Merge remote-tracking branch 'origin/master' into rules_files_not_matching_cpp

.gitlab-ci.yml

@@ -15,8 +15,8 @@ push_github_sourceforge:
     - git remote update
     - git checkout master
     - git pull origin master
-    - git push sourceforge master
+    - git push sourceforge master --tags
-    - git push github master
+    - git push github master --tags
   only:
     - master
 
@@ -55,6 +55,8 @@ build_win:
         cmake %SRC_DIR%
         -G Ninja
         -DCMAKE_INSTALL_PREFIX=%INSTALL_DIR%
+        -DCHECK_RANGE=ON
+        -DUSE_CGNS=ON
         -DUSE_OCC=ON
         -DOCC_LIBRARY=C:/install_opencascade_7.4.0_static/win64/vc14/lib/TKernel.lib
         -DOCC_INCLUDE_DIR=C:/install_opencascade_7.4.0_static/inc
@@ -68,6 +70,7 @@ test_win:
   <<: *win
   stage: test
   script:
+    - pip install pytest-check
     - cd tests\pytest
     - cd %NETGEN_BUILD_DIR%\netgen
     - ctest -C Release -V --output-on-failure
@@ -237,11 +240,13 @@ build_mac:
         cmake $SRC_DIR
         -DCMAKE_INSTALL_PREFIX=$CMAKE_INSTALL_PREFIX
         -DCMAKE_BUILD_TYPE=Release
+        -DCHECK_RANGE=ON
         -DUSE_NATIVE_ARCH=OFF
         -DUSE_CCACHE=ON
         -DENABLE_UNIT_TESTS=ON
         -DCMAKE_OSX_DEPLOYMENT_TARGET=10.12
         -DCMAKE_OSX_SYSROOT=/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk
+        -DUSE_CGNS=ON
         -DUSE_OCC=ON
         -DOCC_LIBRARY=/usr/local/opt/opencascade-7.4.0/lib/libTKernel.a
         -DOCC_INCLUDE_DIR=/usr/local/opt/opencascade-7.4.0/include/opencascade

CMakeLists.txt

@@ -14,9 +14,12 @@ option( USE_NATIVE_ARCH  "build for native cpu architecture" ON)
 option( USE_GUI     "don't build netgen with GUI" ON )
 option( USE_PYTHON  "build with python interface" ON )
 option( USE_MPI     "enable mpi parallelization" OFF )
+option( USE_MPI4PY  "enable mpi4py interface" ON )
 option( USE_OCC     "(not supported) compile with OpenCascade geometry kernel" OFF)
 option( USE_JPEG    "enable snapshots using library libjpeg" OFF )
 option( USE_MPEG    "enable video recording with FFmpeg, uses libavcodec" OFF )
+option( USE_CGNS    "enable CGNS file read/write support" OFF )
+option( USE_NUMA         "compile with NUMA-aware code")
 option( INTEL_MIC        "cross compile for intel xeon phi")
 option( INSTALL_PROFILES "install environment variable settings to /etc/profile.d" OFF )
 option( USE_CCACHE       "use ccache")
@@ -30,11 +33,12 @@ option( BUILD_STUB_FILES "Build stub files for better autocompletion" ON)
 option( BUILD_FOR_CONDA "Link python libraries only to executables" OFF)
 
 option( USE_SUPERBUILD       "use ccache" ON)
+option( TRACE_MEMORY     "Enable memory tracing" OFF)
+
+set(NG_COMPILE_FLAGS "" CACHE STRING "Additional compile flags")
 
 set(CMAKE_MODULE_PATH "${CMAKE_MODULE_PATH}" "${CMAKE_CURRENT_SOURCE_DIR}/cmake/cmake_modules")
 
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-
 if(APPLE)
   set(INSTALL_DIR_DEFAULT /Applications/Netgen.app)
 else(APPLE)
@@ -50,14 +54,14 @@ if(INSTALL_DIR)
   set(INSTALL_DIR_DEFAULT ${INSTALL_DIR})
 endif(INSTALL_DIR)
 
-if(UNIX)
+if(UNIX AND USE_SUPERBUILD)
   message("Checking for write permissions in install directory...")
   execute_process(COMMAND mkdir -p ${CMAKE_INSTALL_PREFIX})
   execute_process(COMMAND test -w ${CMAKE_INSTALL_PREFIX} RESULT_VARIABLE res)
   if(res)
     message(WARNING "No write access at install directory, please set correct permissions")
   endif()
-endif(UNIX)
+endif(UNIX AND USE_SUPERBUILD)
 
 if (USE_SUPERBUILD)
   project (SUPERBUILD)
@@ -76,12 +80,10 @@ else()
   endif()
 endif()
 
-set(NETGEN_VERSION_MAJOR 6)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-set(NETGEN_VERSION_MINOR 2)
+
-string(TIMESTAMP NETGEN_VERSION_PATCH "%y%U%w" )
+include (${CMAKE_CURRENT_LIST_DIR}/cmake/generate_version_file.cmake)
-set(NETGEN_VERSION "${NETGEN_VERSION_MAJOR}.${NETGEN_VERSION_MINOR}-dev")
+set(CPACK_PACKAGE_VERSION "${NETGEN_VERSION}")
-set(PACKAGE_VERSION "${NETGEN_VERSION_MAJOR}.${NETGEN_VERSION_MINOR}-${NETGEN_VERSION_PATCH}")
-set(CPACK_PACKAGE_VERSION "${PACKAGE_VERSION}")
 
 
 #######################################################################
@@ -193,7 +195,6 @@ check_include_files (dlfcn.h HAVE_DLFCN_H)
 if(HAVE_DLFCN_H)
   add_definitions(-DHAVE_DLFCN_H)
 endif()
-add_definitions(-DNETGEN_VERSION="${NETGEN_VERSION}")
 
 include_directories(BEFORE ${CMAKE_CURRENT_BINARY_DIR})
 
@@ -300,6 +301,14 @@ if (USE_MPI)
     target_include_directories(netgen_metis INTERFACE ${METIS_INCLUDE_DIR})
     target_link_libraries(netgen_metis INTERFACE ${METIS_LIBRARY} )
     target_compile_definitions(netgen_metis INTERFACE METIS )
+
+    if(USE_MPI4PY AND USE_PYTHON)
+      execute_process(COMMAND ${PYTHON_EXECUTABLE} -c "import mpi4py;print(mpi4py.get_include())" OUTPUT_VARIABLE mpi4py_path OUTPUT_STRIP_TRAILING_WHITESPACE)
+      find_path(MPI4PY_INCLUDE_DIR mpi4py.h HINTS ${mpi4py_path}/mpi4py NO_DEFAULT_PATH REQUIRED)
+      target_include_directories(netgen_metis INTERFACE ${MPI4PY_INCLUDE_DIR})
+      target_compile_definitions(netgen_metis INTERFACE NG_MPI4PY )
+      message(STATUS "Found mpi4py: ${MPI4PY_INCLUDE_DIR}")
+    endif(USE_MPI4PY AND USE_PYTHON)
 endif (USE_MPI)
 install(TARGETS netgen_mpi netgen_metis ${NG_INSTALL_DIR})
 
@@ -324,6 +333,12 @@ if (USE_MPEG)
     include_directories(${FFMPEG_INCLUDE_DIR})
 endif (USE_MPEG)
 
+#######################################################################
+add_custom_target(ng_generate_version_file
+  ${CMAKE_COMMAND}
+    -DBDIR=${CMAKE_CURRENT_BINARY_DIR}
+    -P ${CMAKE_CURRENT_LIST_DIR}/cmake/generate_version_file.cmake
+    )
 #######################################################################
 if(INSTALL_PROFILES)
     file(WRITE  ${CMAKE_CURRENT_BINARY_DIR}/netgen.sh "#!/bin/sh\n")
@@ -388,12 +403,27 @@ if(ENABLE_CPP_CORE_GUIDELINES_CHECK)
   endif()
 endif(ENABLE_CPP_CORE_GUIDELINES_CHECK)
 
+add_library(netgen_cgns INTERFACE)
+if(USE_CGNS)
+    find_library( CGNS_LIBRARY NAMES cgns cgnsdll )
+    find_path( CGNS_INCLUDE_DIR cgnslib.h )
+    target_compile_definitions(netgen_cgns INTERFACE NG_CGNS)
+    target_include_directories(netgen_cgns INTERFACE ${CGNS_INCLUDE_DIR})
+    target_link_libraries(netgen_cgns INTERFACE ${CGNS_LIBRARY})
+    if(NOT WIN32 AND NOT APPLE) # hdf5 is statically linked into cgns in Windows amd MacOS binaries
+        find_library(HDF5_LIBRARY NAMES hdf5 hdf5_serial)
+        target_link_libraries(netgen_cgns INTERFACE ${HDF5_LIBRARY})
+    endif(NOT WIN32 AND NOT APPLE)
+endif(USE_CGNS)
+
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/netgen_version.hpp ${CMAKE_CURRENT_BINARY_DIR}/netgen_config.hpp DESTINATION ${NG_INSTALL_DIR_INCLUDE}/include COMPONENT netgen_devel)
+
+add_subdirectory(windows)
 add_subdirectory(libsrc)
 add_subdirectory(ng)
 add_subdirectory(tutorials)
 add_subdirectory(py_tutorials)
 add_subdirectory(doc)
-add_subdirectory(windows)
 add_subdirectory(nglib)
 if (USE_PYTHON)
   add_subdirectory(python)
@@ -444,6 +474,8 @@ if(USE_NATIVE_ARCH)
         else()
             message(STATUS "Build for generic CPU")
         endif()
+    elseif(APPLE AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
+	# no flag necessary/available on Apple M1
     else()
         target_compile_options(ngcore PUBLIC "-march=native")
     endif(WIN32)

CONTRIBUTING.md

@@ -0,0 +1,18 @@
+
+# How to Contribute
+
+## Reporting issues
+
+If you have a problem using Netgen/NGSolve consider asking a question in our [forum](https://ngsolve.org/forum).
+
+If you found a bug create an issue in the [Github Issue Tracker](https://github.com/NGSolve/netgen/issues). Please be as specific as possible, issues with a reproducible minimal failing example will get more attention than unspecific one liners :)
+
+## Contributing patches
+
+We love and want to encourage community engagement and will review and accept patches and contributions to this project. There are just a few steps to follow:
+
+On your first contribution, to clear any legal questions, we ask you to sign our [Contributor License Agreement](CLA.pdf). Generally you have to sign this only once for Netgen or NGSolve. Please send the signed agreement to <joachim.schoeberl@tuwien.ac.at>.
+
+Place a pull request on GitHub. From there we will pull it into our internal testing environment and, if approved, merge it into the main codebase.
+
+If you have any questions feel free to ask on the [forum](https://ngsolve.org/forum).

cmake/NetgenConfig.cmake.in

@@ -1,4 +1,8 @@
 set(NETGEN_VERSION "@NETGEN_VERSION@")
+set(NETGEN_VERSION_MAJOR "@NETGEN_VERSION_MAJOR@")
+set(NETGEN_VERSION_MINOR "@NETGEN_VERSION_MINOR@")
+set(NETGEN_VERSION_PATCH "@NETGEN_VERSION_PATCH@")
+set(NETGEN_VERSION_TWEAK "@NETGEN_VERSION_TWEAK@")
 
 get_filename_component(NETGEN_CMAKE_DIR "${CMAKE_CURRENT_LIST_FILE}" PATH)
 
@@ -27,6 +31,7 @@ set(NETGEN_METIS_LIBRARY "@METIS_LIBRARY@")
 set(NETGEN_MKL_LIBRARIES "@MKL_LIBRARIES@")
 set(NETGEN_MPI_CXX_INCLUDE_PATH "@MPI_CXX_INCLUDE_PATH@")
 set(NETGEN_MPI_CXX_LIBRARIES "@MPI_CXX_LIBRARIES@")
+set(NETGEN_NUMA_LIBRARY "@NUMA_LIBRARY@")
 set(NETGEN_OCC_INCLUDE_DIR "@OCC_INCLUDE_DIR@")
 set(NETGEN_OCC_LIBRARIES_BIN "@OCC_LIBRARIES_BIN@")
 set(NETGEN_OCC_LIBRARIES "@OCC_LIBRARIES@")
@@ -51,10 +56,12 @@ set(NETGEN_USE_MPI @USE_MPI@)
 set(NETGEN_USE_OCC @USE_OCC@)
 set(NETGEN_USE_JPEG @USE_JPEG@)
 set(NETGEN_USE_MPEG @USE_MPEG@)
+set(NETGEN_USE_CGNS @USE_CGNS@)
 set(NETGEN_INTEL_MIC @INTEL_MIC@)
 set(NETGEN_INSTALL_PROFILES @INSTALL_PROFILES@)
 set(NETGEN_USE_CCACHE @USE_CCACHE@)
 set(NETGEN_USE_NATIVE_ARCH @USE_NATIVE_ARCH@)
+set(NETGEN_USE_NUMA @USE_NUMA@)
 
 set(NETGEN_PYTHON_RPATH "@NETGEN_PYTHON_RPATH@")
 set(NETGEN_RPATH_TOKEN "@NG_RPATH_TOKEN@")

cmake/SuperBuild.cmake

@@ -15,12 +15,12 @@ macro(set_vars VAR_OUT)
   endforeach()
 endmacro()
 #######################################################################
-if(WIN32)
+set (DEPS_DOWNLOAD_URL "https://github.com/NGSolve/ngsolve_dependencies/releases/download/v1.0.0" CACHE STRING INTERNAL)
-  set (DEPS_DOWNLOAD_URL "https://github.com/NGSolve/ngsolve_dependencies/releases/download/v1.0.0" CACHE STRING INTERNAL)
+set (OCC_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/occ_win64.zip" CACHE STRING INTERNAL)
-  set (OCC_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/occ_win64.zip" CACHE STRING INTERNAL)
+set (TCLTK_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/tcltk_win64.zip" CACHE STRING INTERNAL)
-  set (TCLTK_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/tcltk_win64.zip" CACHE STRING INTERNAL)
+set (ZLIB_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/zlib_win64.zip" CACHE STRING INTERNAL)
-  set (ZLIB_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/zlib_win64.zip" CACHE STRING INTERNAL)
+set (CGNS_DOWNLOAD_URL_WIN "${DEPS_DOWNLOAD_URL}/cgns_win64.zip" CACHE STRING INTERNAL)
-endif(WIN32)
+set (CGNS_DOWNLOAD_URL_MAC "${DEPS_DOWNLOAD_URL}/cgns_mac.zip" CACHE STRING INTERNAL)
 
 if(UNIX)
   message("Checking for write permissions in install directory...")
@@ -86,6 +86,10 @@ if(USE_GUI)
   include(cmake/external_projects/tcltk.cmake)
 endif(USE_GUI)
 
+if(USE_CGNS)
+  include(cmake/external_projects/cgns.cmake)
+endif(USE_CGNS)
+
 #######################################################################
 if(USE_MPI)
   if(UNIX)
@@ -128,6 +132,7 @@ set_vars( NETGEN_CMAKE_ARGS
   USE_OCC
   USE_MPEG
   USE_JPEG
+  USE_CGNS
   USE_INTERNAL_TCL
   INSTALL_PROFILES
   INTEL_MIC
@@ -138,8 +143,10 @@ set_vars( NETGEN_CMAKE_ARGS
   USE_SPDLOG
   DEBUG_LOG
   CHECK_RANGE
+  TRACE_MEMORY
   BUILD_STUB_FILES
   BUILD_FOR_CONDA
+  NG_COMPILE_FLAGS
   )
 
 # propagate all variables set on the command line using cmake -DFOO=BAR

cmake/cmake_modules/FindOpenCasCade.cmake

@@ -25,9 +25,9 @@ else(WIN32)
     )
 endif(WIN32)
 
-if(OCC_LIBRARY)
+if(OCC_LIBRARY AND NOT OCC_LIBRARY_DIR)
     get_filename_component(OCC_LIBRARY_DIR ${OCC_LIBRARY} PATH)
-endif(OCC_LIBRARY)
+endif(OCC_LIBRARY AND NOT OCC_LIBRARY_DIR)
 
 if(OCC_INCLUDE_DIR)
     file(STRINGS ${OCC_INCLUDE_DIR}/Standard_Version.hxx OCC_MAJOR
@@ -89,7 +89,7 @@ if(OCC_VERSION_STRING VERSION_GREATER_EQUAL "7.3.0")
 endif()
 
 foreach( libname ${OCC_LIBRARY_NAMES} )
-    find_library( ${libname} ${libname} ${OCC_LIBRARY_DIR} )
+    find_library( ${libname} ${libname} ${OCC_LIBRARY_DIR} NO_DEFAULT_PATH)
     set(OCC_LIBRARIES ${OCC_LIBRARIES} ${${libname}})
 endforeach()
 

cmake/external_projects/cgns.cmake

@@ -0,0 +1,30 @@
+if(WIN32)
+
+  ExternalProject_Add(project_win_cgns
+    URL ${CGNS_DOWNLOAD_URL_WIN}
+    UPDATE_COMMAND "" # Disable update
+    BUILD_IN_SOURCE 1
+    CONFIGURE_COMMAND ""
+    BUILD_COMMAND ""
+    INSTALL_COMMAND ${CMAKE_COMMAND} -E copy_directory . ${CMAKE_INSTALL_PREFIX}
+    LOG_DOWNLOAD 1
+    )
+
+  list(APPEND NETGEN_DEPENDENCIES project_win_cgns)
+endif(WIN32)
+
+if(APPLE)
+  ExternalProject_Add(project_mac_cgns
+    URL ${CGNS_DOWNLOAD_URL_MAC}
+    UPDATE_COMMAND "" # Disable update
+    BUILD_IN_SOURCE 1
+    CONFIGURE_COMMAND ""
+    BUILD_COMMAND ""
+    INSTALL_COMMAND ${CMAKE_COMMAND} -E copy_directory . ${CMAKE_INSTALL_PREFIX}
+    LOG_DOWNLOAD 1
+    )
+
+  list(APPEND NETGEN_DEPENDENCIES project_mac_cgns)
+  list(APPEND NETGEN_CMAKE_ARGS "-DCGNS_INCLUDE_DIR=${CMAKE_INSTALL_PREFIX}/Contents/Resources/include")
+  list(APPEND NETGEN_CMAKE_ARGS "-DCGNS_LIBRARY=${CMAKE_INSTALL_PREFIX}/Contents/MacOS/libcgns.dylib")
+endif(APPLE)

cmake/generate_version_file.cmake

@@ -0,0 +1,92 @@
+if(NOT BDIR)
+  set(BDIR ${CMAKE_CURRENT_BINARY_DIR})
+endif()
+
+find_package(Git REQUIRED)
+execute_process(COMMAND git describe --tags --match "v[0-9]*" --long --dirty WORKING_DIRECTORY ${CMAKE_CURRENT_LIST_DIR} OUTPUT_VARIABLE git_version_string RESULT_VARIABLE status ERROR_QUIET)
+
+if(status AND NOT status EQUAL 0)
+  if(EXISTS ${CMAKE_CURRENT_LIST_DIR}/../version.txt)
+    # for source package files (generated for ubuntu builds on launchpad) read the version from version.txt
+    if(EXISTS ${CMAKE_CURRENT_LIST_DIR}/../version.txt)
+      file(READ ${CMAKE_CURRENT_LIST_DIR}/../version.txt git_version_string )
+    else()
+      get_filename_component(git_version_string ${CMAKE_CURRENT_LIST_DIR}/.. NAME)
+      string(REGEX REPLACE "^netgen(.*)" "\\1" git_version_string "${git_version_string}")
+    endif()
+  else()
+    MESSAGE(WARNING "Could not determine git-version from source code - assuming 6.2.0.0")
+    set(git_version_string "v6.2.0.0")
+  endif()
+endif()
+
+string(REGEX REPLACE "^v([0-9]+)\\..*" "\\1" NETGEN_VERSION_MAJOR "${git_version_string}")
+string(REGEX REPLACE "^v[0-9]+\\.([0-9]+).*" "\\1" NETGEN_VERSION_MINOR "${git_version_string}")
+string(REGEX REPLACE "^v[0-9]+\\.[0-9]+\\.([0-9]+).*" "\\1" NETGEN_VERSION_PATCH "${git_version_string}")
+string(REGEX REPLACE "^v[0-9]+\\.[0-9]+\\.[0-9]+\\-([0-9]+).*" "\\1" NETGEN_VERSION_TWEAK "${git_version_string}")
+string(REGEX REPLACE "^v[0-9]+\\.[0-9]+\\.[0-9]+\\-[0-9]+\\-([0-9a-z]+).*" "\\1" NETGEN_VERSION_HASH "${git_version_string}")
+
+set(NETGEN_VERSION_SHORT ${NETGEN_VERSION_MAJOR}.${NETGEN_VERSION_MINOR}.${NETGEN_VERSION_PATCH})
+set(NETGEN_VERSION_LONG ${NETGEN_VERSION_SHORT}-${NETGEN_VERSION_TWEAK}-${NETGEN_VERSION_HASH})
+
+if(NETGEN_VERSION_TWEAK)
+  # no release version - nightly build
+  set(NETGEN_VERSION ${NETGEN_VERSION_LONG})
+else()
+  # TWEAK is 0 -> current version has a tag assigned
+  set(NETGEN_VERSION ${NETGEN_VERSION_SHORT})
+endif()
+
+set(NETGEN_VERSION_LONG ${NETGEN_VERSION_SHORT}-${NETGEN_VERSION_TWEAK}-${NETGEN_VERSION_HASH})
+
+set(version_file ${BDIR}/netgen_version.hpp)
+set(new_version_file_string "\
+#ifndef NETGEN_VERSION_HPP_INCLUDED
+#define NETGEN_VERSION_HPP_INCLUDED
+#define NETGEN_VERSION \"${NETGEN_VERSION}\"
+#define NETGEN_VERSION_MAJOR ${NETGEN_VERSION_MAJOR}
+#define NETGEN_VERSION_MINOR ${NETGEN_VERSION_MINOR}
+#define NETGEN_VERSION_PATCH ${NETGEN_VERSION_PATCH}
+#define NETGEN_VERSION_TWEAK ${NETGEN_VERSION_TWEAK}
+#define NETGEN_VERSION_HASH \"${NETGEN_VERSION_HASH}\"
+#endif // NETGEN_VERSION_HPP_INCLUDED
+")
+if(EXISTS ${version_file})
+  file(READ ${version_file} old_version_file_string )
+  if(${old_version_file_string} STREQUAL ${new_version_file_string})
+  else()
+    file(WRITE ${BDIR}/netgen_version.hpp ${new_version_file_string})
+  endif()
+else()
+    file(WRITE ${BDIR}/netgen_version.hpp ${new_version_file_string})
+endif()
+
+file(GENERATE OUTPUT netgen_config.hpp CONTENT
+"\
+#ifndef NETGEN_CONFIG_HPP_INCLUDED___
+#define NETGEN_CONFIG_HPP_INCLUDED___
+
+#define NETGEN_USE_NATIVE_ARCH          $<BOOL:${USE_NATIVE_ARCH}>
+#define NETGEN_USE_GUI                  $<BOOL:${USE_GUI}>
+#define NETGEN_USE_PYTHON               $<BOOL:${USE_PYTHON}>
+#define NETGEN_USE_MPI                  $<BOOL:${USE_MPI}}>
+#define NETGEN_USE_MPI4PY               $<BOOL:${USE_MPI4PY}>
+#define NETGEN_USE_OCC                  $<BOOL:${USE_OCC}}>
+#define NETGEN_USE_JPEG                 $<BOOL:${USE_JPEG}}>
+#define NETGEN_USE_MPEG                 $<BOOL:${USE_MPEG}}>
+#define NETGEN_USE_CGNS                 $<BOOL:${USE_CGNS}}>
+#define NETGEN_USE_NUMA                 $<BOOL:${USE_NUMA}}>
+#define NETGEN_INTEL_MIC                $<BOOL:${USE_INTEL_MIC}}>
+#define NETGEN_INSTALL_PROFILES         $<BOOL:${INSTALL_PROFILES}>
+#define NETGEN_USE_CCACHE               $<BOOL:${USE_CCACHE}}>
+#define NETGEN_USE_INTERNAL_TCL         $<BOOL:${USE_INTERNAL_TCL}>
+#define NETGEN_ENABLE_UNIT_TESTS        $<BOOL:${ENABLE_UNIT_TESTS}>
+#define NETGEN_ENABLE_CPP_CORE_GUIDELINES_CHECK $<BOOL:${ENABLE_CPP_CORE_GUIDELINES_CHECK}>
+#define NETGEN_USE_SPDLOG               $<BOOL:${USE_SPDLOG}>
+#define NETGEN_DEBUG_LOG                $<BOOL:${DEBUG_LOG}>
+#define NETGEN_USE_CHECK_RANGE          $<BOOL:${CHECK_RANGE}>
+#define NETGEN_BUILD_STUB_FILES         $<BOOL:${BUILD_STUB_FILES}>
+#define NETGEN_BUILD_FOR_CONDA          $<BOOL:${BUILD_FOR_CONDA}>
+
+#endif // NETGEN_CONFIG_HPP_INCLUDED___
+")

external_dependencies/pybind11

@@ -1 +1 @@
-Subproject commit deb3cb238a9f299d7c57f810165e90a1b14b3e6f
+Subproject commit c16da993094988141101ac4d96a9b2c92f9ac714

libsrc/core/CMakeLists.txt

@@ -11,8 +11,11 @@ add_library(ngcore SHARED
   table.cpp
   taskmanager.cpp
   utils.cpp
+  version.cpp
   )
 
