[petsc-dev] Cusp & Thrust test

[Satish Balay]

Matt,

With your recent NVCC change - I get the following breakage with
my regular linux build.

Satish

---------

================================================================================
TEST checkCUDAPreprocessor from config.setCompilers(/home/balay/tmp/petsc-dev/config/BuildSystem/config/setCompilers.py:531)
TESTING: checkCUDAPreprocessor from config.setCompilers(config/BuildSystem/config/setCompilers.py:531)
  Locate a functional CUDA preprocessor
*******************************************************************************
        CONFIGURATION CRASH  (Please send configure.log to petsc-maint at mcs.anl.gov)
*******************************************************************************
'Configure' object has no attribute 'CUDACC'  File "/home/balay/tmp/petsc-dev/config/configure.py", line 270, in petsc_configure
    framework.configure(out = sys.stdout)
  File "/home/balay/tmp/petsc-dev/config/BuildSystem/config/framework.py", line 946, in configure
    child.configure()
  File "/home/balay/tmp/petsc-dev/config/BuildSystem/config/setCompilers.py", line 1384, in configure
    self.executeTest(self.checkCUDAPreprocessor)
  File "/home/balay/tmp/petsc-dev/config/BuildSystem/config/base.py", line 97, in executeTest
    ret = apply(test, args,kargs)
  File "/home/balay/tmp/petsc-dev/config/BuildSystem/config/setCompilers.py", line 533, in checkCUDAPreprocessor
    for compiler in self.generateCUDAPreprocessorGuesses():
  File "/home/balay/tmp/petsc-dev/config/BuildSystem/config/setCompilers.py", line 528, in generateCUDAPreprocessorGuesses
    yield self.CUDACC+' -E'
(END) 




On Mon, 26 Jul 2010, Matthew Knepley wrote:

> On Mon, Jul 26, 2010 at 7:50 AM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> >
> > On Jul 25, 2010, at 8:26 AM, Matthew Knepley wrote:
> >
> > > They work with normal compilers if you drop an empty file in 'thrust'
> > called 'host_defines.h'.
> > > I think this should be the prefered checking mechanism since now we have
> > fucked up dependencies
> >
> >    So people who use CUDA have to drop an empty file in that strange system
> > location? Doesn't sound like a good solution to the problem.
> >
> >   Since cusp and thrust are never used with the regular compilers shouldn't
> > they be tested with the compiler that actually uses them, nvcc? Thus you
> > just need to add a new language CUDA to BuildSystem, which it needs anyways
> > and test with that.
> 
> 
> How is it that being right makes more work for me?
> 
>    Matt
> 
> 
> >
> >  Barry
> >
> >
> > > again.
> > >
> > >    Matt
> > >
> > > --
> > > What most experimenters take for granted before they begin their
> > experiments is infinitely more interesting than any results to which their
> > experiments lead.
> > > -- Norbert Wiener
> >
> >
> 
> 
>