diff options
Diffstat (limited to 'src/runtime/NEON/functions/NEFullyConnectedLayer.cpp')
-rw-r--r-- | src/runtime/NEON/functions/NEFullyConnectedLayer.cpp | 77 |
1 files changed, 44 insertions, 33 deletions
diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp index 891487efd3..2656d0fa0f 100644 --- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp +++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp @@ -27,6 +27,7 @@ #include "arm_compute/core/Validate.h" #include "arm_compute/runtime/MemoryGroup.h" #include "arm_compute/runtime/NEON/functions/NEConvertFullyConnectedWeights.h" + #include "src/common/utils/Log.h" #include "src/core/helpers/MemoryHelpers.h" #include "src/cpu/operators/CpuFullyConnected.h" @@ -38,80 +39,90 @@ using namespace arm_compute::experimental; struct NEFullyConnectedLayer::Impl { MemoryGroup memory_group{}; - IWeightsManager *weights_manager{ nullptr }; + IWeightsManager *weights_manager{nullptr}; - std::unique_ptr<cpu::CpuFullyConnected> op{ nullptr }; + std::unique_ptr<cpu::CpuFullyConnected> op{nullptr}; - const ITensor *original_weights{ nullptr }; + const ITensor *original_weights{nullptr}; ITensorPack run_pack{}; WorkspaceData<Tensor> workspace{}; experimental::MemoryRequirements aux_mem_req{}; - bool is_prepared{ false }; - bool dynamic_weights{ false }; + bool is_prepared{false}; + bool dynamic_weights{false}; }; NEFullyConnectedLayer::~NEFullyConnectedLayer() = default; -NEFullyConnectedLayer::NEFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager, IWeightsManager *weights_manager) +NEFullyConnectedLayer::NEFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager, + IWeightsManager *weights_manager) : _impl(std::make_unique<Impl>()) { _impl->memory_group = MemoryGroup(std::move(memory_manager)); _impl->weights_manager = weights_manager; } -void NEFullyConnectedLayer::configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, - FullyConnectedLayerInfo fc_info, const WeightsInfo &weights_info) +void NEFullyConnectedLayer::configure(const ITensor *input, + const ITensor *weights, + const ITensor *biases, + ITensor *output, + FullyConnectedLayerInfo fc_info, + const WeightsInfo &weights_info) { // Perform validate step ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output); - ARM_COMPUTE_ERROR_THROW_ON(NEFullyConnectedLayer::validate(input->info(), - weights->info(), + ARM_COMPUTE_ERROR_THROW_ON(NEFullyConnectedLayer::validate(input->info(), weights->info(), biases != nullptr ? biases->info() : nullptr, - output->info(), - fc_info, - weights_info)); + output->info(), fc_info, weights_info)); ARM_COMPUTE_LOG_PARAMS(input, weights, biases, output, fc_info); _impl->op = std::make_unique<cpu::CpuFullyConnected>(); _impl->original_weights = weights; _impl->is_prepared = false; - _impl->op->configure(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(), fc_info, weights_info); + _impl->op->configure(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(), + fc_info, weights_info); - if(_impl->weights_manager != nullptr) + if (_impl->weights_manager != nullptr) { _impl->weights_manager->manage(_impl->original_weights); } _impl->aux_mem_req = _impl->op->workspace(); - _impl->run_pack = { { ACL_SRC_0, input }, { ACL_SRC_1, weights }, { ACL_SRC_2, biases }, { ACL_DST, output } }; - _impl->workspace = manage_workspace<Tensor>(_impl->aux_mem_req, _impl->memory_group, _impl->run_pack, _impl->run_pack); - - _impl->dynamic_weights = - !weights->info()->are_values_constant() && - fc_info.transpose_weights && - !fc_info.are_weights_reshaped && - !fc_info.retain_internal_weights; + _impl->run_pack = {{ACL_SRC_0, input}, {ACL_SRC_1, weights}, {ACL_SRC_2, biases}, {ACL_DST, output}}; + _impl->workspace = + manage_workspace<Tensor>(_impl->aux_mem_req, _impl->memory_group, _impl->run_pack, _impl->run_pack); + + _impl->dynamic_weights = !weights->info()->are_values_constant() && fc_info.transpose_weights && + !fc_info.are_weights_reshaped && !fc_info.retain_internal_weights; } -Status NEFullyConnectedLayer::has_opt_impl(arm_compute::WeightFormat &expected_weight_format, const ITensorInfo *input, const ITensorInfo *weights, - const ITensorInfo *biases, const ITensorInfo *output, const FullyConnectedLayerInfo &fc_info, - const WeightsInfo &weights_info) +Status NEFullyConnectedLayer::has_opt_impl(arm_compute::WeightFormat &expected_weight_format, + const ITensorInfo *input, + const ITensorInfo *weights, + const ITensorInfo *biases, + const ITensorInfo *output, + const FullyConnectedLayerInfo &fc_info, + const WeightsInfo &weights_info) { - return cpu::CpuFullyConnected::has_opt_impl(expected_weight_format, input, weights, biases, output, fc_info, weights_info); + return cpu::CpuFullyConnected::has_opt_impl(expected_weight_format, input, weights, biases, output, fc_info, + weights_info); } -Status NEFullyConnectedLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, - FullyConnectedLayerInfo fc_info, const WeightsInfo &weights_info) +Status NEFullyConnectedLayer::validate(const ITensorInfo *input, + const ITensorInfo *weights, + const ITensorInfo *biases, + const ITensorInfo *output, + FullyConnectedLayerInfo fc_info, + const WeightsInfo &weights_info) { return cpu::CpuFullyConnected::validate(input, weights, biases, output, fc_info, weights_info); } void NEFullyConnectedLayer::run() { - if(!_impl->dynamic_weights) + if (!_impl->dynamic_weights) { prepare(); } @@ -122,7 +133,7 @@ void NEFullyConnectedLayer::run() void NEFullyConnectedLayer::prepare() { - if(!_impl->is_prepared) + if (!_impl->is_prepared) { _impl->op->prepare(_impl->run_pack); @@ -131,13 +142,13 @@ void NEFullyConnectedLayer::prepare() _impl->is_prepared = true; // Handle weights managed infrastructure - if(_impl->weights_manager != nullptr && _impl->weights_manager->are_weights_managed(_impl->original_weights)) + if (_impl->weights_manager != nullptr && _impl->weights_manager->are_weights_managed(_impl->original_weights)) { // Ensure that b gets marked as unused (memory released) only after the last function which uses b also finishes its prepare // This is for cases where multiple functions share the same b (weights) // Therefore when a function marks original b as unused, we pre-mark it in weights manager, and mark it back to used so that it doesn't get released before its last reference const ITensor *original_b = _impl->original_weights; - if(!original_b->is_used()) + if (!original_b->is_used()) { _impl->weights_manager->pre_mark_as_unused(original_b); } |