aboutsummaryrefslogtreecommitdiff
path: root/src/armnn
diff options
context:
space:
mode:
Diffstat (limited to 'src/armnn')
-rw-r--r--src/armnn/layers/DepthwiseConvolution2dLayer.cpp40
-rw-r--r--src/armnn/test/CreateWorkload.hpp20
-rw-r--r--src/armnn/test/OptimizerTests.cpp2
3 files changed, 34 insertions, 28 deletions
diff --git a/src/armnn/layers/DepthwiseConvolution2dLayer.cpp b/src/armnn/layers/DepthwiseConvolution2dLayer.cpp
index 95d4690d4f..c4edc2022f 100644
--- a/src/armnn/layers/DepthwiseConvolution2dLayer.cpp
+++ b/src/armnn/layers/DepthwiseConvolution2dLayer.cpp
@@ -24,7 +24,7 @@ DepthwiseConvolution2dLayer::DepthwiseConvolution2dLayer(const DepthwiseConvolut
{
}
-std::unique_ptr<IWorkload> DepthwiseConvolution2dLayer::CreateWorkload(const Graph& graph,
+std::unique_ptr<IWorkload> DepthwiseConvolution2dLayer::CreateWorkload(const Graph& graph,
const IWorkloadFactory& factory) const
{
// on this level constant data should not be released..
@@ -59,34 +59,40 @@ std::vector<TensorShape>
DepthwiseConvolution2dLayer::InferOutputShapes(const std::vector<TensorShape>& inputShapes) const
{
BOOST_ASSERT(inputShapes.size() == 2);
- const TensorShape& inputShape = inputShapes[0];
- const TensorShape filterShape = inputShapes[1];
+ const TensorShape& inputShape = inputShapes[0];
+ const TensorShape& filterShape = inputShapes[1];
BOOST_ASSERT_MSG(inputShape.GetNumDimensions() == 4, "Convolutions will always have 4D input.");
DataLayoutIndexed dataLayoutIndex(m_Param.m_DataLayout);
- unsigned int inWidth = inputShape[dataLayoutIndex.GetWidthIndex()];
- unsigned int inHeight = inputShape[dataLayoutIndex.GetHeightIndex()];
- unsigned int inBatchSize = inputShape[0];
+ unsigned int inputBatchSize = inputShape[0];
+ unsigned int inputHeight = inputShape[dataLayoutIndex.GetHeightIndex()];
+ unsigned int inputWidth = inputShape[dataLayoutIndex.GetWidthIndex()];
+ unsigned int inputChannels = inputShape[dataLayoutIndex.GetChannelsIndex()];
- unsigned int filterWidth = filterShape[dataLayoutIndex.GetWidthIndex()];
- unsigned int readWidth = (inWidth + m_Param.m_PadLeft + m_Param.m_PadRight) - (filterWidth);
- unsigned int outWidth = 1 + (readWidth / m_Param.m_StrideX);
+ // Expected filter shape: [ M, I, H, W ] - This shape does NOT depend on the data layout
+ // Namely: [ depth multiplier, input channels, filter height, filter width ]
+ // Output channels = input channels * depthMultiplier
- unsigned int filterHeight = filterShape[dataLayoutIndex.GetHeightIndex()];
- unsigned int readHeight = (inHeight + m_Param.m_PadTop + m_Param.m_PadBottom) - (filterHeight);
- unsigned int outHeight = 1 + (readHeight / m_Param.m_StrideY);
unsigned int depthMultiplier = filterShape[0];
- unsigned int outChannels = filterShape[dataLayoutIndex.GetChannelsIndex()] * depthMultiplier;
- unsigned int outBatchSize = inBatchSize;
+ unsigned int filterHeight = filterShape[2];
+ unsigned int readHeight = (inputHeight + m_Param.m_PadTop + m_Param.m_PadBottom) - filterHeight;
+ unsigned int outputHeight = 1 + (readHeight / m_Param.m_StrideY);
+
+ unsigned int filterWidth = filterShape[3];
+ unsigned int readWidth = (inputWidth + m_Param.m_PadLeft + m_Param.m_PadRight) - filterWidth;
+ unsigned int outputWidth = 1 + (readWidth / m_Param.m_StrideX);
+
+ unsigned int outputChannels = inputChannels * depthMultiplier;
+ unsigned int outputBatchSize = inputBatchSize;
TensorShape tensorShape = m_Param.m_DataLayout == armnn::DataLayout::NHWC ?
- TensorShape( { outBatchSize, outHeight, outWidth, outChannels } ) :
- TensorShape( { outBatchSize, outChannels, outHeight, outWidth });
+ TensorShape{ outputBatchSize, outputHeight, outputWidth, outputChannels } :
+ TensorShape{ outputBatchSize, outputChannels, outputHeight, outputWidth };
- return std::vector<TensorShape>({ tensorShape });
+ return std::vector<TensorShape>{ tensorShape };
}
void DepthwiseConvolution2dLayer::ValidateTensorShapesFromInputs()
diff --git a/src/armnn/test/CreateWorkload.hpp b/src/armnn/test/CreateWorkload.hpp
index 3dc18b9e18..f52f6055ca 100644
--- a/src/armnn/test/CreateWorkload.hpp
+++ b/src/armnn/test/CreateWorkload.hpp
@@ -414,18 +414,18 @@ std::unique_ptr<DepthwiseConvolution2dFloat32Workload> CreateDepthwiseConvolutio
{
// Creates the layer we're testing.
DepthwiseConvolution2dDescriptor layerDesc;
- layerDesc.m_PadLeft = 1;
- layerDesc.m_PadRight = 2;
- layerDesc.m_PadTop = 1;
- layerDesc.m_PadBottom = 2;
- layerDesc.m_StrideX = 1;
- layerDesc.m_StrideY = 1;
- layerDesc.m_BiasEnabled = false;
- layerDesc.m_DataLayout = dataLayout;
+ layerDesc.m_PadLeft = 1;
+ layerDesc.m_PadRight = 2;
+ layerDesc.m_PadTop = 1;
+ layerDesc.m_PadBottom = 2;
+ layerDesc.m_StrideX = 1;
+ layerDesc.m_StrideY = 1;
+ layerDesc.m_BiasEnabled = false;
+ layerDesc.m_DataLayout = dataLayout;
DepthwiseConvolution2dLayer* const layer = graph.AddLayer<DepthwiseConvolution2dLayer>(layerDesc, "layer");
- layer->m_Weight = std::make_unique<ScopedCpuTensorHandle>(TensorInfo({1, 4, 4, 2}, DataType));
+ layer->m_Weight = std::make_unique<ScopedCpuTensorHandle>(TensorInfo({1, 2, 4, 4}, DataType)); // [ M, I, H, W ]
layer->m_Weight->Allocate();
// Creates extra layers.
@@ -457,7 +457,7 @@ std::unique_ptr<DepthwiseConvolution2dFloat32Workload> CreateDepthwiseConvolutio
BOOST_TEST(queueDescriptor.m_Inputs.size() == 1);
BOOST_TEST(queueDescriptor.m_Outputs.size() == 1);
- BOOST_TEST((queueDescriptor.m_Weight->GetTensorInfo() == TensorInfo({1, 4, 4, 2}, DataType)));
+ BOOST_TEST((queueDescriptor.m_Weight->GetTensorInfo() == TensorInfo({1, 2, 4, 4}, DataType)));
// Returns so we can do extra, backend-specific tests.
return workload;
diff --git a/src/armnn/test/OptimizerTests.cpp b/src/armnn/test/OptimizerTests.cpp
index 29d1702c64..80addb4bfd 100644
--- a/src/armnn/test/OptimizerTests.cpp
+++ b/src/armnn/test/OptimizerTests.cpp
@@ -898,7 +898,7 @@ BOOST_AUTO_TEST_CASE(DepthwiseConv2dValidateTensorShapesFromInputsNhwc)
{
Graph graph;
const unsigned int inputShape[] = { 1, 3, 3, 2 };
- const unsigned int weightsShape[] = { 1, 3, 3, 2 };
+ const unsigned int weightsShape[] = { 1, 2, 3, 3 };
const unsigned int outputShape[] = { 1, 1, 1, 2 };
CreateDepthwiseConvolution2dGraph(graph, inputShape, weightsShape, outputShape, DataLayout::NHWC);
generated by cgit v1.2.1 (git 2.23.0) at 2024-07-07 00:45:17 +0000