aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--arm_compute/graph/GraphBuilder.h10
-rw-r--r--arm_compute/graph/Types.h1
-rw-r--r--arm_compute/graph/backends/FunctionHelpers.h35
-rw-r--r--arm_compute/graph/backends/ValidateHelpers.h23
-rw-r--r--arm_compute/graph/frontend/Layers.h25
-rw-r--r--arm_compute/graph/nodes/Nodes.h1
-rw-r--r--arm_compute/graph/nodes/ReorgLayerNode.h67
-rw-r--r--src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp2
-rw-r--r--src/graph/GraphBuilder.cpp5
-rw-r--r--src/graph/backends/CL/CLFunctionsFactory.cpp2
-rw-r--r--src/graph/backends/CL/CLNodeValidator.cpp2
-rw-r--r--src/graph/backends/NEON/NEFunctionFactory.cpp2
-rw-r--r--src/graph/backends/NEON/NENodeValidator.cpp2
-rw-r--r--src/graph/nodes/ReorgLayerNode.cpp97
-rw-r--r--src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp26
15 files changed, 293 insertions, 7 deletions
diff --git a/arm_compute/graph/GraphBuilder.h b/arm_compute/graph/GraphBuilder.h
index 43670f4022..d7502600a4 100644
--- a/arm_compute/graph/GraphBuilder.h
+++ b/arm_compute/graph/GraphBuilder.h
@@ -261,6 +261,16 @@ public:
* @return Node ID of the created node, EmptyNodeID in case of error
*/
static NodeID add_pooling_node(Graph &g, NodeParams params, NodeIdxPair input, PoolingLayerInfo pool_info);
+ /** Adds a reorg layer node to the graph
+ *
+ * @param[in] g Graph to add the node to
+ * @param[in] params Common node parameters
+ * @param[in] input Input to the reorg layer node as a NodeID-Index pair
+ * @param[in] stride Stride value to use for reorganizing the values in the output tensor.
+ *
+ * @return Node ID of the created node, EmptyNodeID in case of error
+ */
+ static NodeID add_reorg_node(Graph &g, NodeParams params, NodeIdxPair input, int stride);
/** Adds a reshape layer node to the graph
*
* @param[in] g Graph to add the node to
diff --git a/arm_compute/graph/Types.h b/arm_compute/graph/Types.h
index 2df9998fcf..0b9a953531 100644
--- a/arm_compute/graph/Types.h
+++ b/arm_compute/graph/Types.h
@@ -143,6 +143,7 @@ enum class NodeType
NormalizationLayer,
PermuteLayer,
PoolingLayer,
+ ReorgLayer,
ReshapeLayer,
ResizeLayer,
SoftmaxLayer,
diff --git a/arm_compute/graph/backends/FunctionHelpers.h b/arm_compute/graph/backends/FunctionHelpers.h
index bf458ae33f..b1446ea493 100644
--- a/arm_compute/graph/backends/FunctionHelpers.h
+++ b/arm_compute/graph/backends/FunctionHelpers.h
@@ -695,6 +695,41 @@ std::unique_ptr<IFunction> create_pooling_layer(PoolingLayerNode &node)
return std::move(func);
}
+/** Create a backend reorg layer function
+ *
+ * @tparam ReorgLayerFunction Backend reshape function
+ * @tparam TargetInfo Target-specific information
+ *
+ * @param[in] node Node to create the backend function for
+ *
+ * @return Backend reshape layer function
+ */
+template <typename ReorgLayerFunction, typename TargetInfo>
+std::unique_ptr<IFunction> create_reorg_layer(ReorgLayerNode &node)
+{
+ validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */);
+
+ // Extract IO and info
+ typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0));
+ typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0));
+ ARM_COMPUTE_ERROR_ON(input == nullptr);
+ ARM_COMPUTE_ERROR_ON(output == nullptr);
+
+ // Create and configure function
+ auto func = support::cpp14::make_unique<ReorgLayerFunction>();
+ func->configure(input, output, node.stride());
+
+ // Log info
+ ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " << node.type()
+ << " Target " << TargetInfo::TargetType
+ << " Data Type: " << input->info()->data_type()
+ << " Input shape: " << input->info()->tensor_shape()
