1# Building and Running ExecuTorch on Xtensa HiFi4 DSP 2 3 4In this tutorial we will walk you through the process of getting setup to build ExecuTorch for an Xtensa HiFi4 DSP and running a simple model on it. 5 6[Cadence](https://www.cadence.com/en_US/home.html) is both a hardware and software vendor, providing solutions for many computational workloads, including to run on power-limited embedded devices. The [Xtensa HiFi4 DSP](https://www.cadence.com/en_US/home/tools/ip/tensilica-ip/hifi-dsps/hifi-4.html) is a Digital Signal Processor (DSP) that is optimized for running audio based neural networks such as wake word detection, Automatic Speech Recognition (ASR), etc. 7 8In addition to the chip, the HiFi4 Neural Network Library ([nnlib](https://github.com/foss-xtensa/nnlib-hifi4)) offers an optimized set of library functions commonly used in NN processing that we utilize in this example to demonstrate how common operations can be accelerated. 9 10On top of being able to run on the Xtensa HiFi4 DSP, another goal of this tutorial is to demonstrate how portable ExecuTorch is and its ability to run on a low-power embedded device such as the Xtensa HiFi4 DSP. This workflow does not require any delegates, it uses custom operators and compiler passes to enhance the model and make it more suitable to running on Xtensa HiFi4 DSPs. A custom [quantizer](https://pytorch.org/tutorials/prototype/quantization_in_pytorch_2_0_export_tutorial.html) is used to represent activations and weights as `uint8` instead of `float`, and call appropriate operators. Finally, custom kernels optimized with Xtensa intrinsics provide runtime acceleration. 11 12::::{grid} 2 13:::{grid-item-card} What you will learn in this tutorial: 14:class-card: card-prerequisites 15* In this tutorial you will learn how to export a quantized model with a linear operation targeted for the Xtensa HiFi4 DSP. 16* You will also learn how to compile and deploy the ExecuTorch runtime with the kernels required for running the quantized model generated in the previous step on the Xtensa HiFi4 DSP. 17::: 18:::{grid-item-card} Tutorials we recommend you complete before this: 19:class-card: card-prerequisites 20* [Introduction to ExecuTorch](intro-how-it-works.md) 21* [Setting up ExecuTorch](getting-started-setup.md) 22* [Building ExecuTorch with CMake](runtime-build-and-cross-compilation.md) 23::: 24:::: 25 26```{note} 27The linux part of this tutorial has been designed and tested on Ubuntu 22.04 LTS, and requires glibc 2.34. Workarounds are available for other distributions, but will not be covered in this tutorial. 28``` 29 30## Prerequisites (Hardware and Software) 31 32In order to be able to succesfully build and run ExecuTorch on a Xtensa HiFi4 DSP you'll need the following hardware and software components. 33 34### Hardware 35 - [i.MX RT600 Evaluation Kit](https://www.nxp.com/design/development-boards/i-mx-evaluation-and-development-boards/i-mx-rt600-evaluation-kit:MIMXRT685-EVK) 36 37### Software 38 - x86-64 Linux system (For compiling the DSP binaries) 39 - [MCUXpresso IDE](https://www.nxp.com/design/software/development-software/mcuxpresso-software-and-tools-/mcuxpresso-integrated-development-environment-ide:MCUXpresso-IDE) 40 - This IDE is supported on multiple platforms including MacOS. You can use it on any of the supported platforms as you'll only be using this to flash the board with the DSP images that you'll be building later on in this tutorial. 41- [J-Link](https://www.segger.com/downloads/jlink/) 42 - Needed to flash the board with the firmware images. You can install this on the same platform that you installed the MCUXpresso IDE on. 43 - Note: depending on the version of the NXP board, another probe than JLink might be installed. In any case, flashing is done using the MCUXpresso IDE in a similar way. 44 - [MCUXpresso SDK](https://mcuxpresso.nxp.com/en/select?device=EVK-MIMXRT685) 45 - Download this SDK to your Linux machine, extract it and take a note of the path where you store it. You'll need this later. 46- [Xtensa compiler](https://tensilicatools.com/platform/i-mx-rt600/) 47 - Download this to your Linux machine. This is needed to build ExecuTorch for the HiFi4 DSP. 48- For cases with optimized kernels, the [nnlib repo](https://github.com/foss-xtensa/nnlib-hifi4). 49 50## Setting up Developer Environment 51 52Step 1. In order to be able to successfully install all the software components specified above users will need to go through the NXP tutorial linked below. Although the tutorial itself walks through a Windows setup, most of the steps translate over to a Linux installation too. 53 54[NXP tutorial on setting up the board and dev environment](https://www.nxp.com/document/guide/getting-started-with-i-mx-rt600-evaluation-kit:GS-MIMXRT685-EVK?section=plug-it-in) 55 56```{note} 57Before proceeding forward to the next section users should be able to succesfullly flash the **dsp_mu_polling_cm33** sample application from the tutorial above and notice output on the UART console indicating that the Cortex-M33 and HiFi4 DSP are talking to each other. 