On Jul 19, 2013, at 10:47 AM, Matthew Knepley <[email protected]> wrote:

On Fri, Jul 19, 2013 at 10:38 AM, Jason Sarich <[email protected]> wrote: Shiva,

This is from my bad merge. PETSc used to allow cuda without cusp/thrust, this is apparently no longer an option. I'll fix this.

We do. You configure with --with-cuda-only

In the long run this is a terrible way to indicate this and needs to be changed. It is inconsistent with how we do everything else by simply adding additional supports using --with-xxx. We should have --with-cuda for basic cuda support and --with-cusp for cusp support (which may automatically add the --with-cuda). Barry

Matt

Jason

On Thu, Jul 18, 2013 at 4:06 PM, Chekuri Choudary <[email protected]> wrote: Apparently, I was using a branch that deviated from petsc-dev before this got fixed. I am testing the master and will let you know the result. Thank you very much.

-----Original Message----- From: Satish Balay [mailto:[email protected]] Sent: Thursday, July 18, 2013 4:56 PM To: petsc-maint Cc: [email protected]; Chekuri Choudary Subject: Re: [petsc-maint] CUDA Version Compatibility

On Thu, 18 Jul 2013, Satish Balay wrote:

...

https://bitbucket.org/petsc/petsc/src/3123b27f26841f5e48c5aadd3e04f879 f1d21372/config/PETSc/packages/cuda.py?at=sarich/rnet-all#cl-52

3ae844c2 (Daniel Lowell 2011-10-26 09:17:02 -0500 29)# self.ThrustVersionStr = str(int(self.ThrustVersion)/100000) + '.' + str(int(self.ThrustVersion)/100%1000) + '.' + str(int(self.ThrustVersio 3ae844c2 (Daniel Lowell 2011-10-26 09:17:02 -0500 30)# self.CUSPVersionStr = str(int(self.CUSPVersion)/100000) + '.' + str(int(self.CUSPVersion)/100%1000) + '.' + str(int(self.CUSPVersion)%100)

ops - I meant to refer to the following lines that got commneted - causing configure to give this error.

3ae844c2 (Daniel Lowell 2011-10-26 09:17:02 -0500 51) #### self.cusp = framework.require('config.packages.cusp', self) 3ae844c2 (Daniel Lowell 2011-10-26 09:17:02 -0500 52) #### self.thrust = framework.require('config.packages.thrust', self)

satish

...

Jason might have to take a look on the rbase?

Satish

On Thu, 18 Jul 2013, Matthew Knepley wrote:

...

On Thu, Jul 18, 2013 at 3:24 PM, Chekuri Choudary <[email protected]>wrote:

...

It appears that there is an issue with checkCUDAVersion. I have downloaded the most recent versions of both CUSP and THRUST and tested with both CUDA 4.0 and CUDA 5.0. The configure.log files are attached.

Send $PETSC_DIR/config/PETSc/packages/cuda.py.

Matt

...

Thanks

-----Original Message----- From: Satish Balay [mailto:[email protected]] Sent: Thursday, July 18, 2013 2:35 PM To: Chekuri Choudary Cc: [email protected] Subject: Re: [petsc-maint] CUDA Version Compatibility

CUDA5.0 should work [and perhaps all cuda 4 versions aswell]

...

...

...

Configure Options: --configModules=PETSc.Configure --optionsModule=PETSc.compilerOptions --with-x=1 --CFLAGS="-O3 -g -ggdb -fopenmp" --FFLAGS="-O3 -g" --with-batch=0 --known-mpi-shared-libraries=no --with-cuda=1 --CUDAC=nvcc --with-cuda-dir=/usr/local/cuda-5.0/ PETSC_ARCH=PFLOTRAN1 <<<<

...

...

'Configure' object has no attribute 'thrust' File "./config/configure.py", line 293, in petsc_configure framework.configure(out = sys.stdout) <<

Perhaps there should be a better error check somewere. You need --with-thrust=1 [which is packaged in cuda] - and also --with-cusp=1 [perhaps with appropriate --with-cusp-dir option - or installed at '/usr/local/cuda-5.0/include/cusp']

Check config/examples/arch-cuda-double.py

Satish

On Thu, 18 Jul 2013, Chekuri Choudary wrote:

...

Hi,

Is it necessary to use CUDA 4.1 or 4.2 to use the GPU capabilities in PETSc. I am having problems compiling PETSc with CUDA. I have both CUDA4.0 and CUDA5.0 installed. Please find the configure.log files attached.

Thanks

-- 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