[petsc-users] SuperLU + GPUs
Mark Adams
mfadams at lbl.gov
Sat Apr 18 09:24:22 CDT 2020
Back to SuperLU + GPUs (adding Sherry)
I get this error (appended) running 'check', as I said before. It looks
like ex19 is *failing* with CUDA but it is not clear it has anything to do
with SuperLU. I can not find these diagnostics that got printed after the
error in PETSc or SuperLU.
So this is a problem, but moving on to my code (plex/ex11 in
mark/feature-xgc-interface-rebase-v2, configure script appended). It runs.
I use superlu and GPUs, but they do not seem to be used in SuperLU:
------------------------------------------------------------------------------------------------------------------------
Event Count Time (sec) Flop
--- Global --- --- Stage ---- Total GPU - CpuToGpu - -
GpuToCpu - GPU
Max Ratio Max Ratio Max Ratio Mess AvgLen
Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s Mflop/s Count Size
Count Size %F
---------------------------------------------------------------------------------------------------------------------------------------------------------------
....
MatLUFactorNum 12 1.0 *2.3416e+01* 1.0 0.00e+00 0.0 0.0e+00 0.0e+00
0.0e+00 31 0 0 0 0 31 0 0 0 0 0 0 *0 0.00e+00 0
0.00e+00 0*
No CUDA version. The times are the same and no GPU communication above. So
SuperLU does not seem to be using GPUs.
------------------------------------------------------------------------------------------------------------------------
Event Count Time (sec) Flop
--- Global --- --- Stage ---- Total
Max Ratio Max Ratio Max Ratio Mess AvgLen
Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s
------------------------------------------------------------------------------------------------------------------------
....
MatLUFactorNum 12 1.0 *2.3421e+01* 1.0 0.00e+00 0.0 0.0e+00 0.0e+00
0.0e+00 5 0 0 0 0 5 0 0 0 0 0
There are some differences: ex19 use DMDA and I use DMPlex, 'check' is run
in my home directory, where files can not be written, and I run my code in
the project areas.
The timings are different without superlu so I think superlu is being
used. THis is how I run this (w and w/o -mat_superlu_equil -dm_mat_type
sell)
jsrun -n 1 -a 1 -c 2 -g 1 ./ex113d_no_cuda -dim 3 -dm_view hdf5:re33d.h5
-vec_view hdf5:re33d.h5::append -test_type spitzer -Ez 0 -petscspace_degree
2 -mass_petscspace_degree 2 -petscspace_poly_tensor 1
-mass_petscspace_poly_tensor 1 -dm_type p8est -ion_masses 4 -ion_charges 2
-thermal_temps 4,4 -n 1,.5 -n_0 1e20 -ts_monitor -ts_adapt_monitor
-snes_rtol 1.e-6 -snes_stol 1.e-9 -snes_monitor -snes_converged_reason
-snes_max_it 15 -ts_type arkimex -ts_exact_final_time stepover
-ts_arkimex_type 1bee -ts_max_snes_failures -1 -ts_rtol 1e-3 -ts_dt 1e-1
-ts_adapt_clip .25,1.05 -ts_adapt_dt_max 10 -ts_adapt_dt_min 2e-2
-ts_max_time 3200 -ts_max_steps 1 -ts_adapt_scale_solve_failed 0.75
-ts_adapt_time_step_increase_delay 5 -pc_type lu -ksp_type preonly
-amr_levels_max 11 -amr_re_levels 0 -amr_z_refine1 0 -amr_z_refine2 0
-amr_post_refine 0 -domain_radius -.95 -re_radius 4 -z_radius1 8 -z_radius2
.1 -plot_dt .10 -impurity_source_type pulse -pulse_start_time 2600
-pulse_width_time 100 -pulse_rate 1e+0 -t_cold .005 -info :dm,tsadapt:
-sub_thread_block_size 4 -options_left -log_view -pc_factor_mat_solver_type
superlu -mat_superlu_equil -dm_mat_type sell
So there is a bug in ex19 on SUMMIT and I am not getting GPUs turned on in
SuperLU.
Thoughts?