+target_compile_options(ngcore PUBLIC "${NG_COMPILE_FLAGS}")
+
 # Pybind11 2.3 Issue https://github.com/pybind/pybind11/issues/1604
 if ("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
   target_compile_options(ngcore PUBLIC -fsized-deallocation -faligned-allocation)
@@ -39,6 +42,11 @@ if(CHECK_RANGE OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL
   target_compile_definitions(ngcore PUBLIC NETGEN_ENABLE_CHECK_RANGE)
 endif(CHECK_RANGE OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
 
+if(TRACE_MEMORY OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
+  target_compile_definitions(ngcore PUBLIC NETGEN_TRACE_MEMORY)
+endif(TRACE_MEMORY OR CMAKE_BUILD_TYPE STREQUAL "Debug" OR CMAKE_BUILD_TYPE STREQUAL "DEBUG")
+
+
 if(USE_SPDLOG)
   include_directories(${SPDLOG_INCLUDE_DIR})
   install(DIRECTORY ${SPDLOG_INCLUDE_DIR}
@@ -51,20 +59,29 @@ if(USE_SPDLOG)
   endif(DEBUG_LOG)
 endif(USE_SPDLOG)
 
+if(USE_NUMA)
+    find_library(NUMA_LIBRARY libnuma.so)
+    target_compile_definitions(ngcore PUBLIC USE_NUMA)
+    target_link_libraries(ngcore PRIVATE ${NUMA_LIBRARY})
+endif(USE_NUMA)
+
 install(TARGETS ngcore DESTINATION ${NG_INSTALL_DIR} COMPONENT netgen)
 
-target_link_libraries(ngcore PUBLIC netgen_mpi PRIVATE netgen_python ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(ngcore PUBLIC netgen_mpi PRIVATE "$<BUILD_INTERFACE:netgen_python>" ${CMAKE_THREAD_LIBS_INIT})
 
 install(FILES ngcore.hpp archive.hpp type_traits.hpp version.hpp ngcore_api.hpp logging.hpp
   exception.hpp symboltable.hpp paje_trace.hpp utils.hpp profiler.hpp mpi_wrapper.hpp
   array.hpp taskmanager.hpp concurrentqueue.h localheap.hpp python_ngcore.hpp flags.hpp
-  xbool.hpp signal.hpp bitarray.hpp table.hpp hashtable.hpp
+  xbool.hpp signal.hpp bitarray.hpp table.hpp hashtable.hpp ranges.hpp
+  simd.hpp simd_avx.hpp simd_avx512.hpp simd_generic.hpp simd_sse.hpp simd_arm64.hpp
   DESTINATION ${NG_INSTALL_DIR_INCLUDE}/core COMPONENT netgen_devel)
 
 if(ENABLE_CPP_CORE_GUIDELINES_CHECK)
   set_target_properties(ngcore PROPERTIES CXX_CLANG_TIDY "${DO_CLANG_TIDY}")
 endif(ENABLE_CPP_CORE_GUIDELINES_CHECK)
 
+add_dependencies(ngcore ng_generate_version_file)
+
 if(USE_PYTHON)
   pybind11_add_module(pyngcore SHARED python_ngcore_export.cpp)
   target_link_libraries(pyngcore PUBLIC ngcore netgen_python)

libsrc/core/archive.cpp

@@ -1,5 +1,6 @@
 
 #include "archive.hpp"
+#include "version.hpp"
 
 #ifndef WIN32
 #include <cxxabi.h>
@@ -7,18 +8,6 @@
 
 namespace ngcore
 {
-  // clang-tidy should ignore this static object
-  static std::map<std::string, VersionInfo> library_versions;  // NOLINT
-  std::map<std::string, VersionInfo>& Archive :: GetLibraryVersions()
-  {
-    return library_versions;
-  }
-  const VersionInfo& GetLibraryVersion(const std::string& library)
-  { return library_versions[library]; }
-
-  void SetLibraryVersion(const std::string& library, const VersionInfo& version)
-  { library_versions[library] = version; }
-
   // clang-tidy should ignore this static object
   static std::unique_ptr<std::map<std::string, detail::ClassArchiveInfo>> type_register;  // NOLINT
   const detail::ClassArchiveInfo& Archive :: GetArchiveRegister(const std::string& classname)

libsrc/core/archive.hpp

@@ -30,9 +30,6 @@ namespace pybind11
 
 namespace ngcore
 {
-  // Libraries using this archive can store their version here to implement backwards compatibility
-  NGCORE_API const VersionInfo& GetLibraryVersion(const std::string& library);
-  NGCORE_API void SetLibraryVersion(const std::string& library, const VersionInfo& version);
 
   class NGCORE_API Archive;
 
@@ -156,6 +153,7 @@ namespace ngcore
     virtual void NeedsVersion(const std::string& /*unused*/, const std::string& /*unused*/) {}
 
     // Pure virtual functions that have to be implemented by In-/OutArchive
+    virtual Archive & operator & (float & d) = 0;
     virtual Archive & operator & (double & d) = 0;
     virtual Archive & operator & (int & i) = 0;
     virtual Archive & operator & (long & i) = 0;
@@ -570,9 +568,10 @@ namespace ngcore
 
     virtual void FlushBuffer() {}
 
-  protected:
+    bool parallel = false;
-    static std::map<std::string, VersionInfo>& GetLibraryVersions();
+    bool IsParallel() const { return parallel; }
-
+    void SetParallel (bool _parallel) { parallel = _parallel; }
+    
   private:
     template<typename T, typename ... Bases>
     friend class RegisterClassForArchive;
@@ -680,6 +679,8 @@ namespace ngcore
     BinaryOutArchive& operator=(BinaryOutArchive&&) = delete;
 
     using Archive::operator&;
+    Archive & operator & (float & f) override
+    { return Write(f); }
     Archive & operator & (double & d) override
     { return Write(d); }
     Archive & operator & (int & i) override
@@ -687,7 +688,11 @@ namespace ngcore
     Archive & operator & (short & i) override
     { return Write(i); }
     Archive & operator & (long & i) override
-    { return Write(i); }
+    {
+      // for platform independence
+      int64_t tmp = i;
+      return Write(tmp);
+    }
     Archive & operator & (size_t & i) override
     { return Write(i); }
     Archive & operator & (unsigned char & i) override
@@ -725,14 +730,13 @@ namespace ngcore
     template <typename T>
     Archive & Write (T x)
     {
+      static_assert(sizeof(T) < BUFFERSIZE, "Cannot write large types with this function!");
       if (unlikely(ptr > BUFFERSIZE-sizeof(T)))
         {
           stream->write(&buffer[0], ptr);
-          *reinterpret_cast<T*>(&buffer[0]) = x; // NOLINT
+          ptr = 0;
-          ptr = sizeof(T);
-          return *this;
         }
-      *reinterpret_cast<T*>(&buffer[ptr]) = x; // NOLINT
+      memcpy(&buffer[ptr], &x, sizeof(T));
       ptr += sizeof(T);
       return *this;
     }
@@ -751,6 +755,8 @@ namespace ngcore
       : BinaryInArchive(std::make_shared<std::ifstream>(filename)) { ; }
 
     using Archive::operator&;
+    Archive & operator & (float & f) override
+    { Read(f); return *this; }
     Archive & operator & (double & d) override
     { Read(d); return *this; }
     Archive & operator & (int & i) override
@@ -758,7 +764,12 @@ namespace ngcore
     Archive & operator & (short & i) override
     { Read(i); return *this; }
     Archive & operator & (long & i) override
-    { Read(i); return *this; }
+    {
+      int64_t tmp;
+      Read(tmp);
+      i = tmp;
+      return *this;
+    }
     Archive & operator & (size_t & i) override
     { Read(i); return *this; }
     Archive & operator & (unsigned char & i) override
@@ -815,6 +826,8 @@ namespace ngcore
       TextOutArchive(std::make_shared<std::ofstream>(filename)) { }
 
     using Archive::operator&;
+    Archive & operator & (float & f) override
+    { *stream << f << '\n'; return *this; }
     Archive & operator & (double & d) override
     { *stream << d << '\n'; return *this; }
     Archive & operator & (int & i) override
@@ -866,6 +879,8 @@ namespace ngcore
       : TextInArchive(std::make_shared<std::ifstream>(filename)) {}
 
     using Archive::operator&;
+    Archive & operator & (float & f) override
+    { *stream >> f; return *this; }
     Archive & operator & (double & d) override
     { *stream >> d; return *this; }
     Archive & operator & (int & i) override
@@ -912,6 +927,55 @@ namespace ngcore
     }
   };
 
+  // HashArchive =================================================================
+  // This class enables to easily create hashes for archivable objects by xoring
+  // threw its data
+
+  class NGCORE_API HashArchive : public Archive
+  {
+    size_t hash_value = 0;
+    char* h;
+    int offset = 0;
+  public:
+    HashArchive() : Archive(true)
+      { h = (char*)&hash_value; }
+
+    using Archive::operator&;
+    Archive & operator & (float & f) override { return ApplyHash(f); }
+    Archive & operator & (double & d) override { return ApplyHash(d); }
+    Archive & operator & (int & i) override { return ApplyHash(i); }
+    Archive & operator & (short & i) override { return ApplyHash(i); }
+    Archive & operator & (long & i) override { return ApplyHash(i); }
+    Archive & operator & (size_t & i) override { return ApplyHash(i); }
+    Archive & operator & (unsigned char & i) override { return ApplyHash(i); }
+    Archive & operator & (bool & b) override { return ApplyHash(b); }
+    Archive & operator & (std::string & str) override
+    { for(auto c : str) ApplyHash(c);  return *this; }
+    Archive & operator & (char *& str) override
+    { char* s = str; while(*s != '\0') ApplyHash(*(s++)); return *this; }
+
+    // HashArchive can be used in const context
+    template<typename T>
+      Archive & operator& (const T& val) const
+    { return (*this) & const_cast<T&>(val); }
+
+    size_t GetHash() const { return hash_value; }
+
+  private:
+    template<typename T>
+      Archive& ApplyHash(T val)
+    {
+      size_t n = sizeof(T);
+      char* pval = (char*)&val;
+      for(size_t i = 0; i < n; i++)
+        {
+          h[offset++] ^= pval[i];
+          offset %= 8;
+        }
+      return *this;
+    }
+  };
+
 } // namespace ngcore
 
 #endif // NETGEN_CORE_ARCHIVE_HPP

libsrc/core/array.hpp

@@ -11,6 +11,7 @@
 #include "archive.hpp"
 #include "exception.hpp"
 #include "localheap.hpp"
+#include "profiler.hpp"
 #include "utils.hpp"
 
 namespace ngcore
@@ -39,7 +40,7 @@ namespace ngcore
   };
 
   template <typename ... ARGS>
-  ostream & operator<< (ostream & ost, Tuple<ARGS...> tup)
+  ostream & operator<< (ostream & ost, Tuple<ARGS...> /* tup */)
   {
     return ost;
   }
@@ -211,6 +212,20 @@ namespace ngcore
   template <typename  T>
   constexpr T IndexBASE () { return T(0); }
 
+
+  class IndexFromEnd
+  {
+    ptrdiff_t i;
+  public:
+    constexpr IndexFromEnd (ptrdiff_t ai) : i(ai) { }
+    IndexFromEnd operator+ (ptrdiff_t inc) const { return i+inc; }
+    IndexFromEnd operator- (ptrdiff_t dec) const { return i-dec; }
+    // operator ptrdiff_t () const { return i; }
+    ptrdiff_t Value() const { return i; }
+  };
+
+  constexpr IndexFromEnd END(0);
+  
   
   template <class T, class IndexType = size_t> class FlatArray;
 
@@ -268,7 +283,7 @@ namespace ngcore
     NETGEN_INLINE T & First() { return first; }
     NETGEN_INLINE T & Next() { return next; }
     NETGEN_INLINE auto Size() const { return next-first; }
-    NETGEN_INLINE T operator[] (T i) const { return first+i; }
+    NETGEN_INLINE T operator[] (size_t i) const { return first+i; }
     NETGEN_INLINE bool Contains (T i) const { return ((i >= first) && (i < next)); }
     NETGEN_INLINE T_Range Modify(int inc_beg, int inc_end) const
     { return T_Range(first+inc_beg, next+inc_end); }
@@ -559,6 +574,12 @@ namespace ngcore
       return FlatArray<T> (end-start, data+start);
     }
 
+    /// takes range starting from position start of end-start elements
+    NETGEN_INLINE FlatArray<T> Range (size_t start, IndexFromEnd indend) const
+    {
+      return this->Range(start, size_t(Size()+indend.Value()));
+    }
+    
     /// takes range starting from position start of end-start elements
     NETGEN_INLINE FlatArray<T> Range (T_Range<size_t> range) const
     {
@@ -599,9 +620,25 @@ namespace ngcore
   template <typename T>
   FlatArray<T> View (FlatArray<T> fa) { return fa; }
 
+  template <typename T, typename TI>
+  auto Max (FlatArray<T,TI> array, T max = std::numeric_limits<T>::min()) -> T
+  {
+    for (auto & v : array)
+      if (v > max) max = v;
+    return max;
+  }
+
+  template <typename T, typename TI>
+  auto Min (FlatArray<T,TI> array, T min = std::numeric_limits<T>::max()) -> T
+  {
+    for (auto & v : array)
+      if (v < min) min = v;
+    return min;
+  }
+  
   /// print array
-  template <class T>
+  template <class T, class TIND>
-  inline ostream & operator<< (ostream & s, const FlatArray<T> & a)
+  inline ostream & operator<< (ostream & s, const FlatArray<T, TIND> & a)
   {
     for (auto i : a.Range())
       s << i << ": " << a[i] << "\n";
@@ -638,6 +675,7 @@ namespace ngcore
     /// that's the data we have to delete, nullptr for not owning the memory
     T * mem_to_delete;
 
+
     using FlatArray<T,IndexType>::size;
     using FlatArray<T,IndexType>::data;
     using FlatArray<T,IndexType>::BASE;
@@ -682,6 +720,8 @@ namespace ngcore
 
     NETGEN_INLINE Array (Array && a2) 
     {
+      mt.Swap(sizeof(T) * allocsize, a2.mt, sizeof(T) * a2.allocsize);
+
       size = a2.size; 
       data = a2.data;
       allocsize = a2.allocsize;
@@ -696,10 +736,15 @@ namespace ngcore
     NETGEN_INLINE explicit Array (const Array & a2)
       : FlatArray<T,IndexType> (a2.Size(), a2.Size() ? new T[a2.Size()] : nullptr)
     {
-      allocsize = size;
+      if constexpr (std::is_copy_assignable<T>::value)
-      mem_to_delete = data;
+        {
-      for (size_t i = 0; i < size; i++)
+          allocsize = size;
-        data[i] = a2.data[i];
+          mem_to_delete = data;
+          for (size_t i = 0; i < size; i++)
+            data[i] = a2.data[i];
+        }
+      else
+        throw Exception(std::string("cannot copy-construct Array of type ") + typeid(T).name());        
     }
 
     
@@ -747,6 +792,8 @@ namespace ngcore
     /// if responsible, deletes memory
     NETGEN_INLINE ~Array()
     {
+      if(mem_to_delete)
+        mt.Free(sizeof(T)*allocsize);
       delete [] mem_to_delete;
     }
 
@@ -801,6 +848,8 @@ namespace ngcore
     /// assigns memory from local heap
     NETGEN_INLINE const Array & Assign (size_t asize, LocalHeap & lh)
     {
+      if(mem_to_delete)
+        mt.Free(sizeof(T)*allocsize);
       delete [] mem_to_delete;
       size = allocsize = asize;
       data = lh.Alloc<T> (asize);
@@ -858,7 +907,7 @@ namespace ngcore
       size++;
     }
     
-    NETGEN_INLINE Array<T> & operator += (const T & el)
+    NETGEN_INLINE Array & operator += (const T & el)
     {
       Append (el);
       return *this;
@@ -908,6 +957,8 @@ namespace ngcore
     /// Deallocate memory
     NETGEN_INLINE void DeleteAll ()
     {
+      if(mem_to_delete)
+        mt.Free(sizeof(T)*allocsize);
       delete [] mem_to_delete;
       mem_to_delete = NULL;
       data = 0;
@@ -924,16 +975,24 @@ namespace ngcore
     /// array copy
     NETGEN_INLINE Array & operator= (const Array & a2)
     {
-      SetSize0 ();
+      if constexpr (std::is_copy_assignable<T>::value)
-      SetSize (a2.Size());
+        {
-      for (size_t i = 0; i < size; i++)
+          SetSize0 ();
-        data[i] = a2.data[i];
+          SetSize (a2.Size());
-      return *this;
+          for (size_t i = 0; i < size; i++)
+            data[i] = a2.data[i];
+          return *this;
+        }
+      else
+        throw Exception(std::string("cannot copy Array of type ") + typeid(T).name());
     }
 
+    
     /// steal array 
     NETGEN_INLINE Array & operator= (Array && a2)
     {
+      mt.Swap(sizeof(T)*allocsize, a2.mt, sizeof(T)*a2.allocsize);
+
       ngcore::Swap (size, a2.size);
       ngcore::Swap (data, a2.data);
       ngcore::Swap (allocsize, a2.allocsize);
@@ -1007,16 +1066,26 @@ namespace ngcore
     
     NETGEN_INLINE void Swap (Array & b)
     {
+      mt.Swap(sizeof(T) * allocsize, b.mt, sizeof(T) * b.allocsize);
+
       ngcore::Swap (size, b.size);
       ngcore::Swap (data, b.data);
       ngcore::Swap (allocsize, b.allocsize);
       ngcore::Swap (mem_to_delete, b.mem_to_delete);
     }
 
+    NETGEN_INLINE void StartMemoryTracing () const
+    {
+      mt.Alloc(sizeof(T) * allocsize);
+    }
+
+    const MemoryTracer& GetMemoryTracer() const { return mt; }
+
   private:
 
     /// resize array, at least to size minsize. copy contents
     NETGEN_INLINE void ReSize (size_t minsize);
+    MemoryTracer mt;
   };
 
   
@@ -1029,6 +1098,7 @@ namespace ngcore
     
     T * hdata = data;
     data = new T[nsize];
+    mt.Alloc(sizeof(T) * nsize);
 
     if (hdata)
       {
@@ -1041,6 +1111,8 @@ namespace ngcore
         else
           for (size_t i = 0; i < mins; i++) data[i] = std::move(hdata[i]);
 #endif
+        if(mem_to_delete)
+          mt.Free(sizeof(T) * allocsize);
         delete [] mem_to_delete;
       }
 
@@ -1124,6 +1196,14 @@ namespace ngcore
         data[cnt++] = val;
     }
   
+    template <typename T2>
+    ArrayMem (const BaseArrayObject<T2> & a2)
+      : ArrayMem (a2.Size())
+    {
+      for (size_t i : ngcore::Range(size))
+        data[i] = a2[i];
+    }
+
     
     ArrayMem & operator= (const T & val)
     {
@@ -1179,7 +1259,7 @@ namespace ngcore
 
 
   template <typename ... ARGS>
-  size_t ArraySize (Tuple<ARGS...> tup)  
+  size_t ArraySize (Tuple<ARGS...> /* tup */)  
   { return 0;}
   
   template <typename ... ARGS>
@@ -1192,7 +1272,7 @@ namespace ngcore
 
   
   template <typename T, typename ... ARGS>
-  void StoreToArray (FlatArray<T> a, Tuple<ARGS...> tup) { ; }
+  void StoreToArray (FlatArray<T> /* a */, Tuple<ARGS...> /* tup */) { ; }
   
   template <typename T, typename ... ARGS>
   void StoreToArray (FlatArray<T> a, Tuple<int,ARGS...> tup)
@@ -1284,7 +1364,7 @@ namespace ngcore
 
   /// bubble sort array
   template <class T, class S>
-  inline void BubbleSort (FlatArray<T> data, FlatArray<S> slave)
+  inline void BubbleSort (FlatArray<T> data, FlatArray<S> index)
   {
     for (size_t i = 0; i < data.Size(); i++)
       for (size_t j = i+1; j < data.Size(); j++)
@@ -1294,9 +1374,9 @@ namespace ngcore
 	    data[i] = data[j];
 	    data[j] = hv;
 
-	    S hvs = slave[i];
+	    S hvs = index[i];
-	    slave[i] = slave[j];
+	    index[i] = index[j];
-	    slave[j] = hvs;
+	    index[j] = hvs;
 	  }
   }
 

libsrc/core/bitarray.cpp

@@ -36,10 +36,15 @@ namespace ngcore
   void BitArray :: SetSize (size_t asize)
   {
     if (size == asize) return;
-    if (owns_data) delete [] data;
+    if (owns_data)
+      {
+        delete [] data;
+        mt.Free(Addr(size)+1);
+      }
 
     size = asize;
     data = new unsigned char [Addr (size)+1];
+    mt.Alloc(Addr(size)+1);
   }
 
   BitArray & BitArray :: Set () throw()
@@ -83,6 +88,18 @@ namespace ngcore
     return *this;
   }
 
+  bool BitArray :: operator==(const BitArray& other) const
+  {
+    if(size != other.Size())
+      return false;
+    for(auto i : Range(size/CHAR_BIT))
+      if(data[i] != other.data[i])
+        return false;
+    for(auto i : Range(size%CHAR_BIT))
+      if(Test(i + CHAR_BIT * (size/CHAR_BIT)) != other.Test(i + CHAR_BIT * (size/CHAR_BIT)))
+        return false;
+    return true;
+  }
 
   BitArray & BitArray :: operator= (const BitArray & ba2)
   {
@@ -115,29 +132,52 @@ namespace ngcore
     return cnt;
   }
 
-  Archive & operator & (Archive & archive, BitArray & ba)
+  void BitArray :: DoArchive(Archive& archive)
   {
-    if (archive.Output())
+    if(archive.GetVersion("netgen") >= "v6.2.2007-62")
       {
-        archive << ba.Size();
+        archive.NeedsVersion("netgen", "v6.2.2007-62");
-        for (size_t i = 0; i < ba.Size(); i++)
+        auto size = Size();
-          archive << ba[i];
+        archive & size;
+        if(archive.Input())
+          SetSize(size);
+        if(archive.GetVersion("netgen") < "v6.2.2009-20")
+          archive.Do(data, size/CHAR_BIT+1);
+        else
+          {
+            archive.NeedsVersion("netgen", "v6.2.2009-20");
+            archive.Do(data, size/CHAR_BIT);
+            for(size_t i = 0; i < size%CHAR_BIT; i++)
+              {
+                size_t index =  CHAR_BIT * (size/CHAR_BIT) + i;
+                bool b = Test(index);
+                archive & b;
+                b ? SetBit(index) : Clear(index);
+              }
+          }
       }
     else
       {
-        int size;
+        if (archive.Output())
-        archive & size;
-        ba.SetSize (size);
-        ba.Clear();
-        for (size_t i = 0; i < size; i++)
           {
-            bool b;
+            throw Exception("should not get here");
-            archive & b;
+            archive << Size();
-            if (b) ba.SetBit(i);
+            for (size_t i = 0; i < Size(); i++)
+              archive << (*this)[i];
+          }
+        else
+          {
+            size_t size;
+            archive & size;
+            SetSize (size);
+            Clear();
+            for (size_t i = 0; i < size; i++)
+              {
+                bool b;
+                archive & b;
+                if (b) SetBit(i);
+              }
           }
       }
-    return archive;
   }
-
+} // namespace ngcore
-
-}

libsrc/core/bitarray.hpp

@@ -131,6 +131,7 @@ public:
     return Test(i);
   }
 
+  NGCORE_API bool operator==(const BitArray& other) const;
 
