Remove map/unmap overhead for input/output accessor when using DummyAccessor

Don't map/unmap when the tensor data is not accessed in the input or output accessor.
This is so to avoid measuring CPU overhead when benchmarking on the GPU backend.

Resolve COMPMID-4712

Change-Id: I5baba1b93e7a51fe13525bcce6c0cfdecb14493e
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6140
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 11 insertions, 0 deletions

diff --git a/src/graph/Utils.cpp b/src/graph/Utils.cpp
index 37af1bff93..ceb5ba95de 100644
--- a/src/graph/Utils.cpp
+++ b/src/graph/Utils.cpp
@@ -116,6 +116,17 @@ void release_default_graph_context(GraphContext &ctx)
     }
 }
 
+void sync_backends()
+{
+    for(const auto &backend : backends::BackendRegistry::get().backends())
+    {
+        if(backend.second->is_backend_supported())
+        {
+            backend.second->sync();
+        }
+    }
+}
+
 void setup_requested_backend_context(GraphContext &ctx, Target target)
 {
     if(backends::BackendRegistry::get().contains(target))