IVGCVSW-2043 - Merger using ACL for innermost concat axis
* Add ClMergerWorkload and NeonMergerWorkload to call ACL for innermost concat axis
* Modify layer support to call ClMergerWorkloadValidate and NeonMergerWorkloadValidate when concat axis is inner most
* Add m_ConcatAxis to MergerDescriptor
* Modify MergerQueueDescriptor::Validate to check sub tensor only when using subtensor
!android-nn-driver:166
Change-Id: I56676b43964c8d6d726387b41b3cc34a512c0f0a
diff --git a/src/backends/neon/workloads/NeonMergerWorkload.hpp b/src/backends/neon/workloads/NeonMergerWorkload.hpp
index 7103d8a..a4f36d1 100644
--- a/src/backends/neon/workloads/NeonMergerWorkload.hpp
+++ b/src/backends/neon/workloads/NeonMergerWorkload.hpp
@@ -6,18 +6,26 @@
#pragma once
#include <backendsCommon/Workload.hpp>
+#include <neon/workloads/NeonWorkloadUtils.hpp>
namespace armnn
{
+arm_compute::Status NeonMergerWorkloadValidate(const std::vector<const TensorInfo*>& inputs,
+ const TensorInfo& output,
+ const MergerDescriptor& descriptor);
+
class NeonMergerWorkload : public BaseWorkload<MergerQueueDescriptor>
{
public:
- using BaseWorkload<MergerQueueDescriptor>::BaseWorkload;
+ NeonMergerWorkload(const MergerQueueDescriptor& descriptor, const WorkloadInfo& info);
- virtual void Execute() const override
- {
- // With subtensors, merger is a no-op.
- }
+ using BaseWorkload<MergerQueueDescriptor>::BaseWorkload;
+ void Execute() const override;
+
+private:
+ mutable arm_compute::NEConcatenateLayer m_Layer;
+ bool m_Execute;
+
};
} //namespace armnn