   /// invert all bits
   NGCORE_API BitArray & Invert ();
@@ -145,6 +146,18 @@ public:
   NGCORE_API BitArray & operator= (const BitArray & ba2);
 
   NGCORE_API size_t NumSet () const;
+
+  NGCORE_API void DoArchive(Archive& archive);
+  
+  NGCORE_API auto * Data() const { return data; }
+
+  const MemoryTracer& GetMemoryTracer() const { return mt; }
+  void StartMemoryTracing() const
+  {
+    if(owns_data)
+      mt.Alloc(Addr(size)+1);
+  }
+
 private:
   ///
   unsigned char Mask (size_t i) const
@@ -154,6 +167,7 @@ private:
   size_t Addr (size_t i) const
   { return (i / CHAR_BIT); }
 
+  MemoryTracer mt;
 };
 
 
@@ -190,11 +204,8 @@ private:
     return res;
   }
 
-
   NGCORE_API std::ostream & operator<<(std::ostream & s, const BitArray & ba);
 
-  NGCORE_API Archive & operator & (Archive & archive, BitArray & ba);
-
 } // namespace ngcore
 
 #endif // NETGEN_CORE_BITARRAY

libsrc/core/exception.cpp

@@ -217,9 +217,12 @@ static void ngcore_signal_handler(int sig)
 // register signal handler when library is loaded
 static bool dummy = []()
 {
-    signal(SIGABRT, ngcore_signal_handler);
+    if(getenv("NG_BACKTRACE"))
-    signal(SIGILL, ngcore_signal_handler);
+    {
-    signal(SIGSEGV, ngcore_signal_handler);
+      signal(SIGABRT, ngcore_signal_handler);
+      signal(SIGILL, ngcore_signal_handler);
+      signal(SIGSEGV, ngcore_signal_handler);
+    }
     return true;
 }();
 

libsrc/core/flags.cpp

@@ -51,6 +51,10 @@ namespace ngcore
 	auto lflags = flags.GetFlagsFlag (i, name);
 	SetFlag (name, lflags);
       }
+    for(auto i : Range(flags.anyflags.Size()))
+      {
+        SetFlag(flags.anyflags.GetName(i), flags.anyflags[i]);
+      }
   }
   
   Flags :: Flags (Flags && flags)
@@ -178,7 +182,11 @@ namespace ngcore
     return *this;
   }
 
-
+  Flags & Flags :: SetFlag (const string & name, const std::any & val)
+  {
+    anyflags.Set(name, val);
+    return *this;
+  }
 
   string Flags :: GetStringFlag (const string & name, const char * def) const
   {
@@ -279,6 +287,14 @@ namespace ngcore
       }
   }
 
+  const std::any& Flags:: GetAnyFlag(const std::string& name) const
+  {
+    if(anyflags.Used(name))
+      return anyflags[name];
+    static std::any empty;
+    return empty;
+  }
+
   bool Flags :: StringFlagDefined (const string & name) const
   {
     return strflags.Used (name);
@@ -304,6 +320,11 @@ namespace ngcore
     return numlistflags.Used (name);
   }
 
+  bool Flags :: AnyFlagDefined (const string& name) const
+  {
+    return anyflags.Used(name);
+  }
+
   void Flags :: SaveFlags (ostream & str) const
   {
     for (int i = 0; i < strflags.Size(); i++)

libsrc/core/flags.hpp

@@ -11,6 +11,7 @@
 #include <iostream>
 #include <memory>
 #include <string>
+#include <any>
 
 #include "array.hpp"
 #include "symboltable.hpp"
@@ -38,6 +39,8 @@ namespace ngcore
     SymbolTable<std::shared_ptr<Array<double>>> numlistflags;
     /// flags list flags
     SymbolTable<Flags> flaglistflags;
+    /// any object can be stored as a flag
+    SymbolTable<std::any> anyflags;
   public:
     /// no flags
     Flags ();
@@ -94,6 +97,8 @@ namespace ngcore
     Flags &  SetFlag (const std::string & name, const Array<std::string> & val);
     /// Sets double array flag
     Flags &  SetFlag (const std::string & name, const Array<double> & val);
+    /// Sets any flag
+    Flags &  SetFlag(const std::string& name, const std::any& val);
 
 
     Flags SetFlag (const char * name, bool b = true) &&;
@@ -135,6 +140,7 @@ namespace ngcore
     const Array<double> & GetNumListFlag (const std::string & name) const;
     /// Returns flag list flag, empty flag if not exist
     const Flags & GetFlagsFlag (const std::string & name) const;
+    const std::any& GetAnyFlag (const std::string& name) const;
 
 
     /// Test, if string flag is defined
@@ -147,6 +153,7 @@ namespace ngcore
     bool StringListFlagDefined (const std::string & name) const;
     /// Test, if num list flag is defined
     bool NumListFlagDefined (const std::string & name) const;
+    bool AnyFlagDefined (const std::string& name) const;
 
     /// number of string flags
     int GetNStringFlags () const { return strflags.Size(); }

libsrc/core/hashtable.hpp

@@ -174,6 +174,13 @@ namespace ngcore
     {
       return MakeTupleFromInt<N>()(*this);
     }
+
+    bool Contains (T val)
+    {
+      for (int j = 0; j < N; j++)
+        if (i[j] == val) return true;
+      return false;
+    }
   };
 
   /// sort 2 integers
@@ -727,6 +734,11 @@ namespace ngcore
       acont = cont[pos];
     }
 
+    T GetData (size_t pos) const
+    {
+      return cont[pos];
+    }
+
     std::pair<T_HASH,T> GetBoth (size_t pos) const
     {
       return std::pair<T_HASH,T> (hash[pos], cont[pos]);

libsrc/core/localheap.hpp

@@ -198,9 +198,9 @@ public:
       return reinterpret_cast<T*> (oldp);
     }
 
-    virtual void Delete(void* p) {}
+  virtual void Delete(void* /* p */) {}
 
-    virtual void ArrayDelete(void* p) {}
+  virtual void ArrayDelete(void* /* p */) {}
   private: 
     ///
 #ifndef __CUDA_ARCH__
@@ -211,7 +211,7 @@ public:
 
   public:
     /// free memory (dummy function)
-    NETGEN_INLINE void Free (void * data) throw () 
+  NETGEN_INLINE void Free (void * /* data */) throw () 
     {
       ;
     }

libsrc/core/mpi_wrapper.hpp

@@ -8,6 +8,7 @@
 
 #include "array.hpp"
 #include "exception.hpp"
+#include "profiler.hpp"
 
 namespace ngcore
 {
@@ -25,6 +26,9 @@ namespace ngcore
   template <> struct MPI_typetrait<char> {
     static MPI_Datatype MPIType () { return MPI_CHAR; } };  
 
+  template <> struct MPI_typetrait<signed char> {
+    static MPI_Datatype MPIType () { return MPI_CHAR; } };  
+  
   template <> struct MPI_typetrait<unsigned char> {
     static MPI_Datatype MPIType () { return MPI_CHAR; } };  
 
@@ -43,6 +47,10 @@ namespace ngcore
     return MPI_typetrait<T>::MPIType();
   }
   
+  template <class T>
+  inline MPI_Datatype GetMPIType (T &) {
+    return GetMPIType<T>();
+  }
 
   class NgMPI_Comm
   {
@@ -59,6 +67,17 @@ namespace ngcore
     NgMPI_Comm (MPI_Comm _comm, bool owns = false)
       : comm(_comm), valid_comm(true)
     {
+      int flag;
+      MPI_Initialized (&flag);
+      if (!flag)
+        {
+          valid_comm = false;
+          refcount = nullptr;
+          rank = 0;
+          size = 1;
+          return;
+        }
+
       if (!owns)
         refcount = nullptr;
       else
@@ -89,6 +108,11 @@ namespace ngcore
           MPI_Comm_free(&comm);
     }
 
+    bool ValidCommunicator() const
+    {
+      return valid_comm;
+    }
+    
     NgMPI_Comm & operator= (const NgMPI_Comm & c)
     {
       if (refcount)
@@ -117,6 +141,7 @@ namespace ngcore
     int Rank() const { return rank; }
     int Size() const { return size; }
     void Barrier() const {
+      static Timer t("MPI - Barrier"); RegionTimer reg(t);
       if (size > 1) MPI_Barrier (comm);
     }
     
@@ -127,9 +152,13 @@ namespace ngcore
     void Send (T & val, int dest, int tag) const {
       MPI_Send (&val, 1, GetMPIType<T>(), dest, tag, comm);
     }
+
+    void Send (const std::string & s, int dest, int tag) const {
+      MPI_Send( const_cast<char*> (&s[0]), s.length(), MPI_CHAR, dest, tag, comm);
+    }
     
-    template<typename T, typename T2 = decltype(GetMPIType<T>())>
+    template<typename T, typename TI, typename T2 = decltype(GetMPIType<T>())>
-    void Send(FlatArray<T> s, int dest, int tag) const {
+    void Send(FlatArray<T,TI> s, int dest, int tag) const {
       MPI_Send (s.Data(), s.Size(), GetMPIType<T>(), dest, tag, comm);
     }
     
@@ -138,13 +167,24 @@ namespace ngcore
       MPI_Recv (&val, 1, GetMPIType<T>(), src, tag, comm, MPI_STATUS_IGNORE);
     }
 
-    template <typename T, typename T2 = decltype(GetMPIType<T>())>
+    void Recv (std::string & s, int src, int tag) const {    
-    void Recv (FlatArray <T> s, int src, int tag) const {
+      MPI_Status status;
+      int len;
+      MPI_Probe (src, tag, comm, &status);
+      MPI_Get_count (&status, MPI_CHAR, &len);
+      // s.assign (len, ' ');
+      s.resize (len);
+      MPI_Recv( &s[0], len, MPI_CHAR, src, tag, comm, MPI_STATUS_IGNORE);
+    }
+    
+
+    template <typename T, typename TI, typename T2 = decltype(GetMPIType<T>())>
+    void Recv (FlatArray <T,TI> s, int src, int tag) const {
       MPI_Recv (s.Data(), s.Size(), GetMPIType<T> (), src, tag, comm, MPI_STATUS_IGNORE);
     }
     
-    template <typename T, typename T2 = decltype(GetMPIType<T>())>
+    template <typename T, typename TI, typename T2 = decltype(GetMPIType<T>())>
-    void Recv (Array <T> & s, int src, int tag) const
+    void Recv (Array <T,TI> & s, int src, int tag) const
     {
       MPI_Status status;
       int len;
@@ -166,7 +206,7 @@ namespace ngcore
     }
 
     template<typename T, typename T2 = decltype(GetMPIType<T>())>
-    MPI_Request ISend (const FlatArray<T> & s, int dest, int tag) const
+    MPI_Request ISend (FlatArray<T> s, int dest, int tag) const
     {
       MPI_Request request;
       MPI_Isend (s.Data(), s.Size(), GetMPIType<T>(), dest, tag, comm, &request);
@@ -182,7 +222,7 @@ namespace ngcore
     }
     
     template<typename T, typename T2 = decltype(GetMPIType<T>())>
-    MPI_Request IRecv (const FlatArray<T> & s, int src, int tag) const
+    MPI_Request IRecv (FlatArray<T> s, int src, int tag) const
     { 
       MPI_Request request;
       MPI_Irecv (s.Data(), s.Size(), GetMPIType<T>(), src, tag, comm, &request);
@@ -195,6 +235,7 @@ namespace ngcore
     template <typename T, typename T2 = decltype(GetMPIType<T>())> 
     T Reduce (T d, const MPI_Op & op, int root = 0) const
     {
+      static Timer t("MPI - Reduce"); RegionTimer reg(t);          
       if (size == 1) return d;
       
       T global_d;
@@ -205,6 +246,7 @@ namespace ngcore
     template <typename T, typename T2 = decltype(GetMPIType<T>())> 
     T AllReduce (T d, const MPI_Op & op) const
     {
+      static Timer t("MPI - AllReduce"); RegionTimer reg(t);
       if (size == 1) return d;
       
       T global_d;
@@ -227,6 +269,44 @@ namespace ngcore
       MPI_Bcast (&s[0], len, MPI_CHAR, root, comm);
     }
 
+    template <typename T>
+    void AllToAll (FlatArray<T> send, FlatArray<T> recv) const
+    {
+      MPI_Alltoall (send.Data(), 1, GetMPIType<T>(), 
+                    recv.Data(), 1, GetMPIType<T>(), comm);
+    }
+
+
+    template <typename T>
+    void ScatterRoot (FlatArray<T> send) const
+    {
+      if (size == 1) return;
+      MPI_Scatter (send.Data(), 1, GetMPIType<T>(),
+                   MPI_IN_PLACE, -1, GetMPIType<T>(), 0, comm);
+    }
+    
+    template <typename T>
+    void Scatter (T & recv) const
+    {
+      if (size == 1) return;      
+      MPI_Scatter (NULL, 0, GetMPIType<T>(),
+                   &recv, 1, GetMPIType<T>(), 0, comm);
+    }
+
+    template <typename T>
+    void AllGather (T val, FlatArray<T> recv) const
+    {
+      if (size == 1)
+        {
+          recv[0] = val;
+          return;
+        }
+      MPI_Allgather (&val, 1, GetMPIType<T>(),
+                     recv.Data(), 1, GetMPIType<T>(), 
+                     comm);
+    }
+    
+    
     NgMPI_Comm SubCommunicator (FlatArray<int> procs) const
     {
       MPI_Comm subcomm;
@@ -241,6 +321,7 @@ namespace ngcore
 
   NETGEN_INLINE void MyMPI_WaitAll (FlatArray<MPI_Request> requests)
   {
+    static Timer t("MPI - WaitAll"); RegionTimer reg(t);    
     if (!requests.Size()) return;
     MPI_Waitall (requests.Size(), requests.Data(), MPI_STATUSES_IGNORE);
   }
@@ -263,9 +344,10 @@ namespace ngcore
   static MPI_Comm MPI_COMM_WORLD = 12345, MPI_COMM_NULL = 10000;
 
   typedef int MPI_Op;
+  typedef int MPI_Datatype;  
   typedef int MPI_Request;
   
-  enum { MPI_SUM = 0, MPI_MIN = 1, MPI_MAX = 2 };
+  enum { MPI_SUM = 0, MPI_MIN = 1, MPI_MAX = 2, MPI_LOR = 4711 };
   
   class NgMPI_Comm
   {
@@ -276,6 +358,7 @@ namespace ngcore
 
     size_t Rank() const { return 0; }
     size_t Size() const { return 1; }
+    bool ValidCommunicator() const { return false; }
     void Barrier() const { ; } 
     operator MPI_Comm() const { return MPI_Comm(); }
 
@@ -298,13 +381,13 @@ namespace ngcore
     MPI_Request ISend (T & val, int dest, int tag) const { return 0; } 
     
     template<typename T>
-    MPI_Request ISend (const FlatArray<T> & s, int dest, int tag) const { return 0; }
+    MPI_Request ISend (FlatArray<T> s, int dest, int tag) const { return 0; }
 
     template<typename T>
     MPI_Request IRecv (T & val, int dest, int tag) const { return 0; } 
     
     template<typename T>
-    MPI_Request IRecv (const FlatArray<T> & s, int src, int tag) const { return 0; }
+    MPI_Request IRecv (FlatArray<T> s, int src, int tag) const { return 0; }
 
     template <typename T>
     T Reduce (T d, const MPI_Op & op, int root = 0) const { return d; }

libsrc/core/ngcore.hpp

@@ -6,14 +6,15 @@
 #include "bitarray.hpp"
 #include "exception.hpp"
 #include "flags.hpp"
+#include "table.hpp"
 #include "hashtable.hpp"
 #include "localheap.hpp"
 #include "logging.hpp"
 #include "mpi_wrapper.hpp"
 #include "profiler.hpp"
 #include "signal.hpp"
+#include "simd.hpp"
 #include "symboltable.hpp"
-#include "table.hpp"
 #include "taskmanager.hpp"
 #include "version.hpp"
 #include "xbool.hpp"

libsrc/core/ngcore_api.hpp

@@ -67,6 +67,18 @@
   #endif
 #endif
 
+#if defined(__amd64__) || defined(_M_AMD64)
+#define NETGEN_ARCH_AMD64
+#endif
+
+#if defined(__aarch64__) || defined(_M_ARM64)
+#define NETGEN_ARCH_ARM64
+#endif
+
+#if defined(__arm__) || defined(_M_ARM)
+#define NETGEN_ARCH_ARM
+#endif
+
 #ifdef __MAC_OS_X_VERSION_MIN_REQUIRED
 #if __MAC_OS_X_VERSION_MIN_REQUIRED < 101400
 // The c++ standard library on MacOS 10.13 and earlier has no aligned new operator,

libsrc/core/paje_trace.cpp

@@ -8,28 +8,46 @@
 #include "archive.hpp"           // for Demangle
 #include "paje_trace.hpp"
 #include "profiler.hpp"
+#include "mpi_wrapper.hpp"
 
 extern const char *header;
 
+constexpr int MPI_PAJE_WRITER = 1;
+
 namespace ngcore
 {
+  static std::string GetTimerName( int id )
+  {
+#ifndef PARALLEL
+    return NgProfiler::GetName(id);
+#else // PARALLEL
+    if(id<NgProfiler::SIZE)
+      return NgProfiler::GetName(id);
+
+    NgMPI_Comm comm(MPI_COMM_WORLD);
+    return NgProfiler::GetName(id-NgProfiler::SIZE*comm.Rank());
+#endif // PARALLEL
+  }
+
+  std::vector<PajeTrace::MemoryEvent> PajeTrace::memory_events;
+
   // Produce no traces by default
   size_t PajeTrace::max_tracefile_size = 0;
 
   // If true, produce variable counting active threads
   // increases trace by a factor of two
-  bool PajeTrace::trace_thread_counter = true;
+  bool PajeTrace::trace_thread_counter = false;
   bool PajeTrace::trace_threads = true;
+  bool PajeTrace::mem_tracing_enabled = true;
 
   PajeTrace :: PajeTrace(int anthreads, std::string aname)
   {
-    start_time = GetTimeCounter();
 
     nthreads = anthreads;
     tracefile_name = std::move(aname);
 
     int bytes_per_event=33;
-    max_num_events_per_thread = std::min( static_cast<size_t>(std::numeric_limits<int>::max()), max_tracefile_size/bytes_per_event/(2*nthreads+1)*10/7);
+    max_num_events_per_thread = std::min( static_cast<size_t>(std::numeric_limits<int>::max()), max_tracefile_size/bytes_per_event/(nthreads+1+trace_thread_counter*nthreads)*10/7);
     if(max_num_events_per_thread>0)
     {
       logger->info( "Tracefile size = {}MB", max_tracefile_size/1024/1024);
@@ -47,14 +65,61 @@ namespace ngcore
 
     jobs.reserve(reserve_size);
     timer_events.reserve(reserve_size);
+    memory_events.reserve(1024*1024);
 
+    // sync start time when running in parallel
+#ifdef PARALLEL
+    NgMPI_Comm comm(MPI_COMM_WORLD);
+    for(auto i : Range(5))
+        comm.Barrier();
+#endif // PARALLEL
+
+    start_time = GetTimeCounter();
     tracing_enabled = true;
+    mem_tracing_enabled = true;
+    n_memory_events_at_start = memory_events.size();
   }
 
   PajeTrace :: ~PajeTrace()
   {
-    if(!tracefile_name.empty())
+    for(auto & ltask : tasks)
+        for(auto & task : ltask)
+          {
+            task.start_time -= start_time;
+            task.stop_time -= start_time;
+          }
+    for(auto & job : jobs)
+      {
+        job.start_time -= start_time;
+        job.stop_time -= start_time;
+      }
+    for(auto & event : timer_events)
+        event.time -= start_time;
+
+    for(auto & llink : links)
+        for(auto & link : llink)
+            link.time -= start_time;
+
+    for(auto i : IntRange(n_memory_events_at_start, memory_events.size()))
+      memory_events[i].time -= start_time;
+
+    NgMPI_Comm comm(MPI_COMM_WORLD);
+
+    if(comm.Size()==1)
+    {
       Write(tracefile_name);
+    }
+    else
+    {
+      // make sure the timer id is unique across all ranks
+      for(auto & event : timer_events)
+        event.timer_id += NgProfiler::SIZE*comm.Rank();
+
+      if(comm.Rank() == MPI_PAJE_WRITER)
+        Write(tracefile_name);
+      else
+        SendData();
+    }
   }
 
 
@@ -90,7 +155,6 @@ namespace ngcore
       int alias_counter;
 
       FILE * ctrace_stream;
-      TTimePoint start_time;
       std::shared_ptr<Logger> logger = GetLogger("PajeTrace");
 
 
@@ -98,7 +162,7 @@ namespace ngcore
           // return time in milliseconds as double
         // return std::chrono::duration<double>(t-start_time).count()*1000.0;
         // return std::chrono::duration<double>(t-start_time).count() / 2.7e3;
-        return 1000.0*static_cast<double>(t-start_time) * seconds_per_tick;
+        return 1000.0*static_cast<double>(t) * seconds_per_tick;
       }
 
       enum PType
@@ -180,10 +244,10 @@ namespace ngcore
       void operator=(const PajeFile &) = delete;
       void operator=(PajeFile &&) = delete;
 
-      PajeFile( const std::string & filename, TTimePoint astart_time )
+      PajeFile( const std::string & filename)
         {
-          start_time = astart_time;
+          std::string fname = filename + ".trace";
-          ctrace_stream = fopen (filename.c_str(),"w"); // NOLINT
+          ctrace_stream = fopen (fname.c_str(),"w"); // NOLINT
           fprintf(ctrace_stream, "%s", header ); // NOLINT
           alias_counter = 0;
         }
@@ -365,39 +429,85 @@ namespace ngcore
             logger->warn("Tracing stopped during computation due to tracefile size limit of {} megabytes.", max_tracefile_size/1024/1024);
         }
 
-      PajeFile paje(filename, start_time);
+      PajeFile paje(filename);
 
       const int container_type_task_manager = paje.DefineContainerType( 0, "Task Manager" );
       const int container_type_node = paje.DefineContainerType( container_type_task_manager, "Node");
       const int container_type_thread = paje.DefineContainerType( container_type_task_manager, "Thread");
       const int container_type_timer = container_type_thread; //paje.DefineContainerType( container_type_task_manager, "Timers");
       const int container_type_jobs = paje.DefineContainerType( container_type_task_manager, "Jobs");
+      const int container_type_memory = paje.DefineContainerType( container_type_task_manager, "Memory usage");
 
       const int state_type_job = paje.DefineStateType( container_type_jobs, "Job" );
       const int state_type_task = paje.DefineStateType( container_type_thread, "Task" );
       const int state_type_timer = paje.DefineStateType( container_type_timer, "Timer state" );
 
-      const int variable_type_active_threads = paje.DefineVariableType( container_type_jobs, "Active threads" );
+      int variable_type_active_threads = 0;
+      if(trace_thread_counter)
+          variable_type_active_threads = paje.DefineVariableType( container_type_jobs, "Active threads" );
 
       const int container_task_manager = paje.CreateContainer( container_type_task_manager, 0, "The task manager" );
       const int container_jobs = paje.CreateContainer( container_type_jobs, container_task_manager, "Jobs" );
-      paje.SetVariable( start_time, variable_type_active_threads, container_jobs, 0.0 );
 
-      const int num_nodes = 1; //task_manager ? task_manager->GetNumNodes() : 1;
+      int variable_type_memory = 0;
+      const int container_memory = paje.CreateContainer( container_type_memory, container_task_manager, "Memory" );
+      if(mem_tracing_enabled)
+      {
+        variable_type_memory = paje.DefineVariableType( container_type_task_manager, "Memory [MB]" );
+      }
 
