bad has extra -L/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs -lcuda
good does not. Try removing the stubs directory and -lcuda from the bad $PETSC_ARCH/lib/petsc/conf/variables and likely the bad will start working. Barry I never liked the stubs stuff. > On Jan 25, 2022, at 11:29 PM, Fande Kong <fdkong...@gmail.com> wrote: > > Hi Junchao, > > I attached a "bad" configure log and a "good" configure log. > > The "bad" one was on produced at 246ba74192519a5f34fb6e227d1c64364e19ce2c > > and the "good" one at 384645a00975869a1aacbd3169de62ba40cad683 > > This good hash is the last good hash that is just the right before the bad > one. > > I think you could do a comparison between these two logs, and check what the > differences were. > > Thanks, > > Fande > > On Tue, Jan 25, 2022 at 8:21 PM Junchao Zhang <junchao.zh...@gmail.com > <mailto:junchao.zh...@gmail.com>> wrote: > Fande, could you send the configure.log that works (i.e., before this > offending commit)? > --Junchao Zhang > > > On Tue, Jan 25, 2022 at 8:21 PM Fande Kong <fdkong...@gmail.com > <mailto:fdkong...@gmail.com>> wrote: > Not sure if this is helpful. I did "git bisect", and here was the result: > > [kongf@sawtooth2 petsc]$ git bisect bad > 246ba74192519a5f34fb6e227d1c64364e19ce2c is the first bad commit > commit 246ba74192519a5f34fb6e227d1c64364e19ce2c > Author: Junchao Zhang <jczh...@mcs.anl.gov <mailto:jczh...@mcs.anl.gov>> > Date: Wed Oct 13 05:32:43 2021 +0000 > > Config: fix CUDA library and header dirs > > :040000 040000 187c86055adb80f53c1d0565a8888704fec43a96 > ea1efd7f594fd5e8df54170bc1bc7b00f35e4d5f M config > > > Started from this commit, and GPU did not work for me on our HPC > > Thanks, > Fande > > On Tue, Jan 25, 2022 at 7:18 PM Fande Kong <fdkong...@gmail.com > <mailto:fdkong...@gmail.com>> wrote: > > > On Tue, Jan 25, 2022 at 9:04 AM Jacob Faibussowitsch <jacob....@gmail.com > <mailto:jacob....@gmail.com>> wrote: > Configure should not have an impact here I think. The reason I had you run > `cudaGetDeviceCount()` is because this is the CUDA call (and in fact the only > CUDA call) in the initialization sequence that returns the error code. There > should be no prior CUDA calls. Maybe this is a problem with oversubscribing > GPU’s? In the runs that crash, how many ranks are using any given GPU at > once? Maybe MPS is required. > > I used one MPI rank. > > Fande > > > > Best regards, > > Jacob Faibussowitsch > (Jacob Fai - booss - oh - vitch) > >> On Jan 21, 2022, at 12:01, Fande Kong <fdkong...@gmail.com >> <mailto:fdkong...@gmail.com>> wrote: >> >> Thanks Jacob, >> >> On Thu, Jan 20, 2022 at 6:25 PM Jacob Faibussowitsch <jacob....@gmail.com >> <mailto:jacob....@gmail.com>> wrote: >> Segfault is caused by the following check at >> src/sys/objects/device/impls/cupm/cupmdevice.cxx:349 being a >> PetscUnlikelyDebug() rather than just PetscUnlikely(): >> >> ``` >> if (PetscUnlikelyDebug(_defaultDevice < 0)) { // _defaultDevice is in fact < >> 0 here and uncaught >> ``` >> >> To clarify: >> >> “lazy” initialization is not that lazy after all, it still does some 50% of >> the initialization that “eager” initialization does. It stops short >> initializing the CUDA runtime, checking