September 2017 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] (no subject)
by Zhang, Hong 21 Sep '17

21 Sep '17

Great news! According to their papers, MLSVM works only in serial. I am not sure what is stopping them using PETSc in parallel. Btw, are there any other cases that use PETSc for machine learning? Hong (Mr.) > On Sep 21, 2017, at 1:02 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > > From: Ilya Safro isafro(a)g.clemson.edu > Date: September 17, 2017 > Subject: MLSVM 1.0, Multilevel Support Vector Machines > > We are pleased to announce the release of MLSVM 1.0, a library of fast > multilevel algorithms for training nonlinear support vector machine > models on large-scale datasets. The library is developed as an > extension of PETSc to support, among other applications, the analysis > of datasets in scientific computing. > > Highlights: > - The best quality/performance trade-off is achieved with algebraic > multigrid coarsening > - Tested on academic, industrial, and healthcare datasets > - Generates multiple models for each training > - Effective on imbalanced datasets > > Download MLSVM at https://github.com/esadr/mlsvm > > Corresponding paper: Sadrfaridpour, Razzaghi and Safro "Engineering > multilevel support vector machines", 2017, > https://arxiv.org/pdf/1707.07657.pdf >

1 0

Re: [petsc-dev] HDF5 error from GAMG
by Barry Smith 21 Sep '17

21 Sep '17

Mark, It doesn't look like this a PETSc example (in a PETSc branch). If you can send me full instructions for building/compiling a code that has this behavior please send me all info so I can run it and debug it. Barry Otherwise it is just speculation city. > On Sep 18, 2017, at 7:44 PM, Mark Adams <mfadams(a)lbl.gov> wrote: > > I get this strange error when I use GAMG, in parallel. I don't see it in serial and I don't see it with the default solver. > > The problem seems to be that I use -vec_view in my code and the guts of Vec seem to be picking this up and causing havoc. In the past I have added a prefix to my -[x2_]vec_view and that seemed to work, but I would like to try to fix this and not shove it under the carpet again, if possible. > > It looks like a call to VecAssemblyEnd in the guts of GAMG is calling PetscObjectViewFromOptions. I don't understand why this only happens in parallel. > > Any ideas? > Thanks, > > 20:27 master= ~/Codes/picell/src$ make run NP=4 > /Users/markadams/Codes/petsc/arch-macosx-gnu-g/bin/mpiexec -n 4 ./picell.arch-macosx-gnu-g -petscspace_poly_tensor -petscspace_order 1 -mstep 2 -debug 3 -snes_monitor -ksp_monitor -snes_converged_reason -dm_refine 4 -dt .1 -use_bsp 10 -num_particles_proc -1 -num_phi_cells 8 -dm_view hdf5:sol.h5 -vec_view hdf5:sol.h5::append -dm_plex_periodic_cut > [0] **** Warning ****, no global point location. multiple processors (4) not supported > [0] npe=4; 2 x 2 x 1 flux tube grid; mpi_send size (chunksize) has 64 particles. ions only, BSP communication, phi section = 6.28319 > createMesh ntheta_total=4. > [0] 961 equations on 4 processors, 256 local cells, (element 0 used for flux tube list) > DM Object: Parallel Mesh 4 MPI processes > type: plex > Parallel Mesh in 2 dimensions: > 0-cells: 289 289 289 289 > 1-cells: 544 544 544 544 > 2-cells: 256 256 256 256 > Labels: > boundary: 1 strata with value/size (1 (65)) > Face Sets: 2 strata with value/size (1 (31), 4 (31)) > marker: 1 strata with value/size (1 (65)) > depth: 3 strata with value/size (0 (289), 1 (544), 2 (256)) > [0]shiftParticles: BSP have 1 for proc 0 > [0]shiftParticles: BSP have 1 particles for proc 0 (1 chunks) > [0]shiftParticles: BSP sent 1 > [2]shiftParticles: BSP sent 0 > [3]shiftParticles: BSP sent 0 > [1]shiftParticles: BSP sent 0 > 0) 1 local particles, 1/1 global, 0. % total particles moved in 0 messages total (to 1 processors local), 4. load imbalance factor > 0 SNES Function norm 6.176014522980e-01 > 0 KSP Residual norm 2.033453504856e+00 > 1 KSP Residual norm 1.247246018913e-01 > 2 KSP Residual norm 7.169223999741e-03 > 3 KSP Residual norm 3.494127469709e-04 > 4 KSP Residual norm 1.539576656127e-05 > 1 SNES Function norm 7.013048909984e-06 > 0 KSP Residual norm 1.539576656641e-05 > 1 KSP Residual norm 1.233278209501e-06 > 2 KSP Residual norm 6.474128793205e-08 > 3 KSP Residual norm 3.824829808334e-09 > 4 KSP Residual norm 1.813659764651e-10 > 5 KSP Residual norm 3.684110934970e-12 > 2 SNES Function norm 2.386983413511e-12 > Nonlinear solve converged due to CONVERGED_FNORM_RELATIVE iterations 2 > [0]shiftParticles: BSP have 0 for proc 0 > [0]shiftParticles: BSP have 1 for proc 2 > [0]shiftParticles: BSP have 1 particles for proc 2 (1 chunks) > [1]shiftParticles: BSP sent 0 > [2]shiftParticles: BSP sent 0 > [3]shiftParticles: BSP sent 0 > [0]shiftParticles: BSP sent 1 > 1) 1 local particles, 1/1 global, 100. % total particles moved in 0 messages total (to 2 processors local), 4. load imbalance factor > 0 SNES Function norm 6.828943707479e-01 > HDF5-DIAG: Error detected in HDF5 (1.8.18) HDF5-DIAG: Error detected in HDF5 (1.8.18) MPI-process 3: > #000: H5F.c line 604 in H5Fopen(): unable to open file > major: File accessibilty > minor: Unable to open file > #001: H5Fint.c line 1087 in H5F_open(): unable to read superblock > major: File accessibilty > minor: Read failed > #002: H5Fsuper.c line 294 in H5F_super_read(): unable to load superblock > major: Object cache > minor: Unable to protect metadata > #003: H5AC.c line 1262 in H5AC_protect(): H5C_protect() failed. > major: Object cache > minor: Unable to protect metadata > #004: H5C.c line 3574 in H5C_protect(): can't load entry > major: Object cache > minor: Unable to load metadata into cache > #005: H5C.c line 7954 in H5C_load_entry(): unable to load entry > major: Object cache > HDF5-DIAG: Error detected in HDF5 (1.8.18) MPI-process 1: > #000: H5F.c line 604 in H5Fopen(): unable to open file > major: File accessibilty > minor: Unable to open file > #001: H5Fint.c line 1087 in H5F_open(): unable to read superblock > major: File accessibilty > minor: Read failed > #002: H5Fsuper.c line 294 in H5F_super_read(): unable to load superblock > major: Object cache > minor: Unable to protect metadata > #003: H5AC.c line 1262 in H5AC_protect(): H5C_protect() failed. > major: Object cache > minor: Unable to protect metadata > #004: H5C.c line 3574 in H5C_protect(): can't load entry > major: Object cache > minor: Unable to load metadata into cache > #005: H5C.c line 7954 in H5C_load_entry(): unable to load entry > major: Object cache > minor: Unable to load metadata into cache > #006: H5Fsuper_cache.c line 476 in H5F_sblock_load(): truncated file: eof = 365072, sblock->base_addr = 0, stored_eoa = 367568 > major: File accessibilty > minor: File has been truncated > minor: Unable to load metadata into cache > #006: H5Fsuper_cache.c line 476 in H5F_sblock_load(): truncated file: eof = 365072, sblock->base_addr = 0, stored_eoa = 367568 > major: File accessibilty > minor: File has been truncated > > [snip] > > [0]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- > [0]PETSC