MLCE-143 Fixed driver crash during CTS tests

MLCE-144 Fix cts MAX_POOL_2D_V1_0 tests

 * Only apply the Optimization when the base ReshapeLayer is connected to
   the child ReshapeLayer and no other Layer.

Signed-off-by: Finn Williams <Finn.Williams@arm.com>
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Id1215e8b1c06d7bdb77905fec9649a8ec26436f0

3 files changed, 40 insertions, 17 deletions

diff --git a/src/armnn/optimizations/OptimizeConsecutiveReshapes.hpp b/src/armnn/optimizations/OptimizeConsecutiveReshapes.hpp
index 5047d5d678..e2d4a2dcc3 100644
--- a/src/armnn/optimizations/OptimizeConsecutiveReshapes.hpp
+++ b/src/armnn/optimizations/OptimizeConsecutiveReshapes.hpp
@@ -14,7 +14,7 @@ namespace optimizations
 class OptimizeConsecutiveReshapesImpl
 {
 public:
-    /// Run for every connection between a base RashapeLayer and a child ReshapeLayer.
+    /// Run for every connection between a base ReshapeLayer and a child ReshapeLayer.
     /// Inserts an equivalent ReshapeLayer that bypasses both for that connection.
     void Run(Graph& graph, InputSlot& connection) const
     {
@@ -29,12 +29,20 @@ public:
         const TensorInfo& inInfo = parentOut->GetTensorInfo();
         const TensorInfo& outInfo = child.GetOutputHandler().GetTensorInfo();
 
+        // This Optimization is only appropriate when the base ReshapeLayer is connected to the child ReshapeLayer
+        // and no other Layer.
+        if (base.GetOutputSlot(0).GetNumConnections() > 1)
+        {
+            return;
+        }
+
         if (inInfo.GetShape() != outInfo.GetShape())
         {
             // Inserts equivalent reshape before base layer.
             const std::string name = std::string("merged-") + base.GetName() + std::string("-with-") + child.GetName();
             const ReshapeDescriptor descriptor{outInfo.GetShape()};
             auto& newReshape = *graph.InsertNewLayer<ReshapeLayer>(base.GetInputSlot(0), descriptor, name.c_str());
+
             // Sets tensor info for new layer.
             newReshape.GetOutputHandler().SetTensorInfo(outInfo);
             // Reconnects base with original parent.
diff --git a/src/backends/backendsCommon/test/Pooling2dTestImpl.hpp b/src/backends/backendsCommon/test/Pooling2dTestImpl.hpp
index 5edf9c802f..da80d64f8d 100644
--- a/src/backends/backendsCommon/test/Pooling2dTestImpl.hpp
+++ b/src/backends/backendsCommon/test/Pooling2dTestImpl.hpp
@@ -223,17 +223,17 @@ LayerTestResult<T, 4> SimpleMaxPooling2dSize3x3Stride2x4TestCommon(
                 0.0f, 9.0f, 7.0f, 9.0f, 9.0f, 3.0f,
                 0.0f, 8.0f, 9.0f, 9.0f, 9.0f, 9.0f,
 
-                0.0f, 0.0f, 0.0f, 0.0f,-3.0f, 0.0f,
-                0.0f,-1.0f, 0.0f, 0.0f, 0.0f, 0.0f,
-                0.0f,-1.0f,-1.0f,-1.0f,-1.0f, 0.0f,
+                0.0f, 0.0f, 0.0f, 0.0f,-3.0f,-3.0f,
+                0.0f,-1.0f, 0.0f, 0.0f, 0.0f,-2.0f,
+                0.0f,-1.0f,-1.0f,-1.0f,-1.0f,-1.0f,
 
                 0.0f, 8.0f, 8.0f, 8.0f, 8.0f, 8.0f,
                 0.0f, 9.0f, 7.0f, 9.0f, 9.0f, 3.0f,
                 0.0f, 8.0f, 9.0f, 9.0f, 9.0f, 9.0f,
 
