[petsc-users] Cannot eagerly initialize cuda, as doing so results in cuda error 35 (cudaErrorInsufficientDriver) : CUDA driver version is insufficient for CUDA runtime version

Fande Kong fdkong.jd at gmail.com
Mon Jan 31 09:40:24 CST 2022


OK,

Finally we resolved the issue.  The issue was that there were two libcuda
libs on a GPU compute node:  /usr/lib64/libcuda
and /apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs/libcuda.
But on a login node there is one libcuda lib:
/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs/libcuda.
We can not see  /usr/lib64/libcuda from a login node where I was compiling
the code.

Before the Junchao's commit, we did not have  "-Wl,-rpath" to force PETSc
take
/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs/libcuda.
A code compiled on a login node could correctly pick up the cuda lib
from  /usr/lib64/libcuda at runtime.  When with "-Wl,-rpath", the code
always  takes the cuda lib from
/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs/libcuda,
wihch was a bad lib.

Right now, I just compiled code on a compute node instead of a login node,
PETSc was able to pick up the correct lib from  /usr/lib64/libcuda, and
everything ran fine.

I am not sure whether or not it is a good idea to search for "stubs" since
the system might have the correct ones in other places.  Should not I do a
batch compiling?

Thanks,

Fande


On Wed, Jan 26, 2022 at 1:49 PM Fande Kong <fdkong.jd at gmail.com> wrote:

> Yes, please see the attached file.
>
> Fande
>
> On Wed, Jan 26, 2022 at 11:49 AM Junchao Zhang <junchao.zhang at gmail.com>
> wrote:
>
>> Do you have the configure.log with main?
>>
>> --Junchao Zhang
>>
>>
>> On Wed, Jan 26, 2022 at 12:26 PM Fande Kong <fdkong.jd at gmail.com> wrote:
>>
>>> I am on the petsc-main
>>>
>>> commit 1390d3a27d88add7d79c9b38bf1a895ae5e67af6
>>>
>>> Merge: 96c919c d5f3255
>>>
>>> Author: Satish Balay <balay at mcs.anl.gov>
>>>
>>> Date:   Wed Jan 26 10:28:32 2022 -0600
>>>
>>>
>>>     Merge remote-tracking branch 'origin/release'
>>>
>>>
>>> It is still broken.
>>>
>>> Thanks,
>>>
>>>
>>> Fande
>>>
>>> On Wed, Jan 26, 2022 at 7:40 AM Junchao Zhang <junchao.zhang at gmail.com>
>>> wrote:
>>>
>>>> The good uses the compiler's default library/header path.  The bad
>>>> searches from cuda toolkit path and uses rpath linking.
>>>> Though the paths look the same on the login node, they could have
>>>> different behavior on a compute node depending on its environment.
>>>> I think we fixed the issue in cuda.py (i.e., first try the compiler's
>>>> default, then toolkit).  That's why I wanted Fande to use petsc/main.
>>>>
>>>> --Junchao Zhang
>>>>
>>>>
>>>> On Tue, Jan 25, 2022 at 11:59 PM Barry Smith <bsmith at petsc.dev> wrote:
>>>>
>>>>>
>>>>> 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
>>>>> $PETSC_ARCH/lib/petsc/conf/variables and likely the bad will start working.
>>>>>
>>>>> Barry
>>>>>
>>>>> I never liked the stubs stuff.
>>>>>
>>>>> On Jan 25, 2022, at 11:29 PM, Fande Kong <fdkong.jd at gmail.com> wrote:
>>>>>
>>>>> Hi Junchao,
>>>>>
>>>>> I attached a "bad" configure log and a "good" configure log.
>>>>>
>>>>> The "bad" one was on produced
>>>>> at 246ba74192519a5f34fb6e227d1c64364e19ce2c
>>>>>
>>>>> and the "good" one at 384645a00975869a1aacbd3169de62ba40cad683
>>>>>
>>>>> This good hash is the last good hash that is just the right before the
>>>>> bad one.
>>>>>
>>>>> I think you could do a comparison  between these two logs, and check
>>>>> what the differences were.
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Fande
>>>>>
>>>>> On Tue, Jan 25, 2022 at 8:21 PM Junchao Zhang <junchao.zhang at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Fande, could you send the configure.log that works (i.e., before this
>>>>>> offending commit)?
>>>>>> --Junchao Zhang
>>>>>>
>>>>>>
>>>>>> On Tue, Jan 25, 2022 at 8:21 PM Fande Kong <fdkong.jd at gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> Not sure if this is helpful. I did "git bisect", and here was the
>>>>>>> result:
>>>>>>>
>>>>>>> [kongf at sawtooth2 petsc]$ git bisect bad
>>>>>>> 246ba74192519a5f34fb6e227d1c64364e19ce2c is the first bad commit
>>>>>>> commit 246ba74192519a5f34fb6e227d1c64364e19ce2c
>>>>>>> Author: Junchao Zhang <jczhang at 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.jd at gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On Tue, Jan 25, 2022 at 9:04 AM Jacob Faibussowitsch <
>>>>>>>> jacob.fai at 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.jd at gmail.com> wrote:
>>>>>>>>>
>>>>>>>>> Thanks Jacob,
>>>>>>>>>
>>>>>>>>> On Thu, Jan 20, 2022 at 6:25 PM Jacob Faibussowitsch <
>>>>>>>>> jacob.fai at 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 <knepley at gmail.com>
>>>>>>>>>> wrote:
>>>>>>>>>>
>>>>>>>>>> On Thu, Jan 20, 2022 at 6:44 PM Fande Kong <fdkong.jd at gmail.com>
>>>>>>>>>> wrote:
>>>>>>>>>>
>>>>>>>>>>> Thanks, Jed
>>>>>>>>>>>
>>>>>>>>>>> On Thu, Jan 20, 2022 at 4:34 PM Jed Brown <jed at 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.jd at 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 at entry=0x2aaab7f37b70
>>>>>>>>>>>> > <CUDADevice>, device=0x115da00, id=-35, id at entry=-1) at
>>>>>>>>>>>> >
>>>>>>>>>>>> /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/impls/cupm/cupmdevice.cxx:344
>>>>>>>>>>>> > #2  0x00002aaab55577de in PetscDeviceCreate (type=type at entry
>>>>>>>>>>>> =PETSC_DEVICE_CUDA,
>>>>>>>>>>>> > devid=devid at entry=-1, device=device at 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 at entry=PETSC_DEVICE_CUDA,
>>>>>>>>>>>> defaultDeviceId=defaultDeviceId at entry=-1)
>>>>>>>>>>>> > at
>>>>>>>>>>>> >
>>>>>>>>>>>> /home/kongf/workhome/sawtooth/moosegpu/petsc/src/sys/objects/device/interface/device.cxx:273
>>>>>>>>>>>> > #4  0x00002aaab5557bf6 in PetscDeviceInitialize
>>>>>>>>>>>> > (type=type at 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 at entry
>>>>>>>>>>>> =0x115d150,
>>>>>>>>>>>> > method=method at 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 at entry
>>>>>>>>>>>> =0x115d150,
>>>>>>>>>>>> > method=method at 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.jd at gmail.com> wrote:
>>>>>>>>>>>> >
>>>>>>>>>>>> >> Thanks, Jed,
>>>>>>>>>>>> >>
>>>>>>>>>>>> >> This worked!
>>>>>>>>>>>> >>
>>>>>>>>>>>> >> Fande
>>>>>>>>>>>> >>
>>>>>>>>>>>> >> On Wed, Jan 19, 2022 at 11:03 PM Jed Brown <jed at jedbrown.org>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>> >>
>>>>>>>>>>>> >>> Fande Kong <fdkong.jd at gmail.com> writes:
>>>>>>>>>>>> >>>
>>>>>>>>>>>> >>> > On Wed, Jan 19, 2022 at 11:39 AM Jacob Faibussowitsch <
>>>>>>>>>>>> >>> jacob.fai at 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>
>>>>>
>>>>>
>>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220131/b1f59aab/attachment-0001.html>


More information about the petsc-users mailing list