July 2012 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] how to use GPU, is my speed up sufficient?
by Barry Smith 12 Jul '12

12 Jul '12

On Jul 12, 2012, at 10:42 AM, Olga Tramontano wrote: > Hi all > I am very new to PETSc. I just learnt its class structure and I just understood that the interface Vec has been reimplemented with the class VecCUSP that is able to support GPU. My question is: if I have a simple code that allocates a vector and sets it with the database options, if I specify -vec_type seqcusp then this vector is allocated AND computed on the GPU? > Because I compared the execution time of the same algorithm with the database options -vec_type seq and -vec_type seqcusp: these two times are very close. > > This is the code: I'm just trying to compute how long does it take to execute the function VecScale: > #include <petscvec.h> > #include <string.h> > > #undef __FUNCT__ > #define __FUNCT__ "main" > int main(int argc,char **argv) > { > Vec x; > PetscInt n; > PetscErrorCode ierr; > PetscBool exists; > PetscLogDouble t1,t2; > PetscInt a=10; > > ierr = PetscInitialize(&argc,&argv,(char*)0,help);CHKERRQ(ierr); > > ierr = PetscOptionsGetInt(PETSC_NULL,"-size",&n,&exists); CHKERRQ(ierr); > The following code creates the vector > ierr = VecCreate(PETSC_COMM_SELF,&x);CHKERRQ(ierr); > ierr = VecSetSizes(x,PETSC_DECIDE,n);CHKERRQ(ierr); > ierr = VecSetFromOptions(x);CHKERRQ(ierr); > The next piece of code fills the vector with numbers ON THE CPU! > ierr = VecSetRandom(x, PETSC_NULL); CHKERRQ(ierr); > > PetscGetTime(&t1); When you run this with a GPU vector it first copies the vector down to the GPU and then does the scaling. This is likely to be slow because copying to the GPU takes lots of time. > ierr = VecScale(x,a); CHKERRQ(ierr); > PetscGetTime(&t2); > ierr = PetscPrintf(PETSC_COMM_WORLD, "%2.5f\n",(t2-t1)); The above code is not useful for computing GPU performance since you are measuring something completely artificial: the time to copy down and then the scale on the GPU time. Try adding an extra VecScale() call before the timing, this will cause the copy down to the GPU and hence the second call to VecScale will not copy down and will be much faster. Please see our paper @article{minden2010preliminary, title={Preliminary implementation of PETSc using GPUs}, author={Minden, V. and Smith, B.F. and Knepley, M.G.}, journal={Proceedings of the 2010 International Workshop of GPU Solutions to Multiscale Problems in Science and Engineering}, year={2010} } it discusses the importance of having many operations take place on the GPU without copying the data back and forth between the CPU and the GPU all the time (which eliminates the advantage of the GPU). Barry > > ierr = VecDestroy(&x);CHKERRQ(ierr); > > ierr = PetscFinalize(); > return 0; > } > > I use a vector of about 500000 numbers. With the database option -vec_type seqcusp , the method VecScale should call the method VecScale_SeqCUSP, right? Am I wrong? Should I do it expressly? > Anyway with the option -vec_type seq , the method VecScale takes 0.05681 seconds, > with the option -vec_type seqcusp, the method VecScale takes 0.04242 seconds. > > Is this speed up sufficient and realistic? > thanks > > Olga

1 0

FastForward UHPC connections
by Barry Smith 12 Jul '12

12 Jul '12

Begin forwarded message: > From: Pavan Balaji <balaji(a)mcs.anl.gov> > Date: July 12, 2012 8:50:57 PM CDT > To: xlab(a)cels.anl.gov > Subject: [Xlab] FastForward UHPC connections > > > I was surprised to read how connected the FastForward awards are to previous UHPC awardees. > > http://www.hpcwire.com/hpcwire/2012-07-12/doe_primes_pump_for_exascale_supe… > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > Xlab mailing list > Xlab(a)lists.cels.anl.gov > https://lists.cels.anl.gov/mailman/listinfo/xlab

1 0

Re: [petsc-dev] PetscSegv_sigaction / PetscCheckPointer don't build using visual studio
by Satish Balay 12 Jul '12

12 Jul '12

On Wed, 11 Jul 2012, Chetan Jhurani wrote: > Hi, > > The recent changes to PetscCheckPointer in petsc-dev > don't build on windows when using Visual Studio 2005 and 2010. > > They have a very bare-bones signal implementation. > http://msdn.microsoft.com/en-us/library/xdkz3x12.aspx > > The first error encountered is due to this new code in > petsc-private/petscimpl.h > > #if defined(PETSC_HAVE_SETJMP_H) > #include <signal.h> > #include <setjmp.h> > PETSC_EXTERN jmp_buf PetscSegvJumpBuf; > PETSC_EXTERN void PetscSegv_sigaction(int, siginfo_t*, void *); > > PETSC_HAVE_SETJMP_H is 1 and both signal.h and setjmp.h > are found but they don't define siginfo_t. I'm sure there > the compiler doesn't implement a lot of other stuff in > PetscCheckPointer. fixed now. http://petsc.cs.iit.edu/petsc/petsc-dev/rev/68976e0f458f http://petsc.cs.iit.edu/petsc/BuildSystem/rev/71be541c881b Satish

