Hi Bruno,
> If I could get involved in this development, I would be helpful to
> optimize the memory transfer between GPU and CPU by testing different CUDA
> C programming schemes like this:
> https://devblogs.nvidia.com/how-overlap-data-transfers-cuda-cc/
> That would be nice. Playing with strea
Daniel,
Using MPI and CUDA together is pretty much work in progress. On the
> developer branch, LinearAlgebra::distributed::Vector supports
> storing its elements on a CUDA device, but the values are copied to the
> CPU for MPI communication.
> The pull request https://github.com/dealii/dealii/pu
!
Sincerely,
Chih-Che
On Thu, Nov 8, 2018 at 10:58 PM Bruno Turcksin
wrote:
> Chih-Che,
>
> Le jeu. 8 nov. 2018 à 09:49, Chih-Che Chueh a écrit :
> > -- The CUDA compiler identification is unknown
> > -- Check for working CUDA compiler: /pkg/cuda/8.0.61/bin/nvcc
> >
Bruno
> See the attached file for your reference and checking.
> The testsuite only runs for CUDA 8.0 and 9.0 can you try one of this
> version of CUDA and see if that helps?
>
When I am using cuda/8.0.61, I get the following error after I do cmake.
[chueh5851a@glogin1 build]$ module list
Curr
Bruno,
> make[1]: ***
> [source/matrix_free/CMakeFiles/obj_matrix_free_debug.dir/all] Error 2
> > make[1]: *** Waiting for unfinished jobs
> It looks like the first error is before what you showed. Also can you
> post the detailed.log file.
>
See the attached file for your reference and che
d for ‘template struct std::is_convertible’
struct is_convertible
^~
/cm/local/apps/gcc/6.3.0/include/c++/6.3.0/tuple:502:1: error: body of
constexpr function ‘static constexpr bool std::_TC<,
_Elements>::_NonNestedTuple() [with _SrcTuple = const std::tuple&; bool
Hi all,
I am trying installing dealII with CUDA in a GPG machine by following this:
https://www.dealii.org/9.0.0/external-libs/cuda.html
But I get this error in terms of matrix_tools.cc.o. Is this because I don't
turn off Trilinos stuff while cmaking deal.II for successful installation
with CUDA?
Hi Wolfgang,
I am happy to say that I can run both step-32.debug and step-32.release
successfully in parallel. Basically, I re-install p4est again by following
the steps:
./configure --enable-shared --enable-mpi
--prefix=/home/chueh5851a/software/p4est-2.0/FAST
make -j12 &> OUT.log
make install
c
Hi Wolfgang,
I don't know what is happening, but I see that all of the cases you show
> run
> the tutorial programs in release mode. Can you run them in debug mode and
> see
> whether you can get a better error message this way?
>
I will try this later after I figure out this warning happening at
Hi Wolfgang,
Hard to tell what's going wrong. What happens if you run it with one MPI
> process (or without the 'mpirun -np 4')?
Still, not working as you can see below.
[chueh5851a@clogin2 step-xx]$ ./step-32
step-32:140628 terminated with signal 11 at PC=2aaac8a02bd2
SP=7fffb240. Backt
, 2018 at 12:16 PM Wolfgang Bangerth
wrote:
> On 10/22/18 8:09 AM, Chih-Che Chueh wrote:
> > -- Could NOT find MPI_C (missing: MPI_C_WORKS)
> > -- Could NOT find MPI_CXX (missing: MPI_CXX_WORKS)
> > -- Could NOT find MPI_Fortran (missing: MPI_Fortran_WORKS)
> > --
Dear all,
In order to make install the deal.II, I load all the necessary modules
first so that I have this
[chueh5851a@clogin2 build]$ module list
Currently Loaded Modulefiles:
1) blacs/openmpi/gcc/64/1.1patch03 4) scalapack/openmpi/gcc/64/2.0.2
7) mlib/OpenBLAS-0.2.20
2) lapack/gcc/64/3.7.0
> I'm going to add that installing a compiler will really only be the
>> first
>> step. After that, of course, comes the task of understanding how to
>> program these graphics cards, and all of the other things that require
>> interacting with the appropriate software levels. You wi
Hi Wolfgang,
I think you are expecting too much of Bruno. He has a day job as well, for
> which he is paid. I think you will need to figure out the installation on
> your system yourself -- otherwise, Bruno (or any other among us) will be
> asked to do so for everyone else on the mailing list who
Hi Bruno,
Our system administrator is busy working on something else so that he is
unable to update GNU C in our system through what you advised. But he is
able to give you an account so that you can have access to our system. As a
result, I am wondering if you are willing to install the newer ver
Hi Bruno,
> You definitely don't want to update your OS. I am also using ubuntu and I
> have several versions of gcc and clang installed. Lately, I have been using
> spack to install everything for me (https://github.com/LLNL/spack). This
> is becoming the standard way to install new programs o
20:49 GMT-04:00 Bruno Turcksin :
>
>> Chih-Che,
>>
>> 2017-08-23 20:09 GMT-04:00 Chih-Che Chueh :
>> > That's still not working with getting the same error. Is this just
>> because
>> > it can't find the BLAS and LAPACK libraies?
