More types for MPSGraph reduction/scan/sort support

[christiangnrd]

Funny enough I tried to add & and | reductions but Base is broken for those until JuliaLang/julia#58418

[maleadt]

Yeah I had removed these because of incompatible NaN propagation, but you're right that it should work for integers.

[christiangnrd]

Edit: I've decided Bool for scans and reductions is future work for someone who thinks they would benefit from it
A few quirks I'm probably overthinking:
- I don't include Bool for scan when it's technically supported and does something for min/max
- bitwise and and or reductions are broken in Base so I didn't add them butmaybe I should?
- Properly encoding support for types/operations will become a mess of lists/unions. Is it worth it?

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 0477d5a	Previous: 59e377f	Ratio
array/accumulate/Float32/1d	521209 ns	573625 ns	0.91
array/accumulate/Float32/dims=1	492084 ns	571125 ns	0.86
array/accumulate/Float32/dims=1L	8949000 ns	8850292 ns	1.01
array/accumulate/Float32/dims=2	535041 ns	625125 ns	0.86
array/accumulate/Float32/dims=2L	2909729 ns	3358312 ns	0.87
array/accumulate/Int64/1d	930167 ns	894208 ns	1.04
array/accumulate/Int64/dims=1	1111333 ns	1095792 ns	1.01
array/accumulate/Int64/dims=1L	9806146 ns	9580896 ns	1.02
array/accumulate/Int64/dims=2	1466708 ns	1462208 ns	1.00
array/accumulate/Int64/dims=2L	6958729 ns	6584667 ns	1.06
array/broadcast	335750 ns	364209 ns	0.92
array/construct	3292 ns	3334 ns	0.99
array/permutedims/2d	613167 ns	633458 ns	0.97
array/permutedims/3d	1098209 ns	1098354 ns	1.00
array/permutedims/4d	1238750 ns	1590750.5 ns	0.78
array/private/copy	386625.5 ns	407917 ns	0.95
array/private/copyto!/cpu_to_gpu	367000 ns	382125 ns	0.96
array/private/copyto!/gpu_to_cpu	354208 ns	377459 ns	0.94
array/private/copyto!/gpu_to_gpu	325000 ns	357750 ns	0.91
array/private/iteration/findall/bool	1190709 ns	1140208 ns	1.04
array/private/iteration/findall/int	1321687.5 ns	1280458 ns	1.03
array/private/iteration/findfirst/bool	1326770.5 ns	1428334 ns	0.93
array/private/iteration/findfirst/int	1379270.5 ns	1466499.5 ns	0.94
array/private/iteration/findmin/1d	1520459 ns	1567125 ns	0.97
array/private/iteration/findmin/2d	1228771 ns	1270709 ns	0.97
array/private/iteration/logical	1984687.5 ns	1933750 ns	1.03
array/private/iteration/scalar	1752645.5 ns	2445416 ns	0.72
array/random/rand/Float32	599750 ns	616666 ns	0.97
array/random/rand/Int64	641333 ns	629708 ns	1.02
array/random/rand!/Float32	497916 ns	530625 ns	0.94
array/random/rand!/Int64	470875 ns	495291.5 ns	0.95
array/random/randn/Float32	557208 ns	570667 ns	0.98
array/random/randn!/Float32	453083 ns	477000 ns	0.95
array/reductions/mapreduce/Float32/1d	622542 ns	715520.5 ns	0.87
array/reductions/mapreduce/Float32/dims=1	448459 ns	492000 ns	0.91
array/reductions/mapreduce/Float32/dims=1L	700792 ns	689083 ns	1.02
array/reductions/mapreduce/Float32/dims=2	452250 ns	494792 ns	0.91
array/reductions/mapreduce/Float32/dims=2L	1005479.5 ns	1335459 ns	0.75
array/reductions/mapreduce/Int64/1d	800875 ns	943459 ns	0.85
array/reductions/mapreduce/Int64/dims=1	742750 ns	812625 ns	0.91
array/reductions/mapreduce/Int64/dims=1L	1101458 ns	1075667 ns	1.02
