This project is a work in progress. Documentation and testing are not complete.
This project implements routines for matrix operations on CPU and AMD GPU processors. The project includes the following components:
- A set of C++ source files for the core algorithms; these can be used independently of R.
- A set of R package files that encapsulate the C++ source and can be installed and used within R.
- A set of Mac OS XCode project files that can be used for developing and testing the C++ files.
- A makefile that can be used to compile the project from the Mac OS X or Linux command line.
This project requires installation of the AMD clBLAS library.
On OS X, the project uses the built-in Accelerate framework. On Linux, the project requires installation of the OpenCL and OpenBLAS libraries.
See http://blog.quadrivio.com for installation instructions.
- C++
To use the C++ source files within a separate C++ project, copy the contents of
multiblas/src. Compile with the preprocessor definition RPROJECT equal to 0 to remove the
R-specific code, i.e., use -DRPROJECT=0 in gcc or insert #define RPROJECT 0 at the
top of the file shim.h. Messages will be sent to stderr. Logical and runtime
errors will be thrown with std::logic_error or std::runtime_error.
- R
To create a package usable within R, build and check the R package as follows:
# OS X
cd path_to_top_directory
export PKG_CPPFLAGS="-I/usr/local/clblas/include"
export PKG_LIBS='$(BLAS_LIBS) -framework OpenCL -L/usr/local/clblas/lib -lclBLAS'
R CMD build multiblas
R CMD check multiblas_0.91-1.tar.gz
# Linux
cd path_to_top_directory
export PKG_CPPFLAGS="-I/usr/lib64/openblas/include -I/opt/AMDAPPSDK-3.0/include -Dnullptr='NULL' -I/opt/clBLAS-2.4.0-Linux-x64/include"
export PKG_LIBS='$(BLAS_LIBS) -L/opt/AMDAPPSDK-3.0/lib/x86_64 -lOpenCL -L/opt/clBLAS-2.4.0-Linux-x64/lib64 -lclBLAS'
R CMD build multiblas
R CMD check multiblas_0.91-1.tar.gz
(If you download and expand the zip file, the top directory is named multiblas-master.) Then, install the package within R as follows:
# OS X
setwd("path_to_top_directory")
Sys.setenv(PKG_CPPFLAGS = "-I/usr/local/clblas/include")
Sys.setenv(PKG_LIBS = "$(BLAS_LIBS) -framework OpenCL -L/usr/local/clblas/lib/ -lclBLAS")
install.packages("multiblas_0.91-1.tar.gz", repos = NULL, type = "source")
library('multiblas')
# Linux
setwd("path_to_top_directory")
Sys.setenv(PKG_CPPFLAGS="-I/usr/lib64/openblas/include -I/opt/AMDAPPSDK-3.0/include -Dnullptr='NULL' -I/opt/clBLAS-2.4.0-Linux-x64/include")
Sys.setenv(PKG_LIBS = "$(BLAS_LIBS) -L/opt/AMDAPPSDK-3.0/lib/x86_64 -lOpenCL -L/opt/clBLAS-2.4.0-Linux-x64/lib64 -lclBLAS")
install.packages("multiblas_0.91-1.tar.gz", repos = NULL, type = "source")
library('multiblas')
When the C++ files are compiled for use in R, the preprocessor parameter RPROJECT is
automatically defined to 1. The code in shim.h then causes messages to be rerouted to
the R function Rprintf() instead of to stderr. Errors are signaled via the R
function error() instead of by throwing exceptions.
- XCode and command-line
The project includes main.cpp to allow the code to run outside of R. Running the project from
XCode or the command-line causes a few sample matrix multiplications to be calculated.
To build using the command line, cd to the directory containing makefile and type make.
Max OS 10.12.5
CentOS 6.9
R 3.3.0
XCode 8.3.3
This project has not been tested or modified for Windows.
BSD