IVGCVSW-3212 Refactor the Reference BatchNormalization workloads to
handle Float32 and QAsymm8 types

 * Removed the type-specific workload implementations
 * Added type-independent RefBatchNormalizationWorkload implementation
 * Reworked BachNormImpl to use decoders/encoders
 * Improved the validation of the BatchNorm queue descriptor
 * Fixed unit tests where necessary

Change-Id: Icf3fa1332292d38ec2fa0b1cb984cab78426034b
Signed-off-by: Matteo Martincigh <matteo.martincigh@arm.com>
diff --git a/src/backends/reference/workloads/BatchNormImpl.cpp b/src/backends/reference/workloads/BatchNormImpl.cpp
new file mode 100644
index 0000000..36e96d3
--- /dev/null
+++ b/src/backends/reference/workloads/BatchNormImpl.cpp
@@ -0,0 +1,82 @@
+//
+// Copyright © 2017 Arm Ltd. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "BatchNormImpl.hpp"
+#include "RefWorkloadUtils.hpp"
+
+#include <armnn/Tensor.hpp>
+
+#include <DataLayoutIndexed.hpp>
+
+#include <cmath>
+
+namespace armnn
+{
+
+void BatchNormImpl(const BatchNormalizationQueueDescriptor& data,
+                   Decoder<float>& meanDecoder,
+                   Decoder<float>& varianceDecoder,
+                   Decoder<float>& betaDecoder,
+                   Decoder<float>& gammaDecoder,
+                   Decoder<float>& inputDecoder,
+                   Encoder<float>& outputEncoder)
+{
+    const TensorInfo& inputInfo = GetTensorInfo(data.m_Inputs[0]);
+    const TensorShape inputShape = inputInfo.GetShape();
+
+    armnnUtils::DataLayoutIndexed dataLayout(data.m_Parameters.m_DataLayout);
+
+    unsigned int inputBatches  = inputShape[0];
+    unsigned int inputHeight   = inputShape[dataLayout.GetHeightIndex()];
+    unsigned int inputWidth    = inputShape[dataLayout.GetWidthIndex()];
+    unsigned int inputChannels = inputShape[dataLayout.GetChannelsIndex()];
+
+    for (unsigned int c = 0; c < inputChannels; c++)
+    {
+        meanDecoder[c];
+        varianceDecoder[c];
+        betaDecoder[c];
+        gammaDecoder[c];
+        float mean  = meanDecoder.Get();
+        float var   = varianceDecoder.Get();
+        float beta  = betaDecoder.Get();
+        float gamma = gammaDecoder.Get();
+
+        float mult = gamma / sqrtf(var + data.m_Parameters.m_Eps);
+        float add  = beta - mult * mean;
+
+        for (unsigned int n = 0; n < inputBatches; n++)
+        {
+            for (unsigned int h = 0; h < inputHeight; h++)
+            {
+                for (unsigned int w = 0; w < inputWidth; w++)
+                {
+                    unsigned int index = 0;
+
+                    if (dataLayout == DataLayout::NHWC)
+                    {
+                        index = n * inputHeight * inputWidth * inputChannels +
+                                h * inputWidth * inputChannels +
+                                w * inputChannels +
+                                c;
+                    }
+                    else // dataLayout == DataLayout::NCHW
+                    {
+                        index = n * inputHeight * inputWidth * inputChannels +
+                                c * inputHeight * inputWidth +
+                                h * inputWidth +
+                                w;
+                    }
+
+                    inputDecoder[index];
+                    outputEncoder[index];
+                    outputEncoder.Set(mult * inputDecoder.Get() + add);
+                }
+            }
+        }
+    }
+}
+
+} // namespace armnn