IVGCVSW-6873 Import inputs but don't export outputs fails.

Only one bool is used to indicate whether inputs should be imported.
However, its possible for the user to want to import inputs but not
export outputs. In addition it's possible for a user to enabled import
during optimize but then pass a memory source that does not require
import.

* Add m_ExportEnabled to INetwork.hpp.
* Modify Network::dNetwork to consider both m_ImportEnabled
  and m_ExportEnabled.
* Add ValidateSourcesMatchOptimizedNetwork to LoadedNetwork to validate
  import options between optimize and network load.
* Update the TfLite delegate consider exportEnabled flag in the
  optimizer.

!armnn-internal-tests:425350
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: I776eab81595898e43f91ab40306962eae61329f4

9 files changed, 45 insertions, 11 deletions

diff --git a/src/backends/backendsCommon/test/CompatibilityTests.cpp b/src/backends/backendsCommon/test/CompatibilityTests.cpp
index c69a4b5f91..9c85ffcfc3 100644
--- a/src/backends/backendsCommon/test/CompatibilityTests.cpp
+++ b/src/backends/backendsCommon/test/CompatibilityTests.cpp
@@ -73,7 +73,7 @@ TEST_CASE("Neon_Cl_DirectCompatibility_Test")
     graph.TopologicalSort();
 
     std::vector<std::string> errors;
-    auto result = SelectTensorHandleStrategy(graph, backends, registry, true, errors);
+    auto result = SelectTensorHandleStrategy(graph, backends, registry, true, true, errors);
 
     CHECK(result.m_Error == false);
     CHECK(result.m_Warning == false);
diff --git a/src/backends/backendsCommon/test/EndToEndTestImpl.hpp b/src/backends/backendsCommon/test/EndToEndTestImpl.hpp
index 77901df444..cc5aa23ca3 100644
--- a/src/backends/backendsCommon/test/EndToEndTestImpl.hpp
+++ b/src/backends/backendsCommon/test/EndToEndTestImpl.hpp
@@ -204,7 +204,9 @@ inline void ImportNonAlignedInputPointerTest(std::vector<BackendId> backends)
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
     // Optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
     CHECK(optNet);
 
     // Loads it into the runtime.
@@ -269,7 +271,10 @@ inline void ExportNonAlignedOutputPointerTest(std::vector<BackendId> backends)
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
     // Optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    optimizedOptions.m_ExportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
     CHECK(optNet);
 
     // Loads it into the runtime.
@@ -340,7 +345,10 @@ inline void ImportAlignedPointerTest(std::vector<BackendId> backends)
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
     // Optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    optimizedOptions.m_ExportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
     CHECK(optNet);
 
     // Loads it into the runtime.
@@ -424,7 +432,9 @@ inline void ImportOnlyWorkload(std::vector<BackendId> backends)
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
     // optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
 
     INFO("Load Network");
     // Load it into the runtime. It should pass.
@@ -514,7 +524,9 @@ inline void ExportOnlyWorkload(std::vector<BackendId> backends)
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
     // optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ExportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
 
     INFO("Load Network");
     // Load it into the runtime. It should pass.
@@ -601,7 +613,10 @@ inline void ImportAndExportWorkload(std::vector<BackendId> backends)
     input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32, 0.0f, 0, true));
     pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32));
 
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    optimizedOptions.m_ExportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
 
     INFO("Load Network");
     // Load it into the runtime. It should pass.
@@ -694,7 +709,10 @@ inline void ExportOutputWithSeveralOutputSlotConnectionsTest(std::vector<Backend
     activation->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 4, 1 }, DataType::Float32));
 
     // Optimize the network
-    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec());
+    OptimizerOptions optimizedOptions;
+    optimizedOptions.m_ImportEnabled = true;
+    optimizedOptions.m_ExportEnabled = true;
+    IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optimizedOptions);
 
     // Loads it into the runtime.
     NetworkId netId;
diff --git a/src/backends/backendsCommon/test/OptimizedNetworkTests.cpp b/src/backends/backendsCommon/test/OptimizedNetworkTests.cpp
index bcea0610db..cd865def71 100644
--- a/src/backends/backendsCommon/test/OptimizedNetworkTests.cpp
+++ b/src/backends/backendsCommon/test/OptimizedNetworkTests.cpp
@@ -421,7 +421,7 @@ TEST_CASE("OptimizeNetworkCopy")
 
     std::vector<armnn::BackendId> preferredBackends { "CpuRef" };
     armnn::ModelOptions modelOptions;
-    armnn::OptimizerOptions optimizerOptions(false, false, false, false, modelOptions);
+    armnn::OptimizerOptions optimizerOptions(false, false, false, false, modelOptions, false);
     std::vector<std::string> errorMessages;
 
