Enable lazy tensor allocation

- The previous ref_model was allocating the memory for all tensors in
the graph upfront which is unnecessary and wasteful.
- This patch changes to only allocate initial input tensors on the entry
point using the allocateInputTensor() function.
- The output tensors are ensured to have been allocated before executing
a node. The output tenosrs are the inputs for the next node.
- When a node's evaluation is finished, its input tensors will be freed
if they will no longer be used by anyone else.

Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ibb3e8c9e6344f6cd9eb20532a03b2097b93247f9

1 files changed, 2 insertions, 1 deletions

diff --git a/reference_model/src/ops/control_flow.cc b/reference_model/src/ops/control_flow.cc
index 0afb7e2..6bbc587 100644
--- a/reference_model/src/ops/control_flow.cc
+++ b/reference_model/src/ops/control_flow.cc
@@ -43,7 +43,8 @@ int OpControlFlow::evalBlock(TosaSerializationBasicBlock* block,
     ERROR_IF(block_sgt.linkTensorsAndNodes(), "evalBlock(): Failed to link tensors and nodes for %s",
              block_name.c_str());
     ERROR_IF(block_sgt.validateGraph(), "evalBlock(): Failed to validate subgraph for %s", block_name.c_str());
-    ERROR_IF(block_sgt.allocateTensor(), "evalBlock(): Failed to allocate tensor for %s", block_name.c_str());
+    ERROR_IF(block_sgt.allocateInputTensors(), "evalBlock(): Failed to allocate input tensors for %s",
+             block_name.c_str());
 
     int num_input_tensors  = block_sgt.getNumInputTensors();
     int num_output_tensors = block_sgt.getNumOutputTensors();