Skip to content

Latest commit

 

History

History
227 lines (153 loc) · 8.16 KB

README.md

File metadata and controls

227 lines (153 loc) · 8.16 KB
Raw

Docker for TornadoVM

We have two docker configurations for TornadoVM using 2 different JDKs:

  • TornadoVM Docker for NVIDIA GPUs: See instructions

    • JDKs supported:
      • TornadoVM with OpenJDK 21
      • TornadoVM with GraalVM 23.1.0 and JDK 21

  • TornadoVM Docker for Intel Integrated Graphics, Intel CPUs, and Intel FPGAs (Emulated Mode): See instructions

    • JDKs supported:
      • TornadoVM with OpenJDK 21
      • TornadoVM with GraalVM 23.1.0 and JDK 21

  • TornadoVM Docker for Polyglot GraalVM Language Implementations: See instructions

    • JDKs supported:
      • TornadoVM with GraalVM 23.1.0 JDK 21

Nvidia GPUs

Prerequisites

The tornadovm-nvidia-openjdk docker image needs the docker nvidia daemon. More info here: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html.

How to run?

1) Pull the image

For the tornadovm-nvidia-openjdk image:

$ docker pull beehivelab/tornadovm-nvidia-openjdk:latest

This image uses the latest TornadoVM for NVIDIA GPUs and OpenJDK 21.

2) Run an experiment

We provide a runner script that compiles and run your Java programs with TornadoVM. Here's an example:

$ git clone https://github.com/beehive-lab/docker-tornadovm
$ cd docker-tornadovm

## Run Matrix Multiplication - provided in the docker-tornadovm repository
$ ./run_nvidia_openjdk.sh tornado -cp example/target/example-1.0-SNAPSHOT.jar example.MatrixMultiplication 2048

Computing MxM of 2048x2048
	CPU Execution: 0.36 GFlops, Total time = 48254 ms
	GPU Execution: 277.09 GFlops, Total Time = 62 ms
	Speedup: 778x

Using TornadoVM with GraalVM for NVIDIA GPUs

With JDK 21:

$ docker pull beehivelab/tornadovm-nvidia-graalvm:latest

Some options

# To see the generated OpenCL kernel
$ ./run_nvidia.sh tornado --printKernel example/MatrixMultiplication

# To check some runtime info about the kernel execution and device
$ ./run_nvidia.sh tornado --debug example/MatrixMultiplication

The tornado command is just an alias to the java command with all the parameters for TornadoVM execution. So you can pass any Java (OpenJDK or Hotspot) parameter.

$ ./run_nvidia.sh tornado --jvm="-Xmx16g -Xms16g" example/MatrixMultiplication

Intel Integrated Graphics

Prerequisites

The beehivelab/tornadovm-intel-openjdk docker image Intel OpenCL driver for the integrated GPU installed. More info here: https://github.com/intel/compute-runtime.

How to run?

1) Pull the image

For the beehivelab/tornadovm-intel-openjdk image:

$ docker pull beehivelab/tornadovm-intel-openjdk:latest

This image uses the latest TornadoVM for Intel integrated graphics and OpenJDK 21.

2) Run an experiment

We provide a runner script that compiles and run your Java programs with TornadoVM. Here's an example:

$ git clone https://github.com/beehive-lab/docker-tornadovm
$ cd docker-tornadovm

## Run Matrix Multiplication - provided in the docker-tornadovm repository
$ ./run_intel_openjdk.sh tornado -cp example/target/example-1.0-SNAPSHOT.jar example.MatrixMultiplication 256

Computing MxM of 256x256
	CPU Execution: 1.53 GFlops, Total time = 22 ms
	GPU Execution: 8.39 GFlops, Total Time = 4 ms
	Speedup: 5x

Running on FPGAs (Emulation mode)?

The TornadoVM docker image for the Intel platforms contain the FPGA in device 1:0. To offload a Java application onto an FPGA, you can use the following command (example running the DFT application).

