k-Wave

1. bastg
   

   Member
   Joined: Jun '19
   Posts: 7

   Hi -- I am trying to compile the CUDA code from scratch since the binaries require the GLIBC_2.27 library which I don't have and is hard to grab on my system.

   The compilation goes well but fails at linking. This is a bit strange... All the *.o files are where they are supposed to be, but I get a ton of "undefined reference" errors. I am using CentOS Linux release 7.7.1908.

   Here is my Makefile:

   ################################################################################
   #         Set following flags based on your compiler and library paths         #
   ################################################################################
   
   # Select compiler
   # GNU is default due to Intel 2018's compatibility issues with Ubuntu 18.04
    COMPILER = GNU
   #COMPILER = Intel
   
   # SEMI static lining is default since it is expected the binary will run on the
   # same system.
   # Everything will be linked statically, may not work on all GPUs
   #LINKING = STATIC
   # Everything will be linked dynamically
   #LINKING = DYNAMIC
   # Everything but CUDA will be linked statically
   LINKING = SEMI
   
   # Set up paths: If using modules, the paths are set up automatically,
   #               otherwise, set paths manually
   CUDA_DIR = /usr/pubsw/packages/CUDA/10.0
   HDF5_DIR = /autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5
   ZLIB_DIR = /autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/zlib-1.2.11
   SZIP_DIR = /autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/szip-2.1.1
   
   # Select CPU architecture (what instruction set to be used).
   # The native architecture will compile and optimize the code for the underlying
   # processor.
   
    CPU_ARCH = native
   #CPU_ARCH = AVX
   #CPU_ARCH = AVX2
   #CPU_ARCH = AVX512
   ############################## Common flags ###################################
   # Git hash of release 1.3
   GIT_HASH       = -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\"
   
   # Replace tabs by spaces
   .RECIPEPREFIX += 
   
   # What CUDA GPU architectures to include in the binary
   CUDA_ARCH = --generate-code arch=compute_30,code=sm_30 \
               --generate-code arch=compute_32,code=sm_32 \
               --generate-code arch=compute_35,code=sm_35 \
               --generate-code arch=compute_37,code=sm_37 \
               --generate-code arch=compute_50,code=sm_50 \
               --generate-code arch=compute_52,code=sm_52 \
               --generate-code arch=compute_53,code=sm_53 \
               --generate-code arch=compute_60,code=sm_60 \
               --generate-code arch=compute_61,code=sm_61 \
               --generate-code arch=compute_62,code=sm_62 \
               --generate-code arch=compute_70,code=sm_70 \
               --generate-code arch=compute_72,code=sm_72 \
               --generate-code arch=compute_75,code=sm_75
   
   # What libraries to link and how
   ifeq ($(LINKING), STATIC)
     LDLIBS = $(HDF5_DIR)/lib/libhdf5_hl.a         \
              $(HDF5_DIR)/lib/libhdf5.a            \
              $(CUDA_DIR)/lib64/libcufft_static.a  \
              $(CUDA_DIR)/lib64/libculibos.a       \
              $(CUDA_DIR)/lib64/libcudart_static.a \
              $(ZLIB_DIR)/lib/libz.a               \
              $(SZIP_DIR)/lib/libsz.a              \
              -ldl
   
   else ifeq ($(LINKING), DYNAMIC)
     LDLIBS = -lhdf5 -lhdf5_hl -lz -lcufft
   
   else ifeq ($(LINKING), SEMI)
     LDLIBS = $(HDF5_DIR)/lib/libhdf5_hl.a \
              $(HDF5_DIR)/lib/libhdf5.a    \
              $(ZLIB_DIR)/lib/libz.a       \
              $(SZIP_DIR)/lib/libsz.a      \
              -lcufft                      \
              -ldl
   endif
   
   ############################## NVCC + GNU g++ ##################################
   ifeq ($(COMPILER), GNU)
     # C++ compiler for CUDA
     CXX       = /usr/pubsw/packages/CUDA/10.0/bin/nvcc
   
     # C++ standard
     CPP_STD   = -std=c++11
   
     # Enable OpenMP
     OPENMP    = -fopenmp
   
     # Set CPU architecture
     # Sandy Bridge, Ivy Bridge
     ifeq ($(CPU_ARCH), AVX)
       CPU_FLAGS = -m64 -mavx
   
     # Haswell, Broadwell
     else ifeq ($(CPU_ARCH), AVX2)
       CPU_FLAGS = -m64 -mavx2
   
     # Skylake-X, Ice Lake, Cannon Lake
     else ifeq ($(CPU_ARCH), AVX512)
       CPU_FLAGS = -m64 -mavx512f
   
     # Maximum performance for this CPU
     else
       CPU_FLAGS = -m64 -march=native -mtune=native
     endif
   
     # Use maximum optimization
     CPU_OPT   = -O3 -ffast-math -fassociative-math
     # Use maximum optimization
     GPU_OPT   = -O3
   
