diff options
| author | Matthew Bentham <Matthew.Bentham@arm.com> | 2023-04-27 12:13:50 +0000 |
|---|---|---|
| committer | Matthew Bentham <matthew.bentham@arm.com> | 2023-04-28 14:56:58 +0000 |
| commit | 34336f9378eca4d39913ac0d9ba411a494631ad5 (patch) | |
| tree | 551e3904ad77c4a15f431516109d55a5cdd63e2a /src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp | |
| parent | 6c53f9fbea7d0b8786e1d29b850ab7bed85e167a (diff) | |
| download | armnn-34336f9378eca4d39913ac0d9ba411a494631ad5.tar.gz | |
Make Convert workloads use arm_compute::NECast in CpuAcc backend
NECast can use conversion instructions where they are available
so this should in general be faster.
Signed-off-by: Matthew Bentham <Matthew.Bentham@arm.com>
Change-Id: I3f259e17b280a4f4c36f363965ffbc8ee8c4c29f
Diffstat (limited to 'src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp')
| -rw-r--r-- | src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp | 7 |
1 files changed, 6 insertions, 1 deletions
diff --git a/src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp b/src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp index 666f48794b..c6fed76e6d 100644 --- a/src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp +++ b/src/backends/neon/workloads/NeonConvertFp32ToFp16Workload.hpp @@ -5,13 +5,17 @@ #pragma once +#include <arm_compute/runtime/NEON/functions/NECast.h> #include <armnn/backends/Workload.hpp> #include <armnn/backends/WorkloadData.hpp> +#include <memory> #include <neon/workloads/NeonWorkloadUtils.hpp> namespace armnn { +arm_compute::Status NeonConvertFp32ToFp16WorkloadValidate(const TensorInfo& input, const TensorInfo& output); + class NeonConvertFp32ToFp16Workload : public Float32ToFloat16Workload<ConvertFp32ToFp16QueueDescriptor> { public: @@ -23,9 +27,10 @@ public: // Replace output tensor handle with the given TensorHandle void ReplaceOutputTensorHandle(ITensorHandle* tensorHandle, unsigned int slot) override; private: + virtual void Reconfigure(); using TensorHandlePair = std::pair<const ITensorHandle*, ITensorHandle*>; std::vector<TensorHandlePair> m_TensorHandlePairs; - virtual void Reconfigure(); + mutable std::unique_ptr<arm_compute::NECast> m_Cast; }; } //namespace armnn |