Thanks,
Mark
09:28 mark/feature-xgc-interface-rebase-v2 *= ~/petsc$ make
PETSC_DIR=/ccs/home/adams/petsc PETSC_ARCH=arch-summit-opt-gnu-cuda-omp
check
Running check examples to verify correct installation
Using PETSC_DIR=/ccs/home/adams/petsc and
PETSC_ARCH=arch-summit-opt-gnu-cuda-omp
C/C++ example src/snes/tutorials/ex19 run successfully with 1 MPI process
C/C++ example src/snes/tutorials/ex19 run successfully with 2 MPI processes
2c2,39
< Number of SNES iterations = 2
---
*> ex19: cudahook.cc:762: CUresult host_free_callback(void*): Assertion
`cacheNode != __null' failed.*> [h50n09:102287] *** Process received signal
***
> CUDA version: v 10010
> CUDA Devices:
>
> 0 : Tesla V100-SXM2-16GB 7 0
> Global memory: 16128 mb
> Shared memory: 48 kb
> Constant memory: 64 kb
> Block registers: 65536
>
> [h50n09:102287] Signal: Aborted (6)
> [h50n09:102287] Associated errno: Unknown error 1072693248 (1072693248)
> [h50n09:102287] Signal code: User function (kill, sigsend, abort, etc.)
(0)
> [h50n09:102287] [ 0] [0x2000000504d8]
> [h50n09:102287] [ 1] /lib64/libc.so.6(abort+0x2b4)[0x200021bf2094]
> [h50n09:102287] [ 2] /lib64/libc.so.6(+0x356d4)[0x200021be56d4]
> [h50n09:102287] [ 3] /lib64/libc.so.6(__assert_fail+0x64)[0x200021be57c4]
> [h50n09:102287] [ 4]
/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/gcc-6.4.0/spectrum-mpi-10.3.1.2-20200121-awz2q5brde7wgdqqw4ugalrkukeub4eb/container/../lib/libpami_cudahook.so(_Z18host_free_callbackPv+0x2d8)[0x2000000cd2c8]
> [h50n09:102287] [ 5]
/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/gcc-6.4.0/spectrum-mpi-10.3.1.2-20200121-awz2q5brde7wgdqqw4ugalrkukeub4eb/container/../lib/libpami_cudahook.so(cuMemFreeHost+0xb0)[0x2000000c3cc0]
> [h50n09:102287] [ 6]
/sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(+0x42f50)[0x20000ed02f50]
> [h50n09:102287] [ 7]
/sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(+0x11db8)[0x20000ecd1db8]
> [h50n09:102287] [ 8]
/sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(cudaFreeHost+0x74)[0x20000ed12ea4]
> [h50n09:102287] [ 9]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libsuperlu_dist.so.6(dDestroy_LU+0xc4)[0x20000195aff4]
> [h50n09:102287] [10]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(+0x7cdb70)[0x2000008bdb70]
> [h50n09:102287] [11]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(MatLUFactorNumeric+0x1ec)[0x2000005f1a8c]
> [h50n09:102287] [12]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(+0xbf8270)[0x200000ce8270]
> [h50n09:102287] [13]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(PCSetUp+0x1a4)[0x200000d8d5a4]
> [h50n09:102287] [14]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(KSPSetUp+0x40c)[0x200000dc498c]
> [h50n09:102287] [15]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(+0xcd56fc)[0x200000dc56fc]
> [h50n09:102287] [16]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(KSPSolve+0x20)[0x200000dc8260]
