diff options
Diffstat (limited to 'src/backends/neon/workloads')
| -rw-r--r-- | src/backends/neon/workloads/CMakeLists.txt | 4 | ||||
| -rw-r--r-- | src/backends/neon/workloads/NeonComparisonWorkload.cpp | 53 | ||||
| -rw-r--r-- | src/backends/neon/workloads/NeonComparisonWorkload.hpp | 31 | ||||
| -rw-r--r-- | src/backends/neon/workloads/NeonGreaterWorkload.cpp | 44 | ||||
| -rw-r--r-- | src/backends/neon/workloads/NeonGreaterWorkload.hpp | 30 | ||||
| -rw-r--r-- | src/backends/neon/workloads/NeonWorkloads.hpp | 2 |
6 files changed, 87 insertions, 77 deletions
diff --git a/src/backends/neon/workloads/CMakeLists.txt b/src/backends/neon/workloads/CMakeLists.txt index 8160710f65..cbe1e3b945 100644 --- a/src/backends/neon/workloads/CMakeLists.txt +++ b/src/backends/neon/workloads/CMakeLists.txt @@ -16,6 +16,8 @@ list(APPEND armnnNeonBackendWorkloads_sources NeonBatchNormalizationWorkload.hpp NeonBatchToSpaceNdWorkload.cpp NeonBatchToSpaceNdWorkload.hpp + NeonComparisonWorkload.cpp + NeonComparisonWorkload.hpp NeonConcatWorkload.cpp NeonConcatWorkload.hpp NeonConstantWorkload.cpp @@ -38,8 +40,6 @@ list(APPEND armnnNeonBackendWorkloads_sources NeonFloorFloatWorkload.hpp NeonFullyConnectedWorkload.cpp NeonFullyConnectedWorkload.hpp - NeonGreaterWorkload.cpp - NeonGreaterWorkload.hpp NeonInstanceNormalizationWorkload.cpp NeonInstanceNormalizationWorkload.hpp NeonL2NormalizationFloatWorkload.cpp diff --git a/src/backends/neon/workloads/NeonComparisonWorkload.cpp b/src/backends/neon/workloads/NeonComparisonWorkload.cpp new file mode 100644 index 0000000000..0edb3328b9 --- /dev/null +++ b/src/backends/neon/workloads/NeonComparisonWorkload.cpp @@ -0,0 +1,53 @@ +// +// Copyright © 2020 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#include "NeonComparisonWorkload.hpp" +#include <aclCommon/ArmComputeUtils.hpp> +#include <aclCommon/ArmComputeTensorUtils.hpp> +#include <backendsCommon/CpuTensorHandle.hpp> + +namespace armnn +{ +using namespace armcomputetensorutils; + +arm_compute::Status NeonComparisonWorkloadValidate(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + const ComparisonDescriptor& descriptor) +{ + const arm_compute::TensorInfo aclInput0 = BuildArmComputeTensorInfo(input0); + const arm_compute::TensorInfo aclInput1 = BuildArmComputeTensorInfo(input1); + const arm_compute::TensorInfo aclOutput = BuildArmComputeTensorInfo(output); + + const arm_compute::ComparisonOperation comparisonOperation = ConvertComparisonOperationToAcl(descriptor); + + const arm_compute::Status aclStatus = arm_compute::NEElementwiseComparison::validate(&aclInput0, + &aclInput1, + &aclOutput, + comparisonOperation); + return aclStatus; +} + +NeonComparisonWorkload::NeonComparisonWorkload(const ComparisonQueueDescriptor& descriptor, const WorkloadInfo& info) + : BaseWorkload<ComparisonQueueDescriptor>(descriptor, info) +{ + m_Data.ValidateInputsOutputs("NeonComparisonWorkload", 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(); + + const arm_compute::ComparisonOperation comparisonOperation = ConvertComparisonOperationToAcl(m_Data.m_Parameters); + + m_ComparisonLayer.configure(&input0, &input1, &output, comparisonOperation); +} + +void NeonComparisonWorkload::Execute() const +{ + ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonComparisonWorkload_Execute"); + m_ComparisonLayer.run(); +} + +} //namespace armnn