+ << " Output shape: " << output->info()->tensor_shape()
+ << std::endl);
+
+ return std::move(func);
+}
+
/** Create a backend reshape layer function
*
* @tparam ReshapeLayerFunction Backend reshape function
diff --git a/arm_compute/graph/backends/ValidateHelpers.h b/arm_compute/graph/backends/ValidateHelpers.h
index 3064db20c3..e8b8af154f 100644
--- a/arm_compute/graph/backends/ValidateHelpers.h
+++ b/arm_compute/graph/backends/ValidateHelpers.h
@@ -201,6 +201,29 @@ Status validate_permute_layer(PermuteLayerNode &node)
return PermuteLayer::validate(input, output, perm);
}
+
+/** Validates a Reorg layer node
+ *
+ * @tparam ReorgLayer Reorg layer type
+ *
+ * @param[in] node Node to validate
+ *
+ * @return Status
+ */
+template <typename ReorgLayer>
+Status validate_reorg_layer(ReorgLayerNode &node)
+{
+ ARM_COMPUTE_LOG_GRAPH_VERBOSE("Validating ReorgLayer node with ID : " << node.id() << " and Name: " << node.name() << std::endl);
+ ARM_COMPUTE_RETURN_ERROR_ON(node.num_inputs() != 1);
+ ARM_COMPUTE_RETURN_ERROR_ON(node.num_outputs() != 1);
+
+ // Extract input and output
+ arm_compute::ITensorInfo *input = detail::get_backing_tensor_info(node.input(0));
+ arm_compute::ITensorInfo *output = get_backing_tensor_info(node.output(0));
+
+ // Validate function
+ return ReorgLayer::validate(input, output, node.stride());
+}
} // namespace detail
} // namespace backends
} // namespace graph
diff --git a/arm_compute/graph/frontend/Layers.h b/arm_compute/graph/frontend/Layers.h
index 054410e4ad..9a268d3d29 100644
--- a/arm_compute/graph/frontend/Layers.h
+++ b/arm_compute/graph/frontend/Layers.h
@@ -589,6 +589,31 @@ private:
PoolingLayerInfo _pool_info;
};
+/** Reorg Layer */
+class ReorgLayer final : public ILayer
+{
+public:
+ /** Construct a reorg layer.
+ *
+ * @param[in] stride Stride value to use for reorganizing the values in the output tensor.
+ * It defines the spatial distance between 2 consecutive pixels in the x and y direction
+ */
+ ReorgLayer(int stride)
+ : _stride(stride)
+ {
+ }
+
+ NodeID create_layer(IStream &s) override
+ {
+ NodeParams common_params = { name(), s.hints().target_hint };
+ NodeIdxPair input = { s.tail_node(), 0 };
+ return GraphBuilder::add_reorg_node(s.graph(), common_params, input, _stride);
+ }
+
+private:
+ int _stride;
+};
+
/** Reshape Layer */
class ReshapeLayer final : public ILayer
{
diff --git a/arm_compute/graph/nodes/Nodes.h b/arm_compute/graph/nodes/Nodes.h
index 8a85159896..438054c4bf 100644
--- a/arm_compute/graph/nodes/Nodes.h
+++ b/arm_compute/graph/nodes/Nodes.h
@@ -41,6 +41,7 @@
#include "arm_compute/graph/nodes/OutputNode.h"
#include "arm_compute/graph/nodes/PermuteLayerNode.h"
#include "arm_compute/graph/nodes/PoolingLayerNode.h"
+#include "arm_compute/graph/nodes/ReorgLayerNode.h"
#include "arm_compute/graph/nodes/ReshapeLayerNode.h"
#include "arm_compute/graph/nodes/ResizeLayerNode.h"
#include "arm_compute/graph/nodes/SoftmaxLayerNode.h"
diff --git a/arm_compute/graph/nodes/ReorgLayerNode.h b/arm_compute/graph/nodes/ReorgLayerNode.h
new file mode 100644
index 0000000000..1cc0698b46
--- /dev/null
+++ b/arm_compute/graph/nodes/ReorgLayerNode.h
@@ -0,0 +1,67 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_GRAPH_REORG_LAYER_NODE_H__
+#define __ARM_COMPUTE_GRAPH_REORG_LAYER_NODE_H__
+
+#include "arm_compute/graph/INode.h"
+
+namespace arm_compute
+{
+namespace graph
+{
+/** Reorg Layer node */
+class ReorgLayerNode final : public INode
+{
+public:
+ /** Constructor
+ *
+ * @param[in] stride Stride value to use for reorganizing the values in the output tensor.
