====== Compilers and Libraries ====== ===== Compiler ===== Following compilers are available on the accelerator server. \\ For more details, see each manual. \\ ^Compiler ^Version ^Notes ^ |Intel Compiler Fortran/C/C++ |2022.3.1\\ 2021.5.0\\ 19.1.3.304\\ 19.1.0.166\\ 19.0.2.187\\ 18.0.3.222\\ 17.0.4.196\\ |default: 17.0.4.196\\ 2021.5.0 is only available when running applications, not compiling. | |PGI Compiler Fortran/C/C++ |20.4\\ 19.10\\ 19.1\\ 18.10\\ 18.5 |default: 19.1 | |NVIDIA HPC Compiler |23.1\\ 22.11\\ 22.5\\ 21.11\\ 20.9\\ |default: 20.9 | |GNU Compiler |4.9.1\\ 6.1.0\\ 8.5.0\\ 11.3.0 |default: 6.1.0 | |nvcc Compiler (CUDA Toolkit) |11.6.2\\ 10.2.89\\ 10.1.243\\ 9.2.148\\ 9.0.176\\ 8.0.44 |default: 9.0.176 | \\ ------ ==== Intel Compiler ==== \\ **Setting of program environment**\\ Intel compiler is the default one on the accelerator server.\\ If you want to change the compiler version, execute the following command.\\ $ module avail intel intel/17.0.4(default) intel/18.0.3 intel/19.0.2 intel/19.1.0 intel/19.1.3 intel/21.5.0 intel/oneapi/22.3.1 $ module switch intel/17.0.4 intel/18.0.3 intel/21.5.0 is only available when running applications, not compiling.\\ \\ **How to compile**\\ \\ **Command**\\ ・serial ^Language ^Command ^Execution type ^ |Fortran |ifort |ifort [options] files | |C |icc |icc [options] files | |C++ |icpc |icpc [options] files | \\ ・MPI ^Language ^Command ^Execution type ^ |Fortran |mpiifort |mpiifort [options] files | |C |mpiicc |mpiicc [options] files | |C++ |mpiicpc |mpiicpc [options] files | \\ **Options**\\ \\ ・Optimization options and others ^Option ^Description ^ |-o outfile |Set output file. If ‘-o’ option is not specified, the default output file is a.out. | |-l library_name |Specify libraries to be linked. | |-L library_path |Specify paths to be searched for the libraries. | |-O0|-O1|-O2|-O3 |Set optimizing options. -O2 is default. | |-fast |Set the whole program execution speed maximum. \\ The following options are set automatically.\\ -ipo, -O3, -no-prec-div, -static, -fp-model fast=2, -xHost | |-parallel |Compile the input source file enabling auto threading. | |-openmp |Compile the input source file enabling OpenMP. | |-xcore-avx512 |Specify Intel AVX-512 instruction as target. \\ (Recommended) | \\ ・Options for Fortran ^Option ^Description ^ |-free|-fixed |Set the format of source files. | \\ ・Debugging Options ^Option ^Description ^ |-g |Output debugging information. | |-traceback |Output traceback when an error occurs. | |-fpe[0-3] |Trap exception handling. (-fpe0 is most detailed) | \\ **Example**\\ \\ ・Make the execution module 'hello.out' from Fortran source file 'hello.f' in fixed form. $ ifort -xcore-avx512 -fixed -o hello.out hello.f \\ ・Make the execution module ’hello.out’ from the Fortran source file ‘hello.f90’ in free form. $ ifort -xcore-avx512 -free -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f' in fixed form. $ ifort -xcore-avx512 -fixed -parallel -o hello.out hello.f \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f90' in free form. $ ifort -xcore-avx512 -free -parallel -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' from C source file 'hello.c'. $ icc -xcore-avx512 -o hello.out hello.c \\ ・Make the execution module 'hello.out' auto threaded from C source file 'hello.c'. $ icc -xcore-avx512 -parallel -o hello.out hello.c \\ ・Make the execution module 'hello.out' from the C++ source file 'hello.cpp'. $ icpc -xcore-avx512 -o hello.out hello.cpp \\ ・Make the execution module 'hello.out' auto threaded from the C++ source file. $ icpc -xcore-avx512 -parallel -o hello.out hello.cpp \\ ------ ==== PGI Compiler ==== \\ **Setting of program environment**\\ When the program environment is changed, execute the following commands. \\ $ module avail PrgEnv-pgi PrgEnv-pgi/18.5 PrgEnv-pgi/18.10 PrgEnv-pgi/19.1(default) PrgEnv-pgi/19.10 PrgEnv-pgi/20.4 $ module switch intel PrgEnv-pgi \\ **How to compile**\\ \\ **Command**\\ ・serial ^Language ^Command ^Execution type ^ |Fortran |pgf90 |pgf90 [options] files | |C |pgcc |pgcc [options] files | |C++ |pgc++ |pgc++ [options] files | \\ ・MPI ^Language ^Command ^Execution type ^ |Fortran |mpif90 |mpif90 [options] files | |C |mpicc |mpicc [options] files | |C++ |mpic++ |mpic++ [options] files | \\ **Options**\\ \\ ・Optimization options and others ^Option ^Description ^ |-o outfile |Set output file. If ‘-o’ option is not specified, the default output file is a.out. | |-l library_name |Specify libraries to be linked. | |-L library_path |Specify paths to be searched for the libraries. | |-O0|-O1|-O2|-O3|-O4 |Set optimizing options. -O2 is default. | |-fast |Enable general optimization flag. | |-Mconcur |Compile the input source file enabling auto threading. | |-mp |Compile the input source file enabling OpenMP. | \\ ・Option for Fortran ^Option ^Description ^ |-Mfree|-Mfixed |Set the format of source files. | \\ ・Debugging options ^Option ^Description ^ |-g|-gopt |Output debugging information. | \\ **Example**\\ \\ ・Make the execution module 'hello.out' from the Fortran source file 'hello.f' in fixed form. $ pgf90 -Mfixed -o hello.out hello.f \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f' in fixed form. $ pgf90 -Mfixed -Mconcur -o hello.out hello.f \\ ・Make the execution module 'hello.out' using OpenMP from the Fortran source file 'hello.f' in fixed form. $ pgf90 -mp -Mfixed -o hello.out hello.f \\ ・Make the execution module ’hello.out’ from the Fortran source file ‘hello.f90’ in free form. $ pgf90 -Mfree -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f90' in free form. $ pgf90 -Mfree -Mconcur -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' using OpenMP from the Fortran source file 'hello.f90' in free form. $ pgf90 -mp -Mfree -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' from C source file 'hello.c'. $ pgcc -o hello.out hello.c \\ ・Make the execution module 'hello.out' auto threaded from C source file 'hello.c'. $ pgcc -Mconcur -o hello.out hello.c \\ ・Make the execution module 'hello.out' using OpenMP from C source file 'hello.c'. $ pgcc -mp -o hello.out hello.c \\ ・Make the execution module 'hello.out' from the C++ source file 'hello.cpp'. $ pgc++ -o hello.out hello.cpp \\ ・Make the execution module 'hello.out' auto threaded from the C++ source file 'hello.cpp' $ pgc++ -Mconcur -o hello.out hello.cpp \\ ・Make the execution module 'hello.out' using OpenMP from the C++ source file 'hello.cpp'. $ pgc++ -mp -o hello.out hello.cpp \\ ------ ==== NVIDIA HPC Compiler ==== \\ **Setting of program environment**\\ The following modules are available.\\ ^Module ^Description ^ |nvhpc |Adds environment variable settings for the NVIDIA HPC Compilers, CUDA libraries, and additional libraries such as MPI, NCCL, and NVSHMEM. | |nvhpc-nompi |Adds environment variable settings for the NVIDIA HPC Compilers, CUDA libraries, and additional libraries such as NCCL and NVSHMEM. | |nvhpc-byo-compilers |Adds environment variable settings for the CUDA libraries and additional libraries such as NCCL and NVSHMEM. | |nvhpc-hpcx |Adds environment variable settings for the such as MPI | \\ When the program environment is changed, execute the following commands. $ module avail nvhpc nvhpc/20.9(default) nvhpc-byo-compiler/20.9(default) nvhpc-nompi/20.9(default) nvhpc/21.11 nvhpc-byo-compiler/21.11 nvhpc-nompi/21.11 nvhpc/22.5 nvhpc-byo-compiler/22.5 nvhpc-nompi/22.5 nvhpc/22.11 nvhpc-byo-compiler/22.11 nvhpc-nompi/22.11 nvhpc/23.1 nvhpc-byo-compiler/23.1 nvhpc-nompi/23.1 nvhpc-hpcx/23.1 $ module switch intel nvhpc \\ **How to compile**\\ \\ **Command**\\ ・serial ^Language ^Command ^Execution type ^ |Fortran |nvfortran |nvfortran [options] files | |C |nvc |nvc [options] files | |C++ |nvc++ |nvc++ [options] files | \\ ・MPI ^Language ^Command ^Execution type ^ |Fortran |mpif90 |mpif90 [options] files | |C |mpicc |mpicc [options] files | |C++ |mpic++ |mpic++ [options] files | \\ **Options**\\ ・Optimization options and others ^Option ^Description ^ |-o outfile |Set output file. If ‘-o’ option is not specified, the default output file is a.out. | |-llibrary_name |Specify libraries to be linked. | |-Llibrary_path |Specify paths to be searched for the libraries. | |-O0|-O1|-O2|-O3|-O4 |Set optimizing options. -O2 is default. | |-fast |Enable general optimization flag. | |-Mconcur |Compile the input source file enabling auto threading. | |-mp |Compile the input source file enabling OpenMP. | ・Option for Fortran ^Option ^Description ^ |-Mfree|-Mfixed |Set the format of source files. | ・Debugging options ^Option ^Description ^ |-g | -gopt |Output debugging information. | \\ **Example**\\ \\ ・Make the execution module 'hello.out' from the Fortran source file 'hello.f' in fixed form. $ nvfortran -Mfixed -o hello.out hello.f \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f' in fixed form $ nvfortran -Mfixed -Mconcur -o hello.out hello.f \\ ・Make the execution module 'hello.out' using OpenMP from the Fortran source file 'hello.f' in fixed form. $ nvfortran -mp -Mfixed -o hello.out hello.f \\ ・Make the execution module ’hello.out’ from the Fortran source file ‘hello.f90’ in free form. $ nvfortran -Mfree -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' auto threaded from the Fortran source file 'hello.f90' in free form. $ nvfortran -Mfree -Mconcur -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' using OpenMP from the Fortran source file 'hello.f90' in free form. $ nvfortran -mp -Mfree -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' from C source file 'hello.c'. $ nvc -o hello.out hello.c \\ ・Make the execution module 'hello.out' auto threaded from C source file 'hello.c'. $ nvc -Mconcur -o hello.out hello.c \\ ・Make the execution module 'hello.out' using OpenMP from C source file 'hello.c'. $ nvc -mp -o hello.out hello.c \\ ・Make the execution module 'hello.out' from the C++ source file 'hello.cpp'. $ nvc++ -o hello.out hello.cpp \\ ・Make the execution module 'hello.out' auto threaded from the C++ source file 'hello.cpp'. $ nvc++ -Mconcur -o hello.out hello.cpp \\ ・Make the execution module 'hello.out' using OpenMP from the C++ source file 'hello.cpp'. $ nvc++ -mp -o hello.out hello.cpp \\ ------ ==== GNU Compiler ==== \\ **Setting of program environment**\\ If you do not make any special settings, the GNU compiler version is OS default 4.8.5.\\ If you use another version, please execute the following command. $ module avail gcc gcc/4.9.1 gcc/6.1.0(default) gcc/8.5.0 gcc/11.3.0 $ module load gcc \\ **How to compile**\\ \\ **Command**\\ ・serial ^Language ^Command ^Execution type ^ |Fortran |gfortran |gfortran [options] files | |C |gcc |gcc [options] files | |C++ |g++ |g++ [options] files | \\ ・MPI ^Language ^Command ^Execution type ^ |Fortran |mpif90 |mpif90 [options] files | |C |mpicc |mpicc [options] files | |C++ |mpicxx |mpicxx [options] files | \\ **Options**\\ \\ ・Optimization options and others ^Option ^Description ^ |-o outfile |Set output file. If ‘-o’ option is not specified, the default output file is a.out. | |-l library_name |Specify libraries to be linked. | |-L library_path |Specify paths to be searched for the libraries. | |-O0|-O1|-O2|-O3|-O4 |Set optimizing options. -O2 is default. | |-fopenmp |Compile the input source file enabling OpenMP. | \\ ・Options for Fortran ^Option ^Description ^ |-ffree-form|-ffixed-form |Set the format of source files. | \\ ・Debugging options ^Option ^Description ^ |-g |Output debugging information. | |-g0|-g1|-g2|-g3 |Manage debugging information.(-g2 = -g) | \\ **Example**\\ \\ ・Make the execution module 'hello.out' from Fortran source file 'hello.f' in fixed form. $ gfortran -ffixed-form -o hello.out hello.f \\ ・Make the execution module ’hello.out’ from the Fortran source file ‘hello.f90’ in free form. $ gfortran -ffree-form -o hello.out hello.f90 \\ ・Make the execution module 'hello.out' from C source file 'hello.c'. $ gcc -o hello.out hello.c \\ ・Make the execution module 'hello.out' from the C++ source file 'hello.cpp'. $ g++ -o hello.out hello.cpp \\ ------ ====nvcc Compiler ==== \\ **Setting of program environment**\\ The CUDA is available as default.\\ If you want to change the version, execute the following command. \\ $ module switch cudatoolkit/9.0.176 cudatoolkit/10.1.243 \\ ・Because the Intel compiler is set up as default, you use Intel compiler for backend compiler. \\ \\ ・Using PGI compiler for backend compiler, execute the following commands to set up PGI compiler. \\ $ module switch intel PrgEnv-pgi \\ **How to compile**\\ \\ **Options** \\ ・Backend compiler, optimization options and others ^Option ^Description ^ |-ccbin Compiler |Specify backend compiler.