From ee33ea5a6e1aa0faac1cc8b5a269bd4f89854821 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Thu, 8 Mar 2018 16:01:29 +0000 Subject: COMPMID-996: Add support for grouped convolution. Change-Id: I279e29ce20b3dde57445264dc11491f127b44d70 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/124429 Reviewed-by: Anthony Barbier Tested-by: Jenkins --- src/graph2/mutators/SplitLayerSubTensorMutator.cpp | 89 ++++++++++++++++++++++ 1 file changed, 89 insertions(+) create mode 100644 src/graph2/mutators/SplitLayerSubTensorMutator.cpp (limited to 'src/graph2/mutators/SplitLayerSubTensorMutator.cpp') diff --git a/src/graph2/mutators/SplitLayerSubTensorMutator.cpp b/src/graph2/mutators/SplitLayerSubTensorMutator.cpp new file mode 100644 index 0000000000..33494ba6bc --- /dev/null +++ b/src/graph2/mutators/SplitLayerSubTensorMutator.cpp @@ -0,0 +1,89 @@ +/* + * 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/graph2/mutators/SplitLayerSubTensorMutator.h" + +#include "arm_compute/graph2/Graph.h" +#include "arm_compute/graph2/Logger.h" +#include "arm_compute/graph2/backends/BackendRegistry.h" +#include "arm_compute/graph2/nodes/SplitLayerNode.h" + +#include "arm_compute/core/utils/misc/Cast.h" +#include "arm_compute/core/utils/misc/Iterable.h" + +namespace arm_compute +{ +namespace graph2 +{ +const char *SplitLayerSubTensorMutator::name() +{ + return "SplitLayerSubTensorMutator"; +} + +void SplitLayerSubTensorMutator::mutate(Graph &g) +{ + // Should be in reverse order of execution + for(auto &node : arm_compute::utils::iterable::reverse_iterate(g.nodes())) + { + if(node && node->type() == NodeType::SplitLayer && node->input(0) != nullptr) + { + // Get output tensor + Tensor *input_tensor = node->input(0); + + // Check that all tensor have the same target and are valid + bool is_valid = std::all_of(node->outputs().cbegin(), node->outputs().cend(), + [&](const TensorID & tid) + { + return (g.tensor(tid) != nullptr) && (g.tensor(tid)->desc().target == input_tensor->desc().target); + }); + + // Create subtensors + if(is_valid && backends::BackendRegistry::get().find_backend(input_tensor->desc().target) != nullptr) + { + ARM_COMPUTE_LOG_GRAPH_VERBOSE("Using sub-tensors for the node with ID : " + << node->id() << " and name : " << node->name() << std::endl); + + auto *split_node = arm_compute::utils::cast::polymorphic_downcast(node.get()); + + const unsigned int axis = split_node->axis(); + const unsigned int num_splits = split_node->num_splits(); + const bool extend_parent = (axis < 2); + + // Create sub-tensor handles + for(unsigned int i = 0; i < node->outputs().size(); ++i) + { + Tensor *output_tensor = node->output(i); + const TensorShape output_shape = output_tensor->desc().shape; + Coordinates coords; + std::tie(std::ignore, coords) = SplitLayerNode::compute_output_shape(input_tensor->desc().shape, num_splits, axis, i); + + backends::IDeviceBackend *backend = backends::BackendRegistry::get().find_backend(output_tensor->desc().target); + std::unique_ptr handle = backend->create_subtensor(input_tensor->handle(), output_shape, coords, extend_parent); + output_tensor->set_handle(std::move(handle)); + } + } + } + } +} +} // namespace graph2 +} // namespace arm_compute -- cgit v1.2.1