> [h50n09:102287] [17]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(+0xe0a170)[0x200000efa170]
> [h50n09:102287] [18]
/ccs/home/adams/petsc/arch-summit-opt-gnu-cuda-omp/lib/libpetsc.so.3.013(SNESSolve+0x814)[0x200000ebd394]
> [h50n09:102287] [19] ./ex19[0x10001a6c]
> [h50n09:102287] [20] /lib64/libc.so.6(+0x25200)[0x200021bd5200]
> [h50n09:102287] [21]
/lib64/libc.so.6(__libc_start_main+0xc4)[0x200021bd53f4]
> [h50n09:102287] *** End of error message ***
> ERROR: One or more process (first noticed rank 0) terminated with signal
6
/ccs/home/adams/petsc/src/snes/tutorials
Possible problem with ex19 running with superlu_dist, diffs above
#!/usr/bin/env python
if __name__ == '__main__':
import sys
import os
sys.path.insert(0, os.path.abspath('config'))
import configure
configure_options = [
'--with-fc=0',
'--COPTFLAGS=-g -O2 -fPIC -fopenmp',
'--CXXOPTFLAGS=-g -O2 -fPIC -fopenmp',
'--FOPTFLAGS=-g -O2 -fPIC -fopenmp',
'--CUDAOPTFLAGS=-O2 -g',
'--with-ssl=0',
'--with-batch=0',
'--with-cxx=mpicxx',
'--with-mpiexec=jsrun -g1',
'--with-cuda=1',
'--with-cudac=nvcc',
'--download-p4est=1',
'--download-zlib',
'--download-hdf5=1',
'--download-metis',
'--download-superlu',
'--download-superlu_dist',
'--with-make-np=16',
# '--with-hwloc=0',
'--download-parmetis',
# '--download-hypre',
'--download-triangle',
# '--download-amgx',
# '--download-fblaslapack',
'--with-blaslapack-lib=-L' + os.environ['OLCF_NETLIB_LAPACK_ROOT'] +
'/lib64 -lblas -llapack',
'--with-cc=mpicc',
# '--with-fc=mpif90',
'--with-shared-libraries=1',
# '--known-mpi-shared-libraries=1',
'--with-x=0',
'--with-64-bit-indices=0',
'--with-debugging=0',
'PETSC_ARCH=arch-summit-opt-gnu-cuda-omp',
'--with-openmp=1',
'--with-threadsaftey=1',
'--with-log=1'
]
configure.petsc_configure(configure_options)
On Wed, Apr 15, 2020 at 9:58 PM Satish Balay <balay at mcs.anl.gov> wrote:
> The crash is inside Superlu_DIST - so don't know what to suggest.
>
> Might have to debug this via debugger and check with Sherry.
>
> Satish
>
> On Wed, 15 Apr 2020, Mark Adams wrote:
>
> > Ah, OK 'check' will test SuperLU. Semi worked:
> >
> > s20:13 mark/feature-xgc-interface-rebase *= ~/petsc$ make
> > PETSC_DIR=/ccs/home/adams/petsc PETSC_ARCH=arch-summit-dbg-gnu-cuda-omp
> > check
> > Running check examples to verify correct installation
> > Using PETSC_DIR=/ccs/home/adams/petsc and
> > PETSC_ARCH=arch-summit-dbg-gnu-cuda-omp
> > C/C++ example src/snes/tutorials/ex19 run successfully with 1 MPI process
> > C/C++ example src/snes/tutorials/ex19 run successfully with 2 MPI
> processes
> > 2c2,38
> > < Number of SNES iterations = 2
> > ---
> > > CUDA version: v 10010
> > > CUDA Devices:
> > >
> > > 0 : Tesla V100-SXM2-16GB 7 0
> > > Global memory: 16128 mb
> > > Shared memory: 48 kb
> > > Constant memory: 64 kb
> > > Block registers: 65536
> > >
> > > ex19: cudahook.cc:762: CUresult host_free_callback(void*): Assertion
> > `cacheNode != __null' failed.
