Homogenize default behavior of higher-dimensional similar to SparseArrays

[alonsoC1s]

As mentioned in #3061 , when calling similar with CUSPARSE arguments and dimensions higher than 2, the fallback CPU method is called. As a result, calls to similar with GPU (sparse) matrices sometimes return host arrays, causing scalar indexing issues down the line

This PR introduces a fallback to similar that creates un-initialized CuArrays when there are 3 or more dimensions given as arguments, which is similar to what Base and SparseArrays do

Important note: I wasn't able to fully test this locally due to some issues instantiating the repo environment with Pkg

[maleadt]

Test failures related.

[alonsoC1s]

@maleadt my bad. I think I got it this time

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 14792c6	Previous: 33ffdef	Ratio
array/accumulate/Float32/1d	100706 ns	101156 ns	1.00
array/accumulate/Float32/dims=1	75725 ns	76494 ns	0.99
array/accumulate/Float32/dims=1L	1583091 ns	1583795 ns	1.00
array/accumulate/Float32/dims=2	142616 ns	143348 ns	0.99
array/accumulate/Float32/dims=2L	656324 ns	656949 ns	1.00
array/accumulate/Int64/1d	118300 ns	118389 ns	1.00
array/accumulate/Int64/dims=1	79226 ns	79916 ns	0.99
array/accumulate/Int64/dims=1L	1693681 ns	1694529 ns	1.00
array/accumulate/Int64/dims=2	155139 ns	155647 ns	1.00
array/accumulate/Int64/dims=2L	961219 ns	961265 ns	1.00
array/broadcast	20390 ns	20513 ns	0.99
array/construct	1314.1 ns	1333.8 ns	0.99
array/copy	18878 ns	18686 ns	1.01
array/copyto!/cpu_to_gpu	214371 ns	212224.5 ns	1.01
array/copyto!/gpu_to_cpu	281379.5 ns	282985 ns	0.99
array/copyto!/gpu_to_gpu	11296 ns	11526 ns	0.98
array/iteration/findall/bool	130863 ns	131867 ns	0.99
array/iteration/findall/int	147506 ns	148383 ns	0.99
array/iteration/findfirst/bool	80977.5 ns	81401 ns	0.99
array/iteration/findfirst/int	83244 ns	83519 ns	1.00
array/iteration/findmin/1d	86011 ns	83773.5 ns	1.03
array/iteration/findmin/2d	116904 ns	117018 ns	1.00
array/iteration/logical	199916 ns	197560.5 ns	1.01
array/iteration/scalar	66771 ns	68375 ns	0.98
array/permutedims/2d	52090 ns	52136 ns	1.00
array/permutedims/3d	52862.5 ns	52817.5 ns	1.00
array/permutedims/4d	51627.5 ns	52006 ns	0.99
array/random/rand/Float32	12917.5 ns	13083 ns	0.99
array/random/rand/Int64	29713.5 ns	37254 ns	0.80
array/random/rand!/Float32	8410.333333333334 ns	8673.666666666666 ns	0.97
array/random/rand!/Int64	34173 ns	33174 ns	1.03
array/random/randn/Float32	37260 ns	43918 ns	0.85
array/random/randn!/Float32	31469 ns	31437 ns	1.00
array/reductions/mapreduce/Float32/1d	34513 ns	35490 ns	0.97
array/reductions/mapreduce/Float32/dims=1	39573 ns	45840 ns	0.86
array/reductions/mapreduce/Float32/dims=1L	51814 ns	51786 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56518.5 ns	56654 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	69201.5 ns	69672 ns	0.99
array/reductions/mapreduce/Int64/1d	42038 ns	43204 ns	0.97
array/reductions/mapreduce/Int64/dims=1	52644 ns	42387 ns	1.24
array/reductions/mapreduce/Int64/dims=1L	87591 ns	87636 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59287 ns	59567 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	84827 ns	85154 ns	1.00
array/reductions/reduce/Float32/1d	34463 ns	35536 ns	0.97
array/reductions/reduce/Float32/dims=1	41870.5 ns	49541 ns	0.85
array/reductions/reduce/Float32/dims=1L	51679 ns	51867 ns	1.00
array/reductions/reduce/Float32/dims=2	56704 ns	56724 ns	1.00
array/reductions/reduce/Float32/dims=2L	69748 ns	70152 ns	0.99
array/reductions/reduce/Int64/1d	42140 ns	43180 ns	0.98
array/reductions/reduce/Int64/dims=1	43394 ns	46092.5 ns	0.94
array/reductions/reduce/Int64/dims=1L	87691 ns	87572 ns	1.00
array/reductions/reduce/Int64/dims=2	59396 ns	59560 ns	1.00
array/reductions/reduce/Int64/dims=2L	84284 ns	84946 ns	0.99
array/reverse/1d	18260 ns	18382 ns	0.99
array/reverse/1dL	68884 ns	68959.5 ns	1.00
array/reverse/1dL_inplace	65921 ns	65940 ns	1.00
array/reverse/1d_inplace	10293.166666666668 ns	8534 ns	1.21
array/reverse/2d	20591 ns	20498 ns	1.00
array/reverse/2dL	72539 ns	72575.5 ns	1.00
array/reverse/2dL_inplace	66040 ns	66219 ns	1.00
array/reverse/2d_inplace	10213 ns	10379 ns	0.98
array/sorting/1d	2735006 ns	2734628.5 ns	1.00
array/sorting/2d	1067915.5 ns	1068430 ns	1.00
array/sorting/by	3303800 ns	3303603 ns	1.00
cuda/synchronization/context/auto	1171.9 ns	1179.7 ns	0.99
cuda/synchronization/context/blocking	935.5588235294117 ns	917.5135135135135 ns	1.02
cuda/synchronization/context/nonblocking	7633.8 ns	7659 ns	1.00
cuda/synchronization/stream/auto	1021.5454545454545 ns	1056.923076923077 ns	0.97
cuda/synchronization/stream/blocking	835.6794871794872 ns	797.961038961039 ns	1.05
cuda/synchronization/stream/nonblocking	7529 ns	8313.1 ns	0.91
integration/byval/reference	144076 ns	144149 ns	1.00
integration/byval/slices=1	145859 ns	146090 ns	1.00
integration/byval/slices=2	284633 ns	284910 ns	1.00
integration/byval/slices=3	422972 ns	423319 ns	1.00
integration/cudadevrt	102614 ns	102707 ns	1.00
integration/volumerhs	9431917 ns	9433360.5 ns	1.00
kernel/indexing	13283 ns	13349 ns	1.00
kernel/indexing_checked	13981 ns	14195 ns	0.98
kernel/launch	2152.1111111111113 ns	2182.222222222222 ns	0.99
kernel/occupancy	671.6194968553459 ns	661.179012345679 ns	1.02
kernel/rand	14483 ns	14520 ns	1.00
latency/import	3816302972 ns	3811597161 ns	1.00
latency/precompile	4606963391 ns	4611897362.5 ns	1.00
latency/ttfp	4392541501.5 ns	4392419951 ns	1.00

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