Loading containers/docker/Dockerfile +20 −2 Original line number Diff line number Diff line Loading @@ -22,6 +22,14 @@ RUN DEBIAN_FRONTEND=noninteractive apt -y install intel-oneapi-compiler-fortran RUN DEBIAN_FRONTEND=noninteractive apt -y install liblapacke-dev liblapacke64-dev libopenblas-dev libopenblas-openmp-dev libopenblas64-dev libopenblas64-openmp-dev # install MPI stack RUN DEBIAN_FRONTEND=noninteractive apt -y install mpi-default-dev mpi-default-bin # install nvidia stuff COPY --chown=root:root containers/docker/dockerstuff/debian-nonfree/debian.sources /etc/apt/sources.list.d/ RUN apt update RUN DEBIAN_FRONTEND=noninteractive apt -y install libquadmath0 libcublaslt11 libgcc-s1 libgomp1 libcudart11.0 libcusparse11 libcublas11 nvidia-cuda-dev # install magma COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/lib/* /usr/lib/x86_64-linux-gnu/ COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/include/* /usr/include/ COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/lib/pkgconfig/* /usr/lib/x86_64-linux-gnu/pkgconfig/ # install packages needed to run python scripts for checks RUN DEBIAN_FRONTEND=noninteractive apt -y install python3 python-is-python3 python3-regex # install packages needed to run doxygen to create html docs Loading @@ -46,7 +54,13 @@ ADD src /root/np-tmcode/src ADD doc /root/np-tmcode/doc ADD build /root/np-tmcode/build ADD test_data /root/np-tmcode/test_data RUN cd np-tmcode/src && make wipe && make -j && cd ../doc/src && doxygen config.dox && cd ../build/latex && make -j RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_magma_mpi RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_magma_serial RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_lapack_mpi && cd ../build/cluster && ln -s np_cluster_lapack_mpi np_cluster RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_lapack_serial RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_legacy_mpi RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_legacy_serial RUN cd np-tmcode/doc/src && doxygen config.dox && cd ../build/latex && make -j # create the container on which the np-tmcode is installed, restarting from Loading @@ -55,8 +69,12 @@ FROM debian:bookworm-slim AS np-tmcode-run-minimal WORKDIR /root # install the strictly needed runtime libraries needed to run the executables # and the python check scripts RUN DEBIAN_FRONTEND=noninteractive apt update && DEBIAN_FRONTEND=noninteractive apt upgrade && DEBIAN_FRONTEND=noninteractive apt -y install libgfortran5 libgcc-s1 libhdf5-103-1 libstdc++6 libssl3 libcurl4 libsz2 zlib1g libnghttp2-14 libidn2-0 librtmp1 libssh2-1 libpsl5 libgssapi-krb5-2 libldap-2.5-0 libzstd1 libbrotli1 libaec0 libunistring2 libgmp10 libkrb5-3 libk5crypto3 libcom-err2 libkrb5support0 libsasl2-2 libp11-kit0 libtasn1-6 libkeyutils1 libffi8 liblapacke64 libopenblas64-0-openmp python3 python-is-python3 python3-regex hdf5-tools mpi-default-bin && rm -rf /var/lib/apt/lists/* COPY --chown=root:root containers/docker/dockerstuff/debian-nonfree/debian.sources /etc/apt/sources.list.d/ RUN apt update RUN DEBIAN_FRONTEND=noninteractive apt update && DEBIAN_FRONTEND=noninteractive apt upgrade && DEBIAN_FRONTEND=noninteractive apt -y install libgfortran5 libgcc-s1 libhdf5-103-1 libstdc++6 libssl3 libcurl4 libsz2 zlib1g libnghttp2-14 libidn2-0 librtmp1 libssh2-1 libpsl5 libgssapi-krb5-2 libldap-2.5-0 libzstd1 libbrotli1 libaec0 libunistring2 libgmp10 libkrb5-3 libk5crypto3 libcom-err2 libkrb5support0 libsasl2-2 libp11-kit0 libtasn1-6 libkeyutils1 libffi8 liblapacke64 libopenblas64-0-openmp python3 python-is-python3 python3-regex hdf5-tools mpi-default-bin libquadmath0 libcublaslt11 libgcc-s1 libgomp1 libcudart11.0 libcusparse11 libcublas11 && rm -rf /var/lib/apt/lists/* COPY --from=np-tmcode-run-dev /root /root COPY --from=np-tmcode-run-dev /usr/lib/x86_64-linux-gnu/libmagma.so /usr/lib/x86_64-linux-gnu/libmagma.so COPY --from=np-tmcode-run-dev /usr/lib/x86_64-linux-gnu/libmagma.so.2 /usr/lib/x86_64-linux-gnu/libmagma.so.2 # remove everything which is not needed to run the codes RUN cd /root/np-tmcode && find build -name "*.o" -exec rm -v \{\} \; && find build -name "*.gcno" -exec rm -v \{\} \; && cd src && rm -rvf cluster libnptm trapping include sphere Makefile make.inc README.md && cd .. && rm -rvf containers && cd doc && rm -rvf src && cd build/latex && rm -rvf *.tex *.out *.sty *.ind *.log *.toc *.ilg *.idx *.aux *.eps Makefile class*.pdf # move the installed software to /usr/local Loading containers/docker/dockerstuff/debian-nonfree/debian.sources 0 → 100644 +14 −0 Original line number