Support sparse matrices from rocSPARSE

[amontoison]

close #254
Very similar to #253.

I requested access to CI with AMD GPUs last week. I should have it but didn't find time to try it yet.
I can do that when I will back in the US.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 57 lines in your changes missing coverage. Please review.
✅ Project coverage is 21.65%. Comparing base (108c7c1) to head (82fa780).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
ext/SparseMatrixColoringsAMDGPUExt.jl	0.00%	57 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (108c7c1) and HEAD (82fa780). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (108c7c1)	HEAD (82fa780)
	3	2

Additional details and impacted files

@@             Coverage Diff              @@
##              main     #255       +/-   ##
============================================
- Coverage   100.00%   21.65%   -78.35%     
============================================
  Files           17       16        -1     
  Lines         1939     1972       +33     
============================================
- Hits          1939      427     -1512     
- Misses           0     1545     +1545     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[amontoison]

Hum I should copy-paste the "conversion" functions that I added in CUDA.jl. I have the following errors with AMD GPUs:

(n = 100, p = 0.05): Error During Test at /home/amontoison/Argonne/SparseMatrixColorings.jl/test/rocm.jl:51
  Got exception outside of a @test
  Scalar indexing is disallowed.
  Invocation of getindex resulted in scalar indexing of a GPU array.
  This is typically caused by calling an iterating implementation of a method.
  Such implementations *do not* execute on the GPU, but very slowly on the CPU,
  and therefore should be avoided.
  
  If you want to allow scalar iteration, use `allowscalar` or `@allowscalar`
  to enable scalar iteration globally or for the operations in question.
  Stacktrace:
    [1] error(s::String)
      @ Base ./error.jl:35
    [2] errorscalar(op::String)
      @ GPUArraysCore ~/.julia/x86_64/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:151
    [3] _assertscalar(op::String, behavior::GPUArraysCore.ScalarIndexing)
      @ GPUArraysCore ~/.julia/x86_64/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:124
    [4] assertscalar(op::String)
      @ GPUArraysCore ~/.julia/x86_64/packages/GPUArraysCore/aNaXo/src/GPUArraysCore.jl:112
    [5] getindex
      @ ~/.julia/x86_64/packages/GPUArrays/uiVyU/src/host/indexing.jl:50 [inlined]
    [6] getindex(A::ROCSparseMatrixCSR{Float64, Int32}, i0::Int64, i1::Int64)
      @ AMDGPU.rocSPARSE ~/.julia/x86_64/packages/AMDGPU/wH6SV/src/sparse/array.jl:352
    [7] _getindex
      @ ./abstractarray.jl:1358 [inlined]
    [8] getindex
      @ ./abstractarray.jl:1312 [inlined]
    [9] iterate
      @ ./abstractarray.jl:1209 [inlined]
   [10] iterate
      @ ./abstractarray.jl:1207 [inlined]
   [11] SparseMatrixCSC{Float64, Int64}(M::ROCSparseMatrixCSR{Float64, Int32})
      @ SparseArrays ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/SparseArrays/src/sparsematrix.jl:891
   [12] SparseMatrixCSC
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/SparseArrays/src/sparsematrix.jl:885 [inlined]
   [13] convert
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/SparseArrays/src/sparsematrix.jl:980 [inlined]
   [14] sparse
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/SparseArrays/src/sparsematrix.jl:1016 [inlined]
   [15] matrix_versions(A::ROCSparseMatrixCSR{Float64, Int32})
      @ SparseMatrixColorings ~/.julia/x86_64/packages/SparseMatrixColorings/FZwji/src/matrices.jl:15
   [16] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1814 [inlined]
   [17] test_coloring_decompression(A0::ROCSparseMatrixCSR{Float64, Int32}, problem::ColoringProblem{:symmetric, :column}, algo::GreedyColoringAlgorithm{:direct, NaturalOrder}; B0::Nothing, color0::Nothing, test_fast::Bool, gpu::Bool)
      @ Main ~/Argonne/SparseMatrixColorings.jl/test/utils.jl:28
   [18] macro expansion
      @ ~/Argonne/SparseMatrixColorings.jl/test/rocm.jl:53 [inlined]
   [19] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1793 [inlined]
   [20] macro expansion
      @ ~/Argonne/SparseMatrixColorings.jl/test/rocm.jl:51 [inlined]
   [21] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1793 [inlined]
   [22] macro expansion
      @ ~/Argonne/SparseMatrixColorings.jl/test/rocm.jl:50 [inlined]
   [23] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1704 [inlined]
   [24] top-level scope
      @ ~/Argonne/SparseMatrixColorings.jl/test/rocm.jl:48
   [25] include(fname::String)
      @ Main ./sysimg.jl:38
   [26] macro expansion
      @ ~/Argonne/SparseMatrixColorings.jl/test/runtests.jl:24 [inlined]
   [27] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1704 [inlined]
   [28] macro expansion
      @ ~/Argonne/SparseMatrixColorings.jl/test/runtests.jl:22 [inlined]
   [29] macro expansion
      @ ~/Applications/julia/x86_64/julia-1.11.5/share/julia/stdlib/v1.11/Test/src/Test.jl:1704 [inlined]
   [30] top-level scope
      @ ~/Argonne/SparseMatrixColorings.jl/test/runtests.jl:14
   [31] include(fname::String)
      @ Main ./sysimg.jl:38
   [32] top-level scope
      @ none:6
   [33] eval
      @ ./boot.jl:430 [inlined]
   [34] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:296
   [35] _start()
      @ Base ./client.jl:531

[gdalle]

This seems to be just in the compress function which is not performance-critical. Maybe as a first step you can do SparseMatrixCSC(Matrix(A_gpu))?

[amontoison]

I am still traveling this week but next week I can fix the issue in AMDGPU.jl.

[amontoison]

@gdalle I finally worked on it tonight.
I needed to add a dispatch SparseArrays.sparse(A::ROCSparseMatrixCSC) = A in AMDGPU.jl.
Require a new release of AMDGPU.jl with the following PR:
JuliaGPU/AMDGPU.jl#816

julia> include("test/runtests.jl")
Test Summary:         | Pass  Total   Time
SparseMatrixColorings |  422    422  10.5s
  CUDA                |           0   0.0s
  ROCm                |  422    422  10.5s