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: f08a059	Previous: e4ac81a	Ratio
array/accumulate/Float32/1d	101247 ns	101073 ns	1.00
array/accumulate/Float32/dims=1	76743 ns	76196 ns	1.01
array/accumulate/Float32/dims=1L	1585917 ns	1585166 ns	1.00
array/accumulate/Float32/dims=2	143733 ns	143846 ns	1.00
array/accumulate/Float32/dims=2L	658278 ns	657343 ns	1.00
array/accumulate/Int64/1d	118799 ns	118428 ns	1.00
array/accumulate/Int64/dims=1	80961 ns	79813 ns	1.01
array/accumulate/Int64/dims=1L	1707513 ns	1706332.5 ns	1.00
array/accumulate/Int64/dims=2	156699.5 ns	155958.5 ns	1.00
array/accumulate/Int64/dims=2L	962521 ns	961689 ns	1.00
array/broadcast	20622 ns	20223 ns	1.02
array/construct	1249.6 ns	1268 ns	0.99
array/copy	18153 ns	18010.5 ns	1.01
array/copyto!/cpu_to_gpu	214356 ns	214386 ns	1.00
array/copyto!/gpu_to_cpu	283694 ns	282599 ns	1.00
array/copyto!/gpu_to_gpu	10887 ns	10725 ns	1.02
array/iteration/findall/bool	135104 ns	133957 ns	1.01
array/iteration/findall/int	150112 ns	148817 ns	1.01
array/iteration/findfirst/bool	81558 ns	80695 ns	1.01
array/iteration/findfirst/int	82867 ns	82681 ns	1.00
array/iteration/findmin/1d	83372 ns	85081 ns	0.98
array/iteration/findmin/2d	116997 ns	116308 ns	1.01
array/iteration/logical	200714.5 ns	196867 ns	1.02
array/iteration/scalar	66145 ns	66869 ns	0.99
array/permutedims/2d	52371.5 ns	51940.5 ns	1.01
array/permutedims/3d	52407 ns	52252 ns	1.00
array/permutedims/4d	52082.5 ns	51176 ns	1.02
array/random/rand/Float32	13014 ns	13404 ns	0.97
array/random/rand/Int64	25431 ns	24711 ns	1.03
array/random/rand!/Float32	10177.666666666666 ns	10187 ns	1.00
array/random/rand!/Int64	21989 ns	21588 ns	1.02
array/random/randn/Float32	37480 ns	43197 ns	0.87
array/random/randn!/Float32	30651 ns	30898 ns	0.99
array/reductions/mapreduce/Float32/1d	33980 ns	34650 ns	0.98
array/reductions/mapreduce/Float32/dims=1	40495.5 ns	39535.5 ns	1.02
array/reductions/mapreduce/Float32/dims=1L	51393.5 ns	51174 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56615 ns	56317 ns	1.01
array/reductions/mapreduce/Float32/dims=2L	69345 ns	69320 ns	1.00
array/reductions/mapreduce/Int64/1d	42363 ns	42198 ns	1.00
array/reductions/mapreduce/Int64/dims=1	42214 ns	41733 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	87299 ns	86905 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59569 ns	59221 ns	1.01
array/reductions/mapreduce/Int64/dims=2L	84785.5 ns	84434 ns	1.00
array/reductions/reduce/Float32/1d	34384 ns	34198.5 ns	1.01
array/reductions/reduce/Float32/dims=1	40444 ns	39152.5 ns	1.03
array/reductions/reduce/Float32/dims=1L	51476 ns	51208.5 ns	1.01
array/reductions/reduce/Float32/dims=2	56764 ns	56419 ns	1.01
array/reductions/reduce/Float32/dims=2L	69652 ns	69565 ns	1.00
array/reductions/reduce/Int64/1d	42501 ns	42368 ns	1.00
array/reductions/reduce/Int64/dims=1	42020 ns	50017 ns	0.84
array/reductions/reduce/Int64/dims=1L	87188 ns	86919 ns	1.00
array/reductions/reduce/Int64/dims=2	59606.5 ns	59687 ns	1.00
array/reductions/reduce/Int64/dims=2L	84660.5 ns	84484 ns	1.00
array/reverse/1d	18051 ns	17716 ns	1.02
array/reverse/1dL	68725 ns	68268 ns	1.01
array/reverse/1dL_inplace	65738 ns	65642 ns	1.00
array/reverse/1d_inplace	8549.333333333334 ns	10197.333333333334 ns	0.84
array/reverse/2d	21092 ns	20523 ns	1.03
array/reverse/2dL	73335 ns	72523 ns	1.01
array/reverse/2dL_inplace	65750 ns	65706 ns	1.00
array/reverse/2d_inplace	9950 ns	9831 ns	1.01
array/sorting/1d	2736444 ns	2735407.5 ns	1.00
array/sorting/2d	1069601 ns	1068528 ns	1.00
array/sorting/by	3306636 ns	3304139 ns	1.00
cuda/synchronization/context/auto	1131.6 ns	1165.7 ns	0.97
cuda/synchronization/context/blocking	920.377358490566 ns	876.8679245283018 ns	1.05
cuda/synchronization/context/nonblocking	7005.6 ns	6853.1 ns	1.02
cuda/synchronization/stream/auto	981.1875 ns	1035.7142857142858 ns	0.95
cuda/synchronization/stream/blocking	802.4257425742575 ns	789.9066666666666 ns	1.02
cuda/synchronization/stream/nonblocking	7178.8 ns	7513.5 ns	0.96
integration/byval/reference	143915 ns	143670 ns	1.00
integration/byval/slices=1	145772 ns	145632 ns	1.00
integration/byval/slices=2	284572 ns	284397 ns	1.00
integration/byval/slices=3	422978 ns	423006 ns	1.00
integration/cudadevrt	102395 ns	102385 ns	1.00
integration/volumerhs	23468528 ns	23431934.5 ns	1.00
kernel/indexing	13234 ns	13020 ns	1.02
kernel/indexing_checked	13957 ns	13828 ns	1.01
kernel/launch	2182.3333333333335 ns	2128.1111111111113 ns	1.03
kernel/occupancy	679.6064516129032 ns	701.7571428571429 ns	0.97
kernel/rand	16515 ns	14086 ns	1.17
latency/import	77746111924 ns	3826333658 ns	20.32
latency/precompile	8301636095.5 ns	4608074622.5 ns	1.80
latency/ttfp	78328318500 ns	4404025901.5 ns	17.79

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?