Diff line number Diff line Types: deb # http://snapshot.debian.org/archive/debian/20240612T000000Z URIs: http://deb.debian.org/debian Suites: bookworm bookworm-updates Components: main non-free contrib non-free-firmware Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg Types: deb # http://snapshot.debian.org/archive/debian-security/20240612T000000Z URIs: http://deb.debian.org/debian-security Suites: bookworm-security Components: main non-free contrib non-free-firmware Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg containers/docker/dockerstuff/magma-compiled/usr/include/magma.h 0 → 100644 +39 −0 Original line number Diff line number Diff line /* -- MAGMA (version 2.7.2) -- Univ. of Tennessee, Knoxville Univ. of California, Berkeley Univ. of Colorado, Denver @date August 2023 */ #ifndef MAGMA_H #define MAGMA_H #ifdef MAGMA_NO_V1 #error "Since MAGMA_NO_V1 is defined, magma.h is invalid; use magma_v2.h" #endif // ============================================================================= // MAGMA configuration #include "magma_config.h" // magma v1 includes cublas.h by default, unless cublas_v2.h has already been included #ifndef CUBLAS_V2_H_ #if defined(MAGMA_HAVE_CUDA) #include <cublas.h> #endif #endif // Include the MAGMA v2 and v1 APIs, // then map names to the v1 API (e.g., magma_zgemm => magma_zgemm_v1). // Some functions (like setmatrix_async) are the same in v1 and v2, // so are provided by the v2 API. #include "magma_v2.h" #include "magmablas_v1.h" #include "magmablas_v1_map.h" #undef MAGMA_API #define MAGMA_API 1 #endif // MAGMA_H containers/docker/dockerstuff/magma-compiled/usr/include/magma.mod 0 → 100644 +106 KiB File added.No diff preview for this file type. View file containers/docker/dockerstuff/magma-compiled/usr/include/magma_auxiliary.h 0 → 100644 +407 −0 Original line number Diff line number Diff line /* -- MAGMA (version 2.7.2) -- Univ. of Tennessee, Knoxville Univ. of California, Berkeley Univ. of Colorado, Denver @date August 2023 */ #ifndef MAGMA_AUXILIARY_H #define MAGMA_AUXILIARY_H #include "magma_types.h" #include <math.h> // sqrtf #ifdef __cplusplus extern "C" { #endif // ============================================================================= // initialization magma_int_t magma_init( void ); magma_int_t magma_finalize( void ); #ifdef MAGMA_HAVE_OPENCL magma_int_t magma_init_opencl( cl_platform_id platform, cl_context context, magma_int_t setup_clBlas ); magma_int_t magma_finalize_opencl( magma_int_t finalize_clBlas ); #endif // ============================================================================= // version information void magma_version( magma_int_t* major, magma_int_t* minor, magma_int_t* micro ); void magma_print_environment(); // ============================================================================= // timing real_Double_t magma_wtime( void ); real_Double_t magma_sync_wtime( magma_queue_t queue ); // ============================================================================= // misc. functions // magma GPU-complex PCIe connection magma_int_t magma_buildconnection_mgpu( magma_int_t gnode[MagmaMaxGPUs+2][MagmaMaxGPUs+2], magma_int_t *ncmplx, magma_int_t ngpu ); void magma_indices_1D_bcyclic( magma_int_t nb, magma_int_t ngpu, magma_int_t dev, magma_int_t j0, magma_int_t j1, magma_int_t* dj0, magma_int_t* dj1 ); void magma_swp2pswp( magma_trans_t trans, magma_int_t n, magma_int_t *ipiv, magma_int_t *newipiv ); // ============================================================================= // get NB blocksize magma_int_t magma_get_smlsize_divideconquer(); // ============================================================================= // memory allocation magma_int_t magma_malloc( magma_ptr *ptr_ptr, size_t bytes ); magma_int_t magma_malloc_cpu( void **ptr_ptr, size_t bytes ); magma_int_t magma_malloc_pinned( void **ptr_ptr, size_t bytes ); magma_int_t magma_free_cpu( void *ptr ); #define magma_free( ptr ) \ magma_free_internal( ptr, __func__, __FILE__, __LINE__ ) #define magma_free_pinned( ptr ) \ magma_free_pinned_internal( ptr, __func__, __FILE__, __LINE__ ) magma_int_t magma_free_internal( magma_ptr ptr, const char* func, const char* file, int line ); magma_int_t magma_free_pinned_internal( void *ptr, const char* func, const char* file, int line ); // returns memory info (basically a wrapper around cudaMemGetInfo magma_int_t magma_mem_info(size_t* freeMem, size_t* totalMem); // wrapper around cudaMemset magma_int_t magma_memset(void * ptr, int value, size_t count); // wrapper around cudaMemsetAsync magma_int_t magma_memset_async(void * ptr, int