58``` 59 60Step 2. Make sure you have completed the ExecuTorch setup tutorials linked to at the top of this page. 61 62## Working Tree Description 63 64The working tree is: 65 66``` 67executorch 68├── backends 69│ └── cadence 70│ ├── aot 71│ ├── ops_registration 72│ ├── tests 73│ ├── utils 74│ ├── hifi 75│ │ ├── kernels 76│ │ ├── operators 77│ │ └── third-party 78│ │ └── hifi4-nnlib 79│ └── [other cadence DSP families] 80│ ├── kernels 81│ ├── operators 82│ └── third-party 83│ └── [any required lib] 84└── examples 85 └── cadence 86 ├── models 87 └── operators 88``` 89 90***AoT (Ahead-of-Time) Components***: 91 92The AoT folder contains all of the python scripts and functions needed to export the model to an ExecuTorch `.pte` file. In our case, [export_example.py](https://github.com/pytorch/executorch/blob/main/backends/cadence/aot/export_example.py) is an API that takes a model (nn.Module) and representative inputs and runs it through the quantizer (from [quantizer.py](https://github.com/pytorch/executorch/blob/main/backends/cadence/aot/quantizer.py)). Then a few compiler passes, also defined in [quantizer.py](https://github.com/pytorch/executorch/blob/main/backends/cadence/aot/quantizer.py), will replace operators with custom ones that are supported and optimized on the chip. Any operator needed to compute things should be defined in [ops_registrations.py](https://github.com/pytorch/executorch/blob/main/backends/cadence/aot/ops_registrations.py) and have corresponding implemetations in the other folders. 93 94***Operators***: 95 96The operators folder contains two kinds of operators: existing operators from the [ExecuTorch portable library](https://github.com/pytorch/executorch/tree/main/kernels/portable/cpu) and new operators that define custom computations. The former is simply dispatching the operator to the relevant ExecuTorch implementation, while the latter acts as an interface, setting up everything needed for the custom kernels to compute the outputs. 97 98***Kernels***: 99 100The kernels folder contains the optimized kernels that will run on the HiFi4 chip. They use Xtensa intrinsics to deliver high performance at low-power. 101 102## Build 103 104In this step, you will generate the ExecuTorch program from different models. You'll then use this Program (the `.pte` file) during the runtime build step to bake this Program into the DSP image. 105 106***Simple Model***: 107 108The first, simple model is meant to test that all components of this tutorial are working properly, and simply does an add operation. The generated file is called `add.pte`. 109 110```bash 111cd executorch 112python3 -m examples.portable.scripts.export --model_name="add" 113``` 114 115***Quantized Operators***: 116 117The other, more complex model are custom operators, including: 118 - a quantized [linear](https://pytorch.org/docs/stable/generated/torch.nn.Linear.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/cadence/operators/quantized_linear_op.py#L28). Linear is the backbone of most Automatic Speech Recognition (ASR) models. 119 - a quantized [conv1d](https://pytorch.org/docs/stable/generated/torch.nn.Conv1d.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/cadence/operators/quantized_conv1d_op.py#L36). Convolutions are important in wake word and many denoising models. 120 121In both cases the generated file is called `CadenceDemoModel.pte`. 122 123```bash 124cd executorch 125python3 -m examples.cadence.operators.quantized_<linear,conv1d>_op 126``` 127 128***Small Model: RNNT predictor***: 129 130The torchaudio [RNNT-emformer](https://pytorch.org/audio/stable/tutorials/online_asr_tutorial.html) model is an Automatic Speech Recognition (ASR) model, comprised of three different submodels: an encoder, a predictor and a joiner. 131The [predictor](https://github.com/pytorch/executorch/blob/main/examples/cadence/models/rnnt_predictor.py) is a sequence of basic ops (embedding, ReLU, linear, layer norm) and can be exported using: 132 133```bash 134cd executorch 135python3 -m examples.cadence.models.rnnt_predictor 136``` 137 138The generated file is called `CadenceDemoModel.pte`. 139 140### Runtime 141 142**Building the DSP firmware image** 143In this step, you'll be building the DSP firmware image that consists of the sample ExecuTorch runner along with the Program generated from the previous step. This image when loaded onto the DSP will run through the model that this Program consists of. 144 145***Step 1***. Configure the environment variables needed to point to the Xtensa toolchain that you have installed in the previous step. The three environment variables that need to be set include: 146```bash 147# Directory in which the Xtensa toolchain was installed 148export XTENSA_TOOLCHAIN=/home/user_name/cadence/XtDevTools/install/tools 149# The version of the toolchain that was installed. This is essentially the name of the directory 150# that is present in the XTENSA_TOOLCHAIN directory from above. 151export TOOLCHAIN_VER=RI-2021.8-linux 152# The Xtensa core that you're targeting. 153export XTENSA_CORE=nxp_rt600_RI2021_8_newlib 154``` 155 156***Step 2***. Clone the [nnlib repo](https://github.com/foss-xtensa/nnlib-hifi4), which contains optimized kernels and primitives for HiFi4 DSPs, with `git clone [email protected]:foss-xtensa/nnlib-hifi4.git`. 157 158***Step 3***. Run the CMake build. 159In order to run the CMake build, you need the path to the following: 160- The Program generated in the previous step 161- Path to the NXP SDK root. This should have been installed already in the [Setting up Developer Environment](#setting-up-developer-environment) section. This is the directory that contains the folders such as boards, components, devices, and other. 162 163```bash 164cd executorch 165rm -rf cmake-out 166# prebuild and install executorch library 167cmake -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/backends/cadence/cadence.cmake \ 168 -DCMAKE_INSTALL_PREFIX=cmake-out \ 169 -DCMAKE_BUILD_TYPE=Debug \ 170 -DPYTHON_EXECUTABLE=python3 \ 171 -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \ 172 -DEXECUTORCH_BUILD_HOST_TARGETS=ON \ 173 -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \ 174 -DEXECUTORCH_BUILD_PTHREADPOOL=OFF \ 175 -DEXECUTORCH_BUILD_CPUINFO=OFF \ 176 -DEXECUTORCH_BUILD_FLATC=OFF \ 177 -DFLATC_EXECUTABLE="$(which flatc)" \ 178 -Bcmake-out . 179 180cmake --build cmake-out -j<num_cores> --target install --config Debug 181# build cadence runner 182cmake -DCMAKE_BUILD_TYPE=Debug \ 183 -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/backends/cadence.cmake \ 184 -DCMAKE_PREFIX_PATH=<path_to_executorch>/cmake-out \ 185 -DMODEL_PATH=<path_to_program_file_generated_in_previous_step> \ 186 -DNXP_SDK_ROOT_DIR=<path_to_nxp_sdk_root> -DEXECUTORCH_BUILD_FLATC=0 \ 187 -DFLATC_EXECUTABLE="$(which flatc)" \ 188 -DNN_LIB_BASE_DIR=<path_to_nnlib_cloned_in_step_2> \ 189 -Bcmake-out/examples/cadence \ 190 examples/cadence 191 192cmake --build cmake-out/examples/cadence -j8 -t cadence_executorch_example 193``` 194 195After having succesfully run the above step you should see two binary files in their CMake output directory. 196```bash 197> ls cmake-xt/*.bin 198cmake-xt/dsp_data_release.bin cmake-xt/dsp_text_release.bin 199``` 200 201## Deploying and Running on Device 202 203***Step 1***. You now take the DSP binary images generated from the previous step and copy them over into your NXP workspace created in the [Setting up Developer Environment](#setting-up-developer-environment) section. Copy the DSP images into the `dsp_binary` section highlighted in the image below. 204 205<img src="_static/img/dsp_binary.png" alt="MCUXpresso IDE" /><br> 206 207```{note} 208As long as binaries have been built using the Xtensa toolchain on Linux, flashing the board and running on the chip can be done only with the MCUXpresso IDE, which is available on all platforms (Linux, MacOS, Windows). 209``` 210 211***Step 2***. Clean your work space 212 213***Step 3***. Click **Debug your Project** which will flash the board with your binaries. 214 215On the UART console connected to your board (at a default baud rate of 115200), you should see an output similar to this: 216 217```bash 218> screen /dev/tty.usbmodem0007288234991 115200 219Executed model 220Model executed successfully. 221First 20 elements of output 0 2220.165528 0.331055 ... 223``` 224 225## Conclusion and Future Work 226 227In this tutorial, you have learned how to export a quantized operation, build the ExecuTorch runtime and run this model on the Xtensa HiFi4 DSP chip. 228 229The (quantized linear) model in this tutorial is a typical operation appearing in ASR models, and can be extended to a complete ASR model by creating the model as a new test and adding the needed operators/kernels to [operators](https://github.com/pytorch/executorch/blob/main/backends/cadence/hifi/operators) and [kernels](https://github.com/pytorch/executorch/blob/main/backends/cadence/hifi/kernels). 230 231Other models can be created following the same structure, always assuming that operators and kernels are available. 232