-                0.0f, 0.0f, 0.0f, 0.0f,-3.0f, 0.0f,
-                0.0f,-1.0f, 0.0f, 0.0f, 0.0f, 0.0f,
-                0.0f,-1.0f,-1.0f,-1.0f,-1.0f, 0.0f
+                0.0f, 0.0f, 0.0f, 0.0f,-3.0f,-3.0f,
+                0.0f,-1.0f, 0.0f, 0.0f, 0.0f,-2.0f,
+                0.0f,-1.0f,-1.0f,-1.0f,-1.0f,-1.0f
         }));
     }
 
diff --git a/src/backends/reference/workloads/Pooling2d.cpp b/src/backends/reference/workloads/Pooling2d.cpp
index f2532cac03..cf83f8ce2b 100644
--- a/src/backends/reference/workloads/Pooling2d.cpp
+++ b/src/backends/reference/workloads/Pooling2d.cpp
@@ -107,9 +107,9 @@ namespace
         }
     }
 
-    bool OnPaddingOnly(int start, int end, int maxRange, int padding)
+    bool OnPaddingOnly(int start, int end, int maxRange)
     {
-        if (end <= 0 || start > (maxRange - padding))
+        if (end <= 0 || start > maxRange)
         {
             return true;
         }
@@ -187,34 +187,49 @@ void Pooling2d(Decoder<float>& rInputDecoder,
         {
             for (int yOutput = 0; yOutput < heightOutput; yOutput++)
             {
+                //  Calculate values independent of the x axis
+                int hstart = (yOutput * strideY) - padTop;
+                int hend = hstart + poolHeight;
+                // Clamp the pooling region inside the valid input area (which includes the padding).
+                // This is necessary because the final pooling in a row may overlap beyond the padding.
+                hend = std::min(hend, heightInput + padBottom);
+
+                int height = hend - hstart;
+                bool hclamped = ClampRange(hstart, hend, heightInput);
+
                 for (int xOutput = 0; xOutput < widthOutput; xOutput++)
                 {
-                    int hstart = (yOutput * strideY) - padTop;
                     int wstart = (xOutput * strideX) - padLeft;
-                    int hend = hstart + poolHeight;
                     int wend = wstart + poolWidth;
 
                     // Clamp the pooling region inside the valid input area (which includes the padding).
                     // This is necessary because the final pooling in a row may overlap beyond the padding.
-                    hend = std::min(hend, heightInput + padBottom);
                     wend = std::min(wend, widthInput + padRight);
 
                     float result = defaultInitializer;
-                    float poolAreaSize = boost::numeric_cast<float>((hend - hstart) * (wend - wstart));
+                    float poolAreaSize = boost::numeric_cast<float>(height * (wend - wstart));
 
                     // Special case: when the pooling kernel is over a padding region and the padding
                     //               size is larger or equal to the kernel and the kernel only covers
                     //               padding and no real values, then we initialize the result as zero
                     //               by convention. This is because we need to choose a value here and
                     //               all values we have are padding, which we ignore.
-                    if (OnPaddingOnly(hstart, hend, heightInput, padBottom) ||
-                        OnPaddingOnly(wstart, wend, widthInput, padRight))
+                    if (OnPaddingOnly(hstart, hend, heightInput) ||
+                        OnPaddingOnly(wstart, wend, widthInput))
                     {
                         result = 0.0f;
+
+                        unsigned int outputIndex = dataLayout.GetIndex(outputShape,
+                                                                       boost::numeric_cast<unsigned int>(n),
+                                                                       boost::numeric_cast<unsigned int>(c),
+                                                                       boost::numeric_cast<unsigned int>(yOutput),
+                                                                       boost::numeric_cast<unsigned int>(xOutput));
+                        rOutputEncoder[outputIndex];
+                        rOutputEncoder.Set(result);
+                        continue;
                     }
 
-                    bool clamped = ClampRange(wstart, wend, widthInput);
-                    clamped |= ClampRange(hstart, hend, heightInput);
+                    bool clamped = hclamped |= ClampRange(wstart, wend, widthInput);
 
                     if (clamped && params.m_PaddingMethod == PaddingMethod::Exclude)
                     {