value, size_t count, magma_queue_t queue); // type-safe convenience functions to avoid using (void**) cast and sizeof(...) // here n is the number of elements (floats, doubles, etc.) not the number of bytes. /******************************************************************************/ /// @addtogroup magma_malloc /// imalloc, smalloc, etc. /// @{ /// Type-safe version of magma_malloc(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc( magmaInt_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc( magmaIndex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc(), for magma_uindex_t arrays. Allocates n*sizeof(magma_uindex_t) bytes. static inline magma_int_t magma_uindex_malloc( magmaUIndex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_uindex_t) ); } /// Type-safe version of magma_malloc(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc( magmaFloat_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc( magmaDouble_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc( magmaFloatComplex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc( magmaDoubleComplex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} /******************************************************************************/ /// @addtogroup magma_malloc_cpu /// imalloc_cpu, smalloc_cpu, etc. /// @{ /// Type-safe version of magma_malloc_cpu(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc_cpu( magma_int_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc_cpu(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc_cpu( magma_index_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc_cpu(), for magma_uindex_t arrays. Allocates n*sizeof(magma_uindex_t) bytes. static inline magma_int_t magma_uindex_malloc_cpu( magma_uindex_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_uindex_t) ); } /// Type-safe version of magma_malloc_cpu(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc_cpu( float **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc_cpu(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc_cpu( double **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc_cpu(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc_cpu( magmaFloatComplex **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc_cpu(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc_cpu( magmaDoubleComplex **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} /******************************************************************************/ /// @addtogroup magma_malloc_pinned /// imalloc_pinned, smalloc_pinned, etc. /// @{ /// Type-safe version of magma_malloc_pinned(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc_pinned( magma_int_t **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc_pinned(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc_pinned( magma_index_t **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc_pinned(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc_pinned( float **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc_pinned(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc_pinned( double **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc_pinned(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc_pinned( magmaFloatComplex **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc_pinned(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc_pinned( magmaDoubleComplex **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} // CUDA MAGMA only magma_int_t magma_is_devptr( const void* ptr ); // ============================================================================= // device support magma_int_t magma_num_gpus( void ); /* todo: num_accelerators */ /* todo: total accelerators? available accelerators? i.e., number to use vs. number available. */ // CUDA MAGMA only magma_int_t magma_getdevice_arch(); /* magma_int_t magma_getdevice_arch( magma_int_t dev or queue ); todo: new */ void magma_getdevices( magma_device_t* devices, magma_int_t size, magma_int_t* num_dev ); void magma_getdevice( magma_device_t* dev ); void magma_setdevice( magma_device_t dev ); size_t magma_mem_size( magma_queue_t queue ); magma_int_t magma_getdevice_multiprocessor_count(); size_t magma_getdevice_shmem_block(); size_t magma_getdevice_shmem_multiprocessor(); // ============================================================================= // queue support // new magma_queue_create adds device #define magma_queue_create( device, queue_ptr ) \ magma_queue_create_internal( device, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_cuda( device, cuda_stream, cublas_handle, cusparse_handle, queue_ptr ) \ magma_queue_create_from_cuda_internal( device, cuda_stream, cublas_handle, cusparse_handle, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_hip( device, hip_stream, hipblas_handle, hipsparse_handle, queue_ptr ) \ magma_queue_create_from_hip_internal( device, hip_stream, hipblas_handle, hipsparse_handle, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_opencl( device, cl_queue, queue_ptr ) \ magma_queue_create_from_opencl_internal( device, cl_queue, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_destroy( queue ) \ magma_queue_destroy_internal( queue, __func__, __FILE__, __LINE__ ) #define magma_queue_sync( queue ) \ magma_queue_sync_internal( queue, __func__, __FILE__, __LINE__ ) void magma_queue_create_internal( magma_device_t device, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #ifdef MAGMA_HAVE_CUDA void magma_queue_create_from_cuda_internal( magma_device_t device, cudaStream_t stream, cublasHandle_t cublas, cusparseHandle_t cusparse, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #endif #ifdef MAGMA_HAVE_HIP void magma_queue_create_from_hip_internal( magma_device_t device, hipStream_t stream, hipblasHandle_t hipblas, hipsparseHandle_t hipsparse, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #endif #ifdef MAGMA_HAVE_OPENCL magma_int_t magma_queue_create_from_opencl_internal( magma_device_t device, cl_command_queue cl_queue, const char* func, const char* file, int line ); #endif void magma_queue_destroy_internal( magma_queue_t queue, const char* func, const char* file, int line ); void magma_queue_sync_internal( magma_queue_t queue, const char* func, const char* file, int line ); magma_int_t magma_queue_get_device( magma_queue_t queue ); // ============================================================================= // event support void magma_event_create( magma_event_t* event_ptr ); void magma_event_create_untimed( magma_event_t* event_ptr ); void magma_event_destroy( magma_event_t event ); void magma_event_record( magma_event_t event, magma_queue_t queue ); void magma_event_query( magma_event_t event ); void magma_event_sync( magma_event_t event ); void magma_queue_wait_event( magma_queue_t queue, magma_event_t event ); // ============================================================================= // error handler void magma_xerbla( const char *name, magma_int_t info ); const char* magma_strerror( magma_int_t error ); // ============================================================================= // string functions size_t magma_strlcpy( char *dst, const char *src, size_t size ); // ============================================================================= // integer functions /// For integers x >= 0, y > 0, returns ceil( x/y ). /// For x == 0, this is 0. /// @ingroup magma_ceildiv __host__ __device__ static inline magma_int_t magma_ceildiv( magma_int_t x, magma_int_t y ) { return (x + y - 1)/y; } /// For integers x >= 0, y > 0, returns x rounded up to multiple of y. /// That is, ceil(x/y)*y. /// For x == 0, this is 0. /// This implementation does not assume y is a power of 2. /// @ingroup magma_ceildiv __host__ __device__ static inline magma_int_t magma_roundup( magma_int_t x, magma_int_t y ) { return magma_ceildiv( x, y ) * y; } // ============================================================================= // scalar functions // real and complex square root // sqrt alone cannot be caught by the generation script because of tsqrt /// @return Square root of x. @ingroup magma_sqrt static inline float magma_ssqrt( float x ) { return sqrtf( x ); } /// @return Square root of x. @ingroup magma_sqrt static inline double magma_dsqrt( double x ) { return sqrt( x ); } /// @return Complex square root of x. @ingroup magma_sqrt magmaFloatComplex magma_csqrt( magmaFloatComplex x ); /// @return Complex square root of x. @ingroup magma_sqrt magmaDoubleComplex magma_zsqrt( magmaDoubleComplex x ); #ifdef __cplusplus } #endif #endif // MAGMA_AUXILIARY_H Loading
