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: 54f12f6	Previous: 0ad0204	Ratio
array/accumulate/Float32/1d	101051 ns	101066 ns	1.00
array/accumulate/Float32/dims=1	76800 ns	76791 ns	1.00
array/accumulate/Float32/dims=1L	1585883 ns	1585160.5 ns	1.00
array/accumulate/Float32/dims=2	144087 ns	142797.5 ns	1.01
array/accumulate/Float32/dims=2L	658505 ns	657399 ns	1.00
array/accumulate/Int64/1d	119114 ns	118248.5 ns	1.01
array/accumulate/Int64/dims=1	80706 ns	79631 ns	1.01
array/accumulate/Int64/dims=1L	1695925.5 ns	1694201.5 ns	1.00
array/accumulate/Int64/dims=2	156668 ns	155693 ns	1.01
array/accumulate/Int64/dims=2L	962510 ns	961345 ns	1.00
array/broadcast	20728 ns	20364 ns	1.02
array/construct	1274.7 ns	1262.1 ns	1.01
array/copy	18167 ns	17799 ns	1.02
array/copyto!/cpu_to_gpu	216068 ns	213023 ns	1.01
array/copyto!/gpu_to_cpu	283881 ns	281078 ns	1.01
array/copyto!/gpu_to_gpu	10895 ns	10549.833333333332 ns	1.03
array/iteration/findall/bool	135031 ns	133910 ns	1.01
array/iteration/findall/int	148800 ns	148076 ns	1.00
array/iteration/findfirst/bool	81943 ns	80909 ns	1.01
array/iteration/findfirst/int	83874 ns	82938 ns	1.01
array/iteration/findmin/1d	85714 ns	82262 ns	1.04
array/iteration/findmin/2d	114989 ns	113580 ns	1.01
array/iteration/logical	200307.5 ns	199223 ns	1.01
array/iteration/scalar	66077.5 ns	67075 ns	0.99
array/permutedims/2d	52804.5 ns	51959 ns	1.02
array/permutedims/3d	53100 ns	52150 ns	1.02
array/permutedims/4d	51570 ns	51448.5 ns	1.00
array/random/rand/Float32	13073 ns	12962 ns	1.01
array/random/rand/Int64	24669 ns	24057 ns	1.03
array/random/rand!/Float32	8569.5 ns	9735.5 ns	0.88
array/random/rand!/Int64	21401 ns	21218 ns	1.01
array/random/randn/Float32	37534.5 ns	43055 ns	0.87
array/random/randn!/Float32	30935 ns	28025 ns	1.10
array/reductions/mapreduce/Float32/1d	35251 ns	33732 ns	1.05
array/reductions/mapreduce/Float32/dims=1	39956 ns	49005 ns	0.82
array/reductions/mapreduce/Float32/dims=1L	51342 ns	51002 ns	1.01
array/reductions/mapreduce/Float32/dims=2	56407 ns	57573.5 ns	0.98
array/reductions/mapreduce/Float32/dims=2L	69581 ns	66928.5 ns	1.04
array/reductions/mapreduce/Int64/1d	43074 ns	42242 ns	1.02
array/reductions/mapreduce/Int64/dims=1	44013 ns	48347 ns	0.91
array/reductions/mapreduce/Int64/dims=1L	87302 ns	86835 ns	1.01
array/reductions/mapreduce/Int64/dims=2	59675 ns	60601 ns	0.98
array/reductions/mapreduce/Int64/dims=2L	84777 ns	83761 ns	1.01
array/reductions/reduce/Float32/1d	35039 ns	34140.5 ns	1.03
array/reductions/reduce/Float32/dims=1	40196 ns	39851 ns	1.01
array/reductions/reduce/Float32/dims=1L	51440 ns	51248 ns	1.00
array/reductions/reduce/Float32/dims=2	56612 ns	58473 ns	0.97
array/reductions/reduce/Float32/dims=2L	70037 ns	67693 ns	1.03
array/reductions/reduce/Int64/1d	42966 ns	42152 ns	1.02
array/reductions/reduce/Int64/dims=1	42138.5 ns	43607.5 ns	0.97
array/reductions/reduce/Int64/dims=1L	87276 ns	86832 ns	1.01
array/reductions/reduce/Int64/dims=2	59434 ns	60330 ns	0.99
array/reductions/reduce/Int64/dims=2L	84738 ns	83394 ns	1.02
array/reverse/1d	18008.5 ns	17656 ns	1.02
array/reverse/1dL	68718 ns	68244 ns	1.01
array/reverse/1dL_inplace	65781 ns	65690 ns	1.00
array/reverse/1d_inplace	10294.666666666666 ns	8405.333333333334 ns	1.22
array/reverse/2d	20925 ns	20682 ns	1.01
array/reverse/2dL	73178 ns	72823 ns	1.00
array/reverse/2dL_inplace	65875 ns	65656 ns	1.00
array/reverse/2d_inplace	10734 ns	9850 ns	1.09
array/sorting/1d	2735825 ns	2735008 ns	1.00
array/sorting/2d	1076226 ns	1068540 ns	1.01
array/sorting/by	3327795 ns	3304170 ns	1.01
cuda/synchronization/context/auto	1183.8 ns	1131.2 ns	1.05
cuda/synchronization/context/blocking	957.2058823529412 ns	882.3673469387755 ns	1.08
cuda/synchronization/context/nonblocking	7488.299999999999 ns	8698.3 ns	0.86
cuda/synchronization/stream/auto	1031.3 ns	994.6 ns	1.04
cuda/synchronization/stream/blocking	839.1375 ns	825.1954022988506 ns	1.02
cuda/synchronization/stream/nonblocking	8112.8 ns	7316.4 ns	1.11
integration/byval/reference	143931 ns	143767 ns	1.00
integration/byval/slices=1	146072.5 ns	145850 ns	1.00
integration/byval/slices=2	284794 ns	284678 ns	1.00
integration/byval/slices=3	423418 ns	423481 ns	1.00
integration/cudadevrt	102495 ns	102411.5 ns	1.00
integration/volumerhs	23430349 ns	23457202 ns	1.00
kernel/indexing	13301 ns	13137 ns	1.01
kernel/indexing_checked	13955 ns	13819 ns	1.01
kernel/launch	2313.4444444444443 ns	2083.777777777778 ns	1.11
kernel/occupancy	745.082191780822 ns	668.00625 ns	1.12
kernel/rand	14310 ns	14198 ns	1.01
latency/import	3836824344.5 ns	3845206986.5 ns	1.00
latency/precompile	4634396037 ns	4633948752 ns	1.00
latency/ttfp	4409220668.5 ns	4453935633 ns	0.99

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?

[maleadt]

Same as #3119, you seem to have many unrelated changes in here that cause CI failures.

[maleadt]

More unrelated stuff... Please rebase that out.