> > > [h16n07:78357] *** Process received signal ***
> > > [h16n07:78357] Signal: Aborted (6)
> > > [h16n07:78357] Signal code: (1704218624)
> > > [h16n07:78357] [ 0] [0x2000000504d8]
> > > [h16n07:78357] [ 1] /lib64/libc.so.6(abort+0x2b4)[0x200023992094]
> > > [h16n07:78357] [ 2] /lib64/libc.so.6(+0x356d4)[0x2000239856d4]
> > > [h16n07:78357] [ 3]
> /lib64/libc.so.6(__assert_fail+0x64)[0x2000239857c4]
> > > [h16n07:78357] [ 4]
> >
> /autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/gcc-6.4.0/spectrum-mpi-10.3.1.2-20200121-awz2q5brde7wgdqqw4ugalrkukeub4eb/container/../lib/libpami_cudahook.so(_Z18host_free_callbackPv+0x2d8)[0x2000000cd2c8]
> > > [h16n07:78357] [ 5]
> >
> /autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/gcc-6.4.0/spectrum-mpi-10.3.1.2-20200121-awz2q5brde7wgdqqw4ugalrkukeub4eb/container/../lib/libpami_cudahook.so(cuMemFreeHost+0xb0)[0x2000000c3cc0]
> > > [h16n07:78357] [ 6]
> >
> /sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(+0x42f50)[0x200010aa2f50]
> > > [h16n07:78357] [ 7]
> >
> /sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(+0x11db8)[0x200010a71db8]
> > > [h16n07:78357] [ 8]
> >
> /sw/summit/cuda/10.1.243/lib64/libcudart.so.10.1(cudaFreeHost+0x74)[0x200010ab2ea4]
> > > [h16n07:78357] [ 9]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libsuperlu_dist.so.6(dDestroy_LU+0x150)[0x200003188058]
> > > [h16n07:78357] [10]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(+0x12ebc6c)[0x2000013dbc6c]
> > > [h16n07:78357] [11]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(MatLUFactorNumeric+0x934)[0x200000d2fae4]
> > > [h16n07:78357] [12]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(+0x1cca7a4)[0x200001dba7a4]
> > > [h16n07:78357] [13]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(PCSetUp+0xde0)[0x200001f3f990]
> > > [h16n07:78357] [14]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(KSPSetUp+0x1848)[0x200001fc5594]
> > > [h16n07:78357] [15]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(+0x1ed9908)[0x200001fc9908]
> > > [h16n07:78357] [16]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(KSPSolve+0x5d0)[0x200001fcc690]
> > > [h16n07:78357] [17]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(+0x21e16ac)[0x2000022d16ac]
> > > [h16n07:78357] [18]
> >
> /ccs/home/adams/petsc/arch-summit-dbg-gnu-cuda-omp/lib/libpetsc.so.3.013(SNESSolve+0x23f4)[0x2000022255c0]
> > > [h16n07:78357] [19] ./ex19[0x10002ac8]
> > > [h16n07:78357] [20] /lib64/libc.so.6(+0x25200)[0x200023975200]
> > > [h16n07:78357] [21]
> > /lib64/libc.so.6(__libc_start_main+0xc4)[0x2000239753f4]
> > > [h16n07:78357] *** End of error message ***
> > > ERROR: One or more process (first noticed rank 0) terminated with
> signal
> > 6
> > /ccs/home/adams/petsc/src/snes/tutorials
> > Possible problem with ex19 running with superlu_dist, diffs above
> > =========================================
> >
> > On Wed, Apr 15, 2020 at 5:58 PM Satish Balay <balay at mcs.anl.gov> wrote:
> >
> > > Please send configure.log
> > >
> > > This is what I get on my linux build:
> > >
> > > [balay at p1 petsc]$ ./configure
> > > --with-mpi-dir=/home/petsc/soft/openmpi-4.0.2-cuda --with-cuda=1
> > > --with-openmp=1 --download-superlu-dist=1 && make && make check
> > > <snip>
> > > Running check examples to verify correct installation
> > > Using PETSC_DIR=/home/balay/petsc and PETSC_ARCH=arch-linux-c-debug
> > > C/C++ example src/snes/tutorials/ex19 run successfully with 1 MPI
> process
> > > C/C++ example src/snes/tutorials/ex19 run successfully with 2 MPI
> processes
> > > 1a2,19
> > > > CUDA version: v 10020
> > > > CUDA Devices:
> > > >
> > > > 0 : Quadro T2000 7 5
> > > > Global memory: 3911 mb
> > > > Shared memory: 48 kb
> > > > Constant memory: 64 kb
> > > > Block registers: 65536
> > > >
> > > > CUDA version: v 10020
> > > > CUDA Devices:
> > > >
> > > > 0 : Quadro T2000 7 5
> > > > Global memory: 3911 mb
> > > > Shared memory: 48 kb
> > > > Constant memory: 64 kb
> > > > Block registers: 65536
> > > >
> > > /home/balay/petsc/src/snes/tutorials
> > > Possible problem with ex19 running with superlu_dist, diffs above
> > > =========================================
> > > Fortran example src/snes/tutorials/ex5f run successfully with 1 MPI
> process
> > > Completed test examples
> > >
> > >
> > > On Wed, 15 Apr 2020, Mark Adams wrote:
> > >
> > > > On Wed, Apr 15, 2020 at 5:17 PM Satish Balay <balay at mcs.anl.gov>
> wrote:
> > > >
> > > > > The build should work. It should give some verbose info [at
> runtime]
> > > > > regarding GPUs - from the following code.
