2 files changed, 29 insertions, 15 deletions

diff --git a/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
index 25023ff0a1..3daedd4efb 100644
--- a/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
+++ b/arm_compute/dynamic_fusion/sketch/MemoryDescriptor.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,12 +32,31 @@ namespace experimental
 {
 namespace dynamic_fusion
 {
-/** Type of memory used by a workload tensor */
+/** Type of memory used by a workload tensor
+ *
+ *  We can classify tensors in 2 dimensions: Topology (where they are in a workload) and Memory allocation:
+ * Topology:
+ *      Argument tensors: "Outer" tensors exposed to the users as inputs and outputs (arguments)
+ *      Intermediate tensors: "Inner" tensors hidden from the users as links between operators
+ * Memory allocation:
+ *      Alloc: Tensors that need to be allocated real backing memory
+ *      No-Alloc: Tensors that don't need to be allocated real backing memory
+ *
+ * We end up with 3 MemoryType based on the product of these two classifications
+ *          |    Argument    |   Intermediate    |
+ * ---------*----------------*-------------------*
+ * Alloc    |     User       |   Auxiliary       |
+ * ---------*----------------*-------------------*
+ * No-Alloc *     N/A        |    Virtual        |
+ * ---------*----------------*-------------------*
+ */
 enum class MemoryType
 {
+    /** Both User and Auxiliary types are of Alloc type. Since they require memory allocation */
     User      = 0, /**< Memory coming directly from users, e.g. for argument tensors */
-    Auxiliary = 1, /**< Additional memory required by the workload tensor, e.g. for temporary tensors */
-    NoAlloc   = 2, /**< Temporary tile which is not allocated as a whole tensor in the memory */
+    Auxiliary = 1, /**< Additional memory required by the workload tensor, e.g. for tensors holding temporary results between kernels */
+    /** Virtual type is of No-Alloc type. Since it doesn't require memory allocation */
+    Virtual = 2, /**< Temporary tile which is not allocated as a whole tensor in the memory. It is mainly used at sketch time to link operators; there should be no Virtual tensors at runtime */
 };
 
 /** Memory information for tensors with @ref MemoryType::Auxiliary.
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h
index f19ad6dfc5..422edb35f1 100644
--- a/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h
+++ b/arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h
@@ -70,18 +70,9 @@ public:
     TensorInfo create_tensor_info(Args &&... args)
     {
         auto tensor_info = TensorInfo(std::forward<Args>(args)...);
-        tensor_info.set_id(allocate_new_tensor_id());
+        register_new_tensor(tensor_info);
         return tensor_info;
     }
-    /** Create a @ref TensorInfo associated with the workload sketch by copying from an existing tensor info
-     * @note The newly copied tensor will have a different identity within the workload than the one copied from
-     *       To copy the identity of @p tensor_info as well, use @ref TensorInfo 's copy constructors instead
-     *
-     * @param[in] tensor_info @ref ITensorInfo to copy from
-     *
-     * @return TensorInfo   Newly created tensor info
-     */
-    TensorInfo create_tensor_info(const ITensorInfo &tensor_info);
     /** Create a default @ref TensorInfo associated with the workload sketch
      * It is usually used by user input or output tensors
      *
@@ -90,7 +81,11 @@ public:
     TensorInfo create_tensor_info();
 
 private:
-    ITensorInfo::Id                 allocate_new_tensor_id();
+    /** Register a new tensor by setting a new id to it and register its memory descriptor in the sketch
+     *
+     * @param[in,out] tensor_info @ref ITensorInfo that will be registered
+     */
+    void register_new_tensor(ITensorInfo &tensor_info);
     std::unique_ptr<Implementation> _impl; /**< Internal opaque implementation*/
 };