duncantl / rcuda Goto Github PK

R interface to CUDA libraries

License: Other

TeX 3.60% R 13.50% Cuda 1.71% C 13.90% Turing 67.09% Makefile 0.06% C++ 0.14% Shell 0.01%

rcuda's Introduction

This RCUDA package provides a reasonably comprehensive interface to the CUDA SDK. This allows one to work with a GPU via R code. We can invoke CUDA code on the GPU, put data on the GPU and bring it back into R. We have access to devices, contexts, streams, events, and all of the CUDA concepts.

The package is now updated to work with CUDA SDK 7.5.

The package requires the RAutoGenRunTime package to be installed. This is available on github and omegahat.net and omegahat.ucdavis.edu.

You may need to set LD_LIBRARY_PATH - or DYLD_LIBRARY_PATH on OS X - to allow the system find the CUDA libraries. They are often installed in /usr/local/cuda/lib.

The code in the package is a mix of manually written and programmatically generated code. The R code in TU/clang.R is used to generate the code and can be re-run for different versions of CUDA. TU/clang.R

rcuda's People

Contributors

Stargazers

Watchers

Forkers

hjyaragon95 rukyhuang xfbingshan wush978 pachoalvarez strategist922 tenaciousjai mardukbp chen0031

rcuda's Issues

problems with cudaversion filename and path

I get an error on configure:

gcc: error: cudaVersion.c: No such file or directory

Because (1) the file seems now to be called cudaversion.c (all lowercase), and (2) at least on my Ubuntu linux, I need to specify the current directory when calling the resulting binary (./cudaVersion instead of cudaVersion)

This is fixed with changing the following two lines (lines 85 & 86) of configure.in

---$CC -o cudaVersion -I$CUDA_INC_DIR cudaVersion.c -L$CUDA_LIB_DIR -lcuda -lcudart $RPATH
---CUDA_VERSION=`cudaVersion`
+++$CC -o cudaVersion -I$CUDA_INC_DIR cudaversion.c -L$CUDA_LIB_DIR -lcuda -lcudart $RPATH
+++CUDA_VERSION=`./cudaVersion`

And then regenerating configure with autoconf

cu_computemode_exclusive no longer exists?

When building/installing RCUDA, I get the following error:

autoEnumConverters.c: In function ‘Renum_convert_CUcomputemode_enum’:
autoEnumConverters.c:1073:13: error: ‘CU_COMPUTEMODE_EXCLUSIVE’ undeclared (first use in this function)
        case CU_COMPUTEMODE_EXCLUSIVE:

There is evidence on the internet that CU_COMPUTEMODE_EXCLUSIVE used to be a member of the CUcomputemode_enum but now it seems there is only CU_COMPUTEMODE_EXCLUSIVE_PROCESS ?

Removing lines 1073-1075 of autoEnumConverters.c allowed for successful build/install on my box (Ubuntu 16.04, Cuda 8.0.61, R 3.4.0):

---       case CU_COMPUTEMODE_EXCLUSIVE:
---           elName = "CU_COMPUTEMODE_EXCLUSIVE";
---       break;

duncantl / rcuda Goto Github PK

rcuda's Introduction

rcuda's People

Contributors

Stargazers

Watchers

Forkers

rcuda's Issues

problems with cudaversion filename and path

cu_computemode_exclusive no longer exists?

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent