Update operator-matrix products with the block-Jacobi preconditioner

src/LinearSolvers/preconditioners.jl

@@ -185,18 +185,21 @@ Base.eltype(::BlockJacobiPreconditioner) = Float64
 # of the blocks, gemm_strided is performing too many unecessary operations,
 # impairing its performance.
 @kernel function mblock_kernel!(y, b, p_len, rp_len, part, blocks)
+    p = size(b, 2)
     i, j = @index(Global, NTuple)
     len = p_len[i]
     rlen = rp_len[i]
 
     if j <= rlen
-        accum = 0.0
-        idxA = @inbounds part[j, i]
-        for k=1:len
-            idxB = @inbounds part[k, i]
-            @inbounds accum = accum + blocks[j, k, i]*b[idxB]
+        for ℓ=1:p
+            accum = 0.0
+            idxA = @inbounds part[j, i]
+            for k=1:len
+                idxB = @inbounds part[k, i]
+                @inbounds accum = accum + blocks[j, k, i]*b[idxB,ℓ]
+            end
+            y[idxA,ℓ] = accum
         end
-        y[idxA] = accum
     end
 end
 
@@ -214,6 +217,22 @@ function mul!(y, C::BlockJacobiPreconditioner, b::Vector{T}) where T
     end
 end
 
+function mul!(Y, C::BlockJacobiPreconditioner, B::Matrix{T}) where T
+    n, p = size(B)
+    fill!(Y, zero(T))
+    for i=1:C.nblocks
+        rlen = C.lpartitions[i]
+        part = C.partitions[1:rlen, i]
+        blck = C.blocks[1:rlen, 1:rlen, i]
+        for rhs=1:p
+            for j=1:C.rest_size[i]
+                idx = part[j]
+                Y[idx,rhs] += dot(blck[j, :], B[part,rhs])
+            end
+        end
+    end
+end
+
 function mul!(y, C::BlockJacobiPreconditioner, b::CuVector{T}) where T
     n = size(b, 1)
     fill!(y, zero(T))
@@ -227,6 +246,19 @@ function mul!(y, C::BlockJacobiPreconditioner, b::CuVector{T}) where T
     KA.synchronize(CUDABackend())
 end
 
+function mul!(Y, C::BlockJacobiPreconditioner, B::CuMatrix{T}) where T
+    n, p = size(B)
+    fill!(Y, zero(T))
+    max_rlen = maximum(C.rest_size)
+    ndrange = (C.nblocks, max_rlen)
+    mblock_kernel!(CUDABackend())(
+        Y, B, C.culpartitions, C.curest_size,
+        C.cupartitions, C.cublocks,
+        ndrange=ndrange,
+    )
+    KA.synchronize(CUDABackend())
+end
+
 """
     build_adjmatrix
 

test/TestLinearSolvers.jl

@@ -19,6 +19,28 @@ function generate_random_system(n::Int, m::Int)
     return spA, b, x♯
 end
 
+function test_preconditioners(device, AT, SMT)
+    n, m = 100, 100
+    A, b, x♯ = generate_random_system(n, m)
+    # Transfer data to device
+    A = A |> SMT
+    b = b |> AT
+    x♯ = x♯ |> AT
+    x = similar(b); r = similar(b)
+    nblocks = 2
+    precond = LS.BlockJacobiPreconditioner(A, nblocks, device)
+    LS.update(precond, A, device)
+    Iₙ = Matrix{Float64}(I, n, n) |> AT
+    Y = zeros(Float64, n, n) |> AT
+    Y2 = zeros(Float64, n, n) |> AT
+    mul!(Y, precond, Iₙ)
+    for i=1:n
+        eᵢ = Iₙ[:,i] |> AT
+        mul!(view(Y2,:,i), precond, eᵢ)
+    end
+    @test Y ≈ Y2
+end
+
 function test_custom_bicgstab(device, AT, SMT)
     n, m = 100, 100
     A, b, x♯  = generate_random_system(n, m)