Initial support for MPSNDArray

[christiangnrd]

Not hooked up to any higher-level features but comes with tests.

[github-actions]

Metal Benchmarks

Benchmark suite	Current: 6940471	Previous: 8c119cf	Ratio
private array/construct	26675.642857142855 ns	27535.714285714286 ns	0.97
private array/broadcast	460709 ns	462667 ns	1.00
private array/random/randn/Float32	726479 ns	801500 ns	0.91
private array/random/randn!/Float32	643875 ns	649125 ns	0.99
private array/random/rand!/Int64	573959 ns	556667 ns	1.03
private array/random/rand!/Float32	599083 ns	589000 ns	1.02
private array/random/rand/Int64	756104 ns	757125 ns	1.00
private array/random/rand/Float32	607333.5 ns	612979.5 ns	0.99
private array/copyto!/gpu_to_gpu	665458 ns	636833 ns	1.04
private array/copyto!/cpu_to_gpu	654208 ns	802250 ns	0.82
private array/copyto!/gpu_to_cpu	801291.5 ns	811416 ns	0.99
private array/accumulate/1d	1323000 ns	1335417 ns	0.99
private array/accumulate/2d	1396792 ns	1385542 ns	1.01
private array/iteration/findall/int	2037208 ns	2069208.5 ns	0.98
private array/iteration/findall/bool	1818687.5 ns	1807625 ns	1.01
private array/iteration/findfirst/int	1702875 ns	1689417 ns	1.01
private array/iteration/findfirst/bool	1658020.5 ns	1654417 ns	1.00
private array/iteration/scalar	3754583 ns	3888771 ns	0.97
private array/iteration/logical	3182584 ns	3174145.5 ns	1.00
private array/iteration/findmin/1d	1757708.5 ns	1740875 ns	1.01
private array/iteration/findmin/2d	1359750 ns	1351417 ns	1.01
private array/reductions/reduce/1d	1037271 ns	1020500 ns	1.02
private array/reductions/reduce/2d	664542 ns	658166 ns	1.01
private array/reductions/mapreduce/1d	1037187 ns	1042750 ns	0.99
private array/reductions/mapreduce/2d	668541 ns	656958 ns	1.02
private array/permutedims/4d	3067083 ns	2558833 ns	1.20
private array/permutedims/2d	1016791.5 ns	1022709 ns	0.99
private array/permutedims/3d	1584083 ns	1587645.5 ns	1.00
private array/copy	589833 ns	591375 ns	1.00
latency/precompile	5234192416 ns	5143804334 ns	1.02
latency/ttfp	6701849458 ns	6880237708.5 ns	0.97
latency/import	1159261542 ns	1161084541.5 ns	1.00
integration/metaldevrt	731583.5 ns	715041 ns	1.02
integration/byval/slices=1	1594125 ns	1632375 ns	0.98
integration/byval/slices=3	10332250 ns	10058875 ns	1.03
integration/byval/reference	1636208.5 ns	1585312.5 ns	1.03
integration/byval/slices=2	2627062.5 ns	2691375 ns	0.98
kernel/indexing	473792 ns	462625 ns	1.02
kernel/indexing_checked	468917 ns	460479 ns	1.02
kernel/launch	8084 ns	10104.166666666668 ns	0.80
metal/synchronization/stream	14875 ns	14834 ns	1.00
metal/synchronization/context	15167 ns	15167 ns	1
shared array/construct	26406.166666666668 ns	26138.916666666664 ns	1.01
shared array/broadcast	478583.5 ns	468542 ns	1.02
shared array/random/randn/Float32	754542 ns	809834 ns	0.93
shared array/random/randn!/Float32	643792 ns	664917 ns	0.97
shared array/random/rand!/Int64	572166 ns	556875 ns	1.03
shared array/random/rand!/Float32	598958 ns	595666 ns	1.01
shared array/random/rand/Int64	757625 ns	787624.5 ns	0.96
shared array/random/rand/Float32	621958 ns	602166 ns	1.03
shared array/copyto!/gpu_to_gpu	89584 ns	87792 ns	1.02
shared array/copyto!/cpu_to_gpu	87875 ns	88708 ns	0.99
shared array/copyto!/gpu_to_cpu	78500 ns	79167 ns	0.99
shared array/accumulate/1d	1345625 ns	1355375.5 ns	0.99
shared array/accumulate/2d	1399959 ns	1387104.5 ns	1.01
shared array/iteration/findall/int	1799000 ns	1754499.5 ns	1.03
shared array/iteration/findall/bool	1561895.5 ns	1593625 ns	0.98
shared array/iteration/findfirst/int	1403458 ns	1384875 ns	1.01
shared array/iteration/findfirst/bool	1373187 ns	1353459 ns	1.01
shared array/iteration/scalar	159375 ns	160834 ns	0.99
shared array/iteration/logical	2989084 ns	2942875 ns	1.02
shared array/iteration/findmin/1d	1463167 ns	1457563 ns	1.00
shared array/iteration/findmin/2d	1365458 ns	1364667 ns	1.00
shared array/reductions/reduce/1d	735291 ns	726000 ns	1.01
shared array/reductions/reduce/2d	657208 ns	654625 ns	1.00
shared array/reductions/mapreduce/1d	737583 ns	736917 ns	1.00
shared array/reductions/mapreduce/2d	674000 ns	664458.5 ns	1.01
shared array/permutedims/4d	2725833 ns	2550125 ns	1.07
shared array/permutedims/2d	1015625 ns	1026542 ns	0.99
shared array/permutedims/3d	1642625 ns	1599229.5 ns	1.03
shared array/copy	243750 ns	237042 ns	1.03

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