ERROR: Error in external library > [0]PETSC ERROR: Error in HDF5 call H5Fopen() Status -1 > [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. > [0]PETSC ERROR: Petsc Development GIT revision: v3.7.6-5541-gd07ad3ccba GIT Date: 2017-09-17 13:49:32 -0400 > [0]PETSC ERROR: ./picell.arch-macosx-gnu-g on a arch-macosx-gnu-g named MarksMac-5.local by markadams Mon Sep 18 20:27:26 2017 > [0]PETSC ERROR: Configure options --with-cc=clang --with-cc++=clang++ COPTFLAGS="-g -mavx2" CXXOPTFLAGS="-g -mavx2" FOPTFLAGS="-g -mavx2" --download-mpich=1 --download-hypre=1 --download-metis=1 --download-parmetis=1 --download-c2html=1 --download-ctetgen --download-p4est=1 --download-superlu_dist --download-superlu --download-triangle=1 --download-hdf5=1 --download-zlib --with-x=0 --with-debugging=1 PETSC_ARCH=arch-macosx-gnu-g --download-chaco --with-viewfromoptions=1 > [0]PETSC ERROR: #1 PetscViewerFileSetName_HDF5() line 242 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/impls/hdf5/hdf5v.c > [0]PETSC ERROR: #2 PetscViewerFileSetName() line 650 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/impls/ascii/filev.c > [0]PETSC ERROR: #3 PetscOptionsGetViewer() line 339 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/interface/viewreg.c > [0]PETSC ERROR: #4 PetscObjectViewFromOptions() line 2773 in /Users/markadams/Codes/petsc/src/sys/objects/options.c > [0]PETSC ERROR: #5 VecAssemblyEnd() line 182 in /Users/markadams/Codes/petsc/src/vec/vec/interface/vector.c > [0]PETSC ERROR: #6 PCGAMGGetDataWithGhosts() line 376 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/util.c > [0]PETSC ERROR: #7 PCGAMGProlongator_AGG() line 1071 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/agg.c > [0]PETSC ERROR: #8 PCSetUp_GAMG() line 519 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/gamg.c > [0]PETSC ERROR: #9 PCSetUp() line 924 in /Users/markadams/Codes/petsc/src/ksp/pc/interface/precon.c > [0]PETSC ERROR: #10 KSPSetUp() line 378 in /Users/markadams/Codes/petsc/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #11 KSPSolve() line 609 in /Users/markadams/Codes/petsc/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #12 SNESSolve_NEWTONLS() line 224 in /Users/markadams/Codes/petsc/src/snes/impls/ls/ls.c > [0]PETSC ERROR: #13 SNESSolve() line 4106 in /Users/markadams/Codes/petsc/src/snes/interface/snes.c > [0]PETSC ERROR: #14 DMPICellSolve() line 18 in /Users/markadams/Codes/petsc/src/snes/utils/dmpicellsnes.c > [0]PETSC ERROR: #15 go() line 1120 in /Users/markadams/Codes/picell/src/main.c > [0]PETSC ERROR: #16 main() line 1304 in /Users/markadams/Codes/picell/src/main.c > [0]PETSC ERROR: PETSc Option Table entries: > [0]PETSC ERROR: -debug 3 > [0]PETSC ERROR: -dm_plex_periodic_cut > [0]PETSC ERROR: -dm_refine 4 > [0]PETSC ERROR: -dm_view hdf5:sol.h5 > [0]PETSC ERROR: -dt .1 > [0]PETSC ERROR: -ksp_monitor > [0]PETSC ERROR: -ksp_type cg > [0]PETSC ERROR: -mstep 2 > [0]PETSC ERROR: -num_particles_proc -1 > [0]PETSC ERROR: -num_phi_cells 8 > [0]PETSC ERROR: -options_left > [0]PETSC ERROR: -pc_type gamg > [0]PETSC ERROR: -petscspace_order 1 > [0]PETSC ERROR: -petscspace_poly_tensor > [0]PETSC ERROR: -snes_converged_reason > [0]PETSC ERROR: -snes_monitor > [0]PETSC ERROR: -use_bsp 10 > [0]PETSC ERROR: -vec_view hdf5:sol.h5::append > [0]PETSC ERROR: ----------------End of Error Message -------send entire error message to petsc-maint(a)mcs.anl.gov---------- > application called MPI_Abort(MPI_COMM_WORLD, 76) - process 0 >