array/reductions/mapreduce/Int64/dims=2	826499.5 ns	987541 ns	0.84
array/reductions/mapreduce/Int64/dims=2L	2212833 ns	2168458 ns	1.02
array/reductions/reduce/Float32/1d	618167 ns	693083 ns	0.89
array/reductions/reduce/Float32/dims=1	451979 ns	496312.5 ns	0.91
array/reductions/reduce/Float32/dims=1L	704520.5 ns	687979.5 ns	1.02
array/reductions/reduce/Float32/dims=2	362458 ns	395834 ns	0.92
array/reductions/reduce/Float32/dims=2L	497583 ns	485542 ns	1.02
array/reductions/reduce/Int64/1d	804292 ns	962083 ns	0.84
array/reductions/reduce/Int64/dims=1	736437 ns	814541.5 ns	0.90
array/reductions/reduce/Int64/dims=1L	1098583 ns	1069500 ns	1.03
array/reductions/reduce/Int64/dims=2	292875 ns	973542 ns	0.30
array/reductions/reduce/Int64/dims=2L	687917 ns	2149542 ns	0.32
array/shared/copy	162833.5 ns	154083 ns	1.06
array/shared/copyto!/cpu_to_gpu	40875 ns	40708 ns	1.00
array/shared/copyto!/gpu_to_cpu	40291 ns	41000 ns	0.98
array/shared/copyto!/gpu_to_gpu	40167 ns	41208 ns	0.97
array/shared/iteration/findall/bool	1199667 ns	1148500 ns	1.04
array/shared/iteration/findall/int	1329167 ns	1285750 ns	1.03
array/shared/iteration/findfirst/bool	1029208 ns	1119667 ns	0.92
array/shared/iteration/findfirst/int	1021375 ns	1207312.5 ns	0.85
array/shared/iteration/findmin/1d	1278750 ns	1317938 ns	0.97
array/shared/iteration/findmin/2d	1229416 ns	1269625 ns	0.97
array/shared/iteration/logical	1793937.5 ns	1709354 ns	1.05
array/shared/iteration/scalar	3598.875 ns	3614.625 ns	1.00
array/sorting/1d	2004750 ns
array/sorting/2d	8449750 ns
integration/byval/reference	1151417 ns	1141958 ns	1.01
integration/byval/slices=1	1152125 ns	1136875 ns	1.01
integration/byval/slices=2	2080542 ns	2055750 ns	1.01
integration/byval/slices=3	18090625 ns	8460958 ns	2.14
integration/metaldevrt	440666.5 ns	451500 ns	0.98
kernel/indexing	329083 ns	351792 ns	0.94
kernel/indexing_checked	488729.5 ns	534209 ns	0.91
kernel/launch	1995.9 ns	1975 ns	1.01
kernel/rand	493667 ns	528250 ns	0.93
latency/import	1815418000 ns	1791534292 ns	1.01
latency/precompile	40257051458.5 ns	39637308833.5 ns	1.02
latency/ttfp	2150084542 ns	2124080375 ns	1.01
metal/synchronization/context	525.8229166666666 ns	523.9637305699482 ns	1.00
metal/synchronization/stream	276.61290322580646 ns	277.5974025974026 ns	1.00

This comment was automatically generated by workflow using github-action-benchmark.

[codecov]

Codecov Report

❌ Patch coverage is 97.61905% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.69%. Comparing base (59e377f) to head (0477d5a).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/mpsgraphs/scan.jl	93.33%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #863      +/-   ##
==========================================
+ Coverage   86.25%   86.69%   +0.44%     
==========================================
  Files          76       76              
  Lines        5130     5142      +12     
==========================================
+ Hits         4425     4458      +33     
+ Misses        705      684      -21     

☔ View full report in Codecov by Harness.
📢 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.

[christiangnrd]

3x regression for the array/accumulate/Int64/dims=1L benchmark. Other Int64 benchmarks seem to be the same or improved.