+ */
+ ReorgLayerNode(int stride);
+ /** Stride value to use for reorganizing the values in the output tensor.
+ *
+ * @return Stride value to use for reorganizing the values in the output tensor.
+ */
+ int stride() const;
+ /** Computes reorg output descriptor
+ *
+ * @param[in] input_descriptor Input descriptor
+ * @param[in] stride Stride value to use for reorganizing the values in the output tensor.
+ *
+ * @return Output descriptor
+ */
+ static TensorDescriptor compute_output_descriptor(const TensorDescriptor &input_descriptor, int stride);
+
+ // Inherited overridden methods:
+ NodeType type() const override;
+ bool forward_descriptors() override;
+ TensorDescriptor configure_output(size_t idx) const override;
+ void accept(INodeVisitor &v) override;
+
+private:
+ int _stride;
+};
+} // namespace graph
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_GRAPH_REORG_LAYER_NODE_H__ */
diff --git a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
index 94b438cb83..88758b523a 100644
--- a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
+++ b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp
@@ -165,7 +165,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights,
const TensorShape output_shape = compute_depthwise_convolution_shape(*input, *weights, conv_info, depth_multiplier);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
- ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_quantized_asymmetric(input->data_type()) && (output->data_type() != DataType::S32));
+ //ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_quantized_asymmetric(input->data_type()) && (output->data_type() != DataType::S32));
ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_float(input->data_type()) && (output->data_type() != DataType::F32));
}
diff --git a/src/graph/GraphBuilder.cpp b/src/graph/GraphBuilder.cpp
index f2eca67f55..55fe5e3def 100644
--- a/src/graph/GraphBuilder.cpp
+++ b/src/graph/GraphBuilder.cpp
@@ -433,6 +433,11 @@ NodeID GraphBuilder::add_pooling_node(Graph &g, NodeParams params, NodeIdxPair i
return create_simple_single_input_output_node<PoolingLayerNode>(g, params, input, pool_info);
}
+NodeID GraphBuilder::add_reorg_node(Graph &g, NodeParams params, NodeIdxPair input, int stride)
+{
+ return create_simple_single_input_output_node<ReorgLayerNode>(g, params, input, stride);
+}
+
NodeID GraphBuilder::add_reshape_node(Graph &g, NodeParams params, NodeIdxPair input, TensorShape shape)
{
return create_simple_single_input_output_node<ReshapeLayerNode>(g, params, input, shape);
diff --git a/src/graph/backends/CL/CLFunctionsFactory.cpp b/src/graph/backends/CL/CLFunctionsFactory.cpp
index bf3dcba224..25456e886f 100644
--- a/src/graph/backends/CL/CLFunctionsFactory.cpp
+++ b/src/graph/backends/CL/CLFunctionsFactory.cpp
@@ -105,6 +105,8 @@ std::unique_ptr<IFunction> CLFunctionFactory::create(INode *node, GraphContext &
return detail::create_permute_layer<CLPermute, CLTargetInfo>(*polymorphic_downcast<PermuteLayerNode *>(node));
case NodeType::PoolingLayer:
return detail::create_pooling_layer<CLPoolingLayer, CLTargetInfo>(*polymorphic_downcast<PoolingLayerNode *>(node));
+ case NodeType::ReorgLayer:
+ return detail::create_reorg_layer<CLReorgLayer, CLTargetInfo>(*polymorphic_downcast<ReorgLayerNode *>(node));
case NodeType::ReshapeLayer:
return detail::create_reshape_layer<CLReshapeLayer, CLTargetInfo>(*polymorphic_downcast<ReshapeLayerNode *>(node));
case NodeType::ResizeLayer:
diff --git a/src/graph/backends/CL/CLNodeValidator.cpp b/src/graph/backends/CL/CLNodeValidator.cpp
index ba5b59d76a..ece0d324f6 100644
--- a/src/graph/backends/CL/CLNodeValidator.cpp
+++ b/src/graph/backends/CL/CLNodeValidator.cpp