$ ./run_intel_openjdk.sh tornado --threadInfo  --jvm="-Ds0.t0.device=1:0" -m tornado.examples/uk.ac.manchester.tornado.examples.dynamic.DFTDynamic 256 default 1
WARNING: Using incubator modules: jdk.incubator.foreign, jdk.incubator.vector
Initialization time:  1066024424 ns
 
Task info: s0.t0
        Backend           : OPENCL
        Device            : Intel(R) FPGA Emulation Device CL_DEVICE_TYPE_ACCELERATOR (available)
        Dims              : 1
        Global work offset: [0]
        Global work size  : [256]
        Local  work size  : [64, 1, 1]
        Number of workgroups  : [4]
 
Total time:  276927741 ns 
 
Is valid?: true
 
Validation: SUCCESS

Using TornadoVM with GraalVM for Intel Integrated Graphics

With JDK 21:

$ docker pull beehivelab/tornadovm-intel-graalvm:latest

Polyglot GraalVM Language Implementations

Prerequisites

Currently, there are three docker images available that combine TornadoVM with polyglot GraalVM language implementations (GraalPy, GraalJS and TruffleRuby) and include the OpenCL drivers for NVIDIA GPUs. The three docker images need the docker nvidia daemon. More info here: https://github.com/NVIDIA/nvidia-docker.

How to run?

1) Pull the images. The images use the latest TornadoVM for NVIDIA GPUs and OpenJDK 21.

  • For the tornadovm-polyglot-graalpy-23.1.0-nvidia-opencl-container image:
$ docker pull beehivelab/tornadovm-polyglot-graalpy-23.1.0-nvidia-opencl-container:latest
  • For the tornadovm-polyglot-graaljs-23.1.0-nvidia-opencl-container image:
$ docker pull beehivelab/tornadovm-polyglot-graaljs-23.1.0-nvidia-opencl-container:latest
  • For the tornadovm-polyglot-truffleruby-23.1.0-nvidia-opencl-container image:
$ docker pull beehivelab/tornadovm-polyglot-truffleruby-23.1.0-nvidia-opencl-container:latest

2) Run an experiment

We provide a runner script for each image in order to compile and run your Python, JavaScript and Ruby programs with TornadoVM. Here's an example taken from TornadoVM documentation that executes a matrix multiplication OpenCL kernel from Python, JavaScript and Ruby:

  • Python:
$ git clone https://github.com/beehive-lab/docker-tornadovm
$ cd docker-tornadovm

## Launch the docker image with the NVIDIA OpenCL runtime
$ ./polyglotImages/polyglot-graalpy/tornadovm-polyglot-nvidia.sh

## Run Matrix Multiplication from a Python program.
$ ./polyglotImages/polyglot-graalpy/tornadovm-polyglot-nvidia.sh tornado --printKernel --truffle python example/polyglot-examples/mxmWithTornadoVM.py

## Launch the docker image with the Intel oneAPI runtime
$ ./polyglotImages/polyglot-graalpy/tornadovm-polyglot-intel.sh

## Run Matrix Multiplication from a Python program.
$ ./polyglotImages/polyglot-graalpy/tornadovm-polyglot-intel.sh tornado --printKernel --truffle python example/polyglot-examples/mxmWithTornadoVM.py
  • JavaScript:
$ git clone https://github.com/beehive-lab/docker-tornadovm
$ cd docker-tornadovm

## Launch the docker image with the NVIDIA OpenCL runtime
$ ./polyglotImages/polyglot-graaljs/tornadovm-polyglot.sh

## Run Matrix Multiplication from a JavaScript program.
$ ./polyglotImages/polyglot-graaljs/tornadovm-polyglot.sh tornado --printKernel --truffle js example/polyglot-examples/mxmWithTornadoVM.js
  • Ruby:
$ git clone https://github.com/beehive-lab/docker-tornadovm
$ cd docker-tornadovm

## Launch the docker image with the NVIDIA OpenCL runtime
$ ./polyglotImages/polyglot-truffleruby/tornadovm-polyglot.sh

## Run Matrix Multiplication from a Python program.
$ ./polyglotImages/polyglot-truffleruby/tornadovm-polyglot.sh tornado --printKernel --truffle ruby example/polyglot-examples/mxmWithTornadoVM.rb

Enjoy TornadoVM!

Docker scripts have been inspired by blang/latex-docker

License

This project is developed at The University of Manchester, and it is fully open source under the Apache 2 license.