Support 5-arg mul! (JuliaArrays#174) (JuliaArrays#175)

* Fix JuliaArrays#174

* v0.16.1

Author: dlfivefifty

Project.toml

@@ -1,14 +1,14 @@
 name = "BlockArrays"
 uuid = "8e7c35d0-a365-5155-bbbb-fb81a777f24e"
-version = "0.16"
+version = "0.16.1"
 
 [deps]
 ArrayLayouts = "4c555306-a7a7-4459-81d9-ec55ddd5c99a"
 FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 
 [compat]
-ArrayLayouts = "0.7.1"
+ArrayLayouts = "0.7.2"
 FillArrays = "0.11.6"
 julia = "1.6"
 

src/blocklinalg.jl

@@ -192,18 +192,16 @@ end
 
 function materialize!(M::MatMulMatAdd{<:AbstractBlockLayout,<:AbstractColumnMajor,<:AbstractColumnMajor})
     α, A, X_in, β, Y_in = M.α, M.A, M.B, M.β, M.C
-    _fill_lmul!(β, Y_in)
     X = PseudoBlockArray(X_in, (axes(A,2), axes(X_in,2)))
-    Y = PseudoBlockArray(Y_in, (axes(A,1), axes(Y_in,2)))
+    Y = PseudoBlockArray(Y_in, (axes(A,1), axes(X_in,2)))
     _block_muladd!(α, A, X, β, Y)
     Y_in
 end
 
 function materialize!(M::MatMulMatAdd{<:AbstractColumnMajor,<:AbstractBlockLayout,<:AbstractColumnMajor})
     α, A_in, X, β, Y_in = M.α, M.A, M.B, M.β, M.C
-    _fill_lmul!(β, Y_in)
     A = PseudoBlockArray(A_in, (axes(A_in,1),axes(X,1)))
-    Y = PseudoBlockArray(Y_in, (axes(Y_in,1),axes(X,2)))
+    Y = PseudoBlockArray(Y_in, (axes(A,1),axes(X,2)))
     _block_muladd!(α, A, X, β, Y)
     Y_in
 end

test/test_blocklinalg.jl

@@ -222,4 +222,16 @@ import ArrayLayouts: DenseRowMajor, ColumnMajor, StridedLayout
         mul!(Z, copy(X), Y)
         @test Z ≈ X*Y
     end
+
+    @testset "5-arg mul! (#174)" begin
+        A = BlockArray(rand(4, 5), [1,3], [2,3])
+        Ã = PseudoBlockArray(A)
+        B = BlockArray(rand(5, 3), [2,3], [1,1,1])
+        B̃ = PseudoBlockArray(B)
+        C = randn(4,3)
+        @test mul!(view(copy(C),:,1:3), A, B, 1, 2) ≈ A*B + 2C
+        @test mul!(view(copy(C),:,1:3), A, B̃, 1, 2) ≈ A*B + 2C
+        @test mul!(view(copy(C),:,1:3), Ã, B, 1, 2) ≈ A*B + 2C
+        @test mul!(view(copy(C),:,1:3), Ã, B̃, 1, 2) ≈ A*B + 2C
+    end
 end