     // optimize the network.
diff --git a/src/backends/cl/test/ClCustomAllocatorTests.cpp b/src/backends/cl/test/ClCustomAllocatorTests.cpp
index 139e688dc2..251c98fcad 100644
--- a/src/backends/cl/test/ClCustomAllocatorTests.cpp
+++ b/src/backends/cl/test/ClCustomAllocatorTests.cpp
@@ -120,6 +120,7 @@ TEST_CASE("ClCustomAllocatorTest")
     // Optimise ArmNN network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     armnn::IOptimizedNetworkPtr optNet = Optimize(*myNetwork, {"GpuAcc"}, run->GetDeviceSpec(), optOptions);
     CHECK(optNet);
 
diff --git a/src/backends/cl/test/ClFallbackTests.cpp b/src/backends/cl/test/ClFallbackTests.cpp
index 6ac94337ba..51a983a681 100644
--- a/src/backends/cl/test/ClFallbackTests.cpp
+++ b/src/backends/cl/test/ClFallbackTests.cpp
@@ -50,6 +50,7 @@ TEST_CASE("ClImportEnabledFallbackToNeon")
     // optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -330,6 +331,7 @@ TEST_CASE("ClImportEnabledFallbackSubgraphToNeon")
     // optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
diff --git a/src/backends/cl/test/ClImportTensorHandleTests.cpp b/src/backends/cl/test/ClImportTensorHandleTests.cpp
index 20537b3c81..9a075d2b7d 100644
--- a/src/backends/cl/test/ClImportTensorHandleTests.cpp
+++ b/src/backends/cl/test/ClImportTensorHandleTests.cpp
@@ -142,6 +142,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClImportEndToEnd")
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -338,6 +339,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportConv2dEndToEnd")
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(*network, backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -470,6 +472,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportConvertFp16toFp32EndToE
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(network.GetGraph(), backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -613,6 +616,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportConvertFp32toFp16EndToE
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(network.GetGraph(), backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -747,6 +751,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportSimpleConvertFp32toFp16
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(network.GetGraph(), backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -896,6 +901,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportRepeatedInferencesEndTo
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(*network, backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
@@ -1117,6 +1123,7 @@ TEST_CASE_FIXTURE(ClContextControlFixture, "ClForceImportRepeatedInferencesInver
     // Optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = false;
+    optOptions.m_ExportEnabled = false;
     std::vector<armnn::BackendId> backends = {armnn::Compute::GpuAcc};
     IOptimizedNetworkPtr optNet = Optimize(*network, backends, runtime->GetDeviceSpec(), optOptions);
     CHECK(optNet);
diff --git a/src/backends/cl/test/ClOptimizedNetworkTests.cpp b/src/backends/cl/test/ClOptimizedNetworkTests.cpp
index cf17eae208..6648759a9a 100644
--- a/src/backends/cl/test/ClOptimizedNetworkTests.cpp
+++ b/src/backends/cl/test/ClOptimizedNetworkTests.cpp
@@ -130,7 +130,7 @@ TEST_CASE("FastMathEnabledTestOnGpuAcc")
 
     auto modelOptionsOut = GetModelOptionsForTesting(optimizedNet.get());
 
-    CHECK(modelOptionsOut.size() == 1);
+    CHECK(modelOptionsOut.size() == 2); // FastMathEnabled and the Global to hold the import export values.
     CHECK(modelOptionsOut[0].GetOption(0).GetName() == "FastMathEnabled");
     CHECK(modelOptionsOut[0].GetOption(0).GetValue().AsBool() == true);
 }
diff --git a/src/backends/neon/test/NeonFallbackTests.cpp b/src/backends/neon/test/NeonFallbackTests.cpp
index d2de843fd9..8e0e0ab99b 100644
--- a/src/backends/neon/test/NeonFallbackTests.cpp
+++ b/src/backends/neon/test/NeonFallbackTests.cpp
@@ -60,6 +60,7 @@ TEST_CASE("FallbackImportToCpuAcc")
     std::vector<BackendId> backends = { "MockRef", Compute::CpuAcc };
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -203,6 +204,7 @@ TEST_CASE("FallbackPaddingCopyToCpuAcc")
     std::vector<BackendId> backends = { "MockRef", Compute::CpuAcc };
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -338,6 +340,7 @@ TEST_CASE("FallbackImportFromCpuAcc")
     std::vector<BackendId> backends = { "MockRef", Compute::CpuAcc };
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -482,6 +485,7 @@ TEST_CASE("FallbackPaddingCopyFromCpuAcc")
     std::vector<BackendId> backends = { "MockRef", Compute::CpuAcc };
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -746,6 +750,7 @@ TEST_CASE("NeonImportEnabledFallbackToCl")
     // optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
@@ -1037,6 +1042,7 @@ TEST_CASE("NeonImportEnabledFallbackSubgraphToCl")
     // optimize the network
     OptimizerOptions optOptions;
     optOptions.m_ImportEnabled = true;
+    optOptions.m_ExportEnabled = true;
     IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec(), optOptions);
 
     Graph& graph = GetGraphForTesting(optNet.get());
diff --git a/src/backends/neon/test/NeonOptimizedNetworkTests.cpp b/src/backends/neon/test/NeonOptimizedNetworkTests.cpp
index 9b448b270d..dcda9bfd07 100644
--- a/src/backends/neon/test/NeonOptimizedNetworkTests.cpp
+++ b/src/backends/neon/test/NeonOptimizedNetworkTests.cpp
@@ -106,7 +106,7 @@ TEST_CASE("FastMathEnabledTestOnCpuAcc")
 
     auto modelOptionsOut = GetModelOptionsForTesting(optimizedNet.get());
 
-    CHECK(modelOptionsOut.size() == 1);
+    CHECK(modelOptionsOut.size() == 2); // FastMathEnabled and the Global to hold the import export values.
     CHECK(modelOptionsOut[0].GetOption(0).GetName() == "FastMathEnabled");
     CHECK(modelOptionsOut[0].GetOption(0).GetValue().AsBool() == true);
 }