\\ Intel compiler: icpc\\ PGI compiler: pgc++ | |-O0|1... |Set optimizing options.　\\ (pass to the backend compiler) | |-Xcompiler options |Set compiler options except for optimizing options.　\\ (pass to the backend compiler) | |-gencode options |Specify CUDA’s version of generating code. | |--machine {32|64}\\ (-m) |Specify the bits. | |-I include_path |Specify include search paths. | |-L library_path |Specify paths to be searched for the libraries. | |-l library_name |Specify libraries to be linked. | |--help (-h) |Display list and description of available options. | |--version (-V) |Display version information. | \\ **Example**\\ \\ ・In case of specifying Intel compiler (icc) for backend compiler, $ nvcc –ccbin icpc -m64 -gencode arch=compute_70,code=compute_70 -o simple simple.cpp \\ ・In case of specifying PGI compiler (pgc++) for backend compiler, $ nvcc -ccbin pgc++ -m64 -gencode arch=compute_70,code=compute_70 -o simple simple.cpp \\ ===== Libraries ===== Following libraries are available for compiling and linking on the accelerator server. \\ For more details, see each manual. \\ \\ ^Library ^Version ^Compiler ^Notes ^ |Intel MKL \\ (Intel Math Kernel Library) |19.1.3.304\\ 19.1.0.166\\ 19.0.2.187\\ 18.0.3.222\\ 17.0.4.196 |Intel compiler | | |cuBLAS |9.0\\ 8.0 |Intel compiler\\ PGI compiler | | |cuDNN |7.6.5 for CUDA10.2\\ 7.6.3 for CUDA10.1\\ 7.6.5 for CUDA9.2\\ 7.6.3 for CUDA9.0 |Intel compiler\\ PGI compiler | | \\ ------ ==== Intel MKL ==== Intel MKL (Intel Math Kernel Library) contains BLAS, LAPACK, SparseBLAS, PARDISO, Iterative Sparse Solver, FFT, random number generation and so on.\\ \\ **Setting program environment**\\ \\ Intel MKL is default.\\ Change the compiler version as follows.\\ $ module switch intel/17.0.4 intel/18.0.3 \\ **Example**\\ \\ ・Make the execution module 'hello.out' from the Fortran source file 'hello.f90' using BLAS in fixed form. $ ifort -mkl -o hello.out -fixed hello.f90 \\ ・Make the execution module 'hello.out' from C source file 'hello.c' using BLAS. $ icc -mkl -o hello.out hello.c \\ ------ ==== cuBLAS ==== The cuBLAS is BLAS library supporting the CUDA.\\ \\ **Setting of program environment**\\ \\ ・The CUDA including the cuBLAS is available as default.\\ If you want to change the version, execute the following command.\\ $ module switch cudatoolkit/9.0.176 cudatoolkit/8.0.44 \\ ・Because the Intel compiler is set up as default, you use Intel compiler for backend compiler.\\ \\ ・Using PGI compiler for backend compiler, execute the following commands to set up PGI compiler.\\ $ module switch intel PrgEnv-pgi \\ **Example**\\ \\ ・In case of specifying Intel compiler (icc) for backend and linking cuBLAS libraries,\\ $ nvcc -ccbin icpc -I../../common/inc -m64 -gencode arch=compute_70,code=compute_70 -o simpleCUBLAS simpleCUBLAS.cpp -lcublas \\ ・In case of specifying PGI compiler (pgc++) for backend and linking cuBLAS libraries,\\ $ nvcc -ccbin pgc++ -I../../common/inc -m64 -gencode arch=compute_70,code=compute_70 -o simpleCUBLAS simpleCUBLAS.cpp -lcublas \\ ------ ==== cuDNN ==== \\ **Setting of program environment**\\ \\ ・The CUDA including the cuDNN is available as default.\\ If you want to change the version, execute the following command.\\ $ module switch cudatoolkit/9.0.176 cudatoolkit/10.1.243 \\ ・Because the Intel compiler is set up as default, you use Intel compiler for backend compiler.\\ \\ ・Using PGI compiler for backend compiler, execute the following commands to set up PGI compiler.\\ $ module switch intel PrgEnv-pgi \\ **Example**\\ \\ ・In case of specifying Intel compiler (icc) for backend and linking cuDNN libraries, $ nvcc -ccbin icpc -I../../common/inc -m64 -gencode arch=compute_70,code=compute_70 -o simpleCUBLAS simpleCUBLAS.cpp -lcudnn \\ ・In case of specifying PGI compiler (pgc++) for backend and linking cuDNN libraries, $ nvcc -ccbin pgc++ -I../../common/inc -m64 -gencode arch=compute_70,code=compute_70 -o simpleCUBLAS simpleCUBLAS.cpp -lcudnn \\ ~~NOCACHE~~