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

Junchao Zhang junchao.zhang at gmail.com
Mon Jan 31 13:10:48 CST 2022


On Mon, Jan 31, 2022 at 10:50 AM Fande Kong <fdkong.jd at gmail.com> wrote:

> Sorry for the confusion. I thought I explained pretty well :-)
>
> Good:
>
> PETSc was linked to  /usr/lib64/libcuda for libcuda
>
> Bad:
>
> PETSc was linked
> to /apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs
> for libcuda
>
> My question would be: where should I look for libcuda?
>
> Our HPC admin told me that I should use the one from  /usr/lib64/libcuda
>
Your admin was correct.

>
> I am trying to understand why we need to link to "stubs"?
>
Kokkos needs libcuda.so, so we added this requirement.


> Just to be clear, I am fine with PETSc-main as is since I can use a
> compute node to compile PETSc.  However, here I am trying really hard to
> understand where I should look for the right libcuda.
>
I need your help to find out: Why on compute nodes, did the petsc test
executable find libcuda.so at /apps/local/spack/software/
gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs?
Note this path is not in the executable's rpath.  Maybe you need to login
to a compute node and do 'env' to list all variables for us to have a look.
>
>
> Thanks for your help
>
> Fande
>
>
> On Mon, Jan 31, 2022 at 9:19 AM Junchao Zhang <junchao.zhang at gmail.com>
> wrote:
>
>> Fande,
>>   From your configure_main.log
>>
>> cuda:
>>   Version:  10.1
>>   Includes:
>> -I/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/include
>>   Library:
>>  -Wl,-rpath,/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64
>> -L/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64
>> -L/apps/local/spack/software/gcc-7.5.0/cuda-10.1.243-v4ymjqcrr7f72qfiuzsstuy5jiajbuey/lib64/stubs
>> -lcudart -lcufft -lcublas -lcusparse -lcusolver -lcurand -lcuda
>>
>>
>> You can see the `stubs` directory is not in rpath. We took a lot of
>> effort to achieve that. You need to double check the reason.
>>
>> --Junchao Zhang
>>
>>
>> On Mon, Jan 31, 2022 at 9:40 AM Fande Kong <fdkong.jd at gmail.com> wrote:
>>
>>> 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/c2cb86ea/attachment-0001.html>


More information about the petsc-users mailing list