installing cp2k-7.1 with gpu shows this file requires compiler and library support for the ISO C++ 2011 standard.

[egbertchang]

Recently , I tried to install cp2k-7.1 with gpu.

intel2017u4

cuda9.1

but the information showed :

gcc520/include/c++/5.2.0/bits/c++0x_warning.h:32:2: error: #error This file requires compiler and library support for the ISO C++ 2011 standard. This support is currently experimental, and must be enabled with the -std=c++11 or -std=gnu++11 compiler options.

there is my file

make -j 32   ARCH=Linux-x86-64-intel VERSION=popt

Linux-x86-64-intel.popt

LIBXSMM  = $HOME/soft/cp2k-7.1-need/libxsmm

LIBXC    = $HOME/soft/cp2k-7.1-need/libxc/intel-hsw

LIBINT   = $HOME/soft/cp2k-7.1-need/libint/intel-skx

LIBELPA  = $HOME/soft/cp2k-7.1-need/elpa/intel-skx-omp

NVCC     = nvcc

CC       = cc

FC       = mpiifort

LD       = mpiifort

CXX      = g++

AR       = ar -r

NVFLAGS  = --std=c++11  --cuda -keep --dryrun

DFLAGS   = -D__MKL -D__FFTW3 -D__LIBXSMM \

           -D__parallel -D__SCALAPACK \

           -D__ELPA=201911 \

           -D__LIBXC \

           -D__LIBINT  \

           -D__MAX_CONTR=4

DFLAGS  += -D__ACC -D__DBCSR_ACC -D__PW_CUDA -D__HAS_PATCHED_CUFFT_70

CFLAGS   = $(DFLAGS) -O2 

FCFLAGS  = $(DFLAGS) -O2 -funroll-loops -fpp -free -qopenmp -auto

GPUVER = V100

FCFLAGS += -fp-model precise

FCFLAGS += -g -traceback

FCFLAGS += -I${MKLROOT}/include -I${MKLROOT}/include/fftw

FCFLAGS += -I$(LIBXSMM)/include

FCFLAGS += -I$(LIBXC)/include

FCFLAGS += -I$(LIBINT)/include

FCFLAGS += -I$(LIBELPA)/include/elpa_openmp-2019.11.001/modules -I$(LIBELPA)/include/elpa_openmp-2019.11.001/elpa

LDFLAGS  = $(FCFLAGS) -static-intel

LDFLAGS_C = $(FCFLAGS) -static-intel -nofor_main

LIBS     = -L$(LIBELPA)/lib -lelpa

MKL_LIB  = ${MKLROOT}/lib/intel64

LIBS    += $(MKL_LIB)/libmkl_scalapack_lp64.a -Wl,--start-group \

           $(MKL_LIB)/libmkl_intel_lp64.a ${MKL_LIB}/libmkl_sequential.a \

           $(MKL_LIB)/libmkl_core.a \

           ${MKL_LIB}/libmkl_blacs_intelmpi_lp64.a -Wl,--end-group \

           -lpthread -lm

LIBS    += -L$(LIBXSMM)/lib -lxsmmf -lxsmm -ldl

LIBS    += -L$(LIBXC)/lib -lxcf03 -lxc

LIBS    += -L$(LIBINT)/lib -lint2 -lstdc++

LIBS    += -lcudart -lcublas -lcufft -lnvrtc -lstdc++ -lcuda

does my popt file or the enivorment have some mistakes? 

Thanks in advance for your assistance!

[Krack Matthias (PSI)]

Did you try to add the flag -std=c++11 or -std=gnu++11 also to the CFLAGS?

 

M.

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/78a99a91-1baa-4377-a667-39cd327bb9e4%40googlegroups.com.

[egbertchang]

yes, i have already tried both of them. but the problem were still  there

在 2020年4月9日星期四 UTC+8下午6:16:26，Matthias Krack写道：

To unsubscribe from this group and stop receiving emails from it, send an email to cp...@googlegroups.com.

[Alfio Lazzaro]

CUDA requires C++, so you have to set the CXXFLAGS.

As example, take a look at the file: https://storage.googleapis.com/cp2k-ci/dashboard_cuda-pascal_report.txt

(the arch file is reported at the beginning)

Basically, the line is:

CXXFLAGS    = -O2 -std=c++11 -D__CUDA

BTW, you have a legacy macro, you are using CUDA 9, so you don't need -D__HAS_PATCHED_CUFFT_70.

Best regards,

Alfio

[egbertchang]

thank you, the problem was fixed after i following your advise,  however now it showed this mistake

ar: cublas.o: No such file or directory

   

but my cuda lib have the cublas lib 

libcublas_device.a     

libcublas.so            

libcublas.so.9.0       

libcublas.so.9.0.176 

在 2020年4月10日星期五 UTC+8下午3:53:23，Alfio Lazzaro写道：

[Alfio Lazzaro]

This is not the library, it is the compilation of the file cublas.cu (the file is at exts/dbcsr/src/acc/cublaswrap).

I assume something went wrong in the compilation.

Could you run a 

make -j 32   ARCH=Linux-x86-64-intel VERSION=popt distclean

and then recompile again? You can also dum the make.log and attach here:

make -j 32   ARCH=Linux-x86-64-intel VERSION=popt > make.log

Alfio

[egbertchang]

Dear Alfio 

here is my make.log 

在 2020年4月12日星期日 UTC+8下午4:31:46，Alfio Lazzaro写道：

[Alfio Lazzaro]

From what I can see, you are compiling cublas.cu via

nvcc -c --std=c++11 -O3 --cuda -keep --dryrun -w -g  -I'/gpfs/users_home/20116/cp2k-7.1-icc17u4/cp2k-7.1/exts/dbcsr/src' /gpfs/users_home/20116/cp2k-7.1-icc17u4/cp2k-7.1/exts/dbcsr/src/acc/cublaswrap/cublas.cu

I'm not a nvcc expert, but from the help I found that:

--cuda  (-cuda)                         

        Compile all .cu input files to .cu.cpp.ii output.

Is there any reason why you added this flag? I think it avoids generating the canonical .o file, hence the error you get...

Could you try without it?

Alfio

[egbertchang]

thank so much for your great advise, the mistake has gone 

however after installation, i Run the cp2k.popt , it shows:

ABORT in dbcsr_lib.F:187 DBCSR compiled w/ threading support while libcusmm com

 piled w/o threading support.

should i compile with CXXOMPFLAGS? 

在 2020年4月13日星期一 UTC+8上午3:42:56，Alfio Lazzaro写道：

[Alfio Lazzaro]

Ah, sorry, I thought you ere doing POPT (i.e. no OpenMP) version...

Well, no just that:

1. Need to add OpenMP flag to CXXFLAGS (yes, you are right)

2. Need to add OpenMP flag for NVFLAGS host compilation, i.e. add -Xcompiler='-fopenmp'

Alfio

[egbertchang]

it's working 

i really appreciate for your help.

thank you so much!

在 2020年4月14日星期二 UTC+8上午3:23:28，Alfio Lazzaro写道：