     # CPU Debug flags
     CPU_DEBUG =
     # Debug flags
     GPU_DEBUG =
     # Profile flags
     PROFILE   =
     # C++ warning flags
     WARNING   = -Wall
    # Add include directories
     INCLUDES  = -I$(HDF5_DIR)/include -I.
     # Add library directories
     LIB_PATHS = -L$(HDF5_DIR)/lib -L$(CUDA_DIR)/lib64
   
     # Set compiler flags and header files directories
     CXXFLAGS  = -Xcompiler="$(CPU_FLAGS) $(CPU_OPT) $(OPENMP)  \
                             $(CPU_DEBUG) $(PROFILE) $(WARNING)"\
                 $(GPU_OPT) $(CPP_STD) $(GPU_DEBUG) \
                 $(GIT_HASH)                        \
                 $(INCLUDES)                        \
                 --device-c --restrict
   
     # Set linker flags and library files directories
     LDFLAGS   = -Xcompiler="$(OPENMP)" \
                 -Xlinker="-rpath,$(HDF5_DIR)/lib:$(CUDA_DIR)/lib64" \
                 -std=c++11             \
                  $(LIB_PATHS)
   endif
   
   ############################ NVCC + Intel icpc #################################
   ifeq ($(COMPILER), Intel)
     # C++ compiler for CUDA
     CXX       = /usr/pubsw/packages/CUDA/10.0/bin/nvcc
   
     # C++ standard
     CPP_STD   = -std=c++11
   
     # Enable OpenMP
     OPENMP    = -qopenmp
   
     # Set CPU architecture
     # Sandy Bridge, Ivy Bridge
     ifeq ($(CPU_ARCH), AVX)
       CPU_FLAGS = -m64 -xAVX
   
     # Haswell, Broadwell
     else ifeq ($(CPU_ARCH), AVX2)
       CPU_FLAGS = -m64 -xCORE-AVX2
   
     # Skylake-X, Ice Lake, Cannon Lake
     else ifeq ($(CPU_ARCH), AVX512)
       CPU_FLAGS = -m64 -xCORE-AVX512
     # Maximum performance for this CPU
     else
       CPU_FLAGS = -m64 -xhost
     endif
   
     # Use maximum optimization
     CPU_OPT   = -Ofast
     # Use maximum optimization
     GPU_OPT   = -O3
   
     # CPU Debug flags
     CPU_DEBUG =
     # Debug flags
     GPU_DEBUG =
     # Profile flags
     PROFILE   =
     # C++ warning flags
     WARNING   = -Wall
   
     # Add include directories
     INCLUDES  = -I$(HDF5_DIR)/include -I.
     # Add library directories
     LIB_PATHS = -L$(HDF5_DIR)/lib -L$(CUDA_DIR)/lib64
   
     # Set compiler flags and header files directories
     CXXFLAGS  = -Xcompiler="$(CPU_FLAGS) $(CPU_OPT) $(OPENMP)   \
                             $(CPU_DEBUG) $(PROFILE) $(WARNING)" \
                 $(GPU_OPT) $(CPP_STD) $(GPU_DEBUG) \
                 $(GIT_HASH)                        \
                 $(INCLUDES)                        \
                 --device-c --restrict -ccbin=icpc
   
     # Set linker flags and library files directories
     ifneq ($(LINKING), DYNAMIC)
       LDFLAGS = -Xcompiler="$(OPENMP) -static-intel -qopenmp-link=static"
     else
       LDFLAGS = -Xcompiler="$(OPENMP)"
     endif
   
     LDFLAGS  += -std=c++11 -ccbin=icpc \
                 -Xlinker="-rpath,$(HDF5_DIR)/lib:$(CUDA_DIR)/lib64" \
                 $(LIB_PATHS)
   endif
   ################################### Build ######################################
   # Target binary name
   TARGET       = kspaceFirstOrder-CUDA
   
   # Units to be compiled
   DEPENDENCIES = main.o                                   \
                  Containers/MatrixContainer.o             \
                  Containers/CudaMatrixContainer.o         \
                  Containers/OutputStreamContainer.o       \
                  Hdf5/Hdf5File.o                          \
                  Hdf5/Hdf5FileHeader.o                    \
                  KSpaceSolver/KSpaceFirstOrderSolver.o    \
                  KSpaceSolver/SolverCudaKernels.o         \
                  Logger/Logger.o                          \
                  MatrixClasses/BaseFloatMatrix.o          \
                  MatrixClasses/BaseIndexMatrix.o          \
                  MatrixClasses/CufftComplexMatrix.o       \
                  MatrixClasses/ComplexMatrix.o            \
                  MatrixClasses/IndexMatrix.o              \
                  MatrixClasses/RealMatrix.o               \
                  MatrixClasses/TransposeCudaKernels.o     \
                  OutputStreams/BaseOutputStream.o         \
                  OutputStreams/IndexOutputStream.o        \
                  OutputStreams/CuboidOutputStream.o       \
                  OutputStreams/WholeDomainOutputStream.o  \
                  OutputStreams/OutputStreamsCudaKernels.o \
                  Parameters/CommandLineParameters.o       \
                  Parameters/Parameters.o                  \
                  Parameters/CudaParameters.o              \
                  Parameters/CudaDeviceConstants.o
   