1 0

Re: [petsc-dev] how to use GPU, is my speed up sufficient?
by Matthew Knepley 12 Jul '12

12 Jul '12

On Thu, Jul 12, 2012 at 10:42 AM, Olga Tramontano <tramontanoolga(a)yahoo.it>wrote: > Hi all > I am very new to PETSc. I just learnt its class structure and I just > understood that the interface Vec has been reimplemented with the class > VecCUSP that is able to support GPU. My question is: if I have a simple > code that allocates a vector and sets it with the database options, if I > specify -vec_type seqcusp then this vector is allocated AND computed on the > GPU? > Because I compared the execution time of the same algorithm with the > database options -vec_type seq and -vec_type seqcusp: these two times are > very close. > For any performance question, you have to send the output of -log_summary for all runs, or we have no idea what you actually did. Matt > This is the code: I'm just trying to compute how long does it take to > execute the function VecScale: > #include <petscvec.h> > #include <string.h> > > #undef __FUNCT__ > #define __FUNCT__ "main" > int main(int argc,char **argv) > { > Vec x; > PetscInt n; > PetscErrorCode ierr; > PetscBool exists; > PetscLogDouble t1,t2; > PetscInt a=10; > > ierr = PetscInitialize(&argc,&argv,(char*)0,help);CHKERRQ(ierr); > > ierr = PetscOptionsGetInt(PETSC_NULL,"-size",&n,&exists); CHKERRQ(ierr); > > ierr = VecCreate(PETSC_COMM_SELF,&x);CHKERRQ(ierr); > ierr = VecSetSizes(x,PETSC_DECIDE,n);CHKERRQ(ierr); > ierr = VecSetFromOptions(x);CHKERRQ(ierr); > > ierr = VecSetRandom(x, PETSC_NULL); CHKERRQ(ierr); > > PetscGetTime(&t1); > ierr = VecScale(x,a); CHKERRQ(ierr); > PetscGetTime(&t2); > ierr = PetscPrintf(PETSC_COMM_WORLD, "%2.5f\n",(t2-t1)); > > ierr = VecDestroy(&x);CHKERRQ(ierr); > > ierr = PetscFinalize(); > return 0; > } > > I use a vector of about 500000 numbers. With the database option > -vec_type seqcusp , the method VecScale should call the > method VecScale_SeqCUSP, right? Am I wrong? Should I do it expressly? > Anyway with the option -vec_type seq , the method VecScale takes 0.05681 > seconds, > with the option -vec_type seqcusp, the method VecScale takes 0.04242 > seconds. > > Is this speed up sufficient and realistic? > thanks > > Olga > -- 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

1 0

how to use GPU, is my speed up sufficient?
by Olga Tramontano 12 Jul '12

12 Jul '12

Hi all I am very new to PETSc. I just learnt its class structure and I just understood that the interface Vec has been reimplemented with the class VecCUSP that is able to support GPU. My question is: if I have a simple code that allocates a vector and sets it with the database options, if I specify -vec_type seqcusp then this vector is allocated AND computed on the GPU? Because I compared the execution time of the same algorithm with the database options -vec_type seq and -vec_type seqcusp: these two times are very close. This is the code: I'm just trying to compute how long does it take to execute the function VecScale: #include <petscvec.h> #include <string.h> #undef __FUNCT__ #define __FUNCT__ "main" int main(int argc,char **argv) { Vec x; PetscInt n; PetscErrorCode ierr; PetscBool exists; PetscLogDouble t1,t2; PetscInt a=10; ierr = PetscInitialize(&argc,&argv,(char*)0,help);CHKERRQ(ierr); ierr = PetscOptionsGetInt(PETSC_NULL,"-size",&n,&exists); CHKERRQ(ierr); ierr = VecCreate(PETSC_COMM_SELF,&x);CHKERRQ(ierr); ierr = VecSetSizes(x,PETSC_DECIDE,n);CHKERRQ(ierr); ierr = VecSetFromOptions(x);CHKERRQ(ierr); ierr = VecSetRandom(x, PETSC_NULL); CHKERRQ(ierr); PetscGetTime(&t1); ierr = VecScale(x,a); CHKERRQ(ierr); PetscGetTime(&t2); ierr = PetscPrintf(PETSC_COMM_WORLD, "%2.5f\n",(t2-t1)); ierr = VecDestroy(&x);CHKERRQ(ierr); ierr = PetscFinalize(); return 0; } I use a vector of about 500000 numbers. With the database option -vec_type seqcusp , the method VecScale should call the method VecScale_SeqCUSP, right? Am I wrong? Should I do it expressly? Anyway with the option -vec_type seq , the method VecScale takes 0.05681 seconds, with the option -vec_type seqcusp, the method VecScale takes 0.04242 seconds. Is this speed up sufficient and realistic? thanks Olga