+
+      int num_nodes = 1; //task_manager ? task_manager->GetNumNodes() : 1;
+      std::vector <int> thread_aliases;
       std::vector<int> container_nodes;
-      container_nodes.reserve(num_nodes);
+
-      for(int i=0; i<num_nodes; i++)
+#ifdef PARALLEL
+      // Hostnames
+      NgMPI_Comm comm(MPI_COMM_WORLD);
+      auto rank = comm.Rank();
+      auto nranks = comm.Size();
+      if(nranks>1)
+      {
+        nthreads = nranks;
+        thread_aliases.reserve(nthreads);
+
+        std::array<char, MPI_MAX_PROCESSOR_NAME+1> ahostname;
+        int len;
+        MPI_Get_processor_name(ahostname.data(), &len);
+        std::string hostname = ahostname.data();
+
+        std::map<std::string, int> host_map;
+
+        std::string name;
+        for(auto i : IntRange(0, nranks))
+        {
+          if(i!=MPI_PAJE_WRITER)
+            comm.Recv(name, i, 0);
+          else
+            name = hostname;
+          if(host_map.count(name)==0)
+          {
+            host_map[name] = container_nodes.size();
+            container_nodes.emplace_back( paje.CreateContainer( container_type_node, container_task_manager, name) );
+          }
+          thread_aliases.emplace_back( paje.CreateContainer( container_type_thread, container_nodes[host_map[name]], "Rank " + ToString(i) ) );
+        }
+      }
+      else
+#endif // PARALLEL
+      {
+        container_nodes.reserve(num_nodes);
+        for(int i=0; i<num_nodes; i++)
           container_nodes.emplace_back( paje.CreateContainer( container_type_node, container_task_manager, "Node " + ToString(i)) );
 
-      std::vector <int> thread_aliases;
+        thread_aliases.reserve(nthreads);
-      thread_aliases.reserve(nthreads);
+        if(trace_threads)
-      if(trace_threads)
+          for (int i=0; i<nthreads; i++)
-        for (int i=0; i<nthreads; i++)
           {
             auto name = "Thread " + ToString(i);
             thread_aliases.emplace_back( paje.CreateContainer( container_type_thread, container_nodes[i*num_nodes/nthreads], name ) );
           }
+      }
 
       std::map<const std::type_info *, int> job_map;
       std::map<const std::type_info *, int> job_task_map;
@@ -416,20 +526,73 @@ namespace ngcore
           paje.PopState( j.stop_time, state_type_job, container_jobs );
         }
 
+      size_t memory_at_start = 0;
+
+      for(const auto & i : IntRange(0, n_memory_events_at_start))
+      {
+        if(memory_events[i].is_alloc)
+            memory_at_start += memory_events[i].size;
+        else
+            memory_at_start -= memory_events[i].size;
+      }
+
+      paje.SetVariable( 0, variable_type_memory, container_memory, 1.0*memory_at_start/(1024*1024));
+
+      for(const auto & i : IntRange(n_memory_events_at_start, memory_events.size()))
+      {
+        auto & m = memory_events[i];
+        if(m.size==0)
+            continue;
+        double size = 1.0*m.size/(1024*1024);
+        if(m.is_alloc)
+          paje.AddVariable( m.time, variable_type_memory, container_memory, size);
+        else
+          paje.SubVariable( m.time, variable_type_memory, container_memory, size);
+      }
+
       std::set<int> timer_ids;
       std::map<int,int> timer_aliases;
+      std::map<int,std::string> timer_names;
 
       for(auto & event : timer_events)
-        timer_ids.insert(event.timer_id);
+          timer_ids.insert(event.timer_id);
 
 
+      // Timer names
       for(auto & vtasks : tasks)
-        for (Task & t : vtasks)
+          for (Task & t : vtasks)
-          if(t.id_type==Task::ID_TIMER)
+              if(t.id_type==Task::ID_TIMER)
-            timer_ids.insert(t.id);
+                  timer_ids.insert(t.id);
 
       for(auto id : timer_ids)
-        timer_aliases[id] = paje.DefineEntityValue( state_type_timer, NgProfiler::GetName(id), -1 );
+          timer_names[id] = GetTimerName(id);
+
+#ifdef PARALLEL
+      if(nranks>1)
+      {
+        for(auto src : IntRange(0, nranks))
+        {
+          if(src==MPI_PAJE_WRITER)
+            continue;
+
+          size_t n_timers;
+          comm.Recv (n_timers, src, 0);
+
+          int id;
+          std::string name;
+          for(auto i : IntRange(n_timers))
+          {
+            comm.Recv (id, src, 0);
+            comm.Recv (name, src, 0);
+            timer_ids.insert(id);
+            timer_names[id] = name;
+          }
+        }
+      }
+#endif // PARALLEL
+
+      for(auto id : timer_ids)
+          timer_aliases[id] = paje.DefineEntityValue( state_type_timer, timer_names[id], -1 );
 
       int timerdepth = 0;
       int maxdepth = 0;
@@ -494,6 +657,50 @@ namespace ngcore
           }
         }
 
+#ifdef PARALLEL
+      if(nranks>1)
+      {
+        for(auto & event : timer_events)
+        {
+          if(event.is_start)
+            paje.PushState( event.time, state_type_timer, thread_aliases[MPI_PAJE_WRITER], timer_aliases[event.timer_id] );
+          else
+            paje.PopState( event.time, state_type_timer, thread_aliases[MPI_PAJE_WRITER] );
+        }
+
+        // Timer events
+        Array<int> timer_id;
+        Array<TTimePoint> time;
+        Array<bool> is_start;
+        Array<int> thread_id;
+
+        for(auto src : IntRange(0, nranks))
+        {
+          if(src==MPI_PAJE_WRITER)
+            continue;
+
+          comm.Recv (timer_id, src, 0);
+          comm.Recv (time, src, 0);
+          comm.Recv (is_start, src, 0);
+          comm.Recv (thread_id, src, 0);
+
+          for(auto i : Range(timer_id.Size()))
+          {
+            TimerEvent event;
+            event.timer_id = timer_id[i];
+            event.time = time[i];
+            event.is_start = is_start[i];
+            event.thread_id = thread_id[i];
+
+            if(event.is_start)
+              paje.PushState( event.time, state_type_timer, thread_aliases[src], timer_aliases[event.timer_id] );
+            else
+              paje.PopState( event.time, state_type_timer, thread_aliases[src] );
+          }
+        }
+      }
+#endif // PARALLEL
+
       // Merge link event
       int nlinks = 0;
       for( auto & l : links)
@@ -552,25 +759,96 @@ namespace ngcore
                 }
             }
         }
-      WriteSunburstHTML();
+      WriteTimingChart();
+#ifdef NETGEN_TRACE_MEMORY
+      WriteMemoryChart("");
+#endif // NETGEN_TRACE_MEMORY
       paje.WriteEvents();
     }
 
+  void PajeTrace::SendData( )
+    {
+#ifdef PARALLEL
+      // Hostname
+      NgMPI_Comm comm(MPI_COMM_WORLD);
+      auto rank = comm.Rank();
+      auto nranks = comm.Size();
+
+      std::string hostname;
+        {
+          std::array<char, MPI_MAX_PROCESSOR_NAME+1> ahostname;
+          int len;
+          MPI_Get_processor_name(ahostname.data(), &len);
+          hostname = ahostname.data();
+        }
+
+      comm.Send(hostname, MPI_PAJE_WRITER, 0);
+
+      // Timer names
+      std::set<int> timer_ids;
+      std::map<int,std::string> timer_names;
+
+      for(auto & event : timer_events)
+          timer_ids.insert(event.timer_id);
+
+      for(auto id : timer_ids)
+          timer_names[id] = GetTimerName(id);
+      size_t size = timer_ids.size();
+      comm.Send(size, MPI_PAJE_WRITER, 0);
+      for(auto id : timer_ids)
+        {
+          comm.Send(id, MPI_PAJE_WRITER, 0);
+          comm.Send(timer_names[id], MPI_PAJE_WRITER, 0);
+        }
+
+
+      // Timer events
+      Array<int> timer_id;
+      Array<TTimePoint> time;
+      Array<bool> is_start;
+      Array<int> thread_id;
+
+      for(auto & event : timer_events)
+        {
+          timer_id.Append(event.timer_id);
+          time.Append(event.time);
+          is_start.Append(event.is_start);
+          thread_id.Append(event.thread_id);
+        }
+
+      comm.Send (timer_id, MPI_PAJE_WRITER, 0);
+      comm.Send (time, MPI_PAJE_WRITER, 0);
+      comm.Send (is_start, MPI_PAJE_WRITER, 0);
+      comm.Send (thread_id, MPI_PAJE_WRITER, 0);
+#endif // PARALLEL
+    }
+
   ///////////////////////////////////////////////////////////////////
   // Write HTML file drawing a sunburst chart with cumulated timings
   struct TreeNode
   {
       int id = 0;
       std::map<int, TreeNode> children;
-      double time = 0.0;
+      double chart_size = 0.0; // time without children (the chart lib accumulates children sizes again)
+      double size = 0.0;
+      double min_size = 1e99;
+      double max_size = 0.0;
       std::string name;
+
+      size_t calls = 0;
       TTimePoint start_time = 0;
   };
 
-  void PrintNode (const TreeNode &n, int &level, std::ofstream & f);
+  void PrintNode (const TreeNode &n, std::ofstream & f)
-  void PrintNode (const TreeNode &n, int &level, std::ofstream & f)
   {
-      f << "{ name: \"" + n.name + "\", size: " + ToString(n.time);
+      f << "{ name: \"" + n.name + "\"";
+      f << ", calls: " << n.calls;
+      f << ", size: " << n.chart_size;
+      f << ", value: " << n.size;
+      f << ", min: " << n.min_size;
+      f << ", max: " << n.max_size;
+      if(n.calls)
+        f << ", avg: " << n.size/n.calls;
       int size = n.children.size();
       if(size>0)
       {
@@ -578,7 +856,7 @@ namespace ngcore
           f << ", children: [";
           for(auto & c : n.children)
           {
-              PrintNode(c.second, level, f);
+              PrintNode(c.second, f);
               if(++i<size)
                   f << " , ";
           }
@@ -587,12 +865,224 @@ namespace ngcore
       f << '}';
   }
 
-  void PajeTrace::WriteSunburstHTML( )
+  void WriteSunburstHTML( TreeNode & root, std::string filename, bool time_or_memory )
+  {
+    std::ofstream f(filename+".html");
+    f.precision(4);
+    f << R"CODE_(
+<head>
+  <script src="https://d3js.org/d3.v5.min.js"></script>
+  <script src="https://unpkg.com/sunburst-chart"></script>
+
+  <style>body { margin: 0 }</style>
+)CODE_";
+    if(!time_or_memory)
+      f << "<title>Maximum Memory Consumption</title>\n";
+    f << R"CODE_(
+</head>
+<body>
+  <div id="chart"></div>
+
+  <script>
+    const data = 
+)CODE_";
+      PrintNode(root, f);
+      f << ";\n\n";
+      if(time_or_memory)
+        f << "const chart_type = 'time';\n";
+      else
+        f << "const chart_type = 'memory';\n";
+      f << R"CODE_(
+    const color = d3.scaleOrdinal(d3.schemePaired);
+
+    let getTime = (t) =>
+    {
+       if(t>=1000)  return (t/1000).toPrecision(4) + '  s';
+       if(t>=0.1)   return t.toPrecision(4) + ' ms';
+       if(t>=1e-4)  return (t*1e3).toPrecision(4) + ' us';
+
+       return (t/1e6).toPrecision(4) + ' ns';
+    };
+
+    const KB_ = 1024;
+    const MB_ = KB_*1024;
+    const GB_ = MB_*1024;
+    let getMemory = (m) =>
+    {
+       if(m>=GB_)  return (m/GB_).toPrecision(4) + ' GB';
+       if(m>=MB_)  return (m/MB_).toPrecision(4) + ' MB';
+       if(m>=KB_)  return (m/KB_).toPrecision(4) + ' KB';
+       return m.toPrecision(4) + ' B';
+    };
+
+    Sunburst()
+      .data(data)
+      .size('size')
+      .color(d => color(d.name))
+      .tooltipTitle((d, node) => { return node.parent ? node.parent.data.name + " &rarr; " + d.name : d.name; })
+      .tooltipContent((d, node) => {
+        if(chart_type=="memory")
+        {
+          return `Total Memory: <i>${getMemory(d.value)}</i> <br>`
+               + `Memory: <i>${getMemory(d.size)}</i>`
+        }
+        else
+        {
+          return `Time: <i>${getTime(d.value)}</i> <br>`
+               + `calls: <i>${d.calls}</i> <br>`
+               + `min: <i>${getTime(d.min)}</i> <br>`
+               + `max: <i>${getTime(d.max)}</i> <br>`
+               + `avg: <i>${getTime(d.avg)}</i>`
+        }
+      })
+      (document.getElementById('chart'));
+
+      // Line breaks in tooltip
+      var all = document.getElementsByClassName('sunbirst-tooltip');
+      for (var i = 0; i < all.length; i++) {
+          all[i].white_space = "";
+      }
+  </script>
+</body>
+)CODE_" << std::endl;
+
+
+  }
+
+#ifdef NETGEN_TRACE_MEMORY
+  void PajeTrace::WriteMemoryChart( std::string fname )
+  {
+    if(fname=="")
+      fname = tracefile_name + "_memory";
+    size_t mem_allocated = 0;
+    size_t max_mem_allocated = 0;
+    size_t imax_mem_allocated = 0;
+
+    const auto & names = MemoryTracer::GetNames();
+    const auto & parents = MemoryTracer::GetParents();
+    size_t N = names.size();
+
+    Array<size_t> mem_allocated_id;
+    mem_allocated_id.SetSize(N);
+    mem_allocated_id = 0;
+
+    // Find point with maximum memory allocation, check for missing allocs/frees
+    for(auto i : IntRange(memory_events.size()))
+    {
+      const auto & ev = memory_events[i];
+
+      if(ev.is_alloc)
+      {
+        mem_allocated += ev.size;
+        mem_allocated_id[ev.id] += ev.size;
+        if(mem_allocated > max_mem_allocated && i>=n_memory_events_at_start)
+        {
+          imax_mem_allocated = i;
+          max_mem_allocated = mem_allocated;
+        }
+      }
+      else
+      {
+        if(ev.size > mem_allocated)
+          {
+            std::cerr << "Error in memory tracer: have total allocated memory < 0" << std::endl;
+            mem_allocated = 0;
+          }
+        else
+          mem_allocated -= ev.size;
+        if(ev.size > mem_allocated_id[ev.id])
+          {
+            std::cerr << "Error in memory tracer: have allocated memory < 0 in tracer " << names[ev.id] << std::endl;
+            mem_allocated_id[ev.id] = 0;
+          }
+        else
+          mem_allocated_id[ev.id] -= ev.size;
+      }
+    }
+
+    // reconstruct again the memory consumption after event imax_mem_allocated
+    mem_allocated_id = 0;
+    for(auto i : IntRange(imax_mem_allocated+1))
+    {
+      const auto & ev = memory_events[i];
+
+      if(ev.is_alloc)
+        mem_allocated_id[ev.id] += ev.size;
+      else
+        {
+          if(ev.size > mem_allocated_id[ev.id])
+            mem_allocated_id[ev.id] = 0;
+          else
+            mem_allocated_id[ev.id] -= ev.size;
+        }
+    }
+
+    TreeNode root;
+    root.name="all";
+
+    Array<TreeNode*> nodes;
+    nodes.SetSize(N);
+    nodes = nullptr;
+    nodes[0] = &root;
+    Array<Array<int>> children(N);
+
+    Array<size_t> sorting; // topological sorting (parents before children)
+    sorting.SetAllocSize(N);
+
+    for(auto i : IntRange(1, N))
+        children[parents[i]].Append(i);
+
+    ArrayMem<size_t, 100> stack;
+    sorting.Append(0);
+    stack.Append(0);
+
+    while(stack.Size())
+    {
+      auto current = stack.Last();
+      stack.DeleteLast();
+
+      for(const auto child : children[current])
+      {
+        sorting.Append(child);
+        if(children[child].Size())
+          stack.Append(child);
+      }
+    }
+
+    for(auto i : sorting)
+    {
+      if(i==0)
+          continue;
+
+      TreeNode * parent = nodes[parents[i]];
+
+      auto & node = parent->children[i];
+      nodes[i] = &node;
+      node.id = i;
+      node.chart_size = mem_allocated_id[i];
+      node.size = mem_allocated_id[i];
+      node.name = names[i];
+    }
+
+    for(auto i_ : Range(sorting))
+    {
+      // reverse topological order to accumulate total memory usage of all children
+      auto i = sorting[sorting.Size()-1-i_];
+      if(i==0)
+          continue;
+      nodes[parents[i]]->size += nodes[i]->size;
+    }
+
+    WriteSunburstHTML( root, fname, false );
+
+  }
+#endif // NETGEN_TRACE_MEMORY
+
+  void PajeTrace::WriteTimingChart( )
   {
       std::vector<TimerEvent> events;
 
       TreeNode root;
-      root.time=0;
       root.name="all";
       TreeNode *current = &root;
 
@@ -629,7 +1119,10 @@ namespace ngcore
 
       std::sort (events.begin(), events.end());
 
-      root.time = 1000.0*static_cast<double>(stop_time-start_time) * seconds_per_tick;
+      root.size = 1000.0*static_cast<double>(stop_time) * seconds_per_tick;
+      root.calls = 1;
+      root.min_size = root.size;
+      root.max_size = root.size;
 
       for(auto & event : events)
       {
@@ -645,8 +1138,8 @@ namespace ngcore
 
               if(need_init)
               {
-                  current->name = is_timer_event ? NgProfiler::GetName(id) : job_names[id];
+                  current->name = is_timer_event ? GetTimerName(id) : job_names[id];
-                  current->time = 0.0;
+                  current->size = 0.0;
                   current->id = id;
               }
 
@@ -658,44 +1151,22 @@ namespace ngcore
                 std::cout << "node stack empty!" << std::endl;
                 break;
               }
-              double time = 1000.0*static_cast<double>(event.time-current->start_time) * seconds_per_tick;
+              double size = 1000.0*static_cast<double>(event.time-current->start_time) * seconds_per_tick;
-              current->time += time;
+              current->size += size;
+              current->chart_size += size;
+              current->min_size = std::min(current->min_size, size);
+              current->max_size = std::max(current->max_size, size);
+              current->calls++;
+
               current = node_stack.back();
-              current->time -= time;
+              current->chart_size -= size;
               node_stack.pop_back();
           }
       }
 
-      int level = 0;
+      root.chart_size = 0.0;
-      std::ofstream f(tracefile_name+".html");
-      f.precision(4);
-      f << R"CODE_(
-<head>
-  <script src="https://d3js.org/d3.v5.min.js"></script>
-  <script src="https://unpkg.com/sunburst-chart"></script>
 
-  <style>body { margin: 0 }</style>
+      ngcore::WriteSunburstHTML( root, tracefile_name, true );
-</head>
-<body>
-  <div id="chart"></div>
-
-  <script>
-    const data = 
-)CODE_";
-      PrintNode(root, level, f);
-      f << R"CODE_( ;
-
-    const color = d3.scaleOrdinal(d3.schemePaired);
-
-    Sunburst()
-      .data(data)
-      .size('size')
-      .color(d => color(d.name))
-      .tooltipContent((d, node) => `Time: <i>${node.value.toPrecision(6)}ms</i>`)
-      (document.getElementById('chart'));
-  </script>
-</body>
-)CODE_" << std::endl;
   }
 
 } // namespace ngcore

libsrc/core/paje_trace.hpp

@@ -23,18 +23,28 @@ namespace ngcore
       NGCORE_API static size_t max_tracefile_size;
       NGCORE_API static bool trace_thread_counter;
       NGCORE_API static bool trace_threads;
+      NGCORE_API static bool mem_tracing_enabled;
 
       bool tracing_enabled;
       TTimePoint start_time;
       int nthreads;
+      size_t n_memory_events_at_start;
 
     public:
-      void WriteSunburstHTML();
+      NGCORE_API void WriteTimingChart();
+#ifdef NETGEN_TRACE_MEMORY
+      NGCORE_API void WriteMemoryChart( std::string fname );
+#endif // NETGEN_TRACE_MEMORY
 
       // Approximate number of events to trace. Tracing will
       // be stopped if any thread reaches this number of events
       unsigned int max_num_events_per_thread;
 
+      static void SetTraceMemory( bool trace_memory )
+        {
+          mem_tracing_enabled = trace_memory;
+        }
+
       static void SetTraceThreads( bool atrace_threads )
         {
           trace_threads = atrace_threads;
@@ -96,10 +106,21 @@ namespace ngcore
           bool operator < (const ThreadLink & other) const { return time < other.time; }
         };
 
+      struct MemoryEvent
+        {
+          TTimePoint time;
+          size_t size;
+          int id;
+          bool is_alloc;
+
+          bool operator < (const MemoryEvent & other) const { return time < other.time; }
+        };
+
       std::vector<std::vector<Task> > tasks;
       std::vector<Job> jobs;
       std::vector<TimerEvent> timer_events;
       std::vector<std::vector<ThreadLink> > links;
+      NGCORE_API static std::vector<MemoryEvent> memory_events;
 
     public:
       NGCORE_API void StopTracing();
@@ -129,6 +150,27 @@ namespace ngcore
           timer_events.push_back(TimerEvent{timer_id, GetTimeCounter(), false});
         }
 
+      void AllocMemory(int id, size_t size)
+        {
+          if(!mem_tracing_enabled) return;
+          memory_events.push_back(MemoryEvent{GetTimeCounter(), size, id, true});
+        }
+
+      void FreeMemory(int id, size_t size)
+        {
+          if(!mem_tracing_enabled) return;
+          memory_events.push_back(MemoryEvent{GetTimeCounter(), size, id, false});
+        }
+
+      void ChangeMemory(int id, long long size)
+        {
+          if(size>0)
+            AllocMemory(id, size);
+          if(size<0)
+            FreeMemory(id, -size);
+        }
+
+
       NETGEN_INLINE int StartTask(int thread_id, int id, int id_type = Task::ID_NONE, int additional_value = -1)
         {
           if(!tracing_enabled) return -1;
@@ -185,6 +227,8 @@ namespace ngcore
 
       void Write( const std::string & filename );
 
+      void SendData(); // MPI parallel data reduction
+
     };
 } // namespace ngcore
 

libsrc/core/profiler.cpp

@@ -113,5 +113,9 @@ namespace ngcore
 
   NgProfiler prof; // NOLINT
 
+#ifdef NETGEN_TRACE_MEMORY
+  std::vector<std::string> MemoryTracer::names{"all"};
+  std::vector<int> MemoryTracer::parents{-1};
+#endif // NETGEN_TRACE_MEMORY
 
 } // namespace ngcore

libsrc/core/profiler.hpp

@@ -3,6 +3,7 @@
 
 #include <array>
 #include <chrono>
+#include <functional>
 #include <string>
 
 #include "logging.hpp"
@@ -299,7 +300,171 @@ namespace ngcore
       return tres;
   }
 
