diff options
Diffstat (limited to 'src/backends/neon/workloads')
-rw-r--r-- | src/backends/neon/workloads/NeonConvolution2dWorkload.cpp | 25 | ||||
-rw-r--r-- | src/backends/neon/workloads/NeonConvolution2dWorkload.hpp | 10 |
2 files changed, 30 insertions, 5 deletions
diff --git a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp index 83f761158a..d35b9685be 100644 --- a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp +++ b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp @@ -59,8 +59,10 @@ arm_compute::Status NeonConvolution2dWorkloadValidate(const TensorInfo& input, } NeonConvolution2dWorkload::NeonConvolution2dWorkload( - const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info, - std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager) + const Convolution2dQueueDescriptor& descriptor, + const WorkloadInfo& info, + std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager, + const bool isFastMathEnabled) : BaseWorkload<Convolution2dQueueDescriptor>(descriptor, info) { using arm_compute::NEDirectConvolutionLayer; @@ -97,7 +99,19 @@ NeonConvolution2dWorkload::NeonConvolution2dWorkload( &output, padStrideInfo, arm_compute::WeightsInfo(), - aclDilationInfo); + aclDilationInfo, + arm_compute::ActivationLayerInfo(), + isFastMathEnabled); + + m_ConvolutionMethod = + convolutionLayer->get_convolution_method(input.info(), + m_KernelTensor->info(), + output.info(), + padStrideInfo, + arm_compute::WeightsInfo(), + aclDilationInfo, + arm_compute::ActivationLayerInfo(), + isFastMathEnabled); m_ConvolutionLayer.reset(convolutionLayer.release()); @@ -120,6 +134,11 @@ void NeonConvolution2dWorkload::Execute() const m_ConvolutionLayer->run(); } +arm_compute::ConvolutionMethod NeonConvolution2dWorkload::GetConvolutionMethod() const +{ + return m_ConvolutionMethod; +} + void NeonConvolution2dWorkload::FreeUnusedTensors() { FreeTensorIfUnused(m_KernelTensor); diff --git a/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp b/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp index 54e08a2042..860d78ba7e 100644 --- a/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp +++ b/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp @@ -28,17 +28,23 @@ class NeonConvolution2dWorkload : public BaseWorkload<Convolution2dQueueDescript public: using BaseWorkload<Convolution2dQueueDescriptor>::m_Data; - NeonConvolution2dWorkload(const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info, - std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager); + NeonConvolution2dWorkload(const Convolution2dQueueDescriptor& descriptor, + const WorkloadInfo& info, + std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager, + const bool isFastMathENabled = false); void Execute() const override; + arm_compute::ConvolutionMethod GetConvolutionMethod() const; + private: std::unique_ptr<arm_compute::IFunction> m_ConvolutionLayer; std::unique_ptr<arm_compute::Tensor> m_KernelTensor; std::unique_ptr<arm_compute::Tensor> m_BiasTensor; + arm_compute::ConvolutionMethod m_ConvolutionMethod; + void FreeUnusedTensors(); }; |