Proper support for similar on CuSparseMats

[kshyatt]

Should address #1667

[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/array.jl b/lib/cusparse/array.jl
index b278b8129..81e34c159 100644
--- a/lib/cusparse/array.jl
+++ b/lib/cusparse/array.jl
@@ -247,21 +247,21 @@ Base.similar(Mat::CuSparseMatrixCSR, T::Type) = CuSparseMatrixCSR(copy(Mat.rowPt
 Base.similar(Mat::CuSparseMatrixBSR, T::Type) = CuSparseMatrixBSR(copy(Mat.rowPtr), copy(Mat.colVal), similar(nonzeros(Mat), T), Mat.blockDim, Mat.dir, nnz(Mat), size(Mat))
 Base.similar(Mat::CuSparseMatrixCOO, T::Type) = CuSparseMatrixCOO(copy(Mat.rowInd), copy(Mat.colInd), similar(nonzeros(Mat), T), size(Mat), nnz(Mat))
 
-Base.similar(Mat::CuSparseMatrixCSC, T::Type, N::Int, M::Int) =  CuSparseMatrixCSC(CUDA.zeros(Int32, 1), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N, M))
-Base.similar(Mat::CuSparseMatrixCSR, T::Type, N::Int, M::Int) =  CuSparseMatrixCSR(CUDA.zeros(Int32, 1), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N,M))
-Base.similar(Mat::CuSparseMatrixCOO, T::Type, N::Int, M::Int) =  CuSparseMatrixCOO(CUDA.zeros(Int32, 0), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N,M))
+Base.similar(Mat::CuSparseMatrixCSC, T::Type, N::Int, M::Int) = CuSparseMatrixCSC(CUDA.zeros(Int32, 1), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N, M))
+Base.similar(Mat::CuSparseMatrixCSR, T::Type, N::Int, M::Int) = CuSparseMatrixCSR(CUDA.zeros(Int32, 1), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N, M))
+Base.similar(Mat::CuSparseMatrixCOO, T::Type, N::Int, M::Int) = CuSparseMatrixCOO(CUDA.zeros(Int32, 0), CUDA.zeros(Int32, 0), CuVector{T}(undef, 0), (N, M))
 
-Base.similar(Mat::CuSparseMatrixCSC{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M) 
-Base.similar(Mat::CuSparseMatrixCSR{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M) 
-Base.similar(Mat::CuSparseMatrixCOO{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M) 
+Base.similar(Mat::CuSparseMatrixCSC{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
+Base.similar(Mat::CuSparseMatrixCSR{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
+Base.similar(Mat::CuSparseMatrixCOO{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
 
-Base.similar(Mat::CuSparseMatrixCSC, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...) 
-Base.similar(Mat::CuSparseMatrixCSR, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...) 
-Base.similar(Mat::CuSparseMatrixCOO, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...) 
+Base.similar(Mat::CuSparseMatrixCSC, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
+Base.similar(Mat::CuSparseMatrixCSR, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
+Base.similar(Mat::CuSparseMatrixCOO, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
 
-Base.similar(Mat::CuSparseMatrixCSC, dims::Tuple{Int, Int}) = similar(Mat, dims...) 
-Base.similar(Mat::CuSparseMatrixCSR, dims::Tuple{Int, Int}) = similar(Mat, dims...) 
-Base.similar(Mat::CuSparseMatrixCOO, dims::Tuple{Int, Int}) = similar(Mat, dims...) 
+Base.similar(Mat::CuSparseMatrixCSC, dims::Tuple{Int, Int}) = similar(Mat, dims...)
+Base.similar(Mat::CuSparseMatrixCSR, dims::Tuple{Int, Int}) = similar(Mat, dims...)
+Base.similar(Mat::CuSparseMatrixCOO, dims::Tuple{Int, Int}) = similar(Mat, dims...)
 
 Base.similar(Mat::CuSparseArrayCSR) = CuSparseArrayCSR(copy(Mat.rowPtr), copy(Mat.colVal), similar(nonzeros(Mat)), size(Mat))
 
diff --git a/test/libraries/cusparse.jl b/test/libraries/cusparse.jl
index 823165c25..1d72972fd 100644
--- a/test/libraries/cusparse.jl
+++ b/test/libraries/cusparse.jl
@@ -153,10 +153,10 @@ end
             @test size(similar(d_x, (3, 4))) == (3, 4)
             @test similar(d_x, Float32) isa CuSparseMatrixCSR{Float32}
         end
-        
+
         @testset "COO" begin
-            x = sprand(elty,m,n, 0.2)
-            d_x  = CuSparseMatrixCOO(x)
+            x = sprand(elty, m, n, 0.2)
+            d_x = CuSparseMatrixCOO(x)
             @test collect(d_x) == collect(x)
             @test similar(d_x) isa CuSparseMatrixCOO{elty}
             @test similar(d_x, (3, 4)) isa CuSparseMatrixCOO{elty}

[github-actions]

