MathGroup Archive 2011

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: CUDA XCompiler

  • To: mathgroup at smc.vnet.net
  • Subject: [mg122832] Re: CUDA XCompiler
  • From: DmitryG <einschlag at gmail.com>
  • Date: Fri, 11 Nov 2011 04:58:30 -0500 (EST)
  • Delivered-to: l-mathgroup@mail-archive0.wolfram.com
  • References: <j76592$2c6$1@smc.vnet.net>

On Oct 13, 2:50 am, Grischika <Grisch... at mail.ru> wrote:
> Hello!
> This is initialization code of CUDA:
> Needs["CUDALink`"]
> Needs["CCompilerDriver`"]
> In[]:=CUDAQ[]
> Out:=True
>
>  The command  CUDADot[RandomInteger[1, {5, 5}], RandomInteger[1, {5,
> 5}]] works as it should.
>
> But evaluation of
> fun = CUDAFunctionLoad["__global__ void zero(int * in, int length) {
>       int index = threadIdx.x + blockIdx.x * blockDim.x;
>
>         if (index < length)
>            in[index] = 0;
>     }", "zero", {{_Integer}, _Integer}, {10}]
> gives the error message:
> CUDAFunctionLoad::invxpth: The "XCompilerInstallation" option set to
> $Failed is not valid. "XCompilerInstallation" must be a string
> pointing to the C compiler directory.
>
> So it means that something wrong with C Compiler. All compilations
> like Compile[... ,CompilationTarget->"C"]  work.
>
> Here information about the compiler . Microsoft Visual Studio 10.0
> Ultimate.
> In[]:=CCompilers[]
> Out[]:={{"Name" -> "Visual Studio",
>   "Compiler" ->
>    CCompilerDriver`VisualStudioCompiler`VisualStudioCompiler,
>   "CompilerInstallation" ->
>    "C:\\Program Files\\Microsoft Visual Studio 10.0",
>   "CompilerName" -> Automatic}, {"Name" -> "NVIDIACUDA Compiler",
>   "Compiler" -> NVCCCompiler,
>   "CompilerInstallation" ->
>    "C:\\Documents and Settings\\Grigory\\Application \
> Data\\Mathematica\\Paclets\\Repository\\CUDAResources-Win-8.0.0.7\\\
> CUDAToolkit\\bin\\", "CompilerName" -> Automatic}}
>
> In[]:=CUDACCompilers[]
> Out[]:={}
>
> How to fix this compilation problem?

I have tested this piece of code on my system (also Microsoft Visual
Studio 10) and I've got the error message

CUDAFunctionLoad::cmpf: The kernel compilation failed. Consider
setting the option "ShellOutputFunction"->Print to display the
compiler error message. >>

I cannot say more to this because I am not a C programmer.

I wanted to use simple CUDA commands like CUDADot but they are slower
the regular commands, so I do not yet see the point in GPU computing
with Mathematica.

Best,

Dmitry



  • Prev by Date: Manipulate variables not getting evaluated
  • Next by Date: Re: Pattern to match element in nested list
  • Previous by thread: Re: Manipulate variables not getting evaluated
  • Next by thread: Re: CUDA XCompiler