Add @inbounds to SparseArrays.nzrange for CuSparseDeviceMatrixCSC

[Zinoex]

Add @inbounds to SparseArrays.nzrange(g::CuSparseDeviceMatrixCSC, col::Integer) to avoid bounds checking the colPtr (that was causing me much register spilling into local mem), consistent with the new SparseArrays functionality of GPUArrays.jl.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic master) to apply these changes.

Click here to view the suggested changes.

diff --git a/lib/cusparse/device.jl b/lib/cusparse/device.jl
index 6fa563552..dea10c5f6 100644
--- a/lib/cusparse/device.jl
+++ b/lib/cusparse/device.jl
@@ -37,7 +37,7 @@ SparseArrays.nnz(g::CuSparseDeviceMatrixCSC) = g.nnz
 SparseArrays.rowvals(g::CuSparseDeviceMatrixCSC) = g.rowVal
 SparseArrays.getcolptr(g::CuSparseDeviceMatrixCSC) = g.colPtr
 SparseArrays.getnzval(g::CuSparseDeviceMatrixCSC) = g.nzVal
-SparseArrays.nzrange(g::CuSparseDeviceMatrixCSC, col::Integer) = @inbounds SparseArrays.getcolptr(g)[col]:(SparseArrays.getcolptr(g)[col+1]-1)
+SparseArrays.nzrange(g::CuSparseDeviceMatrixCSC, col::Integer) = @inbounds SparseArrays.getcolptr(g)[col]:(SparseArrays.getcolptr(g)[col + 1] - 1)
 SparseArrays.nonzeros(g::CuSparseDeviceMatrixCSC) = g.nzVal
 
 const CuSparseDeviceColumnView{Tv, Ti} = SubArray{Tv, 1, <:CuSparseDeviceMatrixCSC{Tv, Ti}, Tuple{Base.Slice{Base.OneTo{Int}}, Int}}

[kshyatt]

Just fyi these types are probably going away soon -- I honestly don't remember if I had @inbounds on the GPUArrays.jl function 🙈

[Zinoex]

I'll cross that bridge when I need to. The GPUArrays.jl sparse device arrays look to be a drop-in replacement, but I also have some quite complex kernels in IntervalMDP.jl operating on CuSparseMatrixCSC, and therefore, I expect to run into issues when transitioning - and I most certainly don't have the time to fix that atm.

[github-actions]

