[petsc-dev] Why did you remove my CUDA version check!!!
Victor Minden
victorminden at gmail.com
Sun Nov 7 11:36:10 CST 2010
It looks like things are fixed now (?) so maybe this is moot but I think it
was mid- to late- september that I originally put a version check in
setCompilers, but it wasn't working correctly so I'd added a line or two,
not realizing Matt was already re-hauling it. Sorry about that.
-Victor
---
Victor L. Minden
Tufts University
School of Engineering
Class of 2012
On Fri, Nov 5, 2010 at 3:07 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> Matt,
>
> I don't see self.setCompilers.compilerVersionCUDA in rev history. So I
> think this code was broken when you added this stuff [and victor must
> have tried to fix it..
>
> You can revert relavent changes:
> BuildSystem
> febbb50e3b78
> 0d934bb871e4
> 051ab1553c96
>
> petsc-dev
> 5a758fb106fc
>
> You'll still see no compilerVersionCUDA in setCompilers.py
>
> [reverting these, and fixing self.setCompilers.compilerVersionCUDA might
> work..]
>
> asterix:/home/balay/tmp/petsc-dev/config/BuildSystem>hg diff
> diff -r 3d1e6fbcdc48 config/setCompilers.py
> --- a/config/setCompilers.py Fri Nov 05 09:03:09 2010 -0500
> +++ b/config/setCompilers.py Fri Nov 05 14:07:32 2010 -0500
> @@ -513,7 +513,6 @@
>
> def checkCUDACompiler(self):
> '''Locate a functional CUDA compiler'''
> - requiredVersion = '3.2'
> if 'with-cudac' in self.framework.argDB and
> self.framework.argDB['with-cudac'] == '0':
> if 'CUDAC' in self.framework.argDB:
> del self.framework.argDB['CUDAC']
> @@ -523,21 +522,7 @@
> if self.getExecutable(compiler, resultName = 'CUDAC'):
> self.checkCompiler('CUDA')
> # Put version info into the log
> - compilerVersion = self.executeShellCommand(self.CUDAC+'
> --version')
> - compilerVersion = compilerVersion[0]
> - compilerVersion = compilerVersion.split()
> - i = 0
> - for word in compilerVersion:
> - i = i+1
> - if word == 'release':
> - break
> - currentVersion = compilerVersion[i].strip(',')
> - if currentVersion != requiredVersion:
> - #raise RuntimeError('CUDA Error: PETSc currently requires nvcc
> version ' + requiredVersion)
> - print
> '==============================================================================='
> - print '** CUDA Error: PETSc requires nvcc version ' +
> requiredVersion + ' when compiling with --withcuda **'
> - print
> '==============================================================================='
> - sys.exit(3)
> + self.executeShellCommand(self.CUDAC+' --version')
> break
> except RuntimeError, e:
> self.logPrint('Error testing CUDA compiler: '+str(e))
>
>
> Satish
>
> On Fri, 5 Nov 2010, Matthew Knepley wrote:
>
> > On Fri, Nov 5, 2010 at 1:52 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> >
> > > because it was broken [as there was no
> > > self.setCompilers.compilerVersionCUDA] - and I noticed a different
> > > version check in setCompilers.py - and assumed thats the proper thing.
> > >
> > > But looks like vicotr made a mistake by adding in that check in
> > > setCompilers.py.
> > >
> > > Still I don't see where self.setCompilers.compilerVersionCUDA was set
> > > - and when that piece of cod removed..
> >
> >
> > Dammit dammit dammit. Who is fucking with my code. It was in
> > setCompilers.py.
> > Everyone stop fucking with this while I fix it.
> >
> > Matt
> >
> >
> > >
> > > Satish
> > >
> > > On Fri, 5 Nov 2010, Matthew Knepley wrote:
> > >
> > > > That was the correct one. I took the one out
> > > > of setCompilers.py for a reason.
> > > >
> > > > Matt
> > > >
> > > >
> > >
> > >
> >
> >
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101107/b295df1e/attachment.html>
More information about the petsc-dev
mailing list