>> Did yo
Hi Bruno
> I have already moved to the trilinos installation. But after installing
> for
> > a while, I get the error that you can see below. Is this because I am
> using
> > an old version of Intel MKL Math Kernel library? If not, does any one
> know
> > how to fix it?
> First you used pgi now y
Hi Daniel,
Thanks for the answer for the successful p4est installation. I have not
seen you since 2013 deal.II workshop. It was nice to meet you and eat
dinner with you with the same table.
Hi Daniel, Bruno or any others,
I have already moved to the trilinos installation. But after installing fo
; Regards,
> Jean-Paul
>
>
> On Wednesday, August 23, 2017 at 9:20:41 AM UTC+2, Chih-Che Chueh wrote:
>>
>> Hi all,
>>
>> When I try to install trilinos with Cmake, I find something strange:
>> while I am using cmake/3.9.1, after the installation, we get th
Hi all,
When I try to install trilinos with Cmake, I find something strange: while
I am using cmake/3.9.1, after the installation, we get the error. But in
reality the version I am using is newer than 2.6.4. Does any one who faced
the same problem know how to fix the problem?
[chueh@icy build]$
Bruno,
> pgc++-Error-Unknown switch: -print-prog-name=ld
> Don't use pgi :-) We aren't done with the pgi port.
>
I just turn off pgi on my linux background while turning on the necessary
intel C compiler etc, and try again to use the online script. It seems to
work well but with some warnings.
hih-Che
On Mon, Aug 21, 2017 at 8:16 PM, Bruno Turcksin
wrote:
> Chih-Che
>
> 2017-08-17 20:39 GMT-04:00 Chih-Che Chueh :
> > I can wait for the tests that are done. No worry.
> Everything is working now.
>
> Best,
>
> Bruno
>
> --
> The deal.II project
PM, Bruno Turcksin
wrote:
> Chih-Che
>
> 2017-08-17 20:39 GMT-04:00 Chih-Che Chueh :
> > I can wait for the tests that are done. No worry.
> Everything is working now.
>
> Best,
>
> Bruno
>
> --
> The deal.II project is located at http://www.dealii.org/
>
Bruno,
> Let me know once your merger is done so that I can start to install the
> > deal.II with CUDA in the server I use now.
> The merge was done but in the mean time we broke three tests for CUDA
> :-(. I think that it is still worth to try to install deal with CUDA
> and just update later.
Bruno,
> Is there any reason why you still used the previous way? Is this because
> > your GPU accelerator was launched some years ago so that you are unable
> to
> > use the new simple way?
> The problem with UVM is that it makes it easy to write bad code. I am
> working on a code where we use UV
e any questions, please ask. We are looking for people to help
> us with CUDA.
>
> Best,
>
> Bruno
>
>
> On Monday, August 14, 2017 at 7:56:45 AM UTC-4, Chih-Che Chueh wrote:
>>
>> Dear deal.II developers and users,
>>
>> Recently, I spent some spare time
Dear deal.II developers and users,
Recently, I spent some spare time assimilating CUDA C programming in the
last few months, and I already know very well how to use CUDA stream events
to let CPU and kernel (GPU) execution work asynchronously with efficiently
overlapping data transfer between CPU a
28 matches
Mail list logo
