Added Base.similar methods for CuSparseMatrixCOO and BSR

[rainerrodrigues]

This PR adds the missing Base.similar methods for CuSparseMatrixCOO and CuSparseMatrixBSR, allowing them to fallback gracefully without converting to dense CPU arrays.

Fixes #3061
Fixes #3055

[kshyatt]

Also, can some tests be added?

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 845e83c	Previous: d08923d	Ratio
array/accumulate/Float32/1d	101140.5 ns	101201.5 ns	1.00
array/accumulate/Float32/dims=1	76322 ns	77253 ns	0.99
array/accumulate/Float32/dims=1L	1584657 ns	1586384.5 ns	1.00
array/accumulate/Float32/dims=2	143544 ns	144050 ns	1.00
array/accumulate/Float32/dims=2L	657740 ns	658315 ns	1.00
array/accumulate/Int64/1d	118704 ns	118535 ns	1.00
array/accumulate/Int64/dims=1	80145 ns	80398 ns	1.00
array/accumulate/Int64/dims=1L	1705722 ns	1695368 ns	1.01
array/accumulate/Int64/dims=2	156343 ns	156400.5 ns	1.00
array/accumulate/Int64/dims=2L	961909 ns	962477 ns	1.00
array/broadcast	20273 ns	20549 ns	0.99
array/construct	1273.6 ns	1244.7 ns	1.02
array/copy	17792 ns	18346 ns	0.97
array/copyto!/cpu_to_gpu	213574 ns	217009 ns	0.98
array/copyto!/gpu_to_cpu	281602 ns	283646 ns	0.99
array/copyto!/gpu_to_gpu	10655 ns	10940 ns	0.97
array/iteration/findall/bool	134322 ns	135015 ns	0.99
array/iteration/findall/int	149464 ns	150942 ns	0.99
array/iteration/findfirst/bool	80978 ns	81514 ns	0.99
array/iteration/findfirst/int	83157 ns	84056 ns	0.99
array/iteration/findmin/1d	84753 ns	87756 ns	0.97
array/iteration/findmin/2d	117102 ns	117744 ns	0.99
array/iteration/logical	197029.5 ns	201505 ns	0.98
array/iteration/scalar	64912 ns	67810 ns	0.96
array/permutedims/2d	52533 ns	52558 ns	1.00
array/permutedims/3d	52825.5 ns	52631 ns	1.00
array/permutedims/4d	51771 ns	51253 ns	1.01
array/random/rand/Float32	12720 ns	12853 ns	0.99
array/random/rand/Int64	25014 ns	25414 ns	0.98
array/random/rand!/Float32	9253 ns	8376.666666666666 ns	1.10
array/random/rand!/Int64	21467 ns	21965 ns	0.98
array/random/randn/Float32	40917.5 ns	38474 ns	1.06
array/random/randn!/Float32	25853 ns	30963 ns	0.83
array/reductions/mapreduce/Float32/1d	34170 ns	34333 ns	1.00
array/reductions/mapreduce/Float32/dims=1	39205.5 ns	40570 ns	0.97
array/reductions/mapreduce/Float32/dims=1L	51195.5 ns	51354 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56360 ns	56637 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	69013 ns	69577 ns	0.99
array/reductions/mapreduce/Int64/1d	41911 ns	42670 ns	0.98
array/reductions/mapreduce/Int64/dims=1	42090 ns	43498 ns	0.97
array/reductions/mapreduce/Int64/dims=1L	87036 ns	87200 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59145 ns	59546 ns	0.99
array/reductions/mapreduce/Int64/dims=2L	84434 ns	84737 ns	1.00
array/reductions/reduce/Float32/1d	34387 ns	34657.5 ns	0.99
array/reductions/reduce/Float32/dims=1	40647 ns	39996.5 ns	1.02
array/reductions/reduce/Float32/dims=1L	51397.5 ns	51457 ns	1.00
array/reductions/reduce/Float32/dims=2	56481 ns	56754 ns	1.00
array/reductions/reduce/Float32/dims=2L	69363.5 ns	70015.5 ns	0.99
array/reductions/reduce/Int64/1d	41982 ns	42974.5 ns	0.98
array/reductions/reduce/Int64/dims=1	50345.5 ns	42327 ns	1.19
array/reductions/reduce/Int64/dims=1L	86951 ns	87167 ns	1.00
array/reductions/reduce/Int64/dims=2	59596 ns	59657 ns	1.00
array/reductions/reduce/Int64/dims=2L	84767 ns	84515 ns	1.00
array/reverse/1d	17643 ns	17882 ns	0.99
array/reverse/1dL	68225.5 ns	68439 ns	1.00
array/reverse/1dL_inplace	65570 ns	65793.5 ns	1.00
array/reverse/1d_inplace	10158.5 ns	8645.333333333334 ns	1.18
array/reverse/2d	20688 ns	21177 ns	0.98
array/reverse/2dL	72784 ns	73131 ns	1.00
array/reverse/2dL_inplace	65691 ns	65813 ns	1.00
array/reverse/2d_inplace	10372 ns	9973 ns	1.04
array/sorting/1d	2733819 ns	2735906 ns	1.00
array/sorting/2d	1068081 ns	1068705.5 ns	1.00
array/sorting/by	3302391 ns	3304477 ns	1.00
cuda/synchronization/context/auto	1162.1 ns	1153.8 ns	1.01
cuda/synchronization/context/blocking	930.5714285714286 ns	920.219512195122 ns	1.01
cuda/synchronization/context/nonblocking	7707.299999999999 ns	7049.8 ns	1.09
cuda/synchronization/stream/auto	989.2105263157895 ns	1045.5 ns	0.95
cuda/synchronization/stream/blocking	807.7676767676768 ns	845.6486486486486 ns	0.96
cuda/synchronization/stream/nonblocking	7204.8 ns	7261.5 ns	0.99
integration/byval/reference	143774 ns	143708 ns	1.00
integration/byval/slices=1	145530 ns	145668 ns	1.00
integration/byval/slices=2	284392 ns	284436 ns	1.00
integration/byval/slices=3	423048 ns	422883 ns	1.00
integration/cudadevrt	102357 ns	102385 ns	1.00
integration/volumerhs	23462010.5 ns	23480818 ns	1.00
kernel/indexing	12997 ns	13249 ns	0.98
kernel/indexing_checked	13782 ns	14040 ns	0.98
kernel/launch	2047.111111111111 ns	2194.8888888888887 ns	0.93
kernel/occupancy	705.3356164383562 ns	704.5137931034483 ns	1.00
kernel/rand	14118 ns	16709 ns	0.84
latency/import	3837187980.5 ns	3825894453 ns	1.00
latency/precompile	4600227064.5 ns	4600509263 ns	1.00
latency/ttfp	4404423502.5 ns	4395653546.5 ns	1.00

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

[rainerrodrigues]

@kshyatt Hi, can you check if this is suitable and extensive enough for testing?