+  class MemoryTracer;
+
+  namespace detail
+  {
+    //Type trait to check if a class implements a 'void SetMemoryTacing(int)' function
+    template<typename T>
+    struct has_StartMemoryTracing
+    {
+    private:
+      template<typename T2>
+      static constexpr auto check(T2*) ->
+        typename std::is_same<decltype(std::declval<T2>().StartMemoryTracing()),void>::type;
+      template<typename>
+      static constexpr std::false_type check(...);
+      using type = decltype(check<T>(nullptr)); // NOLINT
+    public:
+      static constexpr bool value = type::value;
+    };
+  } // namespace detail
+
+  class MemoryTracer
+  {
+    #ifdef NETGEN_TRACE_MEMORY
+    NGCORE_API static std::vector<std::string> names;
+    NGCORE_API static std::vector<int> parents;
+
+    static int CreateId(const std::string& name)
+    {
+      int id = names.size();
+      names.push_back(name);
+      parents.push_back(0);
+      if(id==10*NgProfiler::SIZE)
+        std::cerr << "Allocated " << id << " MemoryTracer objects" << std::endl;
+      return id;
+    }
+    int id;
+
+    public:
+
+    MemoryTracer( std::string name )
+    {
+      id = CreateId(name);
+    }
+
+    // not tracing
+    MemoryTracer() : id(0) {}
+
+    template <typename... TRest>
+    MemoryTracer( std::string name, TRest & ... rest )
+    {
+      id = CreateId(name);
+      Track(rest...);
+    }
+
+    NETGEN_INLINE void Alloc(size_t size) const
+    {
+      if(id && trace)
+        trace->AllocMemory(id, size);
+    }
+
+    void Free(size_t size) const
+    {
+      if(id && trace)
+        trace->FreeMemory(id, size);
+    }
+
+    void Swap(size_t mysize, MemoryTracer& other, size_t other_size) const
+    {
+      if(!trace || (id == 0 && other.id == 0))
+        return;
+      if(id == 0)
+        return trace->ChangeMemory(other.id, mysize - other_size);
+      if(other.id == 0)
+        return trace->ChangeMemory(id, other_size - mysize);
+
+      // first decrease memory, otherwise have artificial/wrong high peak memory usage
+      if(mysize<other_size)
+        {
+          trace->ChangeMemory(other.id, mysize-other_size);
+          trace->ChangeMemory(id, other_size-mysize);
+        }
+      else
+        {
+          trace->ChangeMemory(id, other_size-mysize);
+          trace->ChangeMemory(other.id, mysize-other_size);
+        }
+    }
+
+    int GetId() const { return id; }
+
+    template <typename T1, typename... TRest>
+    void Track( T1 & obj, const std::string& name, TRest & ... rest ) const
+    {
+      Track(obj, name);
+      Track(rest...);
+    }
+
+    template<typename T>
+    void Track( T & obj, const std::string& name ) const
+    {
+      obj.GetMemoryTracer().Activate(obj, name);
+      parents[obj.GetMemoryTracer().GetId()] = id;
+    }
+
+    static std::string GetName(int id)
+    {
+      return names[id];
+    }
+
+    std::string GetName() const
+    {
+      return names[id];
+    }
+
+    template<typename T>
+    void Activate(T& me, const std::string& name) const
+    {
+      if(!id)
+        {
+          const_cast<MemoryTracer*>(this)->id = CreateId(name);
+          if constexpr(detail::has_StartMemoryTracing<T>::value)
+            me.StartMemoryTracing();
+        }
+      else
+        SetName(name);
+    }
+
+    void SetName(const std::string& name) const
+    {
+      names[id] = name;
+    }
+
+
+    static const std::vector<std::string> & GetNames() { return names; }
+    static const std::vector<int> & GetParents() { return parents; }
+#else // NETGEN_TRACE_MEMORY
+  public:
+    MemoryTracer() {}
+    MemoryTracer( std::string /* name */ ) {}
+    template <typename... TRest>
+    MemoryTracer( std::string /* name */, TRest & ... ) {}
+
+    void Alloc(size_t /* size */) const {}
+    void Free(size_t /* size */) const {}
+    void Swap(...) const {}
+    int GetId() const { return 0; }
+
+    template <typename... TRest>
+    void Track(TRest&...) const {}
+
+    static std::string GetName(int /* id */) { return ""; }
+    std::string GetName() const { return ""; }
+    void SetName(std::string /* name */) const {}
+#endif // NETGEN_TRACE_MEMORY
+  };
 } // namespace ngcore
 
+// Helper macro to easily add multiple timers in a function for profiling
+// Usage: NETGEN_TIMER_FROM_HERE("my_timer_name")
+// Effect: define static Timer and RegionTimer with given name and line number
+#define NETGEN_TOKEN_CONCAT(x, y) x ## y
+#define NETGEN_TOKEN_CONCAT2(x, y) NETGEN_TOKEN_CONCAT(x, y)
+#define NETGEN_TIMER_FROM_HERE(name) \
+  static Timer NETGEN_TOKEN_CONCAT2(timer_, __LINE__)( string(name)+"_"+ToString(__LINE__)); \
+  RegionTimer NETGEN_TOKEN_CONCAT2(rt_,__LINE__)(NETGEN_TOKEN_CONCAT2(timer_,__LINE__));
+
 
 #endif // NETGEN_CORE_PROFILER_HPP

libsrc/core/python_ngcore.cpp

@@ -8,6 +8,8 @@ using std::string;
 namespace ngcore
 {
   bool ngcore_have_numpy = false;
+  bool parallel_pickling = true;
+  
   void SetFlag(Flags &flags, string s, py::object value) 
   {
     if (py::isinstance<py::dict>(value))

libsrc/core/python_ngcore.hpp

@@ -5,17 +5,70 @@
 #include <pybind11/pybind11.h>
 #include <pybind11/operators.h>
 #include <pybind11/numpy.h>
+#include <pybind11/stl.h>
 
 #include "array.hpp"
 #include "archive.hpp"
 #include "flags.hpp"
 #include "ngcore_api.hpp"
+#include "profiler.hpp"
 namespace py = pybind11;
 
+////////////////////////////////////////////////////////////////////////////////
+// automatic conversion of python list to Array<>
+namespace pybind11 {
+namespace detail {
+
+template <typename Type, typename Value> struct ngcore_list_caster {
+    using value_conv = make_caster<Value>;
+
+    bool load(handle src, bool convert) {
+        if (!isinstance<sequence>(src) || isinstance<str>(src))
+            return false;
+        auto s = reinterpret_borrow<sequence>(src);
+        value.SetSize(s.size());
+        value.SetSize0();
+        for (auto it : s) {
+            value_conv conv;
+            if (!conv.load(it, convert))
+                return false;
+            value.Append(cast_op<Value &&>(std::move(conv)));
+        }
+        return true;
+    }
+
+public:
+    template <typename T>
+    static handle cast(T &&src, return_value_policy policy, handle parent) {
+        if (!std::is_lvalue_reference<T>::value)
+            policy = return_value_policy_override<Value>::policy(policy);
+        list l(src.Size());
+        size_t index = 0;
+        for (auto &&value : src) {
+            auto value_ = reinterpret_steal<object>(value_conv::cast(forward_like<T>(value), policy, parent));
+            if (!value_)
+                return handle();
+            PyList_SET_ITEM(l.ptr(), (ssize_t) index++, value_.release().ptr()); // steals a reference
+        }
+        return l.release();
+    }
+
+    PYBIND11_TYPE_CASTER(Type, _("Array[") + value_conv::name + _("]"));
+};
+
+template <typename Type> struct type_caster<ngcore::Array<Type>>
+ : ngcore_list_caster<ngcore::Array<Type>, Type> { };
+
+
+} // namespace detail
+} // namespace pybind11
+////////////////////////////////////////////////////////////////////////////////
+
 namespace ngcore
 {
   NGCORE_API extern bool ngcore_have_numpy;
-
+  NGCORE_API extern bool parallel_pickling;
+  
   // Python class name type traits
   template <typename T>
   struct PyNameTraits {
@@ -142,6 +195,10 @@ namespace ngcore
              return py::make_iterator (self.begin(),self.end());
              }, py::keep_alive<0,1>()) // keep array alive while iterator is used
 
+        .def("__str__", [](TFlat& self)
+                        {
+                          return ToString(self);
+                        })
       ;
 
       if constexpr (detail::HasPyFormat<T>::value)
@@ -225,7 +282,6 @@ namespace ngcore
     using ARCHIVE::stream;
     using ARCHIVE::version_map;
     using ARCHIVE::logger;
-    using ARCHIVE::GetLibraryVersions;
   public:
     PyArchive(const pybind11::object& alst = pybind11::none()) :
       ARCHIVE(std::make_shared<std::stringstream>()),
@@ -270,10 +326,11 @@ namespace ngcore
 
     pybind11::list WriteOut()
     {
+      auto version_runtime = GetLibraryVersions();
       FlushBuffer();
       lst.append(pybind11::bytes(std::static_pointer_cast<std::stringstream>(stream)->str()));
       stream = std::make_shared<std::stringstream>();
-      *this & GetLibraryVersions();
+      *this & version_runtime;
       FlushBuffer();
       lst.append(pybind11::bytes(std::static_pointer_cast<std::stringstream>(stream)->str()));
       stream = std::make_shared<std::stringstream>();
@@ -291,19 +348,14 @@ namespace ngcore
     return pybind11::pickle([](T* self)
                       {
                         PyArchive<T_ARCHIVE_OUT> ar;
+                        ar.SetParallel(parallel_pickling);
                         ar & self;
                         auto output = pybind11::make_tuple(ar.WriteOut());
-                        GetLogger("Archive")->trace("Pickling output for object of type {} = {}",
-                                                    Demangle(typeid(T).name()),
-                                                    std::string(pybind11::str(output)));
                         return output;
                       },
                       [](const pybind11::tuple & state)
                       {
                         T* val = nullptr;
-                        GetLogger("Archive")->trace("State for unpickling of object of type {} = {}",
-                                                    Demangle(typeid(T).name()),
-                                                    std::string(pybind11::str(state[0])));
                         PyArchive<T_ARCHIVE_IN> ar(state[0]);
                         ar & val;
                         return val;

libsrc/core/python_ngcore_export.cpp

@@ -30,6 +30,7 @@ PYBIND11_MODULE(pyngcore, m) // NOLINT
                       if (a[i]) ba->SetBit(i);
                     return ba;
                   } ), py::arg("vec"))
+    .def(NGSPickle<BitArray>())
     .def("__str__", &ToString<BitArray>)
     .def("__len__", &BitArray::Size)
     .def("__getitem__", [] (BitArray & self, int i)
@@ -69,6 +70,29 @@ PYBIND11_MODULE(pyngcore, m) // NOLINT
                                              }
                                          }, py::arg("inds"), py::arg("value"), "Clear/Set bit at given positions")
 
+    .def("__setitem__", [] (BitArray & self, py::slice inds, BitArray & ba)
+                                         {
+                                           size_t start, step, stop, n;
+                                           if (!inds.compute(self.Size(), &start, &stop, &step, &n))
+                                             throw py::error_already_set();
+
+                                           if (start == 0 && n == self.Size() && step == 1)
+                                             {
+                                               self = ba;
+                                             }
+                                           else
+                                             {
+                                               for (size_t i = 0; i < n; i++, start += step)
+                                                 {
+                                                   bool b = ba.Test(i);
+                                                   if (b)
+                                                     self.SetBit(start);
+                                                   else
+                                                     self.Clear(start);
+                                                 }
+                                             }
+                                         }, py::arg("inds"), py::arg("ba"), "copy BitArray")
+
     .def("__setitem__", [](BitArray & self,  IntRange range, bool b)
       {
         if (b)
@@ -222,4 +246,30 @@ threads : int
     .def("__timing__", &TaskManager::Timing)
     ;
 
+  py::class_<PajeTrace>(m, "PajeTrace")
+    .def(py::init( [] (string filename, size_t size_mb, bool threads, bool thread_counter, bool memory)
+          {
+              PajeTrace::SetMaxTracefileSize(size_mb*1014*1024);
+              PajeTrace::SetTraceThreads(threads);
+              PajeTrace::SetTraceMemory(memory);
+              PajeTrace::SetTraceThreadCounter(thread_counter);
+              trace = new PajeTrace(TaskManager::GetMaxThreads(), filename);
+              return trace;
+          }), py::arg("filename")="ng.trace", py::arg("size")=1000,
+              py::arg("threads")=true, py::arg("thread_counter")=false,
+              py::arg("memory")=true,
+              "size in Megabytes"
+        )
+    .def("__enter__", [](PajeTrace & self) { })
+    .def("__exit__", [](PajeTrace & self, py::args) { self.StopTracing(); })
+    .def("__del__", [](PajeTrace & self) { trace = nullptr; })
+    .def_static("SetTraceThreads", &PajeTrace::SetTraceThreads)
+    .def_static("SetTraceThreadCounter", &PajeTrace::SetTraceThreadCounter)
+    .def_static("SetMaxTracefileSize", &PajeTrace::SetMaxTracefileSize)
+#ifdef NETGEN_TRACE_MEMORY
+    .def_static("WriteMemoryChart", [](string filename){ if(trace) trace->WriteMemoryChart(filename); }, py::arg("filename")="memory" )
+#endif // NETGEN_TRACE_MEMORY
+    ;
+
+
 }

libsrc/core/ranges.hpp

@@ -0,0 +1,109 @@
+#ifndef NETGEN_CORE_RANGES_HPP
+#define NETGEN_CORE_RANGES_HPP
+
+#include <iterator>
+
+namespace ngcore
+{
+  template<typename Iterator>
+  class AdapterRange
+  {
+    Iterator _begin,_end;
+  public:
+    AdapterRange(Iterator abegin, Iterator aend) : _begin(abegin), _end(aend) { ; }
+    Iterator begin() const { return _begin; }
+    Iterator end() const { return _end; }
+  };
+
+  template<typename FUNC>
+  class FilterAdapter
+  {
+    FUNC f;
+  public:
+    FilterAdapter(FUNC af) : f(af) { ; }
+    FUNC GetFunction() const { return f; }
+  };
+
+  template<typename FUNC, typename Iterator>
+  class FilterIterator
+  {
+    Iterator iter;
+    Iterator end;
+    FUNC f;
+  public:
+    FilterIterator(FUNC af, Iterator aiter, Iterator aend)
+      :  f(af), iter(aiter), end(aend)
+    {
+      while(iter!=end && !f(*iter))
+        ++iter;
+    }
+    inline FilterIterator& operator ++()
+    {
+      ++iter;
+      while(iter!=end && !f(*iter))
+        ++iter;
+      return *this;
+    }
+
+    inline bool operator !=(FilterIterator other)
+    {
+      return iter != other.iter;
+    }
+
+    inline bool operator ==(FilterIterator other)
+    {
+      return iter == other.iter;
+    }
+
+    inline decltype(auto) operator *() const
+    {
+      return *iter;
+    }
+  };
+
+  template<typename FUNC>
+  FilterAdapter<FUNC> filter(FUNC f) { return {f}; }
+
+  template<typename Range, typename FUNC>
+  auto operator |(Range&& range, FilterAdapter<FUNC> adapter)
+    -> AdapterRange<FilterIterator<FUNC,decltype(std::begin(range))>>
+  {
+    return {{adapter.GetFunction(),std::begin(range),std::end(range)},
+        {adapter.GetFunction(), std::end(range), std::end(range)}};
+  }
+
+  template<typename FUNC, typename Iterator>
+  class TransformIterator
+  {
+    FUNC f;
+    Iterator iter;
+  public:
+    TransformIterator(FUNC af, Iterator aiter) : f(af), iter(aiter) { ; }
+
+    TransformIterator& operator++() { ++iter; }
+    bool operator !=(TransformIterator other) { return iter != other.iter; }
+    decltype(auto) operator *() const { return f(*iter); }
+  };
+
+  template<typename FUNC>
+  class TransformAdapter
+  {
+    FUNC f;
+  public:
+    TransformAdapter(FUNC af) : f(af) { ; }
+    FUNC GetFunction() const { return f; }
+  };
+
+  template<typename FUNC>
+  TransformAdapter<FUNC> transform(FUNC f) { return {f}; }
+
+  template<typename Range, typename FUNC>
+  auto operator |(Range&& range, TransformAdapter<FUNC> adapter)
+    -> AdapterRange<TransformIterator<FUNC,decltype(std::begin(range))>>
+  {
+    return {{adapter.GetFunction(), std::begin(range)},
+        {adapter.GetFunction(),std::end(range)}};
+  }
+} // namespace ngcore
+
+#endif // NETGEN_CORE_RANGES_HPP

libsrc/core/simd.hpp

@@ -0,0 +1,75 @@
+#ifndef NETGEN_CORE_SIMD_HPP
+#define NETGEN_CORE_SIMD_HPP
+
+/**************************************************************************/
+/* File:   simd.hpp                                                       */
+/* Author: Joachim Schoeberl, Matthias Hochsteger                         */
+/* Date:   25. Mar. 16                                                    */
+/**************************************************************************/
+
+#include "ngcore_api.hpp"
+
+#include "simd_generic.hpp"
+
+#ifdef NETGEN_ARCH_AMD64
+#ifndef __SSE__
+#define __SSE__
+#endif
+#include "simd_sse.hpp"
+#endif
+
+#ifdef __AVX__
+#include "simd_avx.hpp"
+#endif
+
+#ifdef __AVX512F__
+#include "simd_avx512.hpp"
+#endif
+
+#ifdef __aarch64__
+#include "simd_arm64.hpp"
+#endif
+
+namespace ngcore
+{
+#ifdef NETGEN_ARCH_AMD64
+  NETGEN_INLINE auto HSum (SIMD<double,2> v1, SIMD<double,2> v2, SIMD<double,2> v3, SIMD<double,2> v4)
+  {
+    SIMD<double,2> hsum1 = my_mm_hadd_pd (v1.Data(), v2.Data());
+    SIMD<double,2> hsum2 = my_mm_hadd_pd (v3.Data(), v4.Data());
+    return SIMD<double,4> (hsum1, hsum2);
+  }
+
+  NETGEN_INLINE auto GetMaskFromBits( unsigned int i )
+  {
+    return SIMD<mask64>::GetMaskFromBits(i);
+  }
+#endif
+
+
+  NETGEN_INLINE void SIMDTranspose (SIMD<double,4> a1, SIMD<double,4> a2, SIMD <double,4> a3, SIMD<double,4> a4,
+                             SIMD<double,4> & b1, SIMD<double,4> & b2, SIMD<double,4> & b3, SIMD<double,4> & b4)
+  {
+    SIMD<double,4> h1,h2,h3,h4;
+    std::tie(h1,h2) = Unpack(a1,a2);
+    std::tie(h3,h4) = Unpack(a3,a4);
+    b1 = SIMD<double,4> (h1.Lo(), h3.Lo());
+    b2 = SIMD<double,4> (h2.Lo(), h4.Lo());
+    b3 = SIMD<double,4> (h1.Hi(), h3.Hi());
+    b4 = SIMD<double,4> (h2.Hi(), h4.Hi());
+  }
+
+  template<int N>
+  NETGEN_INLINE auto HSum (SIMD<double,N> s1, SIMD<double,N> s2)
+  {
+    return SIMD<double,2>(HSum(s1), HSum(s2));
+  }
+
+  template<int N>
+  NETGEN_INLINE auto HSum (SIMD<double,N> s1, SIMD<double,N> s2, SIMD<double,N> s3, SIMD<double,N> s4 )
+  {
+    return SIMD<double,4>(HSum(s1), HSum(s2), HSum(s3), HSum(s4));
+  }
+}
+
+#endif // NETGEN_CORE_SIMD_HPP

libsrc/core/simd_arm64.hpp

@@ -0,0 +1,175 @@
+#include "arm_neon.h"
+
+namespace ngcore
+{
+
+  template <>
+  class SIMD<mask64,2>
+  {
+    int64x2_t mask;
+  public:
+    SIMD (int i)
+    {
+      mask[0] = i > 0 ? -1 : 0;
+      mask[1] = i > 1 ? -1 : 0;
+    }
+
+    SIMD (bool i0, bool i1) { mask[0] = i0 ? -1:0; mask[1] = i1 ? -1 : 0; }
+    SIMD (SIMD<mask64,1> i0, SIMD<mask64,1> i1) { mask[0] = i0[0]; mask[1] = i1[0]; }
+    SIMD (float64x2_t _data) : mask{_data} { }
+    auto Data() const { return mask; }
+    static constexpr int Size() { return 2; }
+    // static NETGEN_INLINE SIMD<mask64, 2> GetMaskFromBits (unsigned int i);
+    int64_t operator[] (int i) const { return mask[i]; }
+
+    template <int I>
+    int64_t Get() const { return mask[I]; }
+    
+    auto Lo() const { return mask[0]; }
+    auto Hi() const { return mask[1]; }
+  };
+
+  
+  template<>
+  class SIMD<double,2>
+  {
+    float64x2_t data;
+
+  public:
+    static constexpr int Size() { return 2; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    // SIMD (double v0, double v1) : data{v0,v1} { }
+    SIMD (double v0, double v1) : data{vcombine_f64(float64x1_t{v0}, float64x1_t{v1})} { }
+    SIMD (std::array<double, 2> arr) : data{arr[0], arr[1]} { } 
+
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (double val) : data{val,val} { }
+    SIMD (int val)    : data{double(val),double(val)} { }
+    SIMD (size_t val) : data{double(val),double(val)} { }
+
+    SIMD (double const * p)
+    {
+      data = vld1q_f64(p);      
+      // data[0] = p[0];
+      // data[1] = p[1];
+    }
+    
+    SIMD (double const * p, SIMD<mask64,2> mask)
+      {
+	data[0] = mask[0] ? p[0] : 0;
+	data[1] = mask[1] ? p[1] : 0;
+      }
+    SIMD (float64x2_t _data) { data = _data; }
+    
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
+    SIMD (const T & func)
+    {
+      data[0] = func(0);
+      data[1] = func(1);
+    }
+
+    void Store (double * p)
+    {
+      vst1q_f64(p, data);
+      /*
+      p[0] = data[0];
+      p[1] = data[1];
+      */
+    }
+    
+    void Store (double * p, SIMD<mask64,2> mask)
+    {
+      if (mask[0]) p[0] = data[0];
+      if (mask[1]) p[1] = data[1];
+    }
+    
+    // NETGEN_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
+    NETGEN_INLINE double operator[] (int i) const { return data[i]; }
+    NETGEN_INLINE double & operator[] (int i)  { return ((double*)&data)[i]; }
+
+    template <int I>
+    double Get() const { return data[I]; }
+    
+    NETGEN_INLINE auto Data() const { return data; }
+    NETGEN_INLINE auto & Data() { return data; }
+
+    operator std::tuple<double&,double&> ()
+    {
+      auto pdata = (double*)&data;
+      return std::tuple<double&,double&>(pdata[0], pdata[1]);
+    }
+    
+    double Lo() const { return Get<0>(); } // data[0]; }
+    double Hi() const { return Get<1>(); } // data[1]; }
+    // double Hi() const { return vget_high_f64(data)[0]; }
+  };
+
+
+
+  NETGEN_INLINE double HSum (SIMD<double,2> sd)
+  {
+    return sd.Lo()+sd.Hi();  // sd[0]+sd[1];
+  }
+
+  NETGEN_INLINE SIMD<double,2> HSum (SIMD<double,2> a, SIMD<double,2> b)
+  {
+    // return SIMD<double,2> (a[0]+a[1], b[0]+b[1]);
+    return vpaddq_f64(a.Data(), b.Data());
+    
+  }
+
+  // a*b+c
+  NETGEN_INLINE SIMD<double,2> FMA (SIMD<double,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    return vmlaq_f64(c.Data(), a.Data(), b.Data());
+  }
+  NETGEN_INLINE SIMD<double,2> FMA (const double & a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    return FMA(SIMD<double,2> (a), b, c);
+  }
+  // -a*b+c
+  NETGEN_INLINE SIMD<double,2> FNMA (SIMD<double,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    return vmlsq_f64(c.Data(), a.Data(), b.Data());    
+    // return c-a*b;
+  }
+  NETGEN_INLINE SIMD<double,2> FNMA (const double & a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    return FNMA(SIMD<double,2> (a), b, c);
+  }
+
+  NETGEN_INLINE SIMD<double,2> operator+ (SIMD<double,2> a, SIMD<double,2> b)
+  { return a.Data()+b.Data(); }
+  
+  NETGEN_INLINE SIMD<double,2> operator- (SIMD<double,2> a, SIMD<double,2> b)
+  { return a.Data()-b.Data(); }
+  NETGEN_INLINE SIMD<double,2> operator- (SIMD<double,2> a)
+  { return -a.Data(); }
+  
+  NETGEN_INLINE SIMD<double,2> operator* (SIMD<double,2> a, SIMD<double,2> b)
+  { return a.Data()*b.Data(); }
+  
+  NETGEN_INLINE SIMD<double,2> operator/ (SIMD<double,2> a, SIMD<double,2> b)
+  { return a.Data()/b.Data(); }
+
+
+  
+  NETGEN_INLINE SIMD<double,2> If (SIMD<mask64,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    // return { a[0] ? b[0] : c[0], a[1] ? b[1] : c[1] };
+    return vbslq_f64(a.Data(), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<int64_t,2> If (SIMD<mask64,2> a, SIMD<int64_t,2> b, SIMD<int64_t,2> c)
+  {
+    return SIMD<int64_t,2> (a[0] ? b[0] : c[0], a[1] ? b[1] : c[1]);
+  }
+
+  NETGEN_INLINE SIMD<mask64,2> operator&& (SIMD<mask64,2> a, SIMD<mask64,2> b)
+  {
+    return vandq_u64 (a.Data(), b.Data());
+  }
+  
+}
+