@@ -59,6 +59,8 @@ Status CLNodeValidator::validate(INode *node)
CLDepthwiseConvolutionLayer3x3>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
case NodeType::PermuteLayer:
return detail::validate_permute_layer<CLPermute>(*polymorphic_downcast<PermuteLayerNode *>(node));
+ case NodeType::ReorgLayer:
+ return detail::validate_reorg_layer<CLReorgLayer>(*polymorphic_downcast<ReorgLayerNode *>(node));
default:
return Status{};
}
diff --git a/src/graph/backends/NEON/NEFunctionFactory.cpp b/src/graph/backends/NEON/NEFunctionFactory.cpp
index c6da34f63f..ec70a5ae93 100644
--- a/src/graph/backends/NEON/NEFunctionFactory.cpp
+++ b/src/graph/backends/NEON/NEFunctionFactory.cpp
@@ -207,6 +207,8 @@ std::unique_ptr<IFunction> NEFunctionFactory::create(INode *node, GraphContext &
return detail::create_permute_layer<NEPermute, NETargetInfo>(*polymorphic_downcast<PermuteLayerNode *>(node));
case NodeType::PoolingLayer:
return detail::create_pooling_layer<NEPoolingLayer, NETargetInfo>(*polymorphic_downcast<PoolingLayerNode *>(node));
+ case NodeType::ReorgLayer:
+ return detail::create_reorg_layer<NEReorgLayer, NETargetInfo>(*polymorphic_downcast<ReorgLayerNode *>(node));
case NodeType::ReshapeLayer:
return detail::create_reshape_layer<NEReshapeLayer, NETargetInfo>(*polymorphic_downcast<ReshapeLayerNode *>(node));
case NodeType::ResizeLayer:
diff --git a/src/graph/backends/NEON/NENodeValidator.cpp b/src/graph/backends/NEON/NENodeValidator.cpp
index 58ffaf024b..9f0547f55c 100644
--- a/src/graph/backends/NEON/NENodeValidator.cpp
+++ b/src/graph/backends/NEON/NENodeValidator.cpp
@@ -59,6 +59,8 @@ Status NENodeValidator::validate(INode *node)
NEDepthwiseConvolutionLayer3x3>(*polymorphic_downcast<DepthwiseConvolutionLayerNode *>(node));
case NodeType::PermuteLayer:
return detail::validate_permute_layer<NEPermute>(*polymorphic_downcast<PermuteLayerNode *>(node));
+ case NodeType::ReorgLayer:
+ return detail::validate_reorg_layer<NEReorgLayer>(*polymorphic_downcast<ReorgLayerNode *>(node));
default:
return Status{};
}
diff --git a/src/graph/nodes/ReorgLayerNode.cpp b/src/graph/nodes/ReorgLayerNode.cpp
new file mode 100644
index 0000000000..6b83f6b90c
--- /dev/null
+++ b/src/graph/nodes/ReorgLayerNode.cpp
@@ -0,0 +1,97 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/graph/nodes/ReorgLayerNode.h"
+
+#include "arm_compute/graph/Graph.h"
+#include "arm_compute/graph/INodeVisitor.h"
+#include "arm_compute/graph/Utils.h"
+
+namespace arm_compute
+{
+namespace graph
+{
+ReorgLayerNode::ReorgLayerNode(int stride)
+ : _stride(stride)
+{
+ _input_edges.resize(1, EmptyEdgeID);
+ _outputs.resize(1, NullTensorID);
+}
+
+int ReorgLayerNode::stride() const
+{
+ return _stride;
+}
+
+TensorDescriptor ReorgLayerNode::compute_output_descriptor(const TensorDescriptor &input_descriptor, int stride)
+{
+ const unsigned int input_width = get_dimension_size(input_descriptor, DataLayoutDimension::WIDTH);
+ const unsigned int input_height = get_dimension_size(input_descriptor, DataLayoutDimension::HEIGHT);
+ const unsigned int input_channel = get_dimension_size(input_descriptor, DataLayoutDimension::CHANNEL);
+
+ ARM_COMPUTE_ERROR_ON(stride <= 0);
+ ARM_COMPUTE_ERROR_ON_MSG((input_width % stride != 0), "The width of the input tensor must be a multiple of stride");
+ ARM_COMPUTE_ERROR_ON_MSG((input_height % stride != 0), "The height of the input tensor must be a multiple of stride");
+
+ TensorDescriptor output_descriptor = input_descriptor;
+ output_descriptor.shape.set(get_dimension_idx(output_descriptor, DataLayoutDimension::WIDTH), input_width / stride);
+ output_descriptor.shape.set(get_dimension_idx(output_descriptor, DataLayoutDimension::HEIGHT), input_height / stride);
