onnxruntime/docs/execution-providers/ArmNN-ExecutionProvider.md

---
title: Arm NN
description: Instructions to execute ONNX Runtime with Arm NN on Armv8 cores
parent: Execution Providers
nav_order: 2
redirect_from: /docs/reference/execution-providers/ArmNN-ExecutionProvider
---

# ArmNN Execution Provider
{: .no_toc}

## Contents
{: .no_toc }

* TOC placeholder
{:toc}

Accelerate performance of ONNX model workloads across Armv8 cores with the ArmNN execution provider. [ArmNN](https://github.com/ARM-software/armnn) is an open source inference engine maintained by Arm and Linaro companies. 

## Build
For build instructions, please see the [BUILD page](../build/eps.md#armnn).

## Usage
### C/C++
To use ArmNN as execution provider for inferencing, please register it as below.
```
Ort::Env env = Ort::Env{ORT_LOGGING_LEVEL_ERROR, "Default"};
Ort::SessionOptions so;
bool enable_cpu_mem_arena = true;
Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_ArmNN(so, enable_cpu_mem_arena));
```
The C API details are [here](../get-started/with-c.md).

## Performance Tuning
For performance tuning, please see guidance on this page: [ONNX Runtime Perf Tuning](../performance/tune-performance.md)

When/if using [onnxruntime_perf_test](https://github.com/microsoft/onnxruntime/tree/master/onnxruntime/test/perftest), use the flag -e armnn