libsrc/core/simd_avx.hpp

@@ -0,0 +1,309 @@
+#ifndef NETGEN_CORE_SIMD_AVX_HPP
+#define NETGEN_CORE_SIMD_AVX_HPP
+
+/**************************************************************************/
+/* File:   simd_avx.hpp                                                   */
+/* Author: Joachim Schoeberl, Matthias Hochsteger                         */
+/* Date:   25. Mar. 16                                                    */
+/**************************************************************************/
+
+#include <immintrin.h>
+
+namespace ngcore
+{
+
+#if defined(__GNUC__) && (__GNUC__ == 7)
+  // GCC7 does not have intrinsic _mm256_set_m128i, see
+  // https://stackoverflow.com/questions/32630458/setting-m256i-to-the-value-of-two-m128i-values
+  NETGEN_INLINE auto _mm256_set_m128i(__m128i v0, __m128i v1) {
+      return _mm256_insertf128_si256(_mm256_castsi128_si256(v1), (v0), 1);
+  }
+#endif // defined(__GNUC__) && (__GNUC__ == 7)
+
+#if defined(__AVX2__)
+  NETGEN_INLINE __m256i my_mm256_cmpgt_epi64 (__m256i a, __m256i b)
+  {
+    return _mm256_cmpgt_epi64 (a,b);
+  }
+#else
+  NETGEN_INLINE __m256i my_mm256_cmpgt_epi64 (__m256i a, __m256i b)
+  {
+    __m128i rlo = _mm_cmpgt_epi64(_mm256_extractf128_si256(a, 0),
+                                  _mm256_extractf128_si256(b, 0));
+    __m128i rhi = _mm_cmpgt_epi64(_mm256_extractf128_si256(a, 1),
+                                  _mm256_extractf128_si256(b, 1));
+    return _mm256_insertf128_si256 (_mm256_castsi128_si256(rlo), rhi, 1);
+  }
+#endif
+
+
+  template <>
+  class SIMD<mask64,4>
+  {
+    __m256i mask;
+  public:
+    SIMD (int64_t i)
+      : mask(my_mm256_cmpgt_epi64(_mm256_set1_epi64x(i),
+                                  _mm256_set_epi64x(3, 2, 1, 0)))
+    { ; }
+    SIMD (__m256i _mask) : mask(_mask) { ; }
+    SIMD (__m256d _mask) : mask(_mm256_castpd_si256(_mask)) { ; }
+    __m256i Data() const { return mask; }
+    static constexpr int Size() { return 4; }
+    static SIMD<mask64, 4> GetMaskFromBits (unsigned int i);
+  };
+
+  static SIMD<mask64, 4> masks_from_4bits[16] = {
+    _mm256_set_epi64x (0,0,0,0), _mm256_set_epi64x (0,0,0,-1),
+    _mm256_set_epi64x (0,0,-1,0), _mm256_set_epi64x (0,0,-1,-1),
+    _mm256_set_epi64x (0,-1,0,0), _mm256_set_epi64x (0,-1,0,-1),
+    _mm256_set_epi64x (0,-1,-1,0), _mm256_set_epi64x (0,-1,-1,-1),
+    _mm256_set_epi64x (-1,0,0,0), _mm256_set_epi64x (-1,0,0,-1),
+    _mm256_set_epi64x (-1,0,-1,0), _mm256_set_epi64x (-1,0,-1,-1),
+    _mm256_set_epi64x (-1,-1,0,0), _mm256_set_epi64x (-1,-1,0,-1),
+    _mm256_set_epi64x (-1,-1,-1,0), _mm256_set_epi64x (-1,-1,-1,-1)
+  };
+
+  NETGEN_INLINE SIMD<mask64, 4> SIMD<mask64, 4> :: GetMaskFromBits (unsigned int i)
+  {
+    return masks_from_4bits[i & 15];
+  }
+
+  template<>
+  class alignas(32) SIMD<int64_t,4>
+  {
+    __m256i data;
+
+  public:
+    static constexpr int Size() { return 4; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (int64_t val) { data = _mm256_set1_epi64x(val); }
+    SIMD (int64_t v0, int64_t v1, int64_t v2, int64_t v3) { data = _mm256_set_epi64x(v3,v2,v1,v0); }
+    SIMD (std::array<int64_t,4> a)
+      : data{_mm256_set_epi64x(a[3],a[2],a[1],a[0])}
+    {}
+    SIMD (SIMD<int64_t,2> v0, SIMD<int64_t,2> v1)
+        : data(_mm256_set_m128i(v0.Data(),v1.Data()))
+      {}
+    SIMD (__m256i _data) { data = _data; }
+
+    NETGEN_INLINE auto operator[] (int i) const { return ((int64_t*)(&data))[i]; }
+    NETGEN_INLINE __m256i Data() const { return data; }
+    NETGEN_INLINE __m256i & Data() { return data; }
+
+    SIMD<int64_t,2> Lo() const { return _mm256_extractf128_si256(data, 0); }
+    SIMD<int64_t,2> Hi() const { return _mm256_extractf128_si256(data, 1); }
+    static SIMD FirstInt(int n0=0) { return { n0+0, n0+1, n0+2, n0+3 }; }
+  };
+
+
+  NETGEN_INLINE SIMD<int64_t,4> operator-(SIMD<int64_t,4> a) { return _mm256_sub_epi64(_mm256_setzero_si256(), a.Data()); }
+
+#ifdef __AVX2__
+  NETGEN_INLINE SIMD<int64_t,4> operator+ (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_add_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<int64_t,4> operator- (SIMD<int64_t,4> a, SIMD<int64_t,4> b) { return _mm256_sub_epi64(a.Data(),b.Data()); }
+#endif // __AVX2__
+
+  template<>
+  class alignas(32) SIMD<double,4>
+  {
+    __m256d data;
+
+  public:
+    static constexpr int Size() { return 4; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (double val) { data = _mm256_set1_pd(val); }
+    SIMD (int val)    { data = _mm256_set1_pd(val); }
+    SIMD (size_t val) { data = _mm256_set1_pd(val); }
+    SIMD (double v0, double v1, double v2, double v3) { data = _mm256_set_pd(v3,v2,v1,v0); }
+    SIMD (SIMD<double,2> v0, SIMD<double,2> v1) : SIMD(v0[0], v0[1], v1[0], v1[1]) { ; }
+    SIMD (double const * p) { data = _mm256_loadu_pd(p); }
+    SIMD (double const * p, SIMD<mask64,4> mask) { data = _mm256_maskload_pd(p, mask.Data()); }
+    SIMD (__m256d _data) { data = _data; }
+    SIMD (std::array<double,4> a)
+      : data{_mm256_set_pd(a[3],a[2],a[1],a[0])}
+    {}
+
+    void Store (double * p) { _mm256_storeu_pd(p, data); }
+    void Store (double * p, SIMD<mask64,4> mask) { _mm256_maskstore_pd(p, mask.Data(), data); }
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
+    SIMD (const T & func)
+    {
+      data = _mm256_set_pd(func(3), func(2), func(1), func(0));
+    }
+
+    NETGEN_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
+    NETGEN_INLINE double & operator[] (int i) { return ((double*)(&data))[i]; }
+    // [[deprecated("don't write to individual elments of SIMD")]]
+    // NETGEN_INLINE double & operator[] (int i) { return ((double*)(&data))[i]; }
+    template <int I>
+    double Get() const { return ((double*)(&data))[I]; }
+    NETGEN_INLINE __m256d Data() const { return data; }
+    NETGEN_INLINE __m256d & Data() { return data; }
+
+    SIMD<double,2> Lo() const { return _mm256_extractf128_pd(data, 0); }
+    SIMD<double,2> Hi() const { return _mm256_extractf128_pd(data, 1); }
+
+    operator std::tuple<double&,double&,double&,double&> ()
+    { return std::tuple<double&,double&,double&,double&>((*this)[0], (*this)[1], (*this)[2], (*this)[3]); }
+  };
+
+  NETGEN_INLINE auto Unpack (SIMD<double,4> a, SIMD<double,4> b)
+  {
+    return std::make_tuple(SIMD<double,4>(_mm256_unpacklo_pd(a.Data(),b.Data())),
+                      SIMD<double,4>(_mm256_unpackhi_pd(a.Data(),b.Data())));
+  }
+
+  NETGEN_INLINE SIMD<double,4> operator- (SIMD<double,4> a) { return _mm256_xor_pd(a.Data(), _mm256_set1_pd(-0.0)); }
+  NETGEN_INLINE SIMD<double,4> operator+ (SIMD<double,4> a, SIMD<double,4> b) { return _mm256_add_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,4> operator- (SIMD<double,4> a, SIMD<double,4> b) { return _mm256_sub_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,4> operator* (SIMD<double,4> a, SIMD<double,4> b) { return _mm256_mul_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,4> operator/ (SIMD<double,4> a, SIMD<double,4> b) { return _mm256_div_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,4> operator* (double a, SIMD<double,4> b) { return _mm256_set1_pd(a)*b.Data(); }
+  NETGEN_INLINE SIMD<double,4> operator* (SIMD<double,4> b, double a) { return _mm256_set1_pd(a)*b.Data(); }
+
+  NETGEN_INLINE SIMD<double,4> sqrt (SIMD<double,4> a) { return _mm256_sqrt_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,4> floor (SIMD<double,4> a) { return _mm256_floor_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,4> ceil (SIMD<double,4> a) { return _mm256_ceil_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,4> fabs (SIMD<double,4> a) { return _mm256_max_pd(a.Data(), (-a).Data()); }
+
+
+#ifdef __FMA__
+  NETGEN_INLINE SIMD<double,4> FMA (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    return _mm256_fmadd_pd (a.Data(), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<double,4> FMA (const double & a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    return _mm256_fmadd_pd (_mm256_set1_pd(a), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<double,4> FNMA (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    return _mm256_fnmadd_pd (a.Data(), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<double,4> FNMA (const double & a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    return _mm256_fnmadd_pd (_mm256_set1_pd(a), b.Data(), c.Data());
+  }
+#endif
+
+#if defined(__FMA__) && !defined(__AVX512F__)
+  // make sure to use the update-version of fma
+  // important in matrix kernels using 12 sum-registers, 3 a-values and updated b-value
+  // avx512 has enough registers, and gcc seems to use only the first 16 z-regs
+  NETGEN_INLINE void FMAasm (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> & sum)
+  {
+    asm ("vfmadd231pd %[a], %[b], %[sum]"
+         : [sum] "+x" (sum.Data())
+         : [a] "x" (a.Data()), [b] "x" (b.Data())
+         );
+  }
+
+  NETGEN_INLINE void FNMAasm (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> & sum)
+  {
+    asm ("vfnmadd231pd %[a], %[b], %[sum]"
+         : [sum] "+x" (sum.Data())
+         : [a] "x" (a.Data()), [b] "x" (b.Data())
+         );
+  }
+#endif
+  
+
+
+  
+  NETGEN_INLINE SIMD<mask64,4> operator<= (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_LE_OQ); }
+  NETGEN_INLINE SIMD<mask64,4> operator< (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_LT_OQ); }
+  NETGEN_INLINE SIMD<mask64,4> operator>= (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_GE_OQ); }
+  NETGEN_INLINE SIMD<mask64,4> operator> (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_GT_OQ); }
+  NETGEN_INLINE SIMD<mask64,4> operator== (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_EQ_OQ); }
+  NETGEN_INLINE SIMD<mask64,4> operator!= (SIMD<double,4> a , SIMD<double,4> b)
+  { return _mm256_cmp_pd (a.Data(), b.Data(), _CMP_NEQ_OQ); }
+
+  NETGEN_INLINE SIMD<mask64,4> operator<= (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  _mm256_xor_si256(_mm256_cmpgt_epi64(a.Data(),b.Data()),_mm256_set1_epi32(-1)); }
+  NETGEN_INLINE SIMD<mask64,4> operator< (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  my_mm256_cmpgt_epi64(b.Data(),a.Data()); }
+  NETGEN_INLINE SIMD<mask64,4> operator>= (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  _mm256_xor_si256(_mm256_cmpgt_epi64(b.Data(),a.Data()),_mm256_set1_epi32(-1)); }
+  NETGEN_INLINE SIMD<mask64,4> operator> (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  my_mm256_cmpgt_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<mask64,4> operator== (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  _mm256_cmpeq_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<mask64,4> operator!= (SIMD<int64_t,4> a , SIMD<int64_t,4> b)
+  { return  _mm256_xor_si256(_mm256_cmpeq_epi64(a.Data(),b.Data()),_mm256_set1_epi32(-1)); }
+
+#ifdef __AVX2__
+  NETGEN_INLINE SIMD<mask64,4> operator&& (SIMD<mask64,4> a, SIMD<mask64,4> b)
+  { return _mm256_and_si256 (a.Data(), b.Data()); }
+  NETGEN_INLINE SIMD<mask64,4> operator|| (SIMD<mask64,4> a, SIMD<mask64,4> b)
+  { return _mm256_or_si256 (a.Data(), b.Data()); }
+  NETGEN_INLINE SIMD<mask64,4> operator! (SIMD<mask64,4> a)
+  { return _mm256_xor_si256 (a.Data(), _mm256_cmpeq_epi64(a.Data(),a.Data())); }
+#else //AVX2 is a superset of AVX. Without it, it is necessary to reinterpret the types
+  NETGEN_INLINE SIMD<mask64,4> operator&& (SIMD<mask64,4> a, SIMD<mask64,4> b)
+  { return _mm256_castpd_si256(_mm256_and_pd (_mm256_castsi256_pd(a.Data()),_mm256_castsi256_pd( b.Data()))); }
+  NETGEN_INLINE SIMD<mask64,4> operator|| (SIMD<mask64,4> a, SIMD<mask64,4> b)
+  { return _mm256_castpd_si256(_mm256_or_pd (_mm256_castsi256_pd(a.Data()), _mm256_castsi256_pd(b.Data()))); }
+  NETGEN_INLINE SIMD<mask64,4> operator! (SIMD<mask64,4> a)
+  { return _mm256_castpd_si256(_mm256_xor_pd (_mm256_castsi256_pd(a.Data()),_mm256_castsi256_pd( _mm256_cmpeq_epi64(a.Data(),a.Data())))); }
+#endif
+  NETGEN_INLINE SIMD<double,4> If (SIMD<mask64,4> a, SIMD<double,4> b, SIMD<double,4> c)
+  { return _mm256_blendv_pd(c.Data(), b.Data(), _mm256_castsi256_pd(a.Data())); }
+
+  NETGEN_INLINE SIMD<double,4> IfPos (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    auto cp = _mm256_cmp_pd (a.Data(), _mm256_setzero_pd(), _CMP_GT_OS);
+    return _mm256_blendv_pd(c.Data(), b.Data(), cp);
+  }
+
+  NETGEN_INLINE SIMD<double,4> IfZero (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
+  {
+    auto cp = _mm256_cmp_pd (a.Data(), _mm256_setzero_pd(), _CMP_EQ_OS);
+    return _mm256_blendv_pd(c.Data(), b.Data(), cp);
+  }
+
+  NETGEN_INLINE double HSum (SIMD<double,4> sd)
+  {
+    // __m128d hv = _mm_add_pd (_mm256_extractf128_pd(sd.Data(),0), _mm256_extractf128_pd(sd.Data(),1));
+    __m128d hv = (sd.Lo()+sd.Hi()).Data();
+    return _mm_cvtsd_f64 (_mm_hadd_pd (hv, hv));
+  }
+
+  NETGEN_INLINE auto HSum (SIMD<double,4> sd1, SIMD<double,4> sd2)
+  {
+    __m256d hv = _mm256_hadd_pd(sd1.Data(), sd2.Data());
+    __m128d hv2 = _mm_add_pd (_mm256_extractf128_pd(hv,0), _mm256_extractf128_pd(hv,1));
+    return SIMD<double,2>(_mm_cvtsd_f64 (hv2),  _mm_cvtsd_f64(_mm_shuffle_pd (hv2, hv2, 3)));
+  }
+
+  NETGEN_INLINE auto HSum (SIMD<double,4> v1, SIMD<double,4> v2, SIMD<double,4> v3, SIMD<double,4> v4)
+  {
+    __m256d hsum1 = _mm256_hadd_pd (v1.Data(), v2.Data());
+    __m256d hsum2 = _mm256_hadd_pd (v3.Data(), v4.Data());
+    SIMD<double,4> hsum = _mm256_add_pd (_mm256_permute2f128_pd (hsum1, hsum2, 1+2*16),
+                                         _mm256_blend_pd (hsum1, hsum2, 12));
+    return hsum;
+    // return make_tuple(hsum[0], hsum[1], hsum[2], hsum[3]);
+  }
+
+
+  NETGEN_INLINE SIMD<int64_t,4> If (SIMD<mask64,4> a, SIMD<int64_t,4> b, SIMD<int64_t,4> c)
+  { return _mm256_castpd_si256(_mm256_blendv_pd(_mm256_castsi256_pd(c.Data()), _mm256_castsi256_pd(b.Data()),
+                                                _mm256_castsi256_pd(a.Data()))); }
+
+}
+
+#endif // NETGEN_CORE_SIMD_AVX_HPP
+

libsrc/core/simd_avx512.hpp

@@ -0,0 +1,249 @@
+#ifndef NETGEN_CORE_SIMD_AVX512_HPP
+#define NETGEN_CORE_SIMD_AVX512_HPP
+
+/**************************************************************************/
+/* File:   simd_avx512.hpp                                                */
+/* Author: Joachim Schoeberl, Matthias Hochsteger                         */
+/* Date:   25. Mar. 16                                                    */
+/**************************************************************************/
+
+#include <immintrin.h>
+
+namespace ngcore
+{
+
+  template <>
+  class SIMD<mask64,8>
+  {
+    __mmask8 mask;
+  public:
+    SIMD (size_t i)
+      : mask(_mm512_cmpgt_epi64_mask(_mm512_set1_epi64(i),
+                                     _mm512_set_epi64(7, 6, 5, 4, 3, 2, 1, 0)))
+    { ; }
+    SIMD (int i)
+      : mask(_mm512_cmpgt_epi64_mask(_mm512_set1_epi64(i),
+                                     _mm512_set_epi64(7, 6, 5, 4, 3, 2, 1, 0)))
+    { ; }
+    SIMD (int64_t i)
+      : mask(_mm512_cmpgt_epi64_mask(_mm512_set1_epi64(i),
+                                     _mm512_set_epi64(7, 6, 5, 4, 3, 2, 1, 0)))
+    { ; }
+    SIMD (__mmask8 _mask) : mask(_mask) { ; }
+    __mmask8 Data() const { return mask; }
+    static constexpr int Size() { return 8; }
+    static NETGEN_INLINE SIMD<mask64, 8> GetMaskFromBits (unsigned int i)
+    {
+      return SIMD<mask64, 8>(__mmask8(i));
+    }
+  };
+
+  template<>
+  class alignas(64) SIMD<int64_t,8>
+  {
+    __m512i data;
+
+  public:
+    static constexpr int Size() { return 8; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (int64_t val) { data = _mm512_set1_epi64(val); }
+    SIMD (int64_t v0, int64_t v1, int64_t v2, int64_t v3, int64_t v4, int64_t v5, int64_t v6, int64_t v7) { data = _mm512_set_epi64(v7,v6,v5,v4,v3,v2,v1,v0); }
+    SIMD (__m512i _data) { data = _data; }
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<int64_t(int)>>::value, int>::type = 0>
+      SIMD (const T & func)
+    {
+      data = _mm512_set_epi64(func(7), func(6), func(5), func(4), func(3), func(2), func(1), func(0));
+    }
+
+
+    NETGEN_INLINE auto operator[] (int i) const { return ((int64_t*)(&data))[i]; }
+    NETGEN_INLINE __m512i Data() const { return data; }
+    NETGEN_INLINE __m512i & Data() { return data; }
+    static SIMD FirstInt() { return { 0, 1, 2, 3, 4, 5, 6, 7 }; }
+  };
+
+  NETGEN_INLINE SIMD<int64_t,8> operator-(SIMD<int64_t,8> a) { return _mm512_sub_epi64(_mm512_setzero_si512(), a.Data()); }
+
+  NETGEN_INLINE SIMD<int64_t,8> operator+ (SIMD<int64_t,8> a, SIMD<int64_t,8> b) { return _mm512_add_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<int64_t,8> operator- (SIMD<int64_t,8> a, SIMD<int64_t,8> b) { return _mm512_sub_epi64(a.Data(),b.Data()); }
+
+   NETGEN_INLINE SIMD<int64_t,8> If (SIMD<mask64,8> a, SIMD<int64_t,8> b, SIMD<int64_t,8> c)
+  { return _mm512_mask_blend_epi64(a.Data(), c.Data(), b.Data()); }
+
+
+  template<>
+  class alignas(64) SIMD<double,8>
+  {
+    __m512d data;
+  public:
+    static constexpr int Size() { return 8; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (double val) { data = _mm512_set1_pd(val); }
+    SIMD (int val)    { data = _mm512_set1_pd(val); }
+    SIMD (size_t val) { data = _mm512_set1_pd(val); }
+    SIMD (double const * p) { data = _mm512_loadu_pd(p); }
+    SIMD (double const * p, SIMD<mask64,8> mask)
+      { data = _mm512_mask_loadu_pd(_mm512_setzero_pd(), mask.Data(), p); }
+    SIMD (__m512d _data) { data = _data; }
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
+    SIMD (const T & func)
+    {
+      data = _mm512_set_pd(func(7), func(6), func(5), func(4), func(3), func(2), func(1), func(0));
+    }
+
+    void Store (double * p) { _mm512_storeu_pd(p, data); }
+    void Store (double * p, SIMD<mask64,8> mask) { _mm512_mask_storeu_pd(p, mask.Data(), data); }
+
+    template <typename Function>
+    void SIMD_function (const Function & func, std::true_type)
+    {
+      data = (__m512){ func(7), func(6), func(5), func(4),
+                       func(3), func(2), func(1), func(0) };
+    }
+
+    // not a function
+    void SIMD_function (double const * p, std::false_type)
+    {
+      data = _mm512_loadu_pd(p);
+    }
+
+    void SIMD_function (double val, std::false_type)
+    {
+      data = _mm512_set1_pd(val);
+    }
+
+    void SIMD_function (__m512d _data, std::false_type)
+    {
+      data = _data;
+    }
+
+    NETGEN_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
+    NETGEN_INLINE __m512d Data() const { return data; }
+    NETGEN_INLINE __m512d & Data() { return data; }
+
+  };
+
+  NETGEN_INLINE SIMD<double,8> operator- (SIMD<double,8> a) { return -a.Data(); }
+  NETGEN_INLINE SIMD<double,8> operator+ (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_add_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,8> operator- (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_sub_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,8> operator* (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_mul_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,8> operator/ (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_div_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,8> operator* (double a, SIMD<double,8> b) { return _mm512_set1_pd(a)*b.Data(); }
+  NETGEN_INLINE SIMD<double,8> operator* (SIMD<double,8> b, double a) { return _mm512_set1_pd(a)*b.Data(); }
+
+  NETGEN_INLINE SIMD<double,8> sqrt (SIMD<double,8> a) { return _mm512_sqrt_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,8> floor (SIMD<double,8> a) { return _mm512_floor_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,8> ceil (SIMD<double,8> a) { return _mm512_ceil_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,8> fabs (SIMD<double,8> a) { return _mm512_max_pd(a.Data(), -a.Data()); }
+
+  NETGEN_INLINE SIMD<mask64,8> operator<= (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_LE_OQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator< (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_LT_OQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator>= (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_GE_OQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator> (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_GT_OQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator== (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_EQ_OQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator!= (SIMD<double,8> a , SIMD<double,8> b)
+  { return _mm512_cmp_pd_mask (a.Data(), b.Data(), _CMP_NEQ_OQ); }
+
+  NETGEN_INLINE SIMD<mask64,8> operator<= (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(), _MM_CMPINT_LE); }
+  NETGEN_INLINE SIMD<mask64,8> operator< (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(), _MM_CMPINT_LT); }
+  NETGEN_INLINE SIMD<mask64,8> operator>= (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(),  _MM_CMPINT_NLT); }
+  NETGEN_INLINE SIMD<mask64,8> operator> (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(), _MM_CMPINT_NLE); }
+  NETGEN_INLINE SIMD<mask64,8> operator== (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(), _MM_CMPINT_EQ); }
+  NETGEN_INLINE SIMD<mask64,8> operator!= (SIMD<int64_t,8> a , SIMD<int64_t,8> b)
+  { return _mm512_cmp_epi64_mask (a.Data(), b.Data(), _MM_CMPINT_NE); }
+
+  NETGEN_INLINE SIMD<mask64,8> operator&& (SIMD<mask64,8> a, SIMD<mask64,8> b)
+  { return (__mmask8)(a.Data() & b.Data()); }
+  NETGEN_INLINE SIMD<mask64,8> operator|| (SIMD<mask64,8> a, SIMD<mask64,8> b)
+  { return (__mmask8)(a.Data() | b.Data()); }
+  NETGEN_INLINE SIMD<mask64,8> operator! (SIMD<mask64,8> a)
+  { return (__mmask8)(~a.Data()); }
+
+  NETGEN_INLINE SIMD<double,8> If (SIMD<mask64,8> a, SIMD<double,8> b, SIMD<double,8> c)
+  { return _mm512_mask_blend_pd(a.Data(), c.Data(), b.Data()); }
+
+  NETGEN_INLINE SIMD<double,8> IfPos (SIMD<double,8> a, SIMD<double> b, SIMD<double> c)
+  {
+    auto k = _mm512_cmp_pd_mask(a.Data(),_mm512_setzero_pd(), _CMP_GT_OS);
+    return _mm512_mask_blend_pd(k,c.Data(),b.Data());
+  }
+  NETGEN_INLINE SIMD<double,8> IfZero (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
+  {
+    auto k = _mm512_cmp_pd_mask(a.Data(),_mm512_setzero_pd(), _CMP_EQ_OS);
+    return _mm512_mask_blend_pd(k,c.Data(),b.Data());
+  }
+
+
+  NETGEN_INLINE auto Unpack (SIMD<double,8> a, SIMD<double,8> b)
+  {
+    return std::make_tuple(SIMD<double,8>(_mm512_unpacklo_pd(a.Data(),b.Data())),
+                      SIMD<double,8>(_mm512_unpackhi_pd(a.Data(),b.Data())));
+  }
+
+
+  NETGEN_INLINE double HSum (SIMD<double,8> sd)
+  {
+    SIMD<double,4> low = _mm512_extractf64x4_pd(sd.Data(),0);
+    SIMD<double,4> high = _mm512_extractf64x4_pd(sd.Data(),1);
+    return HSum(low)+HSum(high);
+  }
+
+  NETGEN_INLINE auto HSum (SIMD<double,8> sd1, SIMD<double,8> sd2)
+  {
+    return SIMD<double,2>(HSum(sd1), HSum(sd2));
+  }
+
+  NETGEN_INLINE SIMD<double,4> HSum (SIMD<double,8> v1, SIMD<double,8> v2, SIMD<double,8> v3, SIMD<double,8> v4)
+  {
+    SIMD<double> lo,hi;
+    std::tie(lo,hi) = Unpack(v1, v2);
+    SIMD<double> sum01 = lo+hi;
+    std::tie(lo,hi) = Unpack(v3, v4);
+    SIMD<double> sum23 = lo+hi;
+    // sum01  b a b a b a b a
+    // sum23  d c d c d c d c
+    // __m512 perm = _mm512_permutex2var_pd (sum01.Data(), _mm512_set_epi64(1,2,3,4,5,6,7,8), sum23.Data());
+    __m256d ab =  _mm512_extractf64x4_pd(sum01.Data(),0) + _mm512_extractf64x4_pd(sum01.Data(),1);
+    __m256d cd =  _mm512_extractf64x4_pd(sum23.Data(),0) + _mm512_extractf64x4_pd(sum23.Data(),1);
+    return _mm256_add_pd (_mm256_permute2f128_pd (ab, cd, 1+2*16), _mm256_blend_pd (ab, cd, 12));
+  }
+
+  NETGEN_INLINE SIMD<double,8> FMA (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
+  {
+    return _mm512_fmadd_pd (a.Data(), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<double,8> FMA (const double & a, SIMD<double,8> b, SIMD<double,8> c)
+  {
+    return _mm512_fmadd_pd (_mm512_set1_pd(a), b.Data(), c.Data());
+  }
+
+  NETGEN_INLINE SIMD<double,8> FNMA (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
+  {
+    return _mm512_fnmadd_pd (a.Data(), b.Data(), c.Data());
+  }
+  NETGEN_INLINE SIMD<double,8> FNMA (const double & a, SIMD<double,8> b, SIMD<double,8> c)
+  {
+    return _mm512_fnmadd_pd (_mm512_set1_pd(a), b.Data(), c.Data());
+  }
+
+}
+
+#endif // NETGEN_CORE_SIMD_AVX512_HPP