containers/docker/Dockerfile +20 −2 Original line number Diff line number Diff line Loading @@ -22,6 +22,14 @@ RUN DEBIAN_FRONTEND=noninteractive apt -y install intel-oneapi-compiler-fortran RUN DEBIAN_FRONTEND=noninteractive apt -y install liblapacke-dev liblapacke64-dev libopenblas-dev libopenblas-openmp-dev libopenblas64-dev libopenblas64-openmp-dev # install MPI stack RUN DEBIAN_FRONTEND=noninteractive apt -y install mpi-default-dev mpi-default-bin # install nvidia stuff COPY --chown=root:root containers/docker/dockerstuff/debian-nonfree/debian.sources /etc/apt/sources.list.d/ RUN apt update RUN DEBIAN_FRONTEND=noninteractive apt -y install libquadmath0 libcublaslt11 libgcc-s1 libgomp1 libcudart11.0 libcusparse11 libcublas11 nvidia-cuda-dev # install magma COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/lib/* /usr/lib/x86_64-linux-gnu/ COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/include/* /usr/include/ COPY --chown=root:root containers/docker/dockerstuff/magma-compiled/usr/lib/pkgconfig/* /usr/lib/x86_64-linux-gnu/pkgconfig/ # install packages needed to run python scripts for checks RUN DEBIAN_FRONTEND=noninteractive apt -y install python3 python-is-python3 python3-regex # install packages needed to run doxygen to create html docs Loading @@ -46,7 +54,13 @@ ADD src /root/np-tmcode/src ADD doc /root/np-tmcode/doc ADD build /root/np-tmcode/build ADD test_data /root/np-tmcode/test_data RUN cd np-tmcode/src && make wipe && make -j && cd ../doc/src && doxygen config.dox && cd ../build/latex && make -j RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_magma_mpi RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_MAGMA=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_magma_serial RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_lapack_mpi && cd ../build/cluster && ln -s np_cluster_lapack_mpi np_cluster RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_LAPACK=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_lapack_serial RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 USE_MPI=1 CXX=mpicxx FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_legacy_mpi RUN cd np-tmcode/src && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 CXX=g++ FC=gfortran make wipe && BUILDDIR=../../build BUILDDIR_NPTM=../../build/libnptm LIBNPTM=../../build/libnptm/libnptm.a USE_ILP64=1 USE_OPENMP=1 CXX=g++ FC=gfortran make -j && mv ../build/cluster/np_cluster ../build/cluster/np_cluster_legacy_serial RUN cd np-tmcode/doc/src && doxygen config.dox && cd ../build/latex && make -j # create the container on which the np-tmcode is installed, restarting from Loading @@ -55,8 +69,12 @@ FROM debian:bookworm-slim AS np-tmcode-run-minimal WORKDIR /root # install the strictly needed runtime libraries needed to run the executables # and the python check scripts RUN DEBIAN_FRONTEND=noninteractive apt update && DEBIAN_FRONTEND=noninteractive apt upgrade && DEBIAN_FRONTEND=noninteractive apt -y install libgfortran5 libgcc-s1 libhdf5-103-1 libstdc++6 libssl3 libcurl4 libsz2 zlib1g libnghttp2-14 libidn2-0 librtmp1 libssh2-1 libpsl5 libgssapi-krb5-2 libldap-2.5-0 libzstd1 libbrotli1 libaec0 libunistring2 libgmp10 libkrb5-3 libk5crypto3 libcom-err2 libkrb5support0 libsasl2-2 libp11-kit0 libtasn1-6 libkeyutils1 libffi8 liblapacke64 libopenblas64-0-openmp python3 python-is-python3 python3-regex hdf5-tools mpi-default-bin && rm -rf /var/lib/apt/lists/* COPY --chown=root:root containers/docker/dockerstuff/debian-nonfree/debian.sources /etc/apt/sources.list.d/ RUN apt update RUN DEBIAN_FRONTEND=noninteractive apt update && DEBIAN_FRONTEND=noninteractive apt upgrade && DEBIAN_FRONTEND=noninteractive apt -y install libgfortran5 libgcc-s1 libhdf5-103-1 libstdc++6 libssl3 libcurl4 libsz2 zlib1g libnghttp2-14 libidn2-0 librtmp1 libssh2-1 libpsl5 libgssapi-krb5-2 libldap-2.5-0 libzstd1 libbrotli1 libaec0 libunistring2 libgmp10 libkrb5-3 libk5crypto3 libcom-err2 libkrb5support0 libsasl2-2 libp11-kit0 libtasn1-6 libkeyutils1 libffi8 liblapacke64 libopenblas64-0-openmp python3 python-is-python3 python3-regex hdf5-tools mpi-default-bin libquadmath0 libcublaslt11 libgcc-s1 libgomp1 libcudart11.0 libcusparse11 libcublas11 && rm -rf /var/lib/apt/lists/* COPY --from=np-tmcode-run-dev /root /root COPY --from=np-tmcode-run-dev /usr/lib/x86_64-linux-gnu/libmagma.so /usr/lib/x86_64-linux-gnu/libmagma.so COPY --from=np-tmcode-run-dev /usr/lib/x86_64-linux-gnu/libmagma.so.2 /usr/lib/x86_64-linux-gnu/libmagma.so.2 # remove everything which is not needed to run the codes RUN cd /root/np-tmcode && find build -name "*.o" -exec rm -v \{\} \; && find build -name "*.gcno" -exec rm -v \{\} \; && cd src && rm -rvf cluster libnptm trapping include sphere Makefile make.inc README.md && cd .. && rm -rvf containers && cd doc && rm -rvf src && cd build/latex && rm -rvf *.tex *.out *.sty *.ind *.log *.toc *.ilg *.idx *.aux *.eps Makefile class*.pdf # move the installed software to /usr/local Loading
