Fix broadcasting for differently backed CLArrays of different dimensions (#345)

Author: simeonschaub

src/broadcast.jl

@@ -13,10 +13,10 @@ BroadcastStyle(W::Type{<:WrappedCLArray{T, N}}) where {T, N} =
 # when we are dealing with different buffer styles, we cannot know
 # which one is better, so use shared memory
 BroadcastStyle(
-    ::CLArrayStyle{N, B1},
+    ::CLArrayStyle{M, B1},
     ::CLArrayStyle{N, B2}
-) where {N, B1, B2} =
-    CLArrayStyle{N, cl.UnifiedSharedMemory}()
+) where {M, N, B1, B2} =
+    CLArrayStyle{max(M, N), B1 == B2 ? B1 : cl.UnifiedSharedMemory}()
 
 # allocation of output arrays
 Base.similar(bc::Broadcasted{CLArrayStyle{N, B}}, ::Type{T}, dims) where {T, N, B} =

test/execution.jl

@@ -115,4 +115,22 @@ a = CLArray{Int}(undef, 10)
 
 end
 
+@testset "broadcasting" begin
+    a = rand(Float32, 2, 3)
+    b = rand(Float32, 2)
+
+    c = a .+ b
+    a_cl, b_cl = CLArray(a), CLArray(b)
+    c_cl = a_cl .+ b_cl
+    @test Array(c_cl) == c
+    @test c_cl isa CLArray{Float32, 2, OpenCL.memory_type()}
+
+    if cl.usm_supported(cl.device())
+        a_cl, b_cl = CLMatrix{Float32, cl.UnifiedSharedMemory}(a), CLVector{Float32, OpenCL.memory_type()}(b)
+        c_cl = a_cl .+ b_cl
+        @test Array(c_cl) == c
+        @test c_cl isa CLArray{Float32, 2, cl.UnifiedSharedMemory}
+    end
+end
+
 end