[R-sig-hpc] GPU package crowd-source testing

Charles Determan cdetermanjr at gmail.com
Thu Feb 11 16:36:12 CET 2016


R Users,

My sincere thanks to all those who have been coming forward to test my GPU
package and provide bug reports.  I want to followup on my initial request
with a few qualifiers.

1. I neglected to tell users to also use my github version of 'RViennaCL'
instead of the CRAN version.  I have made some updates that I was
postponing release of until I can solve the multiple device issues with
'gpuR'.

devtools::install_github("cdeterman/RViennaCL")

2. When reporting bugs, either directly to me or ideally in my github
issues (https://github.com/cdeterman/gpuR/issues), please provide your
Operating System, OpenCL version (e.g. 1.0, 1.2, 2.0), OpenCL SDK (e.g.
AMD, CUDA toolkit, etc.) and GPU device.  If you don't know these things
you can get them from Sys.info() for the OS, platformInfo() for the OpenCL
SDK, gpuInfo() for the GPU information, and check your OpenCL header (cl.h)
for the /* OpenCL Version */ section for the highest version number.

3. If you have installed 'gpuR' and it is running without problems I would
still like to know this.  It would be good to begin generating a list of
'tested' devices and associated platform.  I have just created a gitter
account.  I am relatively new to it but I'm hoping it can be used to try
and consolidate responses.  In this case, you can simply reply on the
Tested_GPUs thread (https://gitter.im/cdeterman/gpuR/Tested_GPUs) with your
device and platform backend.

Again, thanks to all for taking the time to try out this package.

Regards,
Charles



On Tue, Feb 9, 2016 at 12:20 PM, Charles Determan <cdetermanjr at gmail.com>
wrote:

> Greetings R users,
>
> I would like to request any users who would be willing to test one of my
> packages.  Normally I would be content using testthat and continuous
> integration services but this particular package is used for GPU computing
> (hence the cross-posting).  It is intended to be as general as possible for
> available devices but I only have access to so much hardware.  I can't
> possibly test it against every GPU available.
>
> As such, I would sincerely appreciate any user that has at least one GPU
> device (Intel, AMD, or NVIDIA) and is willing to experiment with the
> package to try it out.  Note, this will require installing an OpenCL SDK of
> some form.  Installation instructions for the package are found here (
> https://github.com/cdeterman/gpuR/wiki).
>
> At the very least, if you have a valid device, you would only need to
> download the 'development' version of the package and experiment with the
> functions such as a matrix multiplication.
>
> devtools::install_github("cdeterman/gpuR", ref = "develop")
>
> library(gpuR)
> A <- gpuMatrix(rnorm(10000), 100, 100)
> A %*% A
>
> You could also clone my github repo and run all the unit tests I have
> included
>
> git clone -b develop https://github.com/cdeterman/gpuR.git
>
> If using RStudio, just open the package in a new project and press
> 'Ctrl-Shift-T' or more directly run  `devtools::test()`
>
> If using command-line R, switch to the gpuR directory, start R and run
> `devtools::test()`.
>
> If you find any errors or bugs, please report them in my github issues (
> https://github.com/cdeterman/gpuR/issues).  Naturally any recommendations
> on additional features are welcome.
>
> Thank you in advance for any support you can provide.  I want to continue
> improving this package but I am beginning to reach the end of what I can
> accomplish from a hardware perspective.
>
> Best Regards,
> Charles
>
>
>

	[[alternative HTML version deleted]]



More information about the R-sig-hpc mailing list