+ output_descriptor.shape.set(get_dimension_idx(output_descriptor, DataLayoutDimension::CHANNEL), input_channel * stride * stride);
+
+ return output_descriptor;
+}
+
+bool ReorgLayerNode::forward_descriptors()
+{
+ if((input_id(0) != NullTensorID) && (output_id(0) != NullTensorID))
+ {
+ Tensor *dst = output(0);
+ ARM_COMPUTE_ERROR_ON(dst == nullptr);
+ dst->desc() = configure_output(0);
+ return true;
+ }
+ return false;
+}
+
+TensorDescriptor ReorgLayerNode::configure_output(size_t idx) const
+{
+ ARM_COMPUTE_UNUSED(idx);
+ ARM_COMPUTE_ERROR_ON(idx >= _outputs.size());
+
+ const Tensor *src = input(0);
+ ARM_COMPUTE_ERROR_ON(src == nullptr);
+
+ return compute_output_descriptor(src->desc(), _stride);
+}
+
+NodeType ReorgLayerNode::type() const
+{
+ return NodeType::ReorgLayer;
+}
+
+void ReorgLayerNode::accept(INodeVisitor &v)
+{
+ v.visit(*this);
+}
+} // namespace graph
+} // namespace arm_compute \ No newline at end of file
diff --git a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp
index 24b12f4969..ccbd01e2e2 100644
--- a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp
@@ -163,15 +163,26 @@ Status NEDepthwiseConvolutionLayer3x3::validate(const ITensorInfo *input, const
unsigned int depth_multiplier)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
- ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() != DataLayout::NCHW && input->data_layout() != DataLayout::NHWC);
+ ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::UNKNOWN);
if(biases != nullptr)
{
+ const unsigned int channel_idx = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
- ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3));
+ ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(channel_idx));
}
- return NEDepthwiseConvolutionLayer3x3Kernel::validate(input, weights, output, conv_info, depth_multiplier);
+ const bool is_quantized = is_data_type_quantized_asymmetric(input->data_type());
+ TensorInfo accumulator = TensorInfo(output->clone()->set_is_resizable(true).reset_padding().set_data_type(DataType::S32));
+
+ ARM_COMPUTE_RETURN_ON_ERROR(NEDepthwiseConvolutionLayer3x3Kernel::validate(input, weights, is_quantized ? &accumulator : output, conv_info, depth_multiplier));
+
+ if(is_quantized)
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(NEDirectConvolutionLayerOutputStageKernel::validate(&accumulator, biases, output));
+ }
+
+ return Status{};
}
void NEDepthwiseConvolutionLayer3x3::run()
@@ -359,7 +370,10 @@ Status NEDepthwiseConvolutionLayer::validate(const ITensorInfo *input, const ITe
unsigned int depth_multiplier)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
- ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() != DataLayout::NCHW && input->data_layout() != DataLayout::NHWC);
+ ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::UNKNOWN);
+
+ const unsigned int width_idx = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::WIDTH);
+ const unsigned int height_idx = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::HEIGHT);
// Clone output to use auto init
auto output_clone = output->clone();
@@ -391,8 +405,8 @@ Status NEDepthwiseConvolutionLayer::validate(const ITensorInfo *input, const ITe
const size_t weights_w = weights_to_use->dimension(0);
const size_t weights_h = weights_to_use->dimension(1);
const size_t weights_z = weights_to_use->dimension(2);
- const unsigned int conv_w = output_shape.x();
- const unsigned int conv_h = output_shape.y();
+ const unsigned int conv_w = output_shape[width_idx];
+ const unsigned int conv_h = output_shape[height_idx];
const size_t patch_size = weights_w * weights_h + (append_bias ? 1 : 0);
const size_t conv_size = conv_w * conv_h;
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-17 14:25:54 +0000