k-Wave User Forum » Topic: linking error, CUDA code

k-Wave User Forum » Topic: linking error, CUDA code http://www.k-wave.org/forum/topic/linking-error-cuda-code Support for the k-Wave MATLAB toolbox en-US Tue, 12 May 2026 23:27:45 +0000 http://bbpress.org/?v=1.0.2 <![CDATA[Search]]> q http://www.k-wave.org/forum/search.php Hammerben1 on "linking error, CUDA code" http://www.k-wave.org/forum/topic/linking-error-cuda-code#post-7626 Wed, 17 Jun 2020 23:02:33 +0000 Hammerben1 7626@http://www.k-wave.org/forum/ 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?) bastg on "linking error, CUDA code" http://www.k-wave.org/forum/topic/linking-error-cuda-code#post-7505 Mon, 18 May 2020 20:50:19 +0000 bastg 7505@http://www.k-wave.org/forum/ I updated the compiler to 7.3 and that indeed solved the problem... Thanks! Jiri Jaros on "linking error, CUDA code" http://www.k-wave.org/forum/topic/linking-error-cuda-code#post-7297 Tue, 17 Mar 2020 11:44:33 +0000 Jiri Jaros 7297@http://www.k-wave.org/forum/ 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 <code>g++ --version</code> If you don't have at least g++7.3, please install a newer version. <a href="https://linuxhostsupport.com/blog/how-to-install-gcc-on-centos-7/" rel="nofollow">https://linuxhostsupport.com/blog/how-to-install-gcc-on-centos-7/</a> Best wishes Jiri bastg on "linking error, CUDA code" http://www.k-wave.org/forum/topic/linking-error-cuda-code#post-7291 Thu, 12 Mar 2020 23:36:00 +0000 bastg 7291@http://www.k-wave.org/forum/ 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: <pre><code>################################################################################ # 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)</code></pre> Below are the first linking errors: <pre><code>/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</code></pre> Many thanks!