containers/docker/dockerstuff/debian-nonfree/debian.sources 0 → 100644 +14 −0 Original line number Diff line number Diff line Types: deb # http://snapshot.debian.org/archive/debian/20240612T000000Z URIs: http://deb.debian.org/debian Suites: bookworm bookworm-updates Components: main non-free contrib non-free-firmware Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg Types: deb # http://snapshot.debian.org/archive/debian-security/20240612T000000Z URIs: http://deb.debian.org/debian-security Suites: bookworm-security Components: main non-free contrib non-free-firmware Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg
containers/docker/dockerstuff/magma-compiled/usr/include/magma.h 0 → 100644 +39 −0 Original line number Diff line number Diff line /* -- MAGMA (version 2.7.2) -- Univ. of Tennessee, Knoxville Univ. of California, Berkeley Univ. of Colorado, Denver @date August 2023 */ #ifndef MAGMA_H #define MAGMA_H #ifdef MAGMA_NO_V1 #error "Since MAGMA_NO_V1 is defined, magma.h is invalid; use magma_v2.h" #endif // ============================================================================= // MAGMA configuration #include "magma_config.h" // magma v1 includes cublas.h by default, unless cublas_v2.h has already been included #ifndef CUBLAS_V2_H_ #if defined(MAGMA_HAVE_CUDA) #include <cublas.h> #endif #endif // Include the MAGMA v2 and v1 APIs, // then map names to the v1 API (e.g., magma_zgemm => magma_zgemm_v1). // Some functions (like setmatrix_async) are the same in v1 and v2, // so are provided by the v2 API. #include "magma_v2.h" #include "magmablas_v1.h" #include "magmablas_v1_map.h" #undef MAGMA_API #define MAGMA_API 1 #endif // MAGMA_H
containers/docker/dockerstuff/magma-compiled/usr/include/magma.mod 0 → 100644 +106 KiB File added.No diff preview for this file type. View file
containers/docker/dockerstuff/magma-compiled/usr/include/magma_auxiliary.h 0 → 100644 +407 −0 Original line number Diff line number Diff line /* -- MAGMA (version 2.7.2) -- Univ. of Tennessee, Knoxville Univ. of California, Berkeley Univ. of Colorado, Denver @date August 2023 */ #ifndef MAGMA_AUXILIARY_H #define MAGMA_AUXILIARY_H #include "magma_types.h" #include <math.h> // sqrtf #ifdef __cplusplus extern "C" { #endif // ============================================================================= // initialization magma_int_t magma_init( void ); magma_int_t magma_finalize( void ); #ifdef MAGMA_HAVE_OPENCL magma_int_t magma_init_opencl( cl_platform_id platform, cl_context context, magma_int_t setup_clBlas ); magma_int_t magma_finalize_opencl( magma_int_t finalize_clBlas ); #endif // ============================================================================= // version information void magma_version( magma_int_t* major, magma_int_t* minor, magma_int_t* micro ); void magma_print_environment(); // ============================================================================= // timing real_Double_t magma_wtime( void ); real_Double_t magma_sync_wtime( magma_queue_t queue ); // ============================================================================= // misc. functions // magma GPU-complex PCIe connection magma_int_t magma_buildconnection_mgpu( magma_int_t gnode[MagmaMaxGPUs+2][MagmaMaxGPUs+2], magma_int_t *ncmplx, magma_int_t ngpu ); void magma_indices_1D_bcyclic( magma_int_t nb, magma_int_t ngpu, magma_int_t dev, magma_int_t j0, magma_int_t j1, magma_int_t* dj0, magma_int_t* dj1 ); void magma_swp2pswp( magma_trans_t trans, magma_int_t n, magma_int_t *ipiv, magma_int_t *newipiv ); // ============================================================================= // get NB blocksize magma_int_t magma_get_smlsize_divideconquer(); // ============================================================================= // memory allocation magma_int_t magma_malloc( magma_ptr *ptr_ptr, size_t bytes ); magma_int_t magma_malloc_cpu( void **ptr_ptr, size_t bytes ); magma_int_t magma_malloc_pinned( void **ptr_ptr, size_t bytes ); magma_int_t magma_free_cpu( void *ptr ); #define magma_free( ptr ) \ magma_free_internal( ptr, __func__, __FILE__, __LINE__ ) #define magma_free_pinned( ptr ) \ magma_free_pinned_internal( ptr, __func__, __FILE__, __LINE__ ) magma_int_t magma_free_internal( magma_ptr ptr, const char* func, const char* file, int line ); magma_int_t magma_free_pinned_internal( void *ptr, const char* func, const char* file, int line ); // returns memory info (basically a wrapper around cudaMemGetInfo magma_int_t magma_mem_info(size_t* freeMem, size_t* totalMem); // wrapper around cudaMemset magma_int_t magma_memset(void * ptr, int value, size_t count); // wrapper around cudaMemsetAsync magma_int_t magma_memset_async(void * ptr, int value, size_t count, magma_queue_t queue); // type-safe convenience functions to avoid using (void**) cast and sizeof(...) // here n is the number of elements (floats, doubles, etc.) not