Add multiple output support for dynamic fusion

* The dependency graph now can schedule any acyclic graph into
  a sequential list of operators. This is needed as the output
  operators now form branches in the graph.
* Fix the definition of input, output and intermediate tensors
  in GpuKernelComponentGroup to support non-linear but sequential
  list of operators.
* Add constraint on GpuOperatorGroup to enforce strictly linear
  fusion style, but allow output operator as the only form of
  branch.

Resolves: COMPMID-5771
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I68de3a31a2456145081f0a397e4e61dd66327682
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8823
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 5 insertions, 5 deletions

diff --git a/src/dynamic_fusion/sketch/gpu/GpuOperatorGroup.cpp b/src/dynamic_fusion/sketch/gpu/GpuOperatorGroup.cpp
index e8ef835405..7bb14c8698 100644
--- a/src/dynamic_fusion/sketch/gpu/GpuOperatorGroup.cpp
+++ b/src/dynamic_fusion/sketch/gpu/GpuOperatorGroup.cpp
@@ -68,12 +68,12 @@ ArgumentPack<ITensorInfo> Operator::tensors() const
     return _tensors;
 }
 
-bool GpuOperatorGroup::try_add_operator(const Operator &op) const
+bool GpuOperatorGroup::try_add_operator(const Operator &op, bool is_output) const
 {
     const auto src_tensor_ids = get_tensor_ids(op.tensors().get_const_src_tensors());
     const auto dst_tensor_ids = get_tensor_ids(op.tensors().get_const_dst_tensors());
     // Constraint 1
-    if(!_graph.try_add_operator_as_linear(op.id(), src_tensor_ids, dst_tensor_ids))
+    if(!_graph.try_add_operator_as_linear(op.id(), src_tensor_ids, dst_tensor_ids, is_output))
     {
         return false;
     }
@@ -143,12 +143,12 @@ bool GpuOperatorGroup::try_add_operator(const Operator &op) const
     }
     return true;
 }
-void GpuOperatorGroup::add_operator(const Operator &op)
+void GpuOperatorGroup::add_operator(const Operator &op, bool is_output)
 {
-    ARM_COMPUTE_ERROR_ON(!try_add_operator(op));
+    ARM_COMPUTE_ERROR_ON(!try_add_operator(op, is_output));
     const auto src_tensor_ids = get_tensor_ids(op.tensors().get_const_src_tensors());
     const auto dst_tensor_ids = get_tensor_ids(op.tensors().get_const_dst_tensors());
-    _graph.add_operator_as_linear(op.id(), src_tensor_ids, dst_tensor_ids);
+    _graph.add_operator_as_linear(op.id(), src_tensor_ids, dst_tensor_ids, is_output);
     _operators[op.id()] = op;
 }
 Operator GpuOperatorGroup::new_operator(const GpuOperatorType &operator_type, const ArgumentPack<ITensorInfo> &tensors) const