IVGCVSW-1922 Unit test for DepthwiseConvolution with NHWC

Change-Id: I3e6e5b9a62f30d03c05bd7178adea8f4c8275da8

3 files changed, 37 insertions, 2 deletions

diff --git a/src/backends/cl/test/ClCreateWorkloadTests.cpp b/src/backends/cl/test/ClCreateWorkloadTests.cpp
index 67f3e3c5bb..66c2c2aa40 100644
--- a/src/backends/cl/test/ClCreateWorkloadTests.cpp
+++ b/src/backends/cl/test/ClCreateWorkloadTests.cpp
@@ -243,6 +243,36 @@ BOOST_AUTO_TEST_CASE(CreateConvolution2dFloat16NhwcWorkload)
     ClConvolution2dWorkloadTest<ClConvolution2dWorkload, armnn::DataType::Float16>(DataLayout::NHWC);
 }
 
+template <typename DepthwiseConvolutionWorkloadType, typename armnn::DataType DataType>
+static void ClDepthwiseConvolutionWorkloadTest(DataLayout dataLayout)
+{
+    Graph graph;
+    ClWorkloadFactory factory;
+
+    auto workload = CreateDepthwiseConvolution2dWorkloadTest<DepthwiseConvolutionWorkloadType, DataType>
+                    (factory, graph, dataLayout);
+
+    // Checks that inputs/outputs are as we expect them (see definition of CreateDepthwiseConvolution2dWorkloadTest).
+    DepthwiseConvolution2dQueueDescriptor queueDescriptor = workload->GetData();
+    auto inputHandle  = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Inputs[0]);
+    auto outputHandle = boost::polymorphic_downcast<IClTensorHandle*>(queueDescriptor.m_Outputs[0]);
+
+    std::initializer_list<unsigned int> inputShape  = (dataLayout == DataLayout::NCHW)
+            ? std::initializer_list<unsigned int>({ 2, 2, 5, 5 })
+            : std::initializer_list<unsigned int>({ 2, 5, 5, 2 });
+    std::initializer_list<unsigned int> outputShape = (dataLayout == DataLayout::NCHW)
+            ? std::initializer_list<unsigned int>({ 2, 2, 5, 5 })
+            : std::initializer_list<unsigned int>({ 2, 5, 5, 2 });
+
+    BOOST_TEST(CompareIClTensorHandleShape(inputHandle, inputShape));
+    BOOST_TEST(CompareIClTensorHandleShape(outputHandle, outputShape));
+}
+
+BOOST_AUTO_TEST_CASE(CreateDepthwiseConvolutionFloat32NhwcWorkload)
+{
+    ClDepthwiseConvolutionWorkloadTest<ClDepthwiseConvolutionWorkload, DataType::Float32>(DataLayout::NHWC);
+}
+
 template <typename Convolution2dWorkloadType, typename armnn::DataType DataType>
 static void ClDirectConvolution2dWorkloadTest()
 {
diff --git a/src/backends/cl/test/ClLayerTests.cpp b/src/backends/cl/test/ClLayerTests.cpp
index cea7470b76..9934c927e6 100644..100755
--- a/src/backends/cl/test/ClLayerTests.cpp
+++ b/src/backends/cl/test/ClLayerTests.cpp
@@ -65,6 +65,7 @@ ARMNN_AUTO_TEST_CASE(SimpleConvolution2dSquareNhwc, SimpleConvolution2d3x3NhwcTe
 
 // Depthwise Convolution
 ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, true)
+ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthNhwc, DepthwiseConvolution2dDepthNhwcTest, false)
 ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1, DepthwiseConvolution2dDepthMul1Test, false)
 ARMNN_AUTO_TEST_CASE(DepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, true)
 ARMNN_AUTO_TEST_CASE(UnbiasedDepthwiseConvolution2dDepthMul1Uint8, DepthwiseConvolution2dDepthMul1Uint8Test, false)
diff --git a/src/backends/cl/workloads/ClDepthwiseConvolutionWorkload.cpp b/src/backends/cl/workloads/ClDepthwiseConvolutionWorkload.cpp
index 142cbc230f..6fa9ddc6b0 100644
--- a/src/backends/cl/workloads/ClDepthwiseConvolutionWorkload.cpp
+++ b/src/backends/cl/workloads/ClDepthwiseConvolutionWorkload.cpp
@@ -60,12 +60,12 @@ ClDepthwiseConvolutionWorkload::ClDepthwiseConvolutionWorkload(
     auto& weightInfo = m_Data.m_Weight->GetTensorInfo();
 
     m_KernelTensor = std::make_unique<arm_compute::CLTensor>();
-    BuildArmComputeTensor(*m_KernelTensor, weightInfo);
+    BuildArmComputeTensor(*m_KernelTensor, weightInfo, m_Data.m_Parameters.m_DataLayout);
 
     if (m_Data.m_Parameters.m_BiasEnabled)
     {
         m_BiasTensor = std::make_unique<arm_compute::CLTensor>();
-        BuildArmComputeTensor(*m_BiasTensor, m_Data.m_Bias->GetTensorInfo());
+        BuildArmComputeTensor(*m_BiasTensor, m_Data.m_Bias->GetTensorInfo(), m_Data.m_Parameters.m_DataLayout);
     }
 
     arm_compute::PadStrideInfo padStrideInfo(m_Data.m_Parameters.m_StrideX,
@@ -82,6 +82,10 @@ ClDepthwiseConvolutionWorkload::ClDepthwiseConvolutionWorkload(
     arm_compute::ICLTensor& input  = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
     arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
 
+    arm_compute::DataLayout aclDataLayout = ConvertDataLayout(m_Data.m_Parameters.m_DataLayout);
+    input.info()->set_data_layout(aclDataLayout);
+    output.info()->set_data_layout(aclDataLayout);
+
     const unsigned int depthMultiplier = weightInfo.GetShape()[0];
 
     //Check for optimisation opportunities.