1 0

Re: [petsc-dev] HDF5 error from GAMG
by Barry Smith 21 Sep '17

21 Sep '17

> On Sep 18, 2017, at 7:59 PM, Mark Adams <mfadams(a)lbl.gov> wrote: > > Also, I tested this on my Mac (this output) and on Cori at NERSC and the behavior looked identical. > > On Mon, Sep 18, 2017 at 8:44 PM, Mark Adams <mfadams(a)lbl.gov> wrote: > I get this strange error when I use GAMG, in parallel. I don't see it in serial and I don't see it with the default solver. > > The problem seems to be that I use -vec_view in my code and the guts of Vec seem to be picking this up and causing havoc. In the past I have added a prefix to my -[x2_]vec_view and that seemed to work, but I would like to try to fix this and not shove it under the carpet again, if possible. > > It looks like a call to VecAssemblyEnd in the guts of GAMG is calling PetscObjectViewFromOptions. I don't understand why this only happens in parallel. > > Any ideas? > Thanks, > > 20:27 master= ~/Codes/picell/src$ make run NP=4 > /Users/markadams/Codes/petsc/arch-macosx-gnu-g/bin/mpiexec -n 4 ./picell.arch-macosx-gnu-g -petscspace_poly_tensor -petscspace_order 1 -mstep 2 -debug 3 -snes_monitor -ksp_monitor -snes_converged_reason -dm_refine 4 -dt .1 -use_bsp 10 -num_particles_proc -1 -num_phi_cells 8 -dm_view hdf5:sol.h5 -vec_view hdf5:sol.h5::append -dm_plex_periodic_cut > [0] **** Warning ****, no global point location. multiple processors (4) not supported > [0] npe=4; 2 x 2 x 1 flux tube grid; mpi_send size (chunksize) has 64 particles. ions only, BSP communication, phi section = 6.28319 > createMesh ntheta_total=4. > [0] 961 equations on 4 processors, 256 local cells, (element 0 used for flux tube list) > DM Object: Parallel Mesh 4 MPI processes > type: plex > Parallel Mesh in 2 dimensions: > 0-cells: 289 289 289 289 > 1-cells: 544 544 544 544 > 2-cells: 256 256 256 256 > Labels: > boundary: 1 strata with value/size (1 (65)) > Face Sets: 2 strata with value/size (1 (31), 4 (31)) > marker: 1 strata with value/size (1 (65)) > depth: 3 strata with value/size (0 (289), 1 (544), 2 (256)) > [0]shiftParticles: BSP have 1 for proc 0 > [0]shiftParticles: BSP have 1 particles for proc 0 (1 chunks) > [0]shiftParticles: BSP sent 1 > [2]shiftParticles: BSP sent 0 > [3]shiftParticles: BSP sent 0 > [1]shiftParticles: BSP sent 0 > 0) 1 local particles, 1/1 global, 0. % total particles moved in 0 messages total (to 1 processors local), 4. load imbalance factor > 0 SNES Function norm 6.176014522980e-01 > 0 KSP Residual norm 2.033453504856e+00 > 1 KSP Residual norm 1.247246018913e-01 > 2 KSP Residual norm 7.169223999741e-03 > 3 KSP Residual norm 3.494127469709e-04 > 4 KSP Residual norm 1.539576656127e-05 > 1 SNES Function norm 7.013048909984e-06 > 0 KSP Residual norm 1.539576656641e-05 > 1 KSP Residual norm 1.233278209501e-06 > 2 KSP Residual norm 6.474128793205e-08 > 3 KSP Residual norm 3.824829808334e-09 > 4 KSP Residual norm 1.813659764651e-10 > 5 KSP Residual norm 3.684110934970e-12 > 2 SNES Function norm 2.386983413511e-12 > Nonlinear solve converged due to