the number of bytes. /******************************************************************************/ /// @addtogroup magma_malloc /// imalloc, smalloc, etc. /// @{ /// Type-safe version of magma_malloc(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc( magmaInt_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc( magmaIndex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc(), for magma_uindex_t arrays. Allocates n*sizeof(magma_uindex_t) bytes. static inline magma_int_t magma_uindex_malloc( magmaUIndex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magma_uindex_t) ); } /// Type-safe version of magma_malloc(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc( magmaFloat_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc( magmaDouble_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc( magmaFloatComplex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc( magmaDoubleComplex_ptr *ptr_ptr, size_t n ) { return magma_malloc( (magma_ptr*) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} /******************************************************************************/ /// @addtogroup magma_malloc_cpu /// imalloc_cpu, smalloc_cpu, etc. /// @{ /// Type-safe version of magma_malloc_cpu(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc_cpu( magma_int_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc_cpu(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc_cpu( magma_index_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc_cpu(), for magma_uindex_t arrays. Allocates n*sizeof(magma_uindex_t) bytes. static inline magma_int_t magma_uindex_malloc_cpu( magma_uindex_t **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magma_uindex_t) ); } /// Type-safe version of magma_malloc_cpu(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc_cpu( float **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc_cpu(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc_cpu( double **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc_cpu(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc_cpu( magmaFloatComplex **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc_cpu(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc_cpu( magmaDoubleComplex **ptr_ptr, size_t n ) { return magma_malloc_cpu( (void**) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} /******************************************************************************/ /// @addtogroup magma_malloc_pinned /// imalloc_pinned, smalloc_pinned, etc. /// @{ /// Type-safe version of magma_malloc_pinned(), for magma_int_t arrays. Allocates n*sizeof(magma_int_t) bytes. static inline magma_int_t magma_imalloc_pinned( magma_int_t **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magma_int_t) ); } /// Type-safe version of magma_malloc_pinned(), for magma_index_t arrays. Allocates n*sizeof(magma_index_t) bytes. static inline magma_int_t magma_index_malloc_pinned( magma_index_t **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magma_index_t) ); } /// Type-safe version of magma_malloc_pinned(), for float arrays. Allocates n*sizeof(float) bytes. static inline magma_int_t magma_smalloc_pinned( float **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(float) ); } /// Type-safe version of magma_malloc_pinned(), for double arrays. Allocates n*sizeof(double) bytes. static inline magma_int_t magma_dmalloc_pinned( double **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(double) ); } /// Type-safe version of magma_malloc_pinned(), for magmaFloatComplex arrays. Allocates n*sizeof(magmaFloatComplex) bytes. static inline magma_int_t magma_cmalloc_pinned( magmaFloatComplex **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magmaFloatComplex) ); } /// Type-safe version of magma_malloc_pinned(), for magmaDoubleComplex arrays. Allocates n*sizeof(magmaDoubleComplex) bytes. static inline magma_int_t magma_zmalloc_pinned( magmaDoubleComplex **ptr_ptr, size_t n ) { return magma_malloc_pinned( (void**) ptr_ptr, n*sizeof(magmaDoubleComplex) ); } /// @} // CUDA MAGMA only magma_int_t magma_is_devptr( const void* ptr ); // ============================================================================= // device support magma_int_t magma_num_gpus( void ); /* todo: num_accelerators */ /* todo: total accelerators? available accelerators? i.e., number to use vs. number available. */ // CUDA MAGMA only magma_int_t magma_getdevice_arch(); /* magma_int_t magma_getdevice_arch( magma_int_t dev or queue ); todo: new */ void magma_getdevices( magma_device_t* devices, magma_int_t size, magma_int_t* num_dev ); void