   # Build target
   all: $(TARGET)
   
   # Link target
   $(TARGET): $(DEPENDENCIES)
     $(CXX) $(LDFLAGS) $(DEPENDENCIES) $(LDLIBS) -o $@
   
   # Compile CPU units
   %.o: %.cpp
     $(CXX) $(CXXFLAGS) -o $@ -c $<
   
   # Compile CUDA units
   %.o: %.cu
     $(CXX) $(CXXFLAGS) $(CUDA_ARCH) -o $@ -c $<
   
   # Clean repository
   .PHONY: clean
   clean:
     rm -f $(DEPENDENCIES) $(TARGET)

   Below are the first linking errors:
   

   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o main.o -c main.cpp
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o Containers/MatrixContainer.o -c Containers/MatrixContainer.cpp
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict --generate-code arch=compute_30,code=sm_30 --generate-code arch=compute_32,code=sm_32 --generate-code arch=compute_35,code=sm_35 --generate-code arch=compute_37,code=sm_37 --generate-code arch=compute_50,code=sm_50 --generate-code arch=compute_52,code=sm_52 --generate-code arch=compute_53,code=sm_53 --generate-code arch=compute_60,code=sm_60 --generate-code arch=compute_61,code=sm_61 --generate-code arch=compute_62,code=sm_62 --generate-code arch=compute_70,code=sm_70 --generate-code arch=compute_72,code=sm_72 --generate-code arch=compute_75,code=sm_75 -o Containers/CudaMatrixContainer.o -c Containers/CudaMatrixContainer.cu
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o Containers/OutputStreamContainer.o -c Containers/OutputStreamContainer.cpp
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o Hdf5/Hdf5File.o -c Hdf5/Hdf5File.cpp
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o Hdf5/Hdf5FileHeader.o -c Hdf5/Hdf5FileHeader.cpp
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict -o KSpaceSolver/KSpaceFirstOrderSolver.o -c KSpaceSolver/KSpaceFirstOrderSolver.cpp
   KSpaceSolver/KSpaceFirstOrderSolver.cpp:2116:0: warning: ignoring #pragma omp simd [-Wunknown-pragmas]
            #pragma omp simd
    ^
   /usr/pubsw/packages/CUDA/10.0/bin/nvcc -Xcompiler="-m64 -march=native -mtune=native -O3 -ffast-math -fassociative-math -fopenmp   -Wall" -O3 -std=c++11  -D__KWAVE_GIT_HASH__=\"468dc31c2842a7df5f2a07c3a13c16c9b0b2b770\" -I/autofs/space/guerin/USneuromod/KWAVE_1.2.1/src/hdf5-1.12.0/hdf5/include -I. --device-c --restrict --generate-code arch=compute_30,code=sm_30 --generate-code arch=compute_32,code=sm_32 --generate-code arch=compute_35,code=sm_35 --generate-code arch=compute_37,code=sm_37 --generate-code arch=compute_50,code=sm_50 --generate-code arch=compute_52,code=sm_52 --generate-code arch=compute_53,code=sm_53 --generate-code arch=compute_60,code=sm_60 --generate-code arch=compute_61,code=sm_61 --generate-code arch=compute_62,code=sm_62 --generate-code arch=compute_70,code=sm_70 --generate-code arch=compute_72,code=sm_72 --generate-code arch=compute_75,code=sm_75 -o KSpaceSolver/SolverCudaKernels.o -c KSpaceSolver/SolverCudaKernels.cu

   Many thanks!

   Posted 4 years ago #

2. Jiri Jaros
   

   Developer
   Joined: Feb '12
   Posts: 115

   Hi bastg,
   You have a very old compiler (that's also the reason why the binary doesn't work). Since it doesn't know #pragma omp simd, it must be something ancient like g++ 4.9 or so.

   please type
   g++ --version

   If you don't have at least g++7.3, please install a newer version.

   https://linuxhostsupport.com/blog/how-to-install-gcc-on-centos-7/

   Best wishes
   Jiri

   Posted 4 years ago #

3. bastg
   

   Member
   Joined: Jun '19
   Posts: 7

   I updated the compiler to 7.3 and that indeed solved the problem... Thanks!

   Posted 4 years ago #

4. Hammerben1
   

   Member
   Joined: Jan '18
   Posts: 15

   I am having similar problem that CUDA program complains about not having GLIBC-2.27
   So the CUDA program would work if I compiled it under low version GLIBC? (like GLIBC 2.17?)

   Posted 4 years ago #

RSS feed for this topic

Reply

You must log in to post.