Use GPUToolbox.jl

[christiangnrd]

Should be gtg if tests pass

TODO:

• Remove [sources] entry
• Add compat entry

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 74.88%. Comparing base (52d7056) to head (930663c).
Report is 421 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #538      +/-   ##
==========================================
+ Coverage   71.04%   74.88%   +3.84%     
==========================================
  Files          36       57      +21     
  Lines        1143     2708    +1565     
==========================================
+ Hits          812     2028    +1216     
- Misses        331      680     +349     

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

[github-actions]

Metal Benchmarks

Benchmark suite	Current: 930663c	Previous: bec8c71	Ratio
private array/construct	24524.25 ns	24993.166666666668 ns	0.98
private array/broadcast	457958 ns	465708 ns	0.98
private array/random/randn/Float32	812625 ns	758541.5 ns	1.07
private array/random/randn!/Float32	636083 ns	628833.5 ns	1.01
private array/random/rand!/Int64	579229.5 ns	561125 ns	1.03
private array/random/rand!/Float32	592041 ns	594167 ns	1.00
private array/random/rand/Int64	769854.5 ns	782979.5 ns	0.98
private array/random/rand/Float32	607312.5 ns	601042 ns	1.01
private array/copyto!/gpu_to_gpu	656583 ns	666666 ns	0.98
private array/copyto!/cpu_to_gpu	708083 ns	617709 ns	1.15
private array/copyto!/gpu_to_cpu	813333 ns	504083 ns	1.61
private array/accumulate/1d	1344041 ns	1350208 ns	1.00
private array/accumulate/2d	1476875 ns	1390417 ns	1.06
private array/iteration/findall/int	2091500 ns	2066458 ns	1.01
private array/iteration/findall/bool	1848166.5 ns	1826625 ns	1.01
private array/iteration/findfirst/int	1696375 ns	1694583 ns	1.00
private array/iteration/findfirst/bool	1657333.5 ns	1661917 ns	1.00
private array/iteration/scalar	3903458 ns	3825541.5 ns	1.02
private array/iteration/logical	3174750 ns	3192917 ns	0.99
private array/iteration/findmin/1d	1748729.5 ns	1768354 ns	0.99
private array/iteration/findmin/2d	1353208 ns	1347270.5 ns	1.00
private array/reductions/reduce/1d	1031833.5 ns	1037417 ns	0.99
private array/reductions/reduce/2d	668250 ns	664166 ns	1.01
private array/reductions/mapreduce/1d	1044541.5 ns	1038687.5 ns	1.01
private array/reductions/mapreduce/2d	661334 ns	667729.5 ns	0.99
private array/permutedims/4d	2543770.5 ns	2537291.5 ns	1.00
private array/permutedims/2d	1026375 ns	1025229 ns	1.00
private array/permutedims/3d	1591645.5 ns	1574209 ns	1.01
private array/copy	549166.5 ns	569958 ns	0.96
latency/precompile	9100731166 ns	9097851875 ns	1.00
latency/ttfp	3692994125 ns	3676407792 ns	1.00
latency/import	1252891896 ns	1241771750 ns	1.01
integration/metaldevrt	713271 ns	701916 ns	1.02
integration/byval/slices=1	1542375 ns	1647333.5 ns	0.94
integration/byval/slices=3	9548084 ns	10016083 ns	0.95
integration/byval/reference	1580687.5 ns	1621729 ns	0.97
integration/byval/slices=2	2599250 ns	2740896 ns	0.95
kernel/indexing	453334 ns	467166.5 ns	0.97
kernel/indexing_checked	449250 ns	472958 ns	0.95
kernel/launch	9708.333333333334 ns	8042 ns	1.21
metal/synchronization/stream	14917 ns	14166 ns	1.05
metal/synchronization/context	15167 ns	15292 ns	0.99
shared array/construct	23263.833333333332 ns	24368 ns	0.95
shared array/broadcast	466083 ns	462917 ns	1.01
shared array/random/randn/Float32	807833 ns	765792 ns	1.05
shared array/random/randn!/Float32	633750 ns	636834 ns	1.00
shared array/random/rand!/Int64	573917 ns	575708 ns	1.00
shared array/random/rand!/Float32	586917 ns	597854.5 ns	0.98
shared array/random/rand/Int64	784166.5 ns	791979 ns	0.99
shared array/random/rand/Float32	646416 ns	628937.5 ns	1.03
shared array/copyto!/gpu_to_gpu	80625 ns	82917 ns	0.97
shared array/copyto!/cpu_to_gpu	84458 ns	82542 ns	1.02
shared array/copyto!/gpu_to_cpu	76292 ns	81958 ns	0.93
shared array/accumulate/1d	1370042 ns	1344458 ns	1.02
shared array/accumulate/2d	1394625.5 ns	1398542 ns	1.00
shared array/iteration/findall/int	1847500 ns	1827729.5 ns	1.01
shared array/iteration/findall/bool	1603167 ns	1595416 ns	1.00
shared array/iteration/findfirst/int	1429542 ns	1402854 ns	1.02
shared array/iteration/findfirst/bool	1367042 ns	1374000 ns	0.99
shared array/iteration/scalar	160542 ns	152895.5 ns	1.05
shared array/iteration/logical	2975229 ns	2984625 ns	1.00
shared array/iteration/findmin/1d	1462875 ns	1465333 ns	1.00
shared array/iteration/findmin/2d	1362125 ns	1383958 ns	0.98
shared array/reductions/reduce/1d	724708 ns	735375 ns	0.99
shared array/reductions/reduce/2d	672750 ns	675500 ns	1.00
shared array/reductions/mapreduce/1d	744167 ns	739416.5 ns	1.01
shared array/reductions/mapreduce/2d	665375 ns	678250 ns	0.98
shared array/permutedims/4d	2547042 ns	2409542 ns	1.06
shared array/permutedims/2d	1025375 ns	1023021 ns	1.00
shared array/permutedims/3d	1590250 ns	1582666 ns	1.00
shared array/copy	247167 ns	241792 ns	1.02

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

[christiangnrd]

I'll merge this one since the only code from GPUToolbox that this uses is SimpleVersion and I'm very confident that it's the same code.