\ No newline at end of file diff --git a/src/backends/neon/workloads/NeonComparisonWorkload.hpp b/src/backends/neon/workloads/NeonComparisonWorkload.hpp new file mode 100644 index 0000000000..1b01053d00 --- /dev/null +++ b/src/backends/neon/workloads/NeonComparisonWorkload.hpp @@ -0,0 +1,31 @@ +// +// Copyright © 2020 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#pragma once + +#include <neon/workloads/NeonWorkloadUtils.hpp> + +#include <arm_compute/runtime/NEON/functions/NEElementwiseOperations.h> + +namespace armnn +{ + +arm_compute::Status NeonComparisonWorkloadValidate(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + const ComparisonDescriptor& descriptor); + +class NeonComparisonWorkload : public BaseWorkload<ComparisonQueueDescriptor> +{ +public: + NeonComparisonWorkload(const ComparisonQueueDescriptor& descriptor, const WorkloadInfo& info); + + virtual void Execute() const override; + +private: + mutable arm_compute::NEElementwiseComparison m_ComparisonLayer; +}; + +} //namespace armnn
\ No newline at end of file diff --git a/src/backends/neon/workloads/NeonGreaterWorkload.cpp b/src/backends/neon/workloads/NeonGreaterWorkload.cpp deleted file mode 100644 index 1ec5ac4be0..0000000000 --- a/src/backends/neon/workloads/NeonGreaterWorkload.cpp +++ /dev/null @@ -1,44 +0,0 @@ -// -// Copyright © 2017 Arm Ltd. All rights reserved. -// SPDX-License-Identifier: MIT -// - -#include "NeonGreaterWorkload.hpp" -#include <aclCommon/ArmComputeTensorUtils.hpp> -#include <backendsCommon/CpuTensorHandle.hpp> - -namespace armnn -{ - -arm_compute::Status NeonGreaterWorkloadValidate(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::NEGreater::validate(&aclInput0, - &aclInput1, - &aclOutput); -} - -NeonGreaterWorkload::NeonGreaterWorkload(const GreaterQueueDescriptor& descriptor, const WorkloadInfo& info) - : BaseWorkload<GreaterQueueDescriptor>(descriptor, info) -{ - m_Data.ValidateInputsOutputs("NeonGreaterWorkload", 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_GreaterLayer.configure(&input0, &input1, &output); -} - -void NeonGreaterWorkload::Execute() const -{ - ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonGreaterWorkload_Execute"); - m_GreaterLayer.run(); -} - -} //namespace armnn
\ No newline at end of file diff --git a/src/backends/neon/workloads/NeonGreaterWorkload.hpp b/src/backends/neon/workloads/NeonGreaterWorkload.hpp deleted file mode 100644 index 503e60e784..0000000000 --- a/src/backends/neon/workloads/NeonGreaterWorkload.hpp +++ /dev/null @@ -1,30 +0,0 @@ -// -// Copyright © 2017 Arm Ltd. All rights reserved. -// SPDX-License-Identifier: MIT -// - -#pragma once - -#include <neon/workloads/NeonWorkloadUtils.hpp> - -#include <arm_compute/runtime/NEON/functions/NEElementwiseOperations.h> - -namespace armnn -{ - -arm_compute::Status NeonGreaterWorkloadValidate(const TensorInfo& input0, - const TensorInfo& input1, - const TensorInfo& output); - -class NeonGreaterWorkload : public BaseWorkload<GreaterQueueDescriptor> -{ -public: - NeonGreaterWorkload(const GreaterQueueDescriptor& descriptor, const WorkloadInfo& info); - - virtual void Execute() const override; - -private: - mutable arm_compute::NEGreater m_GreaterLayer; -}; - -} //namespace armnn
\ No newline at end of file diff --git a/src/backends/neon/workloads/NeonWorkloads.hpp b/src/backends/neon/workloads/NeonWorkloads.hpp index e28d120de7..2b7eabeb0d 100644 --- a/src/backends/neon/workloads/NeonWorkloads.hpp +++ b/src/backends/neon/workloads/NeonWorkloads.hpp @@ -11,6 +11,7 @@ #include "NeonArgMinMaxWorkload.hpp" #include "NeonBatchNormalizationWorkload.hpp" #include "NeonBatchToSpaceNdWorkload.hpp" +#include "NeonComparisonWorkload.hpp" #include "NeonConstantWorkload.hpp" #include "NeonConvertFp16ToFp32Workload.hpp" #include "NeonConvertFp32ToFp16Workload.hpp" @@ -21,7 +22,6 @@ #include "NeonDetectionPostProcessWorkload.hpp" #include "NeonFloorFloatWorkload.hpp" #include "NeonFullyConnectedWorkload.hpp" -#include "NeonGreaterWorkload.hpp" #include "NeonInstanceNormalizationWorkload.hpp" #include "NeonL2NormalizationFloatWorkload.hpp" #include "NeonLstmFloatWorkload.hpp" |