CONVERGED_FNORM_RELATIVE iterations 2 > [0]shiftParticles: BSP have 0 for proc 0 > [0]shiftParticles: BSP have 1 for proc 2 > [0]shiftParticles: BSP have 1 particles for proc 2 (1 chunks) > [1]shiftParticles: BSP sent 0 > [2]shiftParticles: BSP sent 0 > [3]shiftParticles: BSP sent 0 > [0]shiftParticles: BSP sent 1 > 1) 1 local particles, 1/1 global, 100. % total particles moved in 0 messages total (to 2 processors local), 4. load imbalance factor > 0 SNES Function norm 6.828943707479e-01 > HDF5-DIAG: Error detected in HDF5 (1.8.18) HDF5-DIAG: Error detected in HDF5 (1.8.18) MPI-process 3: > #000: H5F.c line 604 in H5Fopen(): unable to open file > major: File accessibilty > minor: Unable to open file > #001: H5Fint.c line 1087 in H5F_open(): unable to read superblock > major: File accessibilty > minor: Read failed > #002: H5Fsuper.c line 294 in H5F_super_read(): unable to load superblock > major: Object cache > minor: Unable to protect metadata > #003: H5AC.c line 1262 in H5AC_protect(): H5C_protect() failed. > major: Object cache > minor: Unable to protect metadata > #004: H5C.c line 3574 in H5C_protect(): can't load entry > major: Object cache > minor: Unable to load metadata into cache > #005: H5C.c line 7954 in H5C_load_entry(): unable to load entry > major: Object cache > HDF5-DIAG: Error detected in HDF5 (1.8.18) MPI-process 1: > #000: H5F.c line 604 in H5Fopen(): unable to open file > major: File accessibilty > minor: Unable to open file > #001: H5Fint.c line 1087 in H5F_open(): unable to read superblock > major: File accessibilty > minor: Read failed > #002: H5Fsuper.c line 294 in H5F_super_read(): unable to load superblock > major: Object cache > minor: Unable to protect metadata > #003: H5AC.c line 1262 in H5AC_protect(): H5C_protect() failed. > major: Object cache > minor: Unable to protect metadata > #004: H5C.c line 3574 in H5C_protect(): can't load entry > major: Object cache > minor: Unable to load metadata into cache > #005: H5C.c line 7954 in H5C_load_entry(): unable to load entry > major: Object cache > minor: Unable to load metadata into cache > #006: H5Fsuper_cache.c line 476 in H5F_sblock_load(): truncated file: eof = 365072, sblock->base_addr = 0, stored_eoa = 367568 > major: File accessibilty > minor: File has been truncated > minor: Unable to load metadata into cache > #006: H5Fsuper_cache.c line 476 in H5F_sblock_load(): truncated file: eof = 365072, sblock->base_addr = 0, stored_eoa = 367568 > major: File accessibilty > minor: File has been truncated > > [snip] > > [0]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- > [0]PETSC ERROR: Error in external library > [0]PETSC ERROR: Error in HDF5 call H5Fopen() Status -1 > [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. > [0]PETSC ERROR: Petsc Development GIT revision: v3.7.6-5541-gd07ad3ccba GIT Date: 2017-09-17 13:49:32 -0400 > [0]PETSC ERROR: ./picell.arch-macosx-gnu-g on a arch-macosx-gnu-g named MarksMac-5.local by markadams Mon Sep 18 20:27:26 2017 > [0]PETSC ERROR: Configure options --with-cc=clang --with-cc++=clang++ COPTFLAGS="-g -mavx2" CXXOPTFLAGS="-g -mavx2" FOPTFLAGS="-g -mavx2" --download-mpich=1 --download-hypre=1 --download-metis=1 --download-parmetis=1 --download-c2html=1 --download-ctetgen --download-p4est=1 --download-superlu_dist --download-superlu --download-triangle=1 --download-hdf5=1 --download-zlib --with-x=0 --with-debugging=1 PETSC_ARCH=arch-macosx-gnu-g --download-chaco --with-viewfromoptions=1 > [0]PETSC ERROR: #1 PetscViewerFileSetName_HDF5() line 242 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/impls/hdf5/hdf5v.c > [0]PETSC ERROR: #2 PetscViewerFileSetName() line 650 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/impls/ascii/filev.c > [0]PETSC ERROR: #3 PetscOptionsGetViewer() line 339 in /Users/markadams/Codes/petsc/src/sys/classes/viewer/interface/viewreg.c > [0]PETSC ERROR: #4 PetscObjectViewFromOptions() line 2773 in /Users/markadams/Codes/petsc/src/sys/objects/options.c > [0]PETSC ERROR: #5 VecAssemblyEnd() line 182 in /Users/markadams/Codes/petsc/src/vec/vec/interface/vector.c > [0]PETSC ERROR: #6 PCGAMGGetDataWithGhosts() line 376 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/util.c > [0]PETSC ERROR: #7 PCGAMGProlongator_AGG() line 1071 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/agg.c > [0]PETSC ERROR: #8 PCSetUp_GAMG() line 519 in /Users/markadams/Codes/petsc/src/ksp/pc/impls/gamg/gamg.c > [0]PETSC ERROR: #9 PCSetUp() line 924 in /Users/markadams/Codes/petsc/src/ksp/pc/interface/precon.c > [0]PETSC ERROR: #10 KSPSetUp() line 378 in /Users/markadams/Codes/petsc/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #11 KSPSolve() line 609 in /Users/markadams/Codes/petsc/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #12 SNESSolve_NEWTONLS() line 224 in /Users/markadams/Codes/petsc/src/snes/impls/ls/ls.c > [0]PETSC ERROR: #13 SNESSolve() line 4106 in /Users/markadams/Codes/petsc/src/snes/interface/snes.c > [0]PETSC ERROR: #14 DMPICellSolve() line 18 in /Users/markadams/Codes/petsc/src/snes/utils/dmpicellsnes.c > [0]PETSC ERROR: #15 go() line 1120 in /Users/markadams/Codes/picell/src/main.c > [0]PETSC ERROR: #16 main() line 1304 in /Users/markadams/Codes/picell/src/main.c > [0]PETSC ERROR: PETSc Option Table entries: > [0]PETSC ERROR: -debug 3 > [0]PETSC ERROR: -dm_plex_periodic_cut > [0]PETSC ERROR: -dm_refine 4 > [0]PETSC ERROR: -dm_view hdf5:sol.h5 > [0]PETSC ERROR: -dt .1 > [0]PETSC ERROR: -ksp_monitor > [0]PETSC ERROR: -ksp_type cg > [0]PETSC ERROR: -mstep 2 > [0]PETSC ERROR: -num_particles_proc -1 > [0]PETSC ERROR: -num_phi_cells 8 > [0]PETSC ERROR: -options_left > [0]PETSC ERROR: -pc_type gamg > [0]PETSC ERROR: -petscspace_order 1 > [0]PETSC ERROR: -petscspace_poly_tensor > [0]PETSC ERROR: -snes_converged_reason > [0]PETSC ERROR: -snes_monitor > [0]PETSC ERROR: -use_bsp 10 > [0]PETSC ERROR: -vec_view hdf5:sol.h5::append > [0]PETSC ERROR: ----------------End of Error Message -------send entire error message to petsc-maint(a)mcs.anl.gov---------- > application called MPI_Abort(MPI_COMM_WORLD, 76) - process 0 > >