magma_getdevice( magma_device_t* dev ); void magma_setdevice( magma_device_t dev ); size_t magma_mem_size( magma_queue_t queue ); magma_int_t magma_getdevice_multiprocessor_count(); size_t magma_getdevice_shmem_block(); size_t magma_getdevice_shmem_multiprocessor(); // ============================================================================= // queue support // new magma_queue_create adds device #define magma_queue_create( device, queue_ptr ) \ magma_queue_create_internal( device, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_cuda( device, cuda_stream, cublas_handle, cusparse_handle, queue_ptr ) \ magma_queue_create_from_cuda_internal( device, cuda_stream, cublas_handle, cusparse_handle, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_hip( device, hip_stream, hipblas_handle, hipsparse_handle, queue_ptr ) \ magma_queue_create_from_hip_internal( device, hip_stream, hipblas_handle, hipsparse_handle, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_create_from_opencl( device, cl_queue, queue_ptr ) \ magma_queue_create_from_opencl_internal( device, cl_queue, queue_ptr, __func__, __FILE__, __LINE__ ) #define magma_queue_destroy( queue ) \ magma_queue_destroy_internal( queue, __func__, __FILE__, __LINE__ ) #define magma_queue_sync( queue ) \ magma_queue_sync_internal( queue, __func__, __FILE__, __LINE__ ) void magma_queue_create_internal( magma_device_t device, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #ifdef MAGMA_HAVE_CUDA void magma_queue_create_from_cuda_internal( magma_device_t device, cudaStream_t stream, cublasHandle_t cublas, cusparseHandle_t cusparse, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #endif #ifdef MAGMA_HAVE_HIP void magma_queue_create_from_hip_internal( magma_device_t device, hipStream_t stream, hipblasHandle_t hipblas, hipsparseHandle_t hipsparse, magma_queue_t* queue_ptr, const char* func, const char* file, int line ); #endif #ifdef MAGMA_HAVE_OPENCL magma_int_t magma_queue_create_from_opencl_internal( magma_device_t device, cl_command_queue cl_queue, const char* func, const char* file, int line ); #endif void magma_queue_destroy_internal( magma_queue_t queue, const char* func, const char* file, int line ); void magma_queue_sync_internal( magma_queue_t queue, const char* func, const char* file, int line ); magma_int_t magma_queue_get_device( magma_queue_t queue ); // ============================================================================= // event support void magma_event_create( magma_event_t* event_ptr ); void magma_event_create_untimed( magma_event_t* event_ptr ); void magma_event_destroy( magma_event_t event ); void magma_event_record( magma_event_t event, magma_queue_t queue ); void magma_event_query( magma_event_t event ); void magma_event_sync( magma_event_t event ); void magma_queue_wait_event( magma_queue_t queue, magma_event_t event ); // ============================================================================= // error handler void magma_xerbla( const char *name, magma_int_t info ); const char* magma_strerror( magma_int_t error ); // ============================================================================= // string functions size_t magma_strlcpy( char *dst, const char *src, size_t size ); // ============================================================================= // integer functions /// For integers x >= 0, y > 0, returns ceil( x/y ). /// For x == 0, this is 0. /// @ingroup magma_ceildiv __host__ __device__ static inline magma_int_t magma_ceildiv( magma_int_t x, magma_int_t y ) { return (x + y - 1)/y; } /// For integers x >= 0, y > 0, returns x rounded up to multiple of y. /// That is, ceil(x/y)*y. /// For x == 0, this is 0. /// This implementation does not assume y is a power of 2. /// @ingroup magma_ceildiv __host__ __device__ static inline magma_int_t magma_roundup( magma_int_t x, magma_int_t y ) { return magma_ceildiv( x, y ) * y; } // ============================================================================= // scalar functions // real and complex square root // sqrt alone cannot be caught by the generation script because of tsqrt /// @return Square root of x. @ingroup magma_sqrt static inline float magma_ssqrt( float x ) { return sqrtf( x ); } /// @return Square root of x. @ingroup magma_sqrt static inline double magma_dsqrt( double x ) { return sqrt( x ); } /// @return Complex square root of x. @ingroup magma_sqrt magmaFloatComplex magma_csqrt( magmaFloatComplex x ); /// @return Complex square root of x. @ingroup magma_sqrt magmaDoubleComplex magma_zsqrt( magmaDoubleComplex x ); #ifdef __cplusplus } #endif #endif // MAGMA_AUXILIARY_H