CUDA aware MPI, gathering device >> data, and initializing cublas and friends. Lazy also importantly swallows >> any errors that crop up during initialization, storing the resulting error >> code for later (specifically _defaultDevice = -init_error_value;). >> >> So whether you initialize lazily or eagerly makes no difference here, as >> _defaultDevice will always contain -35. >> >> The bigger question is why cudaGetDeviceCount() is returning >> cudaErrorInsufficientDriver. Can you compile and run >> >> ``` >> #include <cuda_runtime.h> >> >> int main() >> { >> int ndev; >> return cudaGetDeviceCount(&ndev): >> } >> ``` >> >> Then show the value of "echo $?”? >> >> Modify your code a little to get more information. >> >> #include <cuda_runtime.h> >> #include <cstdio> >> >> int main() >> { >> int ndev; >> int error = cudaGetDeviceCount(&ndev); >> printf("ndev %d \n", ndev); >> printf("error %d \n", error); >> return 0; >> } >> >> Results: >> >> $ ./a.out >> ndev 4 >> error 0 >> >> >> I have not read the PETSc cuda initialization code yet. If I need to guess >> at what was happening. I will naively think that PETSc did not get correct >> GPU information in the configuration because the compiler node does not have >> GPUs, and there was no way to get any GPU device information. >> >> During the runtime on GPU nodes, PETSc might have incorrect information >> grabbed during configuration and had this kind of false error message. >> >> Thanks, >> >> Fande >> >> >> >> Best regards, >> >> Jacob Faibussowitsch >> (Jacob Fai - booss - oh - vitch) >> >>> On Jan 20, 2022, at 17:47, Matthew Knepley <knep...@gmail.com >>> <mailto:knep...@gmail.com>> wrote: >>> >>> On Thu, Jan 20, 2022 at 6:44 PM Fande Kong <fdkong...@gmail.com >>> <mailto:fdkong...@gmail.com>> wrote: >>> Thanks, Jed >>> >>> On Thu, Jan 20, 2022 at 4:34 PM Jed Brown <j...@jedbrown.org >>> <mailto:j...@jedbrown.org>> wrote: >>> You can't create CUDA or Kokkos Vecs if you're running on a node without a >>> GPU. >>> >>> I am running the code on compute nodes that do have GPUs. >>> >>> If you are actually running on GPUs, why would you need lazy >>> initialization? It would not break with GPUs present. >>> >>> Matt >>> >>> With PETSc-3.16.1, I got good speedup by running GAMG on GPUs. That might >>> be a bug of PETSc-main. >>> >>> Thanks, >>> >>> Fande >>> >>> >>> >>> KSPSetUp 13 1.0 6.4400e-01 1.0 2.02e+09 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 5 0 0 0 0 5 0 0 0 3140 64630 15 1.05e+02 5 >>> 3.49e+01 100 >>> KSPSolve 1 1.0 1.0109e+00 1.0 3.49e+10 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 87 0 0 0 0 87 0 0 0 34522 69556 4 4.35e-03 1 >>> 2.38e-03 100 >>> KSPGMRESOrthog 142 1.0 1.2674e-01 1.0 1.06e+10 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 27 0 0 0 0 27 0 0 0 83755 87801 0 0.00e+00 0 >>> 0.00e+00 100 >>> SNESSolve 1 1.0 4.4402e+01 1.0 4.00e+10 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 21100 0 0 0 21100 0 0 0 901 51365 57 1.10e+03 52 >>> 8.78e+02 100 >>> SNESSetUp 1 1.0 3.9101e-05 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> SNESFunctionEval 2 1.0 1.7097e+01 1.0 1.60e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 8 0 0 0 0 8 0 0 0 0 1 0 0 