Peter Doak
Peter Doak
Using gcc 6.5.0 on x86 [33/61] Building CXX object kokkos/CMakeFiles/kokkos.dir/core/src/impl/Kokkos_HostSpace.cpp.o FAILED: /lustre/or-hydra/cades-cnms/epd/dev/kokkos/bin/nvcc_wrapper -I/lustre/or-hydra/cades-cnms/epd/dev/kokkos/core/src -I/lustre/or-hydra/cades-cnms/epd/dev/kokkos/containers/src -I/lustre/or-hydra/cades-cnms/epd/dev/kokkos/algorithms/src -Ikokkos -I/software/dev_tools/swtree/cs400_centos7.2_pe2016-08/cuda/9.2/centos7.2_binary/include -Drestrict=__restrict__ -D__forceinline=inline -O3 -DNDEBUG --std=c++11 -Xcudafe --diag_suppress=esa_on_defaulted_function_ignored -expt-extended-lambda -march=core-avx2 -mtune=core-avx2 -arch=sm_60 -MD -MT...
KOKKOS_ENABLE_CUDA=false KOKKOS_ENABLE_OPENMP=false causing ```shell [31/54] Building CXX object src/Drivers/CMakeFiles/miniwfs.dir/__/QMCWaveFunctions/SPOSet_builder.cpp.o FAILED: src/Drivers/CMakeFiles/miniwfs.dir/__/QMCWaveFunctions/SPOSet_builder.cpp.o /data/epd/spack/opt/spack/linux-rhel7-x86_64/gcc-7.3.0/llvm-7.0.0-amm2tanx4cbnqcpwd3mnf52fbes3eg7j/bin/clang++ -DADD_ -DH5_USE_16_API -DHAVE_CONFIG_H -DINLINE_ALL=inline -DUSE_REAL_STRUCT_FACTOR -I/scratch/epd/KOKKOS/kokkos/core/src -I/scratch/epd/KOKKOS/kokkos/containers/src -I/scratch/epd/KOKKOS/kokkos/algorithms/src -Ikokkos -I/opt/intel2018/mkl/include -I../src -Isrc -I../external_codes/hana/include -DFUTURE_WAVEFUNCTIONS -Drestrict=__restrict__ -D__forceinline=inline -g --std=c++11...
Removed strangely justified static const, with llvm this causes a segfault.
Currently DCA does little checking of user input for validity or match with the built executable model etc. This needs to improve even for expert users. 1. If working directory...
Meaning they only match on daint. Those from summit cause CI to fail on daint. Also its not a good idea to have binaries like this in the github repository....
With regards to the life cycle of the object, RAII could be better. If we're going to throw for a bad file and the purpose of close file is it...
Currently if you type "single-band-hubbard-model" instead of "single-band-Hubbard-model" no error is indicated. If no model is found error should occur before the loop starts.
Many clusters do not use GPU aware MPI currently the distributed G4 code requires it. This needs to be dealt with for compilation, testing, and runtime.
With distG4 and the need to support linear indexing into large tensors int is not always going to cut it as a type. Make this a compile time choice i.e....
With the update of the interpolation of G0 in TpEqualTimeAccumulator, I went looking for similar code segments. We have the following places in the code where we might should consider...