add mat * dense_vec operation

src/abstractblocksparsearray/arraylayouts.jl

@@ -1,4 +1,4 @@
-using ArrayLayouts: ArrayLayouts, DualLayout, MemoryLayout, MulAdd
+using ArrayLayouts: ArrayLayouts, DenseColumnMajor, DualLayout, MemoryLayout, MulAdd
 using BlockArrays: BlockLayout
 using SparseArraysBase: SparseLayout
 using TypeParameterAccessors: parenttype, similartype
@@ -56,6 +56,20 @@ function Base.similar(
     return similar(BlockSparseArray{elt, length(axes), output_blocktype′}, axes)
 end
 
+# BlockSparseMatrix * dense Vector → dense Vector
+# Returns a plain Vector instead of a BlockedVector
+function Base.similar(
+        mul::MulAdd{
+            <:BlockLayout{<:SparseLayout},
+            <:DenseColumnMajor,
+            <:Any,
+        },
+        elt::Type,
+        axes::Tuple{<:AbstractUnitRange},
+    )
+    return Vector{elt}(undef, length(only(axes)))
+end
+
 # Materialize a SubArray view.
 function ArrayLayouts.sub_materialize(layout::BlockLayout{<:SparseLayout}, a, axes)
     # TODO: Define `blocktype`/`blockstype` for `SubArray` wrapping `BlockSparseArray`.

src/blocksparsearrayinterface/arraylayouts.jl

@@ -1,16 +1,25 @@
-using ArrayLayouts: ArrayLayouts, Dot, MatMulMatAdd, MatMulVecAdd, MulAdd
+using ArrayLayouts: ArrayLayouts, DenseColumnMajor, Dot, MatMulMatAdd, MatMulVecAdd, MulAdd
 using BlockArrays: BlockArrays, BlockLayout, muladd!
 using SparseArraysBase: SparseLayout
 using LinearAlgebra: LinearAlgebra, dot, mul!
 
 const muladd!_blocksparse = blocksparse_style(muladd!)
+# Matrix-matrix case
 function muladd!_blocksparse(
-        α::Number, a1::AbstractArray, a2::AbstractArray, β::Number, a_dest::AbstractArray
+        α::Number, a1::AbstractMatrix, a2::AbstractMatrix, β::Number, a_dest::AbstractMatrix
     )
     mul!(blocks(a_dest), blocks(a1), blocks(a2), α, β)
     return a_dest
 end
 
+# Matrix-vector case: BlockSparseMatrix * dense Vector
+function muladd!_blocksparse(
+        α::Number, a1::AbstractMatrix, a2::AbstractVector, β::Number, a_dest::AbstractVector
+    )
+    mul!_blocksparse(a_dest, a1, a2, α, β)
+    return a_dest
+end
+
 function ArrayLayouts.materialize!(
         m::MatMulMatAdd{
             <:BlockLayout{<:SparseLayout},
@@ -21,15 +30,28 @@ function ArrayLayouts.materialize!(
     muladd!_blocksparse(m.α, m.A, m.B, m.β, m.C)
     return m.C
 end
+
+# BlockSparseMatrix * dense Vector → dense Vector
+function ArrayLayouts.materialize!(
+        m::MatMulVecAdd{
+            <:BlockLayout{<:SparseLayout},
+            <:DenseColumnMajor,
+            <:DenseColumnMajor,
+        },
+    )
+    muladd!_blocksparse(m.α, m.A, m.B, m.β, m.C)
+    return m.C
+end
+
+# BlockSparseMatrix * BlockSparseVector (not yet implemented)
 function ArrayLayouts.materialize!(
         m::MatMulVecAdd{
             <:BlockLayout{<:SparseLayout},
             <:BlockLayout{<:SparseLayout},
             <:BlockLayout{<:SparseLayout},
         },
     )
-    error("Not implemented.")
-    matmul!(m)
+    error("BlockSparseMatrix * BlockSparseVector not implemented.")
     return m.C
 end
 

src/blocksparsearrayinterface/linearalgebra.jl

@@ -1,3 +1,4 @@
+using BlockArrays: Block
 using LinearAlgebra: LinearAlgebra, mul!
 
 const mul!_blocksparse = blocksparse_style(mul!)
@@ -11,3 +12,31 @@ function mul!_blocksparse(
     mul!(blocks(a_dest), blocks(a1), blocks(a2), α, β)
     return a_dest
 end
+
+# Matrix-vector multiplication: BlockSparseMatrix * dense Vector
+function mul!_blocksparse(
+        a_dest::AbstractVector,
+        a1::AbstractMatrix,
+        a2::AbstractVector,
+        α::Number = true,
+        β::Number = false,
+    )
+    # Scale destination by β (or zero it if β == 0)
+    if iszero(β)
+        fill!(a_dest, zero(eltype(a_dest)))
+    elseif !isone(β)
+        a_dest .*= β
+    end
+
+    # Accumulate A[i,j] * v[j_range] into C[i_range]
+    for I in eachblockstoredindex(a1)
+        i, j = Int.(Tuple(I))
+        block_A = a1[I]
+        row_range = axes(a1, 1)[Block(i)]
+        col_range = axes(a1, 2)[Block(j)]
+        v_block = @view a2[col_range]
+        c_block = @view a_dest[row_range]
+        mul!(c_block, block_A, v_block, α, true)  # β=true to accumulate
+    end
+    return a_dest
+end

test/test_basics.jl

@@ -426,6 +426,22 @@ arrayts = (Array, JLArray)
         @test a1' * a2 ≈ Array(a1)' * Array(a2)
         @test dot(a1, a2) ≈ a1' * a2
     end
+    @testset "Matrix-vector multiplication" begin
+        a = dev(BlockSparseArray{elt}(undef, [2, 3], [2, 3]))
+        a[Block(1, 2)] = dev(randn(elt, size(@view(a[Block(1, 2)]))))
+        a[Block(2, 1)] = dev(randn(elt, size(@view(a[Block(2, 1)]))))
+        v = dev(randn(elt, 5))  # Dense vector matching column dimension
+
+        c = a * v
+        @test Array(c) ≈ Array(a) * Array(v)
+        @test c isa Vector{elt}  # Result should be a plain Vector
+        @test length(c) == size(a, 1)
+
+        # Test with adjoint/transpose
+        v2 = dev(randn(elt, 5))  # Vector matching row dimension for transposed multiplication
+        @test Array(a' * v2) ≈ Array(a)' * Array(v2)
+        @test Array(transpose(a) * v2) ≈ transpose(Array(a)) * Array(v2)
+    end
     @testset "cat" begin
         a1 = dev(BlockSparseArray{elt}(undef, [2, 3], [2, 3]))
         a1[Block(2, 1)] = dev(randn(elt, size(@view(a1[Block(2, 1)]))))