diff options
author | Pablo Tello <pablo.tello@arm.com> | 2020-01-23 10:37:17 +0000 |
---|---|---|
committer | Pablo Tello <pablo.tello@arm.com> | 2020-01-24 14:50:49 +0000 |
commit | e61f0712dc480f70372fcfe0921a85a7de53661b (patch) | |
tree | a6a87887b8269a104b898a226c2d7cc31433de21 /src/backends/neon/workloads/NeonDivisionWorkload.cpp | |
parent | 0be3a88518a73819aaf71ef71f4fb2fbfc14bc88 (diff) | |
download | armnn-e61f0712dc480f70372fcfe0921a85a7de53661b.tar.gz |
MLCE-156: Add Division workload in CpuAcc
Change-Id: I1f228fcaf1077867d9755a2b850c6703387fab34
Signed-off-by: Pablo Tello <pablo.tello@arm.com>
Diffstat (limited to 'src/backends/neon/workloads/NeonDivisionWorkload.cpp')
-rw-r--r-- | src/backends/neon/workloads/NeonDivisionWorkload.cpp | 45 |
1 files changed, 45 insertions, 0 deletions
diff --git a/src/backends/neon/workloads/NeonDivisionWorkload.cpp b/src/backends/neon/workloads/NeonDivisionWorkload.cpp new file mode 100644 index 0000000000..6fdb455f25 --- /dev/null +++ b/src/backends/neon/workloads/NeonDivisionWorkload.cpp @@ -0,0 +1,45 @@ +// +// Copyright © 2020 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#include "NeonDivisionWorkload.hpp" +#include <aclCommon/ArmComputeTensorUtils.hpp> +#include <backendsCommon/CpuTensorHandle.hpp> + +namespace armnn +{ + +arm_compute::Status NeonDivisionWorkloadValidate(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output) +{ + const arm_compute::TensorInfo aclInput0 = armcomputetensorutils::BuildArmComputeTensorInfo(input0); + const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input1); + const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output); + + return arm_compute::NEElementwiseDivision::validate(&aclInput0, + &aclInput1, + &aclOutput); +} + +NeonDivisionWorkload::NeonDivisionWorkload(const DivisionQueueDescriptor& descriptor, + const WorkloadInfo& info) + : BaseWorkload<DivisionQueueDescriptor>(descriptor, info) +{ + m_Data.ValidateInputsOutputs("NeonDivisionWorkload", 2, 1); + + arm_compute::ITensor& input0 = boost::polymorphic_downcast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor(); + arm_compute::ITensor& input1 = boost::polymorphic_downcast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor(); + arm_compute::ITensor& output = boost::polymorphic_downcast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor(); + + m_DivLayer.configure(&input0, &input1, &output); +} + +void NeonDivisionWorkload::Execute() const +{ + ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonDivisionWorkload_Execute"); + m_DivLayer.run(); +} + +} //namespace armnn |