diff --git a/src/mapreduce.jl b/src/mapreduce.jl
index 645db2cd..6813fdca 100644
--- a/src/mapreduce.jl
+++ b/src/mapreduce.jl
@@ -117,6 +117,7 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::oneWrappedArray{T},
     Base.check_reducedims(R, A)
     length(A) == 0 && return R # isempty(::Broadcasted) iterates
 
+    R_old = R
     # add singleton dimensions to the output container, if needed
     if ndims(R) < ndims(A)
         dims = Base.fill_to_length(size(R), 1, Val(ndims(A)))
@@ -199,5 +200,5 @@ function GPUArrays.mapreducedim!(f::F, op::OP, R::oneWrappedArray{T},
         GPUArrays.mapreducedim!(identity, op, R′, partial; init=init)
     end
 
-    return R
+    return R_old
 end
diff --git a/test/array.jl b/test/array.jl
index 1ca66453..53eaba96 100644
--- a/test/array.jl
+++ b/test/array.jl
@@ -139,3 +139,9 @@ end
   resize!(b, 1)
   @test length(b) == 1
 end
+
+@testset "mapreducedim! returning same type" begin
+  R = transpose(oneAPI.zeros(Float32, 2, 3))
+  A = oneArray(rand(Float32, 3, 2, 10))
+  @test @inferred(oneAPI.GPUArrays.mapreducedim!(identity, +, R, A)) === R
+end