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 <[email protected]> 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 <[email protected]> > 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 <[email protected]> wrote: > >> >> >> On Tue, Jan 25, 2022 at 9:04 AM Jacob Faibussowitsch <[email protected]> >> 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 <[email protected]> wrote: >>> >>> Thanks Jacob, >>> >>> On Thu, Jan 20, 2022 at 6:25 PM Jacob Faibussowitsch < >>> [email protected]> 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 <[email protected]> wrote: >>>> >>>> On Thu, Jan 20, 2022 at 6:44 PM Fande Kong <[email protected]> wrote: >>>> >>>>> Thanks, Jed >>>>> >>>>> On Thu, Jan 20, 2022 at 4:34 PM Jed Brown <[email protected]> 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 <[email protected]> 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 <[email protected]> >>>>>> wrote: >>>>>> > >>>>>> >> Thanks, Jed, >>>>>> >> >>>>>> >> This worked! >>>>>> >> >>>>>> >> Fande >>>>>> >> >>>>>> >> On Wed, Jan 19, 2022 at 11:03 PM Jed Brown <[email protected]> >>>>>> wrote: >>>>>> >> >>>>>> >>> Fande Kong <[email protected]> writes: >>>>>> >>> >>>>>> >>> > On Wed, Jan 19, 2022 at 11:39 AM Jacob Faibussowitsch < >>>>>> >>> [email protected]> >>>>>> >>> > 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/> >>>> >>>> >>>