libsrc/core/simd_generic.hpp

@@ -0,0 +1,707 @@
+#ifndef NETGEN_CORE_SIMD_GENERIC_HPP
+#define NETGEN_CORE_SIMD_GENERIC_HPP
+
+/**************************************************************************/
+/* File:   simd_base.hpp                                                  */
+/* Author: Joachim Schoeberl, Matthias Hochsteger                         */
+/* Date:   25. Mar. 16                                                    */
+/**************************************************************************/
+
+#include <type_traits>
+#include <functional>
+#include <tuple>
+
+#include "array.hpp"
+
+namespace ngcore
+{
+
+  constexpr int GetDefaultSIMDSize() {
+#if defined __AVX512F__
+    return 8;
+#elif defined __AVX__
+    return 4;
+#elif defined NETGEN_ARCH_AMD64
+    return 2;
+#else
+    return 2;
+#endif
+  }
+
+
+  template <typename T, int N=GetDefaultSIMDSize()> class SIMD;
+
+  class mask64;
+
+  ////////////////////////////////////////////////////////////////////////////
+  namespace detail {
+    template <typename T, size_t N, size_t... I>
+    auto array_range_impl(std::array<T, N> const& arr,
+                   size_t first,
+                   std::index_sequence<I...>)
+    -> std::array<T, sizeof...(I)> {
+        return {arr[first + I]...};
+    }
+
+    template <size_t S, typename T, size_t N>
+    auto array_range(std::array<T, N> const& arr, size_t first) {
+      return array_range_impl(arr, first, std::make_index_sequence<S>{});
+    }
+  
+  } // namespace detail
+
+  ////////////////////////////////////////////////////////////////////////////
+  // mask
+
+  template <>
+  class SIMD<mask64,1>
+  {
+    int64_t mask;
+  public:
+    SIMD (int64_t i)
+      : mask(i > 0 ? -1 : 0) { ; }
+    bool Data() const { return mask; }
+    static constexpr int Size() { return 1; }
+    auto operator[] (int /* i */) const { return mask; }
+  };
+
+
+  template <int N>
+  class alignas(GetDefaultSIMDSize()*sizeof(int64_t)) SIMD<mask64,N>
+  {
+    static constexpr int N1 = std::min(GetDefaultSIMDSize(), N/2);
+    static constexpr int N2 = N-N1;
+
+    SIMD<mask64,N1> lo;
+    SIMD<mask64,N2> hi;
+  public:
+
+    SIMD (int64_t i) : lo(i), hi(i-N1 ) { ; }
+    SIMD (SIMD<mask64,N1> lo_, SIMD<mask64,N2> hi_) : lo(lo_), hi(hi_) { ; }
+    SIMD<mask64,N1> Lo() const { return lo; }
+    SIMD<mask64,N2> Hi() const { return hi; }
+    static constexpr int Size() { return N; }
+  };
+
+  template<int N>
+  NETGEN_INLINE SIMD<mask64,N> operator&& (SIMD<mask64,N> a, SIMD<mask64,N> b)
+    {
+      if constexpr(N==1) return a.Data() && b.Data();
+      else               return { a.Lo() && b.Lo(), a.Hi() && b.Hi() };
+    }
+
+
+  ////////////////////////////////////////////////////////////////////////////
+  // int64
+
+  template<>
+  class SIMD<int64_t,1>
+  {
+    int64_t data;
+
+  public:
+    static constexpr int Size() { return 1; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+    SIMD (int val) : data{val} {}
+    SIMD (int64_t val) : data{val} {}
+    SIMD (size_t val) : data(val) {}
+    explicit SIMD (std::array<int64_t, 1> arr)
+        : data{arr[0]}
+    {}
+
+    int64_t operator[] (int i) const { return ((int64_t*)(&data))[i]; }
+    auto Data() const { return data; }
+    static SIMD FirstInt(int64_t n0=0) { return {n0}; }
+    template <int I>
+    int64_t Get()
+    {
+      static_assert(I==0);
+      return data;
+    }
+  };
+
+  template<int N>
+  class alignas(GetDefaultSIMDSize()*sizeof(int64_t)) SIMD<int64_t,N>
+  {
+    static constexpr int N1 = std::min(GetDefaultSIMDSize(), N/2);
+    static constexpr int N2 = N-N1;
+
+    SIMD<int64_t,N1> lo;
+    SIMD<int64_t,N2> high;
+
+  public:
+    static constexpr int Size() { return N; }
+
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (int val) : lo{val}, high{val} { ; }
+    SIMD (int64_t val) : lo{val}, high{val} { ; }
+    SIMD (size_t val) : lo{val}, high{val} { ; }
+    SIMD (SIMD<int64_t,N1> lo_, SIMD<int64_t,N2> high_) : lo(lo_), high(high_) { ; }
+
+    explicit SIMD( std::array<int64_t, N> arr )
+        : lo(detail::array_range<N1>(arr, 0)),
+          high(detail::array_range<N2>(arr, N1))
+      {}
+
+    template<typename ...T>
+    explicit SIMD(const T... vals)
+    : lo(detail::array_range<N1>(std::array<int64_t, N>{vals...}, 0)),
+      high(detail::array_range<N2>(std::array<int64_t, N>{vals...}, N1))
+      {
+        static_assert(sizeof...(vals)==N, "wrong number of arguments");
+      }
+
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<int64_t(int)>>::value, int>::type = 0>
+      SIMD (const T & func)
+    {
+      for(auto i : IntRange(N1))
+          lo[i] = func(i);
+      for(auto i : IntRange(N2))
+          high[i] = func(N1+i);
+    }
+
+    auto Lo() const { return lo; }
+    auto Hi() const { return high; }
+
+    int64_t operator[] (int i) const { return ((int64_t*)(&lo))[i]; }
+
+    /*
+    operator tuple<int64_t&,int64_t&,int64_t&,int64_t&> ()
+    { return tuple<int64_t&,int64_t&,int64_t&,int64_t&>((*this)[0], (*this)[1], (*this)[2], (*this)[3]); }
+    */
+
+    /*
+    static SIMD FirstInt() { return { 0, 1, 2, 3 }; }
+    */
+    static SIMD FirstInt(int64_t n0=0) { return {SIMD<int64_t,N1>::FirstInt(n0), SIMD<int64_t,N2>::FirstInt(n0+N1)}; }
+    template <int I>
+    int64_t Get()
+    {
+      static_assert(I>=0 && I<N, "Index out of range");
+      if constexpr(I<N1) return lo.template Get<I>();
+      else               return high.template Get<I-N1>();
+    }
+  };
+
+
+  ////////////////////////////////////////////////////////////////////////////
+  // double
+
+  template<>
+  class SIMD<double,1>
+  {
+    double data;
+
+  public:
+    static constexpr int Size() { return 1; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD & operator= (const SIMD &) = default;
+    SIMD (double val) { data = val; }
+    SIMD (int val)    { data = val; }
+    SIMD (size_t val) { data = val; }
+    SIMD (double const * p) { data = *p; }
+    SIMD (double const * p, SIMD<mask64,1> mask) { data = mask.Data() ? *p : 0.0; }
+    explicit SIMD (std::array<double, 1> arr)
+        : data{arr[0]}
+    {}
+
+    template <typename T, typename std::enable_if<std::is_convertible<T,std::function<double(int)>>::value,int>::type = 0>
+    SIMD (const T & func)
+    {
+      data = func(0);
+    }
+
+    template <typename T, typename std::enable_if<std::is_convertible<T,std::function<double(int)>>::value,int>::type = 0>
+    SIMD & operator= (const T & func)
+    {
+      data = func(0);
+      return *this;
+    }
+
+    void Store (double * p) { *p = data; }
+    void Store (double * p, SIMD<mask64,1> mask) { if (mask.Data()) *p = data; }
+
+    double operator[] (int i) const { return ((double*)(&data))[i]; }
+    double Data() const { return data; }
+    template <int I>
+    double Get()
+    {
+      static_assert(I==0);
+      return data;
+    }
+  };
+
+
+  template<int N>
+  class alignas(GetDefaultSIMDSize()*sizeof(double)) SIMD<double, N>
+  {
+    static constexpr int N1 = std::min(GetDefaultSIMDSize(), N/2);
+    static constexpr int N2 = N-N1;
+
+    SIMD<double, N1> lo;
+    SIMD<double, N2> high;
+
+  public:
+    static constexpr int Size() { return N; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD (SIMD<double,N1> lo_, SIMD<double,N2> hi_) : lo(lo_), high(hi_) { ; }
+
+    template <typename T, typename std::enable_if<std::is_convertible<T,std::function<double(int)>>::value,int>::type = 0>
+    SIMD (const T & func)
+    {
+      double  *p = (double*)this;
+      for(auto i : IntRange(N))
+          p[i] = func(i);
+    }
+
+    template <typename T, typename std::enable_if<std::is_convertible<T,std::function<double(int)>>::value,int>::type = 0>
+    SIMD & operator= (const T & func)
+    {
+      double  *p = (double*)this;
+      for(auto i : IntRange(N))
+          p[i] = func(i);
+      return *this;
+    }
+
+
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (double val) : lo{val}, high{val} { ; }
+    SIMD (int val)    : lo{val}, high{val} { ; }
+    SIMD (size_t val) : lo{val}, high{val} { ; }
+
+    SIMD (double const * p) : lo{p}, high{p+N1} { ; }
+    SIMD (double const * p, SIMD<mask64,N> mask)
+        : lo{p, mask.Lo()}, high{p+N1, mask.Hi()}
+      { }
+    SIMD (double * p) : lo{p}, high{p+N1} { ; }
+    SIMD (double * p, SIMD<mask64,N> mask)
+        : lo{p, mask.Lo()}, high{p+N1, mask.Hi()}
+      { }
+
+    explicit SIMD( std::array<double, N> arr )
+        : lo(detail::array_range<N1>(arr, 0)),
+          high(detail::array_range<N2>(arr, N1))
+      {}
+
+    template<typename ...T>
+    explicit SIMD(const T... vals)
+    : lo(detail::array_range<N1>(std::array<double, N>{vals...}, 0)),
+      high(detail::array_range<N2>(std::array<double, N>{vals...}, N1))
+      {
+        static_assert(sizeof...(vals)==N, "wrong number of arguments");
+      }
+
+    void Store (double * p) { lo.Store(p); high.Store(p+N1); }
+    void Store (double * p, SIMD<mask64,N> mask)
+    {
+      lo.Store(p, mask.Lo());
+      high.Store(p+N1, mask.Hi());
+    }
+
+    auto Lo() const { return lo; }
+    auto Hi() const { return high; }
+
+    double operator[] (int i) const { return ((double*)(&lo))[i]; }
+
+    template<typename=std::enable_if<N==2>>
+    operator std::tuple<double&,double&> ()
+    { 
+	double *p = (double*)this;
+	return std::tuple<double&,double&>(p[0], p[1]); 
+    }
+
+    template<typename=std::enable_if<N==4>>
+    operator std::tuple<double&,double&,double&,double&> ()
+    { return std::tuple<double&,double&,double&,double&>((*this)[0], (*this)[1], (*this)[2], (*this)[3]); }
+
+    template <int I>
+    double Get()
+    {
+      static_assert(I>=0 && I<N, "Index out of range");
+      if constexpr(I<N1) return lo.template Get<I>();
+      else               return high.template Get<I-N1>();
+    }
+    auto Data() const { return *this; }
+  };
+
+
+  // Generic operators for any arithmetic type/simd width
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<T,N> operator+ (SIMD<T,N> a, SIMD<T,N> b) {
+      if constexpr(N==1) return a.Data()+b.Data();
+      else               return { a.Lo()+b.Lo(), a.Hi()+b.Hi() };
+  }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<T,N> operator- (SIMD<T,N> a, SIMD<T,N> b) {
+      if constexpr(N==1) return a.Data()-b.Data();
+      else               return { a.Lo()-b.Lo(), a.Hi()-b.Hi() };
+  }
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<T,N> operator- (SIMD<T,N> a) {
+      if constexpr(N==1) return -a.Data();
+      else               return { -a.Lo(), -a.Hi() };
+  }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<T,N> operator* (SIMD<T,N> a, SIMD<T,N> b) {
+      if constexpr(N==1) return a.Data()*b.Data();
+      else               return { a.Lo()*b.Lo(), a.Hi()*b.Hi() };
+  }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<T,N> operator/ (SIMD<T,N> a, SIMD<T,N> b) {
+      if constexpr(N==1) return a.Data()/b.Data();
+      else               return { a.Lo()/b.Lo(), a.Hi()/b.Hi() };
+  }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator< (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() < b.Data();
+      else               return { a.Lo()<b.Lo(), a.Hi()<b.Hi() };
+    }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator<= (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() <= b.Data();
+      else               return { a.Lo()<=b.Lo(), a.Hi()<=b.Hi() };
+    }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator> (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() > b.Data();
+      else               return { a.Lo()>b.Lo(), a.Hi()>b.Hi() };
+    }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator>= (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() >= b.Data();
+      else               return { a.Lo()>=b.Lo(), a.Hi()>=b.Hi() };
+    }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator== (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() == b.Data();
+      else               return { a.Lo()==b.Lo(), a.Hi()==b.Hi() };
+    }
+
+  template <typename T, int N>
+  NETGEN_INLINE SIMD<mask64,N> operator!= (SIMD<T,N> a, SIMD<T,N> b)
+    {
+      if constexpr(N==1) return a.Data() != b.Data();
+      else               return { a.Lo()!=b.Lo(), a.Hi()!=b.Hi() };
+    }
+
+  // int64_t operators with scalar operand (implement overloads to allow implicit casts for second operand)
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator+ (SIMD<int64_t,N> a, int64_t b) { return a+SIMD<int64_t,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator+ (int64_t a, SIMD<int64_t,N> b) { return SIMD<int64_t,N>(a)+b; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator- (int64_t a, SIMD<int64_t,N> b) { return SIMD<int64_t,N>(a)-b; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator- (SIMD<int64_t,N> a, int64_t b) { return a-SIMD<int64_t,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator* (int64_t a, SIMD<int64_t,N> b) { return SIMD<int64_t,N>(a)*b; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator* (SIMD<int64_t,N> b, int64_t a) { return SIMD<int64_t,N>(a)*b; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator/ (SIMD<int64_t,N> a, int64_t b) { return a/SIMD<int64_t,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> operator/ (int64_t a, SIMD<int64_t,N> b) { return SIMD<int64_t,N>(a)/b; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator+= (SIMD<int64_t,N> & a, SIMD<int64_t,N> b) { a=a+b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator+= (SIMD<int64_t,N> & a, int64_t b) { a+=SIMD<int64_t,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator-= (SIMD<int64_t,N> & a, SIMD<int64_t,N> b) { a = a-b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator-= (SIMD<int64_t,N> & a, int64_t b) { a-=SIMD<int64_t,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator*= (SIMD<int64_t,N> & a, SIMD<int64_t,N> b) { a=a*b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator*= (SIMD<int64_t,N> & a, int64_t b) { a*=SIMD<int64_t,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<int64_t,N> & operator/= (SIMD<int64_t,N> & a, SIMD<int64_t,N> b) { a = a/b; return a; }
+
+  // double operators with scalar operand (implement overloads to allow implicit casts for second operand)
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator+ (SIMD<double,N> a, double b) { return a+SIMD<double,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator+ (double a, SIMD<double,N> b) { return SIMD<double,N>(a)+b; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator- (double a, SIMD<double,N> b) { return SIMD<double,N>(a)-b; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator- (SIMD<double,N> a, double b) { return a-SIMD<double,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator* (double a, SIMD<double,N> b) { return SIMD<double,N>(a)*b; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator* (SIMD<double,N> b, double a) { return SIMD<double,N>(a)*b; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator/ (SIMD<double,N> a, double b) { return a/SIMD<double,N>(b); }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> operator/ (double a, SIMD<double,N> b) { return SIMD<double,N>(a)/b; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator+= (SIMD<double,N> & a, SIMD<double,N> b) { a=a+b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator+= (SIMD<double,N> & a, double b) { a+=SIMD<double,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator-= (SIMD<double,N> & a, SIMD<double,N> b) { a = a-b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator-= (SIMD<double,N> & a, double b) { a-=SIMD<double,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator*= (SIMD<double,N> & a, SIMD<double,N> b) { a=a*b; return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator*= (SIMD<double,N> & a, double b) { a*=SIMD<double,N>(b); return a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> & operator/= (SIMD<double,N> & a, SIMD<double,N> b) { a = a/b; return a; }
+
+  // double functions
+
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> L2Norm2 (SIMD<double,N> a) { return a*a; }
+  template <int N>
+  NETGEN_INLINE SIMD<double,N> Trans (SIMD<double,N> a) { return a; }
+
+  template <int N>
+  NETGEN_INLINE double HSum (SIMD<double,N> a)
+    {
+      if constexpr(N==1)
+          return a.Data();
+      else
+          return HSum(a.Lo()) + HSum(a.Hi());
+    }
+
+  NETGEN_INLINE double IfPos (double a, double b, double c) { return a>0 ? b : c; }
+  NETGEN_INLINE double IfZero (double a, double b, double c) { return a==0. ? b : c; }
+
+  template<typename T, int N>
+  NETGEN_INLINE SIMD<T,N> IfPos (SIMD<T,N> a, SIMD<T,N> b, SIMD<T,N> c)
+    {
+      if constexpr(N==1) return a.Data()>0.0 ? b : c;
+      else               return { IfPos(a.Lo(), b.Lo(), c.Lo()), IfPos(a.Hi(), b.Hi(), c.Hi())};
+
+    }
+
+  template<typename T, int N>
+  NETGEN_INLINE SIMD<T,N> IfZero (SIMD<T,N> a, SIMD<T,N> b, SIMD<T,N> c)
+    {
+      if constexpr(N==1) return a.Data()==0.0 ? b : c;
+      else               return { IfZero(a.Lo(), b.Lo(), c.Lo()), IfZero(a.Hi(), b.Hi(), c.Hi())};
+
+    }
+
+  template<typename T, int N>
+  NETGEN_INLINE SIMD<T,N> If (SIMD<mask64,N> a, SIMD<T,N> b, SIMD<T,N> c)
+    {
+      if constexpr(N==1) return a.Data() ? b : c;
+      else               return { If(a.Lo(), b.Lo(), c.Lo()), If(a.Hi(), b.Hi(), c.Hi())};
+
+    }
+
+  // a*b+c
+  template <typename T1, typename T2, typename T3>
+  NETGEN_INLINE auto FMA(T1 a, T2 b, T3 c)
+  {
+    return c+a*b;
+  }
+
+  template <typename T1, typename T2, typename T3>
+  NETGEN_INLINE auto FNMA(T1 a, T2 b, T3 c)
+  {
+    return c-a*b;
+  }
+
+  // update form of fma
+  template <int N>
+  void FMAasm (SIMD<double,N> a, SIMD<double,N> b, SIMD<double,N> & sum)
+  {
+    sum = FMA(a,b,sum);
+  }
+
+  // update form of fms
+  template <int N>
+  void FNMAasm (SIMD<double,N> a, SIMD<double,N> b, SIMD<double,N> & sum)
+  {
+    // sum -= a*b;
+    sum = FNMA(a,b,sum);
+  }
+
+  
+  template <int i, typename T, int N>
+  T get(SIMD<T,N> a) { return a[i]; }
+
+  template <int NUM, typename FUNC>
+  NETGEN_INLINE void Iterate2 (FUNC f)
+  {
+    if constexpr (NUM > 1) Iterate2<NUM-1> (f);
+    if constexpr (NUM >= 1) f(std::integral_constant<int,NUM-1>());
+  }
+
+
+  template <typename T, int N>
+  ostream & operator<< (ostream & ost, SIMD<T,N> simd)
+  {
+    /*
+    ost << simd[0];
+    for (int i = 1; i < simd.Size(); i++)
+      ost << " " << simd[i];
+    */
+    Iterate2<simd.Size()> ([&] (auto I) {
+        if (I.value != 0) ost << " ";
+        ost << get<I.value>(simd);
+      });
+    return ost;
+  }
+
+  using std::sqrt;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> sqrt (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double>([a](int i)->double { return sqrt(a[i]); } );
+  }
+
+  using std::fabs;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> fabs (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double>([a](int i)->double { return fabs(a[i]); } );
+  }
+
+  using std::floor;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> floor (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double>([a](int i)->double { return floor(a[i]); } );
+  }
+
+  using std::ceil;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> ceil (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double>([a](int i)->double { return ceil(a[i]); } );
+  }
+
+  using std::exp;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> exp (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double>([a](int i)->double { return exp(a[i]); } );
+  }
+
+  using std::log;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> log (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return log(a[i]); } );
+  }
+
+  using std::pow;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> pow (ngcore::SIMD<double,N> a, double x) {
+    return ngcore::SIMD<double,N>([a,x](int i)->double { return pow(a[i],x); } );
+  }
+
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> pow (ngcore::SIMD<double,N> a, ngcore::SIMD<double,N> b) {
+    return ngcore::SIMD<double,N>([a,b](int i)->double { return pow(a[i],b[i]); } );
+  }
+
+  using std::sin;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> sin (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return sin(a[i]); } );
+  }
+
+  using std::cos;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> cos (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return cos(a[i]); } );
+  }
+
+  using std::tan;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> tan (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return tan(a[i]); } );
+  }
+
+  using std::atan;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> atan (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return atan(a[i]); } );
+  }
+
+  using std::atan2;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> atan2 (ngcore::SIMD<double,N> y, ngcore::SIMD<double,N> x) {
+    return ngcore::SIMD<double,N>([y,x](int i)->double { return atan2(y[i], x[i]); } );
+  }
+
+  using std::acos;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> acos (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return acos(a[i]); } );
+  }
+
+  using std::asin;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> asin (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return asin(a[i]); } );
+  }
+
+  using std::sinh;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> sinh (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return sinh(a[i]); } );
+  }
+
+  using std::cosh;
+  template <int N>
+  NETGEN_INLINE ngcore::SIMD<double,N> cosh (ngcore::SIMD<double,N> a) {
+    return ngcore::SIMD<double,N>([a](int i)->double { return cosh(a[i]); } );
+  }
+
+  template<int N, typename T>
+  using MultiSIMD = SIMD<T, N*GetDefaultSIMDSize()>;
+
+  template<int N>
+  NETGEN_INLINE auto Unpack (SIMD<double,N> a, SIMD<double,N> b)
+  {
+    if constexpr(N==1)
+      {
+        return std::make_tuple(SIMD<double,N>{a.Data()}, SIMD<double,N>{b.Data()} );
+      }
+    else if constexpr(N==2)
+      {
+        return std::make_tuple(SIMD<double,N>{ a.Lo(), b.Lo() },
+            SIMD<double,N>{ a.Hi(), b.Hi() });
+      }
+    else
+      {
+        auto [a1,b1] = Unpack(a.Lo(), b.Lo());
+        auto [a2,b2] = Unpack(a.Hi(), b.Hi());
+        return std::make_tuple(SIMD<double,N>{ a1, a2 },
+            SIMD<double,N>{ b1, b2 });
+      }
+  }
+
+}
+
+
+namespace std
+{
+  // structured binding support
+  template <typename T, int N >
+  struct tuple_size<ngcore::SIMD<T,N>> : std::integral_constant<std::size_t, N> {};
+  template<size_t N, typename T, int M> struct tuple_element<N,ngcore::SIMD<T,M>> { using type = T; };
+}
+
+#endif // NETGEN_CORE_SIMD_GENERIC_HPP

