consistency

Project.toml

@@ -1,7 +1,7 @@
 name = "AcceleratedKernels"
 uuid = "6a4ca0a5-0e36-4168-a932-d9be78d558f1"
 authors = ["Andrei-Leonard Nicusan <[email protected]> and contributors"]
-version = "0.2.2"
+version = "0.2.3"
 
 [deps]
 ArgCheck = "dce04be8-c92d-5529-be00-80e4d2c0e197"

src/accumulate/accumulate.jl

@@ -174,28 +174,26 @@ function _accumulate_impl!(
 )
     if backend isa GPU
         if isnothing(dims)
-            accumulate_1d!(
+            return accumulate_1d!(
                 op, v, backend, alg,
                 init=init, inclusive=inclusive,
                 block_size=block_size, temp=temp, temp_flags=temp_flags,
             )
-            return v
         else
-            accumulate_nd!(
+            return accumulate_nd!(
                 op, v, backend,
                 init=init, dims=dims, inclusive=inclusive,
                 block_size=block_size,
             )
         end
     else
         if isnothing(dims)
-            accumulate_1d!(
+            return accumulate_1d!(
                 op, v,
                 init=init, inclusive=inclusive,
             )
-            return v
         else
-            accumulate_nd!(
+            return accumulate_nd!(
                 op, v,
                 init=init, dims=dims, inclusive=inclusive,
             )

src/accumulate/accumulate_nd.jl

@@ -218,18 +218,19 @@ end
             end
         end
 
+        _running_prefix = running_prefix[0x1]
         @synchronize()
 
         if block_offset + ai < length_dims
-            total = op(running_prefix[0x1], temp[ai + bank_offset_a + 0x1])
+            total = op(_running_prefix, temp[ai + bank_offset_a + 0x1])
             v[
                 input_base_idx +
                 (block_offset + ai) * vstrides[dims] +
                 0x1
             ] = total
         end
         if block_offset + bi < length_dims
-            total = op(running_prefix[0x1], temp[bi + bank_offset_b + 0x1])
+            total = op(_running_prefix, temp[bi + bank_offset_b + 0x1])
             v[
                 input_base_idx +
                 (block_offset + bi) * vstrides[dims] +