1 0

(no subject)
by Barry Smith 21 Sep '17

21 Sep '17

From: Ilya Safro isafro(a)g.clemson.edu Date: September 17, 2017 Subject: MLSVM 1.0, Multilevel Support Vector Machines We are pleased to announce the release of MLSVM 1.0, a library of fast multilevel algorithms for training nonlinear support vector machine models on large-scale datasets. The library is developed as an extension of PETSc to support, among other applications, the analysis of datasets in scientific computing. Highlights: - The best quality/performance trade-off is achieved with algebraic multigrid coarsening - Tested on academic, industrial, and healthcare datasets - Generates multiple models for each training - Effective on imbalanced datasets Download MLSVM at https://github.com/esadr/mlsvm Corresponding paper: Sadrfaridpour, Razzaghi and Safro "Engineering multilevel support vector machines", 2017, https://arxiv.org/pdf/1707.07657.pdf

1 0

Re: [petsc-dev] MatTransposeMatMult
by Jed Brown 21 Sep '17

21 Sep '17

Hong <hzhang(a)mcs.anl.gov> writes: > Jed, > I reproduced what you observed. > 30x time is spent on numerical C = At*B. > We compute A^T*B using outer product C=(A^T)[:,i]*B[i,:]. > For your large matrix A: rows=574902, cols=184446, > outer product takes long long time to insert values to C. > > We may use At=A^T and C=At*B instead. Right. My point is that users call MatTransposeMatMult and conclude that PETSc is super slow compared to other software. The could call MatTranspose then MatMatMult and get fast results, but that's a sucky and confusing interface. Is there a compelling reason to keep the outer product implementation around? Are there any matrices for which it is competitive?

1 0

Re: [petsc-dev] MatTransposeMatMult
by Hong 21 Sep '17

21 Sep '17

Jed, I reproduced what you observed. 30x time is spent on numerical C = At*B. We compute A^T*B using outer product C=(A^T)[:,i]*B[i,:]. For your large matrix A: rows=574902, cols=184446, outer product takes long long time to insert values to C. We may use At=A^T and C=At*B instead. Hong On Thu, Sep 21, 2017 at 9:59 AM, Hong <hzhang(a)mcs.anl.gov> wrote: > Jed : > >> ~jedbrown/ceres_solver_iteration_001_A.petsc on mcs.anl.gov. >> > I got this file. How many matrices in this file? How to load them? > It seems the first matrix is rectangular. > > Hong > >> >> The issue is that MatTransposeMatMult is implemented using sparse outer >> products which is quite inefficient compared to the standard MatMatMult >> algorithm. >> >> Barry Smith <bsmith(a)mcs.anl.gov> writes: >> >> > Send the runable case so that we can determine the cause of the slow >> down and find possible cures. It could be that the "optimized" version was >> optimized for a particular size of problems or matrix structure. >> > >> > Barry >> > >> >> On Sep 20, 2017, at 4:32 PM, Jed Brown <jed(a)jedbrown.org> wrote: >> >> >> >> I have a simple SeqAIJ example where MatTransposeMatMult is 30x slower >> >> than MatTranspose followed by MatMatMult. This is embarrassing when >> >> people try to compare performance. Are there any cases where these >> >> custom implementations provide value rather than just confusing people >> >> or turning them away from PETSc? >> > >

1 0

Re: [petsc-dev] MatTransposeMatMult
by Hong 21 Sep '17

21 Sep '17

Jed : > ~jedbrown/ceres_solver_iteration_001_A.petsc on mcs.anl.gov. > I got this file. How many matrices in this file? How to load them? It seems the first matrix is rectangular. Hong > > The issue is that MatTransposeMatMult is implemented using sparse outer > products which is quite inefficient compared to the standard MatMatMult > algorithm. > > Barry Smith <bsmith(a)mcs.anl.gov> writes: > > > Send the runable case so that we can determine the cause of the slow > down and find possible cures. It could be that the "optimized" version was > optimized for a particular size of problems or matrix structure. > > > > Barry > > > >> On Sep 20, 2017, at 4:32 PM, Jed Brown <jed(a)jedbrown.org> wrote: > >> > >> I have a simple SeqAIJ example where MatTransposeMatMult is 30x slower > >> than MatTranspose followed by MatMatMult. This is embarrassing when > >> people try to compare performance. Are there any cases where these > >> custom implementations provide value rather than just confusing people > >> or turning them away from PETSc? >