1 0

Re: [petsc-dev] Lastet commit slepc-dev (Rev 2895): "Fix bug in PSGHIEP_QR" causes compilation error
by Jose E. Roman 12 Jul '12

12 Jul '12

El 12/07/2012, a las 10:15, Thomas Hisch escribió: > I'm not sure if this is the appropriate list for slepc bugs. If not, > please fwd this message to the appropriate list. > > I tried to compile slepc-dev (Rev 2895) with both petsc-3.3-p1 and > petsc-dev but I always get this compilation error: > > [ 0%] Building CXX object CMakeFiles/slepc.dir/src/ps/psghiep.c.o > /home/thomas/local/src/slepc-dev/src/ps/psghiep.c: In function > ‘PetscErrorCode PSEigenVectorsPseudoOrthog(PS, PSMatType, > PetscScalar*, PetscScalar*, PetscBool)’: > /home/thomas/local/src/slepc-dev/src/ps/psghiep.c:780:40: error: > cannot convert ‘PetscScalar {aka std::complex<double>}’ to ‘PetscReal > {aka double}’ in assignment > > > Note that I compiled PETSc with complex scalars. Reverting the last > commit fixes the compilation error. > > Regards > Thomas Thanks. We have a nightly build system where we catch these problems in one or two days at most. I would suggest not to update too often. Jose

1 0

Lastet commit slepc-dev (Rev 2895): "Fix bug in PSGHIEP_QR" causes compilation error
by Thomas Hisch 12 Jul '12

12 Jul '12

I'm not sure if this is the appropriate list for slepc bugs. If not, please fwd this message to the appropriate list. I tried to compile slepc-dev (Rev 2895) with both petsc-3.3-p1 and petsc-dev but I always get this compilation error: [ 0%] Building CXX object CMakeFiles/slepc.dir/src/ps/psghiep.c.o /home/thomas/local/src/slepc-dev/src/ps/psghiep.c: In function ‘PetscErrorCode PSEigenVectorsPseudoOrthog(PS, PSMatType, PetscScalar*, PetscScalar*, PetscBool)’: /home/thomas/local/src/slepc-dev/src/ps/psghiep.c:780:40: error: cannot convert ‘PetscScalar {aka std::complex<double>}’ to ‘PetscReal {aka double}’ in assignment Note that I compiled PETSc with complex scalars. Reverting the last commit fixes the compilation error. Regards Thomas

1 0

Re: [petsc-dev] c2html and sowing fail on windows.
by Satish Balay 11 Jul '12

11 Jul '12

On Wed, 11 Jul 2012, Satish Balay wrote: > Hg repo installs have optional extra requirements sowing,c2html > documented at http://www.mcs.anl.gov/petsc/developers/index.html > > I can add in a windows/cygwin note there.. The above url is updated now. Satish

1 0

Re: [petsc-dev] c2html and sowing fail on windows.
by Satish Balay 11 Jul '12

11 Jul '12

On Wed, 11 Jul 2012, Chetan Jhurani wrote: > > -----Original Message----- > > From: petsc-dev-bounces(a)mcs.anl.gov [mailto:[email protected]] On Behalf Of Satish Balay > > Sent: Wednesday, July 11, 2012 12:04 PM > > To: For users of the development version of PETSc > > Subject: Re: [petsc-dev] c2html and sowing fail on windows. > > > > On Wed, 11 Jul 2012, Chetan Jhurani wrote: > > > > > > > > Of course, using "--with-sowing=0 --with-c2html=0" > > > avoids the issue for now, but both the packages don't > > > work on windows out of the box. > > > > These packages need cygwin gcc/g++ installed [and flex for c2html] > > Thanks Satish. I had kept the least number of packages described here > http://www.mcs.anl.gov/petsc/documentation/installation.html#windows > so as to have a fixed portable but standard cygwin, which could be > very small when zipped up. Perhaps the requirement of flex/gcc/g++ > (for these two packages) can be added there. those instructions are for tarball installs. Hg repo installs have optional extra requirements sowing,c2html documented at http://www.mcs.anl.gov/petsc/developers/index.html I can add in a windows/cygwin note there.. Satish

1 0

Re: [petsc-dev] c2html and sowing fail on windows.
by Chetan Jhurani 11 Jul '12

11 Jul '12

> -----Original Message----- > From: petsc-dev-bounces(a)mcs.anl.gov [mailto:[email protected]] On Behalf Of Satish Balay > Sent: Wednesday, July 11, 2012 12:04 PM > To: For users of the development version of PETSc > Subject: Re: [petsc-dev] c2html and sowing fail on windows. > > On Wed, 11 Jul 2012, Chetan Jhurani wrote: > > > > > Of course, using "--with-sowing=0 --with-c2html=0" > > avoids the issue for now, but both the packages don't > > work on windows out of the box. > > These packages need cygwin gcc/g++ installed [and flex for c2html] Thanks Satish. I had kept the least number of packages described here http://www.mcs.anl.gov/petsc/documentation/installation.html#windows so as to have a fixed portable but standard cygwin, which could be very small when zipped up. Perhaps the requirement of flex/gcc/g++ (for these two packages) can be added there. Chetan

1 0