> > > > >
> > > > >
> > > > I don't see that and I am running GPUs in my code and have gotten
> > > cusparse
> > > > LU to run. Should I use '-info :sys:' ?
> > > >
> > > >
> > > > > >>>>> SRC/cublas_utils.c >>>>>>>>>>>
> > > > > void DisplayHeader()
> > > > > {
> > > > > const int kb = 1024;
> > > > > const int mb = kb * kb;
> > > > > // cout << "NBody.GPU" << endl << "=========" << endl << endl;
> > > > >
> > > > > printf("CUDA version: v %d\n",CUDART_VERSION);
> > > > > //cout << "Thrust version: v" << THRUST_MAJOR_VERSION << "." <<
> > > > > THRUST_MINOR_VERSION << endl << endl;
> > > > >
> > > > > int devCount;
> > > > > cudaGetDeviceCount(&devCount);
> > > > > printf( "CUDA Devices: \n \n");
> > > > > <snip>
> > > > > <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
> > > > >
> > > > > Satish
> > > > >
> > > > > On Wed, 15 Apr 2020, Junchao Zhang wrote:
> > > > >
> > > > > > I remember Barry said superlu gpu support is broken.
> > > > > > --Junchao Zhang
> > > > > >
> > > > > >
> > > > > > On Wed, Apr 15, 2020 at 3:47 PM Mark Adams <mfadams at lbl.gov>
> wrote:
> > > > > >
> > > > > > > How does one use SuperLU with GPUs. I don't seem to get any GPU
> > > > > > > performance data so I assume GPUs are not getting turned on.
> Am I
> > > wrong
> > > > > > > about that?
> > > > > > >
> > > > > > > I configure with:
> > > > > > > configure options: --with-fc=0 --COPTFLAGS="-g -O2 -fPIC
> -fopenmp"
> > > > > > > --CXXOPTFLAGS="-g -O2 -fPIC -fopenmp" --FOPTFLAGS="-g -O2 -fPIC
> > > > > -fopenmp"
> > > > > > > --CUDAOPTFLAGS="-O2 -g" --with-ssl=0 --with-batch=0
> > > --with-cxx=mpicxx
> > > > > > > --with-mpiexec="jsrun -g1" --with-cuda=1 --with-cudac=nvcc
> > > > > > > --download-p4est=1 --download-zlib --download-hdf5=1
> > > --download-metis
> > > > > > > --download-superlu --download-superlu_dist --with-make-np=16
> > > > > > > --download-parmetis --download-triangle
> > > > > > >
> > > > >
> > >
> --with-blaslapack-lib="-L/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/gcc-6.4.0/netlib-lapack-3.8.0-wcabdyqhdi5rooxbkqa6x5d7hxyxwdkm/lib64
> > > > > > > -lblas -llapack" --with-cc=mpicc --with-shared-libraries=1
> > > --with-x=0
> > > > > > > --with-64-bit-indices=0 --with-debugging=0
> > > > > > > PETSC_ARCH=arch-summit-opt-gnu-cuda-omp --with-openmp=1
> > > > > > > --with-threadsaftey=1 --with-log=1
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Mark
> > > > > > >
> > > > > >
> > > > >
> > > > >
> > > >
> > >
> > >
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20200418/a004470f/attachment-0001.html>
More information about the petsc-users
mailing list