This is a fork of FastR with GPU support. The GPU backend is an extension of GraalVM and Graal Compiler with OpenCL code generation and data management for GPU computing.
The goal of the FastR-GPU compiler is to automatically execute R expressions on the GPU. It extends Truffle, Graal and a GPU backend for Graal (Marawacc) for compiling an R program into OpenCL. It profiles and specialises R input programs to OpenCL using the Graal Partial Evaluator and compiles the resulting intermediate representation to OpenCL at runtime. This project is a research prototype.
> marawacc.deviceInfo()
NAME : Hawaii
VENDOR : Advanced Micro Devices, Inc.
TYPE : GPU
DRIVER : 1598.5 (VM)
MAX COMPUTE UNITS: 44
MAX FREQUENCY : 1030
GLOBAL MEMORY : 3221225472
LOCAL MEMORY : 32768
ENDIANESS : LITTLE_ENDIAN
> marawacc.init()
> size <- 100000
> # It is executed on the GPU when the R program is JIT by Graal
> mapply(function(x, y) x * y, runif(size), runif(size))
Get the mx tool:
$ mkdir ~/fastr-gpu/
$ cd fastr-gpu
$ git clone https://github.com/graalvm/mx
$ cd mx
$ git checkout 900cc06
$ cd -
$ export PATH=$PWD/mx:$PATHCreate source file:
export PATH=/path/to/mx/mx/:$PATH
export JAVA_HOME=/path/to/jdk1.8x/
export DEFAULT_VM="server"Get the R & Marawacc (OpenCL) JIT compilers:
$ mkdir fastr-gpu
$ cd fastr-gpu
$ git clone [email protected]:jjfumero/jvmci-marawacc.git jvmci ## Download JVMCI dependency
$ git clone [email protected]:jjfumero/truffle-marawacc.git truffle ## Download Truffle dependency
$ git clone https://github.com/jjfumero/graal-marawacc graal ## Download Graal dependency
$ git clone [email protected]:jjfumero/marawacc.git marawacc
$ export JAVA_HOME=/home/juan/bin/jdk1.8.0_91
$ export DEFAULT_VM="server"
$ export PATH=/home/juan/bin/gcc/gcc-5.4.0/bin/:$PATH
$ export LD_LIBRARY_PATH=/home/juan/bin/gcc/gcc-5.4.0/lib64/:$LD_LIBRARY_PATH
$ export PATH=$PWD/mx:$PATH
$ cd marawacc
$ make Build fastr-gpu:
$ cd ..
$ git clone [email protected]:jjfumero/fastr-gpu.git
$ cd fastr-gpu
$ make
$ . source.sh Build done!!!
To generate the Eclipse files:
$ make eclipse Then import the projects into eclipse
This compiler has been tested on Linux Fedora 21/22/23, CentOS 7.4 - CentOS 7.9 and OpenSuse 13 with OpenJDK >= 1.8_61.
Current implementation with JDK 8 u91.
-
Juan Fumero. Accelerating interpreted programming languages on GPUs with just-in-time compilation and runtime optimisations. PhD Dissertation, 30/11/2017 - Link
-
Juan Fumero, Michel Steuwer, Lukas Stadler, and Christophe Dubach. Just-In-Time GPU Compilation for Interpreted Languages with Partial Evaluation. In Proceedings of the 13th ACM SIGPLAN/SIGOPS International Conference on Virtual Execution Environments (VEE '17). ACM, New York, NY, USA, 60-73. DOI: https://doi.org/10.1145/3050748.3050761
-
Juan Fumero, Michel Steuwer, Lukas Stadler, Christophe Dubach. OpenCL JIT Compilation for Dynamic Programming Language. MoreVMs 2017. http://conf.researchr.org/event/MoreVMs-2017/morevms-2017-papers-opencl-jit-compilation-for-dynamic-programming-languages
GPL V2
This project is a research prototype implemented at The University of Edinburgh. The project was partially funded by Oracle Labs.
Juan Fumero < juan.fumero @ manchester.ac.uk >
Christophe Dubach < christophe.dubach @ ed.ac.uk >
Michel Steuwer < michel.steuwer @ ed.ac.uk >
Lukas Stadler < lukas.stadler @ oracle.com >
FastR is an implementation of the R Language in Java atop Truffle and Graal. Truffle is a framework for building self-optimizing AST interpreters. Graal is a dynamic compiler that is used to generate efficient machine code from partially evaluated Truffle ASTs.
FastR is an open-source effort of Purdue University, Johannes Kepler University Linz, and Oracle Labs.
For more details and instructions for downloading and building the system, please visit the FastR Wiki.