Reputation: 2561
I have installed a Nvidia graphic card ( GeForce GT 640 M) and CUDA 5.0. Then I installed the gputools package in R. The package installs perfectly but when I try to use a function in it, it gives me an invalid device function error.
require("gputools")
mat = matrix(runif(25),5,5)
gpuCor(mat)
Error in gpuCor(mat) : invalid device function
In addition: Warning message:
In gpuCor(mat) : PMCC function : kernel finish and memcpy
I have ensured that the cuda samples work fine and my own cuda scripts also work properly. The gpuMatMult function works in the package(I checked the source file and it does not have any device function in it) but functions with device functions is going for a toss.
Following are my system details OS - Ubuntu 12.04 Cuda - CUDA - 5.0 Graphic Card - NVIDIA GEFORCE GT 640 M RAM - 12 GB
Anyone else faced this issue?
Upvotes: 2
Views: 1190
Reputation: 2561
I have found a work around for this. The problem was in the package's makefile, the compilation architectures are passed to the compiler using (arch_compute = sm_10) where the author of the package has put in a list of architectures that should take care of most of the devices. GeForce 640 M for some reason is not there in that list. I removed the whole list and did not pass any architecture arguments and the package installed fine.
Upvotes: 2