rocAL Introduction#

Today’s deep learning applications require loading and pre-processing data efficiently to achieve high processing throughput. This requires creating efficient processing pipelines fully utilizing the underlying hardware capabilities. Some examples are load and decode data, do a variety of augmentations, color-format conversions, etc. Deep learning frameworks require supporting multiple data formats and augmentations to adapt to a variety of data-sets and models.

AMD ROCm Augmentation Library (rocAL) is designed to efficiently do such processing pipelines from both images and video as well as from a variety of storage formats. These pipelines are programmable by the user using both C++ and Python APIs.

Key Components of rocAL#

  • Full processing pipeline support for data_loading, meta-data loading, augmentations, and data-format conversions for training and inference.

  • Being able to do processing on CPU or Radeon GPU (with OpenCL or HIP backend)

  • Ease of integration with framework plugins in Python

  • Support variety of augmentation operations through AMD’s ROCm Performance Primitives (RPP).

  • All available public and open-sourced under ROCm.

Prerequisites#

Refer to the rocAL to follow and install pre-requisites.

Build instructions#

Follow the build instructions in rocAL

rocAL Python#

  • rocAL Python package has been created using Pybind11 which enables data transfer between rocAL C++ API and Python API.

  • Module imports are made similar to other data loaders like NVidia’s DALI.

  • rocal_pybind package has both PyTorch and TensorFlow framework support.

  • Various reader format support including FileReader, COCOReader, and TFRecordReader.

  • example folder contains sample implementations for each reader variation as well as sample training script for PyTorch

  • rocAL is integrated into MLPerf Resnet-50 Pytorch classification example on the ImageNet dataset.

rocAL Python API#

amd.rocal.fn#

  • Contains the image augmentations & file read and decode operations which are linked to rocAL C++ API

  • All ops (listed below) are supported for the single input image and batched inputs.

Image Augmentation

Reader and Decoder

Geometric Ops

ColorTwist

File Reader

CropMirrorNormalize

Brightness

ImageDecoder

Resize

Gamma Correction

ImageDecoderRandomCrop

ResizeCrop

Snow

COCOReader

WarpAffine

Rain

TFRecordReader

FishEye

Blur

LensCorrection

Jitter

Rotate

Hue

Saturation

Fog

Contrast

Vignette

SNPNoise

Pixelate

Blend

amd.rocal.pipeline#

  • Contains Pipeline class which has all the data needed to build and run the rocAL graph.

  • Contains support for context/graph creation, verify and run the graph.

  • Has data transfer functions to exchange data between frameworks and rocAL

  • define_graph functionality has been implemented to add nodes to build a pipeline graph.

amd.rocal.types#

amd.rocal.types are enums exported from C++ API to python. Some examples include CPU, GPU, FLOAT, FLOAT16, RGB, GRAY, etc…

amd.rocal.plugin.pytorch#

  • Contains ROCALGenericIterator for Pytorch.

  • ROCALClassificationIterator class implements iterator for image classification and return images with corresponding labels.

  • From the above classes, any hybrid iterator pipeline can be created by adding augmentations.

  • see example PyTorch Simple Example. Requires PyTorch.

installing rocAL python plugin (Python 3.6)#

  • Build and install RPP

  • Build and install MIVisionX which installs rocAL c++ lib

  • Go to rocal_pybind folder

  • sudo ./run.sh

Steps to run MLPerf Resnet50 classification training with rocAL on a system with MI50+ and ROCm#

  • Step 1: Ensure you have downloaded ILSVRC2012_img_val.tar (6.3GB) and ILSVRC2012_img_train.tar (138 GB) files and unzip into train and val folders

  • Step 2: Build MIVisionX Pytorch docker

  • Step 3: Install rocAL python_pybind plugin as described above

  • Step 4: Clone MLPerf branch and checkout mlperf-v1.1-rocal branch

git clone -b mlperf-v1.1-rocal https://github.com/rrawther/MLPerf-mGPU
  • Step 5: Modify RN50_AMP_LARS_8GPUS_NCHW.sh or RN50_AMP_LARS_8GPUS_NHWC.sh to reflect correct path for imagenet directory

  • Step 8: Run RN50_AMP_LARS_8GPUS_NCHC.sh or RN50_AMP_LARS_8GPUS_NHWC.sh

./RN50_AMP_LARS_8GPUS_NCHW.sh 
(or)
./RN50_AMP_LARS_8GPUS_NHWC.sh

MIVisionX Pytorch Docker#

  • Refer to the docker page for prerequisites and information on building the docker

  • Step 1: Run the docker image*

sudo docker run -it -v <Path-To-Data-HostSystem>:/data -v /<Path-to-GitRepo>:/dockerx -w /dockerx --privileged --device=/dev/kfd --device=/dev/dri --group-add video --shm-size=4g --ipc="host" --network=host <docker-name>
  • Optional: Map localhost directory on the docker image

    • option to map the localhost directory with imagenet dataset folder to be accessed on the docker image.

    • usage: -v {LOCAL_HOST_DIRECTORY_PATH}:{DOCKER_DIRECTORY_PATH}