Merge branch 'JK/linear-solvers' into 'develop'

option(WITH_TESTS "Build tests" ON)

option(WITH_PROFILING "Enable code profiling compiler flags" OFF )

option(WITH_COVERAGE "Enable code coverage reports from unit tests" OFF)

option(WITH_EXAMPLES "Compile the 'examples' directory" ON)

option(WITH_EXAMPLES "Compile the 'src/Examples' directory" ON)

option(WITH_TOOLS "Compile the 'src/Tools' directory" ON)

option(WITH_BENCHMARKS "Compile the 'src/Benchmarks' directory" ON)

option(WITH_PYTHON "Compile the Python bindings" ON)

option(WITH_DOC "Generate documentation" ON)

option(WITH_DOC "Build examples included in the documentation" ON)

# install paths relative to the cmake's prefix

set( TNL_TARGET_INCLUDE_DIRECTORY "include/TNL" )

set( CMAKE_CXX_STANDARD_REQUIRED ON )

set( CMAKE_CXX_EXTENSIONS OFF )

# set Debug/Release options

# set default build options

set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread -Wall -Wno-unused-local-typedefs -Wno-unused-variable -Wno-unknown-pragmas" )

set( CMAKE_CXX_FLAGS_DEBUG "-g" )

set( CMAKE_CXX_FLAGS_RELEASE "-O3 -march=native -mtune=native -DNDEBUG" )

set( CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG" )

# pass -rdynamic only in Debug mode

set( CMAKE_SHARED_LIBRARY_LINK_C_FLAGS "" )

set( CMAKE_SHARED_LIBRARY_LINK_C_FLAGS_DEBUG "-rdynamic" )

set( CMAKE_SHARED_LINKER_FLAGS_DEBUG "-rdynamic" )

set( CMAKE_SHARED_LINKER_FLAGS_RELEASE "" )

# set additional Debug/Release options using generator expressions

# (that way we can exclude some options for specific targets, see https://stackoverflow.com/a/59734798 for details)

add_compile_options(

   # GOTCHA: CMake uses semicolons as list item separator, spaces would lead to a single argument inside double-quotes on the command line

   "$<$<CONFIG:RELEASE>:-march=native;-mtune=native>"

)

# disable GCC's infamous "maybe-uninitialized" warning (it produces mostly false positives)

if( CMAKE_CXX_COMPILER_ID STREQUAL "GNU" )

   set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-maybe-uninitialized" )

add_subdirectory( share )

# Add subdirectories for examples included in the documentation

if( ${WITH_DOC} OR ${WITH_EXAMPLES} )

if( ${WITH_DOC} )

   set( TNL_DOCUMENTATION_OUTPUT_SNIPPETS_PATH "${CMAKE_SOURCE_DIR}/Documentation/output_snippets" )

   file(MAKE_DIRECTORY ${TNL_DOCUMENTATION_OUTPUT_SNIPPETS_PATH})

   add_subdirectory( Documentation/Examples )

    echo "   --with-profiling=yes/no               Enables code profiling compiler falgs. 'no' by default."

    echo "   --with-coverage=yes/no                Enables code coverage reports for unit tests. 'no' by default (lcov is required)."

    echo "   --with-doc=yes/no                     Build documentation. 'yes' by default."

    echo "   --with-examples=yes/no                Compile the 'examples' directory. 'yes' by default."

    echo "   --with-examples=yes/no                Compile the 'src/Examples' directory. 'yes' by default."

    echo "   --with-tools=yes/no                   Compile the 'src/Tools' directory. 'yes' by default."

    echo "   --with-python=yes/no                  Compile the Python bindings. 'yes' by default."

    echo "   --with-benchmarks=yes/no              Compile the 'src/Benchmarks' directory. 'yes' by default."

    echo "   --with-doc=yes/no                     Generate the documentation. 'yes' by default."

    echo "   --cmake=CMAKE                         Path to cmake. 'cmake' by default."

    echo "   --verbose                             It enables verbose build."

    echo "   --root-dir=PATH                       Path to the TNL source code root dir."

    add_executable( tnl-benchmark-blas tnl-benchmark-blas.cpp )

endif()

find_library( CBLAS_LIBRARY NAMES cblas

           PATHS /usr/lib

                 /usr/lib64

                 /usr/lib/x86_64-linux-gnu

                 /usr/local/lib

                 /usr/local/lib64 )

find_library( CBLAS_LIBRARY NAMES cblas )

# fallback for Centos 7.5 - libcblas.so does not exist, link to libtatlas.so or libsatlas.so

# https://forums.centos.org/viewtopic.php?t=48543

find_library( TATLAS_LIBRARY NAMES tatlas

              PATH_SUFFIXES atlas )

find_library( SATLAS_LIBRARY NAMES satlas

              PATH_SUFFIXES atlas )

if( CBLAS_LIBRARY )

   target_compile_definitions( tnl-benchmark-blas PUBLIC "-DHAVE_BLAS" )

   target_link_libraries( tnl-benchmark-blas ${CBLAS_LIBRARY} )

elseif( TATLAS_LIBRARY )

   target_compile_definitions( tnl-benchmark-blas PUBLIC "-DHAVE_BLAS" )

   target_link_libraries( tnl-benchmark-blas ${TATLAS_LIBRARY} )

elseif( SATLAS_LIBRARY )

   target_compile_definitions( tnl-benchmark-blas PUBLIC "-DHAVE_BLAS" )

   target_link_libraries( tnl-benchmark-blas ${SATLAS_LIBRARY} )

else()

   # FIXME: We require the CBLAS interface, but CMake's FindBLAS cannot detect that,

   #        so this fails unless the BLAS implementation includes it in the same

   #        shared library file as the Fortran implementation (e.g. OpenBLAS does that).

   find_package( BLAS )

   if( BLAS_FOUND )

      target_compile_definitions( tnl-benchmark-blas PUBLIC "-DHAVE_BLAS" )

#ifdef HAVE_BLAS

// HOTFIX: cblas.h from the atlas-devel package (version 3.10.1-12.el7) on CentOS 7

// does not declare the functions as `extern "C"`, which breaks name mangling.

// Note that nested `extern "C"` is valid and correct:

// https://stackoverflow.com/questions/48099828/what-happens-if-you-nest-extern-c

extern "C" {

#include <cblas.h>

}

inline int blasIgamax( int n, const float *x, int incx )

{

      virtual HeaderElements getTableHeader() const override

      {

         return HeaderElements({"time", "speedup", "converged", "iterations", "residue_precond", "residue_true"});

         return HeaderElements({"time", "stddev", "stddev/time", "speedup", "converged", "iterations", "residue_precond", "residue_true"});

      }

      virtual RowElements getRowElements() const override

         r = b - r;

         const double residue_true = lpNorm( r, 2.0 ) / lpNorm( b, 2.0 );

         return RowElements({ time, speedup, (double) converged, (double) iterations,

         return RowElements({ time, stddev, stddev/time, speedup, (double) converged, (double) iterations,

                              residue_precond, residue_true });

      }

   };