0.00e+00 6 >>> 1.92e+02 0 >>> SNESJacobianEval 1 1.0 1.6213e+01 1.0 2.80e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 8 0 0 0 0 8 0 0 0 0 2 0 0 0.00e+00 1 >>> 3.20e+01 0 >>> SNESLineSearch 1 1.0 8.5582e+00 1.0 1.24e+08 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 4 0 0 0 0 4 0 0 0 0 14 64153 1 3.20e+01 3 >>> 9.61e+01 94 >>> PCGAMGGraph_AGG 5 1.0 3.0509e+00 1.0 8.19e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 1 0 0 0 0 1 0 0 0 0 27 0 5 3.49e+01 9 >>> 7.43e+01 0 >>> PCGAMGCoarse_AGG 5 1.0 3.8711e+00 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 2 0 0 0 0 2 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> PCGAMGProl_AGG 5 1.0 7.0748e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> PCGAMGPOpt_AGG 5 1.0 1.2904e+00 1.0 2.14e+09 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 1 5 0 0 0 1 5 0 0 0 1661 29807 26 7.15e+02 20 >>> 2.90e+02 99 >>> GAMG: createProl 5 1.0 8.9489e+00 1.0 2.22e+09 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 4 6 0 0 0 4 6 0 0 0 249 29666 31 7.50e+02 29 >>> 3.64e+02 96 >>> Graph 10 1.0 3.0478e+00 1.0 8.19e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 1 0 0 0 0 1 0 0 0 0 27 0 5 3.49e+01 9 >>> 7.43e+01 0 >>> MIS/Agg 5 1.0 4.1290e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> SA: col data 5 1.0 1.9127e-02 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> SA: frmProl0 5 1.0 6.2662e-01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> SA: smooth 5 1.0 4.9595e-01 1.0 1.21e+08 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 244 2709 15 1.97e+02 15 >>> 2.55e+02 90 >>> GAMG: partLevel 5 1.0 4.7330e-01 1.0 6.98e+08 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 2 0 0 0 0 2 0 0 0 1475 4120 5 1.78e+02 10 >>> 2.55e+02 100 >>> PCGAMG Squ l00 1 1.0 2.6027e+00 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 >>> 0.0e+00 1 0 0 0 0 1 0 0 0 0 0 0 0 0.00e+00 0 >>> 0.00e+00 0 >>> PCGAMG Gal l00 1 1.0 3.8406e-01 1.0 5.48e+08 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 1 0 0 0 0 1 0 0 0 1426 4270 1 1.48e+02 2 >>> 2.11e+02 100 >>> PCGAMG Opt l00 1 1.0 2.4932e-01 1.0 7.20e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 289 2653 1 6.41e+01 1 >>> 1.13e+02 100 >>> PCGAMG Gal l01 1 1.0 6.6279e-02 1.0 1.09e+08 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 1645 3851 1 2.40e+01 2 >>> 3.64e+01 100 >>> PCGAMG Opt l01 1 1.0 2.9544e-02 1.0 7.15e+06 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 242 1671 1 4.84e+00 1 >>> 1.23e+01 100 >>> PCGAMG Gal l02 1 1.0 1.8874e-02 1.0 3.72e+07 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 1974 3636 1 5.04e+00 2 >>> 6.58e+00 100 >>> PCGAMG Opt l02 1 1.0 7.4353e-03 1.0 2.40e+06 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 323 1457 1 7.71e-01 1 >>> 2.30e+00 100 >>> PCGAMG Gal l03 1 1.0 2.8479e-03 1.0 4.10e+06 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 1440 2266 1 4.44e-01 2 >>> 5.51e-01 100 >>> PCGAMG Opt l03 1 1.0 8.2684e-04 1.0 2.80e+05 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 339 1667 1 6.72e-02 1 >>> 2.03e-01 100 >>> PCGAMG Gal l04 1 1.0 1.2238e-03 1.0 2.09e+05 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 170 244 1 2.05e-02 2 >>> 2.53e-02 100 >>> PCGAMG Opt l04 1 1.0 4.1008e-04 1.0 1.77e+04 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 0 0 0 0 0 0 0 0 0 0 43 165 1 4.49e-03 1 >>> 1.19e-02 100 >>> PCSetUp 2 1.0 9.9632e+00 1.0 4.95e+09 1.0 0.0e+00 0.0e+00 >>> 0.0e+00 5 12 0 0 0 5 12 0 0 0 496 17826 55 1.03e+03 45 >>> 6.54e+02 98 >>> PCSetUpOnBlocks 44 1.0 9.9087e-04 1.0 2.88e+03 1.0 >>> >>> >>> >>> >>> >>> The point of lazy initialization is to make it possible to run a solve that >>> doesn't use a GPU in PETSC_ARCH that supports GPUs, regardless of whether a >>> GPU is actually present. >>> >>> Fande Kong <fdkong...@gmail.com <mailto:fdkong...@gmail.com>> writes: >>> >>> > I spoke too soon. It seems that we have trouble creating cuda/kokkos vecs >>> > now. Got Segmentation fault. >>> > >>> > Thanks, >>> > >>> > Fande >>> > >>> > Program received signal SIGSEGV, Segmentation fault. >>> > 0x00002aaab5558b11 in >>> > Petsc::CUPMDevice<(Petsc::CUPMDeviceType)0>::CUPMDeviceInternal::initialize >>> > (this=0x1) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/impls/cupm/cupmdevice.cxx:54 >>> > 54 PetscErrorCode CUPMDevice<T>::CUPMDeviceInternal::initialize() noexcept >>> > Missing separate debuginfos, use: debuginfo-install >>> > bzip2-libs-1.0.6-13.el7.x86_64 elfutils-libelf-0.176-5.el7.x86_64 >>> > elfutils-libs-0.176-5.el7.x86_64 glibc-2.17-325.el7_9.x86_64 >>> > libX11-1.6.7-4.el7_9.x86_64 libXau-1.0.8-2.1.el7.x86_64 >>> > libattr-2.4.46-13.el7.x86_64 libcap-2.22-11.el7.x86_64 >>> > libibmad-5.4.0.MLNX20190423.1d917ae-0.1.49224.x86_64 >>> > libibumad-43.1.1.MLNX20200211.078947f-0.1.49224.x86_64 >>> > libibverbs-41mlnx1-OFED.4.9.0.0.7.49224.x86_64 >>> > libmlx4-41mlnx1-OFED.4.7.3.0.3.49224.x86_64 >>> > libmlx5-41mlnx1-OFED.4.9.0.1.2.49224.x86_64 libnl3-3.2.28-4.el7.x86_64 >>> > librdmacm-41mlnx1-OFED.4.7.3.0.6.49224.x86_64 >>> > librxe-41mlnx1-OFED.4.4.2.4.6.49224.x86_64 libxcb-1.13-1.el7.x86_64 >>> > libxml2-2.9.1-6.el7_9.6.x86_64 numactl-libs-2.0.12-5.el7.x86_64 >>> > systemd-libs-219-78.el7_9.3.x86_64 xz-libs-5.2.2-1.el7.x86_64 >>> > zlib-1.2.7-19.el7_9.x86_64 >>> > (gdb) bt >>> > #0 0x00002aaab5558b11 in >>> > Petsc::CUPMDevice<(Petsc::CUPMDeviceType)0>::CUPMDeviceInternal::initialize >>> > (this=0x1) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/impls/cupm/cupmdevice.cxx:54 >>> > #1 0x00002aaab5558db7 in >>> > Petsc::CUPMDevice<(Petsc::CUPMDeviceType)0>::getDevice >>> > (this=this@entry=0x2aaab7f37b70 >>> > <CUDADevice>, device=0x115da00, id=-35, id@entry=-1) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/impls/cupm/cupmdevice.cxx:344 >>> > #2 0x00002aaab55577de in PetscDeviceCreate >>> > (type=type@entry=PETSC_DEVICE_CUDA, >>> > devid=devid@entry=-1, device=device@entry=0x2aaab7f37b48 >>> > <defaultDevices+8>) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/interface/device.cxx:107 >>> > #3 0x00002aaab5557b3a in PetscDeviceInitializeDefaultDevice_Internal >>> > (type=type@entry=PETSC_DEVICE_CUDA, >>> > defaultDeviceId=defaultDeviceId@entry=-1) >>> > at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/interface/device.cxx:273 >>> > #4 0x00002aaab5557bf6 in PetscDeviceInitialize >>> > (type=type@entry=PETSC_DEVICE_CUDA) >>> > at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/interface/device.cxx:234 >>> > #5 0x00002aaab5661fcd in VecCreate_SeqCUDA (V=0x115d150) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/impls/seq/seqcuda/veccuda.c:244 >>> > #6 0x00002aaab5649b40 in VecSetType (vec=vec@entry=0x115d150, >>> > method=method@entry=0x2aaab70b45b8 "seqcuda") at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/interface/vecreg.c:93 >>> > #7 0x00002aaab579c33f in VecCreate_CUDA (v=0x115d150) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/impls/mpi/mpicuda/ >>> > mpicuda.cu:214 <http://mpicuda.cu:214/> >>> > #8 0x00002aaab5649b40 in VecSetType (vec=vec@entry=0x115d150, >>> > method=method@entry=0x7fffffff9260 "cuda") at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/interface/vecreg.c:93 >>> > #9 0x00002aaab5648bf1 in VecSetTypeFromOptions_Private (vec=0x115d150, >>> > PetscOptionsObject=0x7fffffff9210) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/interface/vector.c:1263 >>> > #10 VecSetFromOptions (vec=0x115d150) at >>> > /home/kongf/workhome/sawtooth/moosegpu/petsc/src/vec/vec/interface/vector.c:1297 >>> > #11 0x00002aaab02ef227 in libMesh::PetscVector<double>::init >>> > (this=0x11cd1a0, n=441, n_local=441, fast=false, ptype=libMesh::PARALLEL) >>> > at >>> > /home/kongf/workhome/sawtooth/moosegpu/scripts/../libmesh/installed/include/libmesh/petsc_vector.h:693 >>> > >>> > On Thu, Jan 20, 2022 at 1:09 PM Fande Kong <fdkong...@gmail.com >>> > <mailto:fdkong...@gmail.com>> wrote: >>> > >>> >> Thanks, Jed, >>> >> >>> >> This worked! >>> >> >>> >> Fande >>> >> >>> >> On Wed, Jan 19, 2022 at 11:03 PM Jed Brown <j...@jedbrown.org >>> >> <mailto:j...@jedbrown.org>> wrote: >>> >> >>> >>> Fande Kong <fdkong...@gmail.com <mailto:fdkong...@gmail.com>> writes: >>> >>> >>> >>> > On Wed, Jan 19, 2022 at 11:39 AM Jacob Faibussowitsch < >>> >>> jacob....@gmail.com <mailto:jacob....@gmail.com>> >>> >>> > wrote: >>> >>> > >>> >>> >> Are you running on login nodes or compute nodes (I can’t seem to tell >>> >>> from >>> >>> >> the configure.log)? >>> >>> >> >>> >>> > >>> >>> > I was compiling codes on login nodes, and running codes on compute >>> >>> nodes. >>> >>> > Login nodes do not have GPUs, but compute nodes do have GPUs. >>> >>> > >>> >>> > Just to be clear, the same thing (code, machine) with PETSc-3.16.1 >>> >>> worked >>> >>> > perfectly. I have this trouble with PETSc-main. >>> >>> >>> >>> I assume you can >>> >>> >>> >>> export PETSC_OPTIONS='-device_enable lazy' >>> >>> >>> >>> and it'll work. >>> >>> >>> >>> I think this should be the default. The main complaint is that timing >>> >>> the >>> >>> first GPU-using event isn't accurate if it includes initialization, but >>> >>> I >>> >>> think this is mostly hypothetical because you can't trust any timing >>> >>> that >>> >>> doesn't preload in some form and the first GPU-using event will almost >>> >>> always be something uninteresting so I think it will rarely lead to >>> >>> confusion. Meanwhile, eager initialization is viscerally disruptive for >>> >>> lots of people. >>> >>> >>> >> >>> >>> >>> -- >>> 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 >>> >>> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/> > <configure_bad.log><configure_good.log>
