[petsc-users] Kokkos error at NERSC

Barry Smith bsmith at petsc.dev
Sat Apr 24 14:22:05 CDT 2021


   I got this error when I used Kokkos master branch. When I used the branch we commonly used I did not get this problem. 

   I think it is glitch in Kokkos that the Kokkos team has to deal with if you want to use that version of Kokkos. I think it only comes up on relative new NVIDIA systems so the Kokkos team may not have seen it. Nothing to do with PETSc.

   Barry



> On Apr 24, 2021, at 2:02 PM, Mark Adams <mfadams at lbl.gov> wrote:
> 
> Thanks,
> I now see this. I recall this error being posted recently ...
> 
> nvcc warning : The 'compute_35', 'compute_37', 'compute_50', 'sm_35', 'sm_37' and 'sm_50' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
> /global/homes/m/madams/petsc/arch-cori-gpu-opt-kokkos-gcc/include/Cuda/Kokkos_Cuda_Half.hpp(220): error: more than one conversion function from "Kokkos::Experimental::half_t::impl_type" to a built-in type applies:
>             function "__half::operator float() const"
>             function "__half::operator short() const"
>             function "__half::operator unsigned short() const"
>             function "__half::operator int() const"
>             function "__half::operator unsigned int() const"
>             function "__half::operator long long() const"
>             function "__half::operator unsigned long long() const"
>             function "__half::operator __nv_bool() const"
> 
> 
> On Sat, Apr 24, 2021 at 1:09 PM Barry Smith <bsmith at petsc.dev <mailto:bsmith at petsc.dev>> wrote:
> 
>    In kokkos-kernels.py comment out 
> 
> #self.includes         = ['KokkosBlas.hpp','KokkosSparse_CrsMatrix.hpp']
> 
> 
>    I don't know why this didn't hit has a long long time ago.
> 
> 
> 
>> On Apr 24, 2021, at 8:21 AM, Mark Adams <mfadams at lbl.gov <mailto:mfadams at lbl.gov>> wrote:
>> 
>> I don't see this in kokkos-kernels.py.
>> I seem to get past Kokkos but it does in KK.
>> 
>> On Fri, Apr 23, 2021 at 4:41 PM Junchao Zhang <junchao.zhang at gmail.com <mailto:junchao.zhang at gmail.com>> wrote:
>> Hi, Mark,
>>   It is because in configure check of CUDA, we compile conftest.cc <http://conftest.cc/> (which includes kokkos headers) with mpicxx. Kokkos complains it was configured with cuda but now is not compiled with cudac.
>> 
>>   A workaround is in kokkos.py and kokkos-kernels.py, comment out lines as follows, to disable the check.
>>   #self.includes = ['Kokkos_Macros.hpp']
>> 
>> --Junchao Zhang
>> 
>> 
>> On Fri, Apr 23, 2021 at 1:59 PM Mark Adams <mfadams at lbl.gov <mailto:mfadams at lbl.gov>> wrote:
>> I had Kokkos working here but I seemed to have broken it:
>> 
>> global/homes/m/madams/petsc/arch-cori-gpu-opt-kokkos-gcc/include/setup/Kokkos_Setup_Cuda.hpp:54:2: error: #error "KOKKOS_ENABLE_CUDA defined but the compiler is not defining the __CUDACC__ macro as expected"
>>  #error \
>>  
>> Any ideas?
>> <configure.log>
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20210424/e65f9979/attachment-0001.html>


More information about the petsc-users mailing list