4
Processed 100 events 32 times
Average time per event 151.89809874999995 ± 5.095528318872066 μs
Lowest time per event 143.37917 μs

so this is same as nightly, and slower than Julia 1.11