IVGCVSW-5085 Updates to CL and NEON TensorHandleFactory
* Update the CL and Neon TensorHandleFactories to not use SubTensors if
Axis is on x or y
Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: I782b89f50a92b21fdcbe68dab0281ad265fb3b63
diff --git a/src/backends/cl/ClTensorHandleFactory.cpp b/src/backends/cl/ClTensorHandleFactory.cpp
index 8af97f4..e92913f 100644
--- a/src/backends/cl/ClTensorHandleFactory.cpp
+++ b/src/backends/cl/ClTensorHandleFactory.cpp
@@ -36,6 +36,18 @@
const arm_compute::TensorShape parentShape = armcomputetensorutils::BuildArmComputeTensorShape(
parent.GetShape());
+
+ // In order for ACL to support subtensors the concat axis cannot be on x or y and the values of x and y
+ // must match the parent shapes
+ if (coords.x() != 0 || coords.y() != 0)
+ {
+ return nullptr;
+ }
+ if ((parentShape.x() != shape.x()) || (parentShape.y() != shape.y()))
+ {
+ return nullptr;
+ }
+
if (!::arm_compute::error_on_invalid_subtensor(__func__, __FILE__, __LINE__, parentShape, coords, shape))
{
return nullptr;