libsrc/core/simd_sse.hpp

@@ -0,0 +1,266 @@
+#ifndef NETGEN_CORE_SIMD_SSE_HPP
+#define NETGEN_CORE_SIMD_SSE_HPP
+
+/**************************************************************************/
+/* File:   simd_sse.hpp                                                   */
+/* Author: Joachim Schoeberl, Matthias Hochsteger                         */
+/* Date:   25. Mar. 16                                                    */
+/**************************************************************************/
+
+#include <immintrin.h>
+
+namespace ngcore
+{
+
+  template <>
+  class SIMD<mask64,2>
+  {
+    __m128i mask;
+  public:
+    SIMD (int i)
+      : mask(_mm_cmpgt_epi32(_mm_set1_epi32(i),
+                             _mm_set_epi32(1, 1, 0, 0)))
+    { ; }
+    SIMD (__m128i _mask) : mask(_mask) { ; }
+    __m128i Data() const { return mask; }
+    static constexpr int Size() { return 2; }
+    static NETGEN_INLINE SIMD<mask64, 2> GetMaskFromBits (unsigned int i);
+  };
+
+  static SIMD<mask64, 2> masks_from_2bits[4] = {
+    _mm_set_epi32 (0,0,0,0), _mm_set_epi32 (0,0,-1,0),
+    _mm_set_epi32 (-1,0,0,0), _mm_set_epi32 (-1,0,-1,0),
+  };
+
+  NETGEN_INLINE SIMD<mask64, 2> SIMD<mask64, 2> :: GetMaskFromBits (unsigned int i)
+  {
+    return masks_from_2bits[i & 3];
+  }
+
+
+  template<>
+  class alignas(16) SIMD<int64_t,2>
+  {
+    __m128i data;
+
+  public:
+    static constexpr int Size() { return 2; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD (int64_t v0, int64_t v1) { data = _mm_set_epi64x(v1,v0); }
+    SIMD (std::array<int64_t, 2> arr)
+        : data{_mm_set_epi64x(arr[1],arr[0])}
+    {}
+
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (int64_t val) { data = _mm_set1_epi64x(val); }
+    SIMD (__m128i _data) { data = _data; }
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<int64_t(int)>>::value, int>::type = 0>
+    SIMD (const T & func)
+    {
+      data = _mm_set_epi64(func(1), func(0));
+    }
+
+    NETGEN_INLINE auto operator[] (int i) const { return ((int64_t*)(&data))[i]; }
+    NETGEN_INLINE __m128i Data() const { return data; }
+    NETGEN_INLINE __m128i & Data() { return data; }
+    static SIMD FirstInt(int n0=0) { return { n0, n0+1 }; }
+  };
+
+
+
+NETGEN_INLINE SIMD<int64_t,2> operator-(SIMD<int64_t,2> a) { return _mm_sub_epi64(_mm_setzero_si128(), a.Data()); }
+NETGEN_INLINE SIMD<int64_t,2> operator+ (SIMD<int64_t,2> a, SIMD<int64_t,2> b) { return _mm_add_epi64(a.Data(),b.Data()); }
+NETGEN_INLINE SIMD<int64_t,2> operator- (SIMD<int64_t,2> a, SIMD<int64_t,2> b) { return _mm_sub_epi64(a.Data(),b.Data()); }
+
+
+  template<>
+  class alignas(16) SIMD<double,2>
+  {
+    __m128d data;
+
+  public:
+    static constexpr int Size() { return 2; }
+    SIMD () {}
+    SIMD (const SIMD &) = default;
+    SIMD (double v0, double v1) { data = _mm_set_pd(v1,v0); }
+    SIMD (std::array<double, 2> arr)
+        : data{_mm_set_pd(arr[1], arr[0])}
+    {}
+
+    SIMD & operator= (const SIMD &) = default;
+
+    SIMD (double val) { data = _mm_set1_pd(val); }
+    SIMD (int val)    { data = _mm_set1_pd(val); }
+    SIMD (size_t val) { data = _mm_set1_pd(val); }
+
+    SIMD (double const * p) { data = _mm_loadu_pd(p); }
+    SIMD (double const * p, SIMD<mask64,2> mask)
+      {
+#ifdef __AVX__
+        data = _mm_maskload_pd(p, mask.Data());
+#else
+        // this versions segfaults if p points to the last allowed element
+        // happened on Mac with the new SparseCholesky-factorization
+        // data = _mm_and_pd(_mm_castsi128_pd(mask.Data()), _mm_loadu_pd(p));
+        auto pmask = (int64_t*)&mask;
+        data = _mm_set_pd (pmask[1] ? p[1] : 0.0, pmask[0] ? p[0] : 0.0);
+#endif
+      }
+    SIMD (__m128d _data) { data = _data; }
+
+    void Store (double * p) { _mm_storeu_pd(p, data); }
+    void Store (double * p, SIMD<mask64,2> mask)
+    {
+#ifdef __AVX__
+      _mm_maskstore_pd(p, mask.Data(), data);
+#else
+      /*
+      _mm_storeu_pd (p, _mm_or_pd (_mm_and_pd(_mm_castsi128_pd(mask.Data()), data),
+                                   _mm_andnot_pd(_mm_castsi128_pd(mask.Data()), _mm_loadu_pd(p))));
+      */
+      auto pmask = (int64_t*)&mask;
+      if (pmask[0]) p[0] = (*this)[0];
+      if (pmask[1]) p[1] = (*this)[1];
+#endif
+    }
+
+    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
+    SIMD (const T & func)
+    {
+      data = _mm_set_pd(func(1), func(0));
+    }
+
+    NETGEN_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
+    NETGEN_INLINE __m128d Data() const { return data; }
+    NETGEN_INLINE __m128d & Data() { return data; }
+
+    operator std::tuple<double&,double&> ()
+    {
+      auto pdata = (double*)&data;
+      return std::tuple<double&,double&>(pdata[0], pdata[1]);
+    }
+  };
+
+  NETGEN_INLINE SIMD<double,2> operator- (SIMD<double,2> a) { return _mm_xor_pd(a.Data(), _mm_set1_pd(-0.0)); }
+  NETGEN_INLINE SIMD<double,2> operator+ (SIMD<double,2> a, SIMD<double,2> b) { return _mm_add_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,2> operator- (SIMD<double,2> a, SIMD<double,2> b) { return _mm_sub_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,2> operator* (SIMD<double,2> a, SIMD<double,2> b) { return _mm_mul_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,2> operator/ (SIMD<double,2> a, SIMD<double,2> b) { return _mm_div_pd(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<double,2> operator* (double a, SIMD<double,2> b) { return _mm_set1_pd(a)*b; }
+  NETGEN_INLINE SIMD<double,2> operator* (SIMD<double,2> b, double a) { return _mm_set1_pd(a)*b; }
+
+  template<>
+  NETGEN_INLINE auto Unpack (SIMD<double,2> a, SIMD<double,2> b)
+  {
+    return std::make_tuple(SIMD<double,2>(_mm_unpacklo_pd(a.Data(),b.Data())),
+                      SIMD<double,2>(_mm_unpackhi_pd(a.Data(),b.Data())));
+  }
+
+  NETGEN_INLINE __m128d my_mm_hadd_pd(__m128d a, __m128d b) {
+#if defined(__SSE3__) || defined(__AVX__)
+    return _mm_hadd_pd(a,b);
+#else
+    return _mm_add_pd( _mm_unpacklo_pd(a,b), _mm_unpackhi_pd(a,b) );
+#endif
+  }
+
+#ifndef __AVX__
+  NETGEN_INLINE __m128i my_mm_cmpgt_epi64(__m128i a, __m128i b) {
+    auto  res_lo = _mm_cvtsi128_si64(a)  > _mm_cvtsi128_si64(b) ? -1:0;
+    auto  res_hi = _mm_cvtsi128_si64(_mm_srli_si128(a,8)) > _mm_cvtsi128_si64(_mm_srli_si128(b,8)) ? -1 : 0;
+    return _mm_set_epi64x(res_hi,res_lo);
+  }
+#else
+  NETGEN_INLINE __m128i my_mm_cmpgt_epi64(__m128i a, __m128i b) {
+    return _mm_cmpgt_epi64(a,b);
+  }
+#endif
+
+
+  NETGEN_INLINE SIMD<double,2> sqrt (SIMD<double,2> a) { return _mm_sqrt_pd(a.Data()); }
+  NETGEN_INLINE SIMD<double,2> fabs (SIMD<double,2> a) { return _mm_max_pd(a.Data(), (-a).Data()); }
+  using std::floor;
+  NETGEN_INLINE SIMD<double,2> floor (SIMD<double,2> a)
+  { return ngcore::SIMD<double,2>([&](int i)->double { return floor(a[i]); } ); }
+  using std::ceil;
+  NETGEN_INLINE SIMD<double,2> ceil (SIMD<double,2> a)
+  { return ngcore::SIMD<double,2>([&](int i)->double { return ceil(a[i]); } ); }
+
+  NETGEN_INLINE SIMD<mask64,2> operator<= (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmple_pd(a.Data(),b.Data())); }
+  NETGEN_INLINE SIMD<mask64,2> operator< (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmplt_pd(a.Data(),b.Data())); }
+  NETGEN_INLINE SIMD<mask64,2> operator>= (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmpge_pd(a.Data(),b.Data())); }
+  NETGEN_INLINE SIMD<mask64,2> operator> (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmpgt_pd(a.Data(),b.Data())); }
+  NETGEN_INLINE SIMD<mask64,2> operator== (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmpeq_pd(a.Data(),b.Data())); }
+  NETGEN_INLINE SIMD<mask64,2> operator!= (SIMD<double,2> a , SIMD<double,2> b)
+  { return _mm_castpd_si128( _mm_cmpneq_pd(a.Data(),b.Data())); }
+
+  NETGEN_INLINE SIMD<mask64,2> operator<= (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  _mm_xor_si128(_mm_cmpgt_epi64(a.Data(),b.Data()),_mm_set1_epi32(-1)); }
+  NETGEN_INLINE SIMD<mask64,2> operator< (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  my_mm_cmpgt_epi64(b.Data(),a.Data()); }
+  NETGEN_INLINE SIMD<mask64,2> operator>= (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  _mm_xor_si128(_mm_cmpgt_epi64(b.Data(),a.Data()),_mm_set1_epi32(-1)); }
+  NETGEN_INLINE SIMD<mask64,2> operator> (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  my_mm_cmpgt_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<mask64,2> operator== (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  _mm_cmpeq_epi64(a.Data(),b.Data()); }
+  NETGEN_INLINE SIMD<mask64,2> operator!= (SIMD<int64_t,2> a , SIMD<int64_t,2> b)
+  { return  _mm_xor_si128(_mm_cmpeq_epi64(a.Data(),b.Data()),_mm_set1_epi32(-1)); }
+
+
+
+ NETGEN_INLINE SIMD<mask64,2> operator&& (SIMD<mask64,2> a, SIMD<mask64,2> b)
+  { return _mm_castpd_si128(_mm_and_pd (_mm_castsi128_pd(a.Data()),_mm_castsi128_pd( b.Data()))); }
+  NETGEN_INLINE SIMD<mask64,2> operator|| (SIMD<mask64,2> a, SIMD<mask64,2> b)
+  { return _mm_castpd_si128(_mm_or_pd (_mm_castsi128_pd(a.Data()), _mm_castsi128_pd(b.Data()))); }
+  NETGEN_INLINE SIMD<mask64,2> operator! (SIMD<mask64,2> a)
+  { return _mm_castpd_si128(_mm_xor_pd (_mm_castsi128_pd(a.Data()),_mm_castsi128_pd( _mm_cmpeq_epi64(a.Data(),a.Data())))); }
+#ifdef __SSE4_1__
+  NETGEN_INLINE SIMD<double,2> If (SIMD<mask64,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  { return _mm_blendv_pd(c.Data(), b.Data(), _mm_castsi128_pd(a.Data())); }
+#else
+  NETGEN_INLINE SIMD<double,2> If (SIMD<mask64,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  {
+    return _mm_or_pd(
+                      _mm_andnot_pd(_mm_castsi128_pd(a.Data()),c.Data()),
+                      _mm_and_pd(b.Data(),_mm_castsi128_pd(a.Data()))
+                      );}
+#endif // __SSE4_1__
+
+  NETGEN_INLINE SIMD<double,2> IfPos (SIMD<double,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  { return ngcore::SIMD<double,2>([&](int i)->double { return a[i]>0 ? b[i] : c[i]; }); }
+  NETGEN_INLINE SIMD<double,2> IfZero (SIMD<double,2> a, SIMD<double,2> b, SIMD<double,2> c)
+  { return ngcore::SIMD<double,2>([&](int i)->double { return a[i]==0. ? b[i] : c[i]; }); }
+
+
+  NETGEN_INLINE double HSum (SIMD<double,2> sd)
+  {
+    return _mm_cvtsd_f64 (my_mm_hadd_pd (sd.Data(), sd.Data()));
+  }
+
+  NETGEN_INLINE auto HSum (SIMD<double,2> sd1, SIMD<double,2> sd2)
+  {
+    __m128d hv2 = my_mm_hadd_pd(sd1.Data(), sd2.Data());
+    return SIMD<double,2> (hv2);
+    // return SIMD<double,2>(_mm_cvtsd_f64 (hv2),  _mm_cvtsd_f64(_mm_shuffle_pd (hv2, hv2, 3)));
+  }
+
+  NETGEN_INLINE SIMD<int64_t, 2> If(SIMD<mask64, 2> a, SIMD<int64_t, 2> b,
+                             SIMD<int64_t, 2> c) {
+    return _mm_or_si128(
+                        _mm_andnot_si128(a.Data(),c.Data()),
+                        _mm_and_si128(b.Data(),a.Data())
+                        );
+  }
+
+}
+
+#endif // NETGEN_CORE_SIMD_SSE_HPP

libsrc/core/table.cpp

@@ -18,6 +18,19 @@ namespace ngcore
     size_t size  = entrysize.Size();
     size_t * index = new size_t[size+1];
 
+    if (entrysize.Size() < 100)
+      {
+        size_t mysum = 0;
+        for (size_t i = 0; i < entrysize.Size(); i++)
+          {
+            index[i] = mysum;
+            mysum += entrysize[i];
+          }
+        index[entrysize.Size()] = mysum;
+        return index;
+      }
+
+    
     Array<size_t> partial_sums(TaskManager::GetNumThreads()+1);
     partial_sums[0] = 0;
     ParallelJob
@@ -54,7 +67,7 @@ namespace ngcore
   NGCORE_API size_t * TablePrefixSum64 (FlatArray<size_t> entrysize)
   { return TablePrefixSum2 (entrysize); }
 
-
+  /*
   BaseDynamicTable :: BaseDynamicTable (int size)
     : data(size)
   {
@@ -88,7 +101,6 @@ namespace ngcore
       }
   }
 
-
   BaseDynamicTable :: ~BaseDynamicTable ()
   {
     if (oneblock)
@@ -112,7 +124,7 @@ namespace ngcore
       }
   }
 
-  void BaseDynamicTable :: IncSize (int i, int elsize)
+  void BaseDynamicTable :: IncSize (IndexType i, int elsize)
   {
     if (i < 0 || i >= data.Size())
       {
@@ -135,7 +147,7 @@ namespace ngcore
     line.size++;
   }
 
-  void BaseDynamicTable :: DecSize (int i)
+  void BaseDynamicTable :: DecSize (IndexType i)
   {
     if (i < 0 || i >= data.Size())
       {
@@ -153,6 +165,7 @@ namespace ngcore
 
     line.size--;
   }
+  */
 
   void FilteredTableCreator::Add (size_t blocknr, int data)
   {

libsrc/core/table.hpp

@@ -20,76 +20,76 @@ namespace ngcore
 
 
   template <class T, class IndexType = size_t>
-class FlatTable
+  class FlatTable
-{
-protected:
-  static constexpr IndexType BASE = IndexBASE<IndexType>();  
-  /// number of rows
-  size_t size;
-  /// pointer to first in row
-  size_t * index;
-  /// array of data
-  T * data;
-
-public:
-  FlatTable() = delete;
-
-  NETGEN_INLINE FlatTable(size_t as, size_t * aindex, T * adata)
-    : size(as), index(aindex), data(adata) { ; }
-
-  /// Size of table
-  NETGEN_INLINE size_t Size() const { return size; }
-
-  /// Access entry
-  NETGEN_INLINE const FlatArray<T> operator[] (IndexType i) const
   {
-    i = i-BASE;
+  protected:
-    return FlatArray<T> (index[i+1]-index[i], data+index[i]);
+    static constexpr IndexType BASE = IndexBASE<IndexType>();  
-  }
+    /// number of rows
+    size_t size;
+    /// pointer to first in row
+    size_t * index;
+    /// array of data
+    T * data;
 
-  NETGEN_INLINE T * Data() const { return data; }
-
-  NETGEN_INLINE FlatArray<T> AsArray() const
-  {
-    return FlatArray<T> (index[size]-index[0], data+index[0]);
-  }
-
-  NETGEN_INLINE FlatArray<size_t> IndexArray() const
-  {
-    return FlatArray<size_t, IndexType> (size+1, index);
-  }
-
-  /// takes range starting from position start of end-start elements
-  NETGEN_INLINE FlatTable<T> Range (size_t start, size_t end) const
-  {
-    return FlatTable<T> (end-start, index+start-BASE, data);
-  }
-
-  /// takes range starting from position start of end-start elements
-  NETGEN_INLINE FlatTable<T> Range (T_Range<size_t> range) const
-  {
-    return FlatTable<T> (range.Size(), index+range.First()-BASE, data);
-  }
-
-  NETGEN_INLINE T_Range<IndexType> Range () const
-  {
-    return T_Range<IndexType> (BASE, size+BASE);
-  }
-  
-  class Iterator
-  {
-    const FlatTable & tab;
-    size_t row;
   public:
-    Iterator (const FlatTable & _tab, size_t _row) : tab(_tab), row(_row) { ; }
+    FlatTable() = delete;
-    Iterator & operator++ () { ++row; return *this; }
-    FlatArray<T> operator* () const { return tab[row]; }
-    bool operator!= (const Iterator & it2) { return row != it2.row; }
-  };
 
-  Iterator begin() const { return Iterator(*this, BASE); }
+    NETGEN_INLINE FlatTable(size_t as, size_t * aindex, T * adata)
-  Iterator end() const { return Iterator(*this, BASE+size); }
+      : size(as), index(aindex), data(adata) { ; }
-};
+
+    /// Size of table
+    NETGEN_INLINE size_t Size() const { return size; }
+
+    /// Access entry
+    NETGEN_INLINE const FlatArray<T> operator[] (IndexType i) const
+    {
+      i = i-BASE;
+      return FlatArray<T> (index[i+1]-index[i], data+index[i]);
+    }
+
+    NETGEN_INLINE T * Data() const { return data; }
+
+    NETGEN_INLINE FlatArray<T> AsArray() const
+    {
+      return FlatArray<T> (index[size]-index[0], data+index[0]);
+    }
+
+    NETGEN_INLINE FlatArray<size_t> IndexArray() const
+    {
+      return FlatArray<size_t, IndexType> (size+1, index);
+    }
+
+    /// takes range starting from position start of end-start elements
+    NETGEN_INLINE FlatTable<T> Range (size_t start, size_t end) const
+    {
+      return FlatTable<T> (end-start, index+start-BASE, data);
+    }
+
+    /// takes range starting from position start of end-start elements
+    NETGEN_INLINE FlatTable<T> Range (T_Range<size_t> range) const
+    {
+      return FlatTable<T> (range.Size(), index+range.First()-BASE, data);
+    }
+
+    NETGEN_INLINE T_Range<IndexType> Range () const
+    {
+      return T_Range<IndexType> (BASE, size+BASE);
+    }
+  
+    class Iterator
+    {
+      const FlatTable & tab;
+      size_t row;
+    public:
+      