Merge pull request #365 from martinholters/mh/fixmul

Fix indexing bug in `mul_unrolled_chunks!`

Author: andyferris

src/matrix_multiply.jl

@@ -398,7 +398,7 @@ end
 
     # Do a custom b[:, k2] to return a SVector (an isbits type) rather than a mutable type. Avoids allocation == faster
     tmp_type = SVector{sb[1], eltype(c)}
-    vect_exprs = [:($(Symbol("tmp_$k2")) = partly_unrolled_multiply($(Size(sa)), $(Size(sb[1])), a, $(Expr(:call, tmp_type, [Expr(:ref, :b, sub2ind(sc, i, k2)) for i = 1:sb[1]]...)))) for k2 = 1:sb[2]]
+    vect_exprs = [:($(Symbol("tmp_$k2")) = partly_unrolled_multiply($(Size(sa)), $(Size(sb[1])), a, $(Expr(:call, tmp_type, [Expr(:ref, :b, sub2ind(sb, i, k2)) for i = 1:sb[1]]...)))) for k2 = 1:sb[2]]
 
     exprs = [:(c[$(sub2ind(sc, k1, k2))] = $(Symbol("tmp_$k2"))[$k1]) for k1 = 1:sa[1], k2 = 1:sb[2]]
 

test/matrix_multiply.jl

@@ -295,6 +295,16 @@
         A_mul_B!(a_5, m_5, n_5)
         @test a_5 ≈ a_array_5
 
+        m_array_6 = rand(1:10, 8, 10)
+        n_array_6 = rand(1:10, 10, 8)
+        a_array_6 = m_array_6*n_array_6
+
+        m_6 = MMatrix{8,10}(m_array_6)
+        n_6 = MMatrix{10,8}(n_array_6)
+        a_6 = MMatrix{8,8,Int}()
+        A_mul_B!(a_6, m_6, n_6)
+        @test a_6 == a_array_6
+
         # Float64
         vf = @SVector [2.0, 4.0]
         vf2 = [2.0, 4.0]