1 0

Re: [petsc-dev] PetscSF in Fortran
by Jed Brown 20 Sep '17

20 Sep '17

Unfortunately PetscSFSetGraph/PetscSFGetGraph need custom bindings for Fortran and they have not been written. Shouldn't be difficult, but will take a little work/testing. I would probably make the array of PetscSFNode be a PetscInt array of length 2n (or of shape (2,n)) -- that's an equivalent memory representation to what we're using now. Adrian Croucher <a.croucher(a)auckland.ac.nz> writes: > hi > > For the dual-porosity stuff I'm working on, I'm creating a modified > DMPlex. Looking at e.g. TS ex11.c (in the SplitFaces() routine), it > appears I am going to have to set up a PetscSF for the new DM (from what > I can gather, that is needed for managing parallel communication of > partition ghost cell values). > > The way I've set up the modified DMPlex, the partition ghost cells are > in the same locations as they were in the original single-porosity DM. > So I'm thinking I basically just have to copy the SF straight over from > the original DM. > > The routine DMGetPointSF() seems to work OK in Fortran, but I'm having a > bit of trouble with PetscSFGetGraph(). > > When I try to declare an array of type PetscSFNode for the iremote > parameter, it doesn't seem to know about PetscSFNode, even though I have > a 'use petsc' in my code, which normally gives access to everything. > > Also I'm not sure of exactly how to set up the ilocal and iremote array > parameters for this function in Fortran. How should they be declared- as > pointer arrays? > > - Adrian > > -- > Dr Adrian Croucher > Senior Research Fellow > Department of Engineering Science > University of Auckland, New Zealand > email: a.croucher(a)auckland.ac.nz > tel: +64 (0)9 923 4611

1 0

Re: [petsc-dev] MatTransposeMatMult
by Jed Brown 20 Sep '17

20 Sep '17

~jedbrown/ceres_solver_iteration_001_A.petsc on mcs.anl.gov. The issue is that MatTransposeMatMult is implemented using sparse outer products which is quite inefficient compared to the standard MatMatMult algorithm. Barry Smith <bsmith(a)mcs.anl.gov> writes: > Send the runable case so that we can determine the cause of the slow down and find possible cures. It could be that the "optimized" version was optimized for a particular size of problems or matrix structure. > > Barry > >> On Sep 20, 2017, at 4:32 PM, Jed Brown <jed(a)jedbrown.org> wrote: >> >> I have a simple SeqAIJ example where MatTransposeMatMult is 30x slower >> than MatTranspose followed by MatMatMult. This is embarrassing when >> people try to compare performance. Are there any cases where these >> custom implementations provide value rather than just confusing people >> or turning them away from PETSc?

1 0

PetscSF in Fortran
by Adrian Croucher 20 Sep '17

20 Sep '17

hi For the dual-porosity stuff I'm working on, I'm creating a modified DMPlex. Looking at e.g. TS ex11.c (in the SplitFaces() routine), it appears I am going to have to set up a PetscSF for the new DM (from what I can gather, that is needed for managing parallel communication of partition ghost cell values). The way I've set up the modified DMPlex, the partition ghost cells are in the same locations as they were in the original single-porosity DM. So I'm thinking I basically just have to copy the SF straight over from the original DM. The routine DMGetPointSF() seems to work OK in Fortran, but I'm having a bit of trouble with PetscSFGetGraph(). When I try to declare an array of type PetscSFNode for the iremote parameter, it doesn't seem to know about PetscSFNode, even though I have a 'use petsc' in my code, which normally gives access to everything. Also I'm not sure of exactly how to set up the ilocal and iremote array parameters for this function in Fortran. How should they be declared- as pointer arrays? - Adrian -- Dr Adrian Croucher Senior Research Fellow Department of Engineering Science University of Auckland, New Zealand email: a.croucher(a)auckland.ac.nz tel: +64 (0)9 923 4611

1 0