diff options
author | Aron Virginas-Tar <Aron.Virginas-Tar@arm.com> | 2019-09-10 12:32:08 +0100 |
---|---|---|
committer | Áron Virginás-Tar <aron.virginas-tar@arm.com> | 2019-09-10 14:58:18 +0000 |
commit | 1a763dd350e80f26a661dc5216fdef1256e5b32a (patch) | |
tree | 866c985af88766b2f08f317eed12cbf37b618362 /src/backends/cl/workloads/ClRsqrtWorkload.cpp | |
parent | 0dd3b43e1497ab406acfdefca425143fbefb1ee4 (diff) | |
download | armnn-1a763dd350e80f26a661dc5216fdef1256e5b32a.tar.gz |
IVGCVSW-3829 Add CL workload for RSQRT
Signed-off-by: Aron Virginas-Tar <Aron.Virginas-Tar@arm.com>
Change-Id: Idf40ed0ebf2af3170137bf3a3cb5a9635a203d4a
Diffstat (limited to 'src/backends/cl/workloads/ClRsqrtWorkload.cpp')
-rw-r--r-- | src/backends/cl/workloads/ClRsqrtWorkload.cpp | 44 |
1 files changed, 44 insertions, 0 deletions
diff --git a/src/backends/cl/workloads/ClRsqrtWorkload.cpp b/src/backends/cl/workloads/ClRsqrtWorkload.cpp new file mode 100644 index 0000000000..be687595f7 --- /dev/null +++ b/src/backends/cl/workloads/ClRsqrtWorkload.cpp @@ -0,0 +1,44 @@ +// +// Copyright © 2019 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#include "ClRsqrtWorkload.hpp" + +#include "ClWorkloadUtils.hpp" + +#include <aclCommon/ArmComputeTensorUtils.hpp> + +#include <cl/ClTensorHandle.hpp> + +#include <boost/cast.hpp> + +namespace armnn +{ + +arm_compute::Status ClRsqrtWorkloadValidate(const TensorInfo& input, const TensorInfo& output) +{ + const arm_compute::TensorInfo aclInput = armcomputetensorutils::BuildArmComputeTensorInfo(input); + const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output); + + return arm_compute::CLRsqrtLayer::validate(&aclInput, &aclOutput); +} + +ClRsqrtWorkload::ClRsqrtWorkload(const RsqrtQueueDescriptor& descriptor, const WorkloadInfo& info) + : BaseWorkload<RsqrtQueueDescriptor>(descriptor, info) +{ + m_Data.ValidateInputsOutputs("ClRsqrtWorkload", 1, 1); + + arm_compute::ICLTensor& input = boost::polymorphic_downcast<ClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor(); + arm_compute::ICLTensor& output = boost::polymorphic_downcast<ClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor(); + + m_RsqrtLayer.configure(&input, &output); +} + +void ClRsqrtWorkload::Execute() const +{ + ARMNN_SCOPED_PROFILING_EVENT_CL("ClRsqrtWorkload_Execute"); + RunClFunction(m_RsqrtLayer, CHECK_LOCATION()); +} + +} // namespace armnn |