CUDA.jl Benchmarks

Benchmark suite	Current: 48ee0d2	Previous: 2e983fe	Ratio
latency/precompile	56920777390 ns	56427085830.5 ns	1.01
latency/ttfp	8317020437.5 ns	8362501410 ns	0.99
latency/import	4498206426 ns	4521778039 ns	0.99
integration/volumerhs	9624244.5 ns	9624952.5 ns	1.00
integration/byval/slices=1	147221 ns	146870 ns	1.00
integration/byval/slices=3	426220 ns	425790 ns	1.00
integration/byval/reference	145136 ns	144866 ns	1.00
integration/byval/slices=2	286528.5 ns	286021 ns	1.00
integration/cudadevrt	103621 ns	103323 ns	1.00
kernel/indexing	14182 ns	14090 ns	1.01
kernel/indexing_checked	14943 ns	14977.5 ns	1.00
kernel/occupancy	691.2333333333333 ns	670.5886075949367 ns	1.03
kernel/launch	2199.3333333333335 ns	2115.8 ns	1.04
kernel/rand	18671 ns	16842 ns	1.11
array/reverse/1d	19950 ns	19633 ns	1.02
array/reverse/2dL_inplace	66907 ns	66698 ns	1.00
array/reverse/1dL	70183 ns	69881 ns	1.00
array/reverse/2d	21758 ns	21367 ns	1.02
array/reverse/1d_inplace	9638 ns	9601 ns	1.00
array/reverse/2d_inplace	13400 ns	13220 ns	1.01
array/reverse/2dL	73895 ns	73483 ns	1.01
array/reverse/1dL_inplace	66908 ns	66751 ns	1.00
array/copy	20830 ns	20712 ns	1.01
array/iteration/findall/int	157172.5 ns	156846 ns	1.00
array/iteration/findall/bool	139875 ns	139935.5 ns	1.00
array/iteration/findfirst/int	161160 ns	160606 ns	1.00
array/iteration/findfirst/bool	161959 ns	161405 ns	1.00
array/iteration/scalar	73837 ns	72218 ns	1.02
array/iteration/logical	216845.5 ns	215761.5 ns	1.01
array/iteration/findmin/1d	50519 ns	49669 ns	1.02
array/iteration/findmin/2d	96561.5 ns	96275.5 ns	1.00
array/reductions/reduce/Int64/1d	43137 ns	43492 ns	0.99
array/reductions/reduce/Int64/dims=1	44422 ns	44664.5 ns	0.99
array/reductions/reduce/Int64/dims=2	61610 ns	61641 ns	1.00
array/reductions/reduce/Int64/dims=1L	88803 ns	88640 ns	1.00
array/reductions/reduce/Int64/dims=2L	87915.5 ns	87635.5 ns	1.00
array/reductions/reduce/Float32/1d	37370 ns	36681 ns	1.02
array/reductions/reduce/Float32/dims=1	51991 ns	48806 ns	1.07
array/reductions/reduce/Float32/dims=2	59879 ns	59459 ns	1.01
array/reductions/reduce/Float32/dims=1L	52434 ns	52065 ns	1.01
array/reductions/reduce/Float32/dims=2L	72183 ns	71664 ns	1.01
array/reductions/mapreduce/Int64/1d	43399 ns	43256 ns	1.00
array/reductions/mapreduce/Int64/dims=1	44879.5 ns	44863 ns	1.00
array/reductions/mapreduce/Int64/dims=2	61690 ns	61500 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	89152 ns	88638 ns	1.01
array/reductions/mapreduce/Int64/dims=2L	88167 ns	87897.5 ns	1.00
array/reductions/mapreduce/Float32/1d	37366 ns	36277.5 ns	1.03
array/reductions/mapreduce/Float32/dims=1	48730 ns	41259 ns	1.18
array/reductions/mapreduce/Float32/dims=2	60130 ns	59440 ns	1.01
array/reductions/mapreduce/Float32/dims=1L	52689 ns	52331.5 ns	1.01
array/reductions/mapreduce/Float32/dims=2L	72524.5 ns	71656.5 ns	1.01
array/broadcast	20031 ns	19817 ns	1.01
array/copyto!/gpu_to_gpu	12986 ns	11436 ns	1.14
array/copyto!/cpu_to_gpu	214181 ns	215179 ns	1.00
array/copyto!/gpu_to_cpu	282787 ns	282618 ns	1.00
array/accumulate/Int64/1d	124556 ns	124273 ns	1.00
array/accumulate/Int64/dims=1	83102 ns	83182 ns	1.00
array/accumulate/Int64/dims=2	157715 ns	157485 ns	1.00
array/accumulate/Int64/dims=1L	1709359 ns	1709450 ns	1.00
array/accumulate/Int64/dims=2L	966166 ns	966304 ns	1.00
array/accumulate/Float32/1d	109001 ns	108932 ns	1.00
array/accumulate/Float32/dims=1	80062 ns	80065 ns	1.00
array/accumulate/Float32/dims=2	147542.5 ns	146929 ns	1.00
array/accumulate/Float32/dims=1L	1618641 ns	1618534.5 ns	1.00
array/accumulate/Float32/dims=2L	698238 ns	697506 ns	1.00
array/construct	1282.2 ns	1270.6 ns	1.01
array/random/randn/Float32	45167.5 ns	47947 ns	0.94
array/random/randn!/Float32	24926 ns	24918 ns	1.00
array/random/rand!/Int64	27311 ns	27167 ns	1.01
array/random/rand!/Float32	8903.666666666666 ns	8884.333333333334 ns	1.00
array/random/rand/Int64	29812 ns	37695.5 ns	0.79
array/random/rand/Float32	13240.5 ns	12943 ns	1.02
array/permutedims/4d	55770.5 ns	59797.5 ns	0.93
array/permutedims/2d	54046 ns	53660 ns	1.01
array/permutedims/3d	54951 ns	54666 ns	1.01
array/sorting/1d	2757753 ns	2757791.5 ns	1.00
array/sorting/by	3344532 ns	3344326 ns	1.00
array/sorting/2d	1080947 ns	1080588 ns	1.00
cuda/synchronization/stream/auto	1044.8 ns	1040 ns	1.00
cuda/synchronization/stream/nonblocking	7844.4 ns	6879.299999999999 ns	1.14
cuda/synchronization/stream/blocking	857.6823529411764 ns	805.0612244897959 ns	1.07
cuda/synchronization/context/auto	1196 ns	1175.2 ns	1.02
cuda/synchronization/context/nonblocking	7480 ns	7439.7 ns	1.01
cuda/synchronization/context/blocking	932.9230769230769 ns	896.560975609756 ns	1.04

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