CUDA.jl Benchmarks

Benchmark suite	Current: ef8f9af	Previous: c735d30	Ratio
latency/precompile	46590943591.5 ns	46632963038 ns	1.00
latency/ttfp	7008753919 ns	6945348552 ns	1.01
latency/import	3651469460 ns	3627556110 ns	1.01
integration/volumerhs	9624046.5 ns	9624924.5 ns	1.00
integration/byval/slices=1	147045 ns	146732 ns	1.00
integration/byval/slices=3	425236 ns	425273.5 ns	1.00
integration/byval/reference	145031 ns	144844 ns	1.00
integration/byval/slices=2	285986 ns	285928 ns	1.00
integration/cudadevrt	103563.5 ns	103266 ns	1.00
kernel/indexing	14055 ns	14149 ns	0.99
kernel/indexing_checked	14776 ns	14787 ns	1.00
kernel/occupancy	630.9941520467836 ns	643.4260355029586 ns	0.98
kernel/launch	2017.2 ns	2148.4444444444443 ns	0.94
kernel/rand	14516 ns	16462 ns	0.88
array/reverse/1d	19888.5 ns	19527 ns	1.02
array/reverse/2d	25080 ns	24746 ns	1.01
array/reverse/1d_inplace	11372 ns	10303 ns	1.10
array/reverse/2d_inplace	13200 ns	11908 ns	1.11
array/copy	21004 ns	20974 ns	1.00
array/iteration/findall/int	159106 ns	158927 ns	1.00
array/iteration/findall/bool	139328 ns	139468 ns	1.00
array/iteration/findfirst/int	153717 ns	154085 ns	1.00
array/iteration/findfirst/bool	154809 ns	155140 ns	1.00
array/iteration/scalar	71086 ns	71174 ns	1.00
array/iteration/logical	207585.5 ns	215769 ns	0.96
array/iteration/findmin/1d	41284 ns	41611 ns	0.99
array/iteration/findmin/2d	94234.5 ns	94406 ns	1.00
array/reductions/reduce/1d	45800 ns	40170 ns	1.14
array/reductions/reduce/2d	50576.5 ns	45595.5 ns	1.11
array/reductions/mapreduce/1d	40950 ns	37717 ns	1.09
array/reductions/mapreduce/2d	48838.5 ns	50935 ns	0.96
array/broadcast	20825.5 ns	20980 ns	0.99
array/copyto!/gpu_to_gpu	13517 ns	11688 ns	1.16
array/copyto!/cpu_to_gpu	208316 ns	209993 ns	0.99
array/copyto!/gpu_to_cpu	243999 ns	244189.5 ns	1.00
array/accumulate/1d	108328 ns	108613 ns	1.00
array/accumulate/2d	80024 ns	80233 ns	1.00
array/construct	1296.85 ns	1265 ns	1.03
array/random/randn/Float32	43291.5 ns	44268 ns	0.98
array/random/randn!/Float32	26491 ns	26318 ns	1.01
array/random/rand!/Int64	26944 ns	27025 ns	1.00
array/random/rand!/Float32	8891.666666666666 ns	8630 ns	1.03
array/random/rand/Int64	29836 ns	29758 ns	1.00
array/random/rand/Float32	13198 ns	12929 ns	1.02
array/permutedims/4d	60846 ns	61037 ns	1.00
array/permutedims/2d	54854 ns	55629 ns	0.99
array/permutedims/3d	55949 ns	56014 ns	1.00
array/sorting/1d	2764445 ns	2765122 ns	1.00
array/sorting/by	3352239 ns	3354655 ns	1.00
array/sorting/2d	1081434 ns	1084634.5 ns	1.00
cuda/synchronization/stream/auto	1039 ns	1031.1 ns	1.01
cuda/synchronization/stream/nonblocking	6296.8 ns	6343.8 ns	0.99
cuda/synchronization/stream/blocking	815.6382978723404 ns	793.3529411764706 ns	1.03
cuda/synchronization/context/auto	1180.6 ns	1192.4 ns	0.99
cuda/synchronization/context/nonblocking	6630.4 ns	6640 ns	1.00
cuda/synchronization/context/blocking	940.936170212766 ns	927.8409090909091 ns	1.01

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

[maleadt]

CI failures look related.

[kshyatt]

Modified the test as I think it was only passing due to the behind-the-scenes similar creating a host array 😬 . Also modified the sized similar because in cases where N * M < nnz(Mat), the current setup is invalid.

[codecov]

Codecov Report

Attention: Patch coverage is 75.00000% with 3 lines in your changes missing coverage. Please review.

Project coverage is 73.55%. Comparing base (c735d30) to head (ef8f9af).
Report is 8 commits behind head on master.

Files with missing lines	Patch %	Lines
lib/cusparse/array.jl	75.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2652      +/-   ##
==========================================
- Coverage   73.56%   73.55%   -0.01%     
==========================================
  Files         158      158              
  Lines       15326    15335       +9     
==========================================
+ Hits        11274    11280       +6     
- Misses       4052     4055       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kshyatt]

Look ok to merge?