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> 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> wrote: > > > On Tue, Jan 25, 2022 at 9:04 AM Jacob Faibussowitsch <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> wrote: >> >> Thanks Jacob, >> >> On Thu, Jan 20, 2022 at 6:25 PM Jacob Faibussowitsch <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> wrote: >>> >>> On Thu, Jan 20, 2022 at 6:44 PM Fande Kong <fdkong...@gmail.com> wrote: >>> >>>> Thanks, Jed >>>> >>>> On Thu, Jan 20, 2022 at 4:34 PM Jed Brown <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> 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 >>>>> > #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> >>>>> wrote: >>>>> > >>>>> >> Thanks, Jed, >>>>> >> >>>>> >> This worked! >>>>> >> >>>>> >> Fande >>>>> >> >>>>> >> On Wed, Jan 19, 2022 at 11:03 PM Jed Brown <j...@jedbrown.org> >>>>> wrote: >>>>> >> >>>>> >>> Fande Kong <fdkong...@gmail.com> writes: >>>>> >>> >>>>> >>> > On Wed, Jan 19, 2022 at 11:39 AM Jacob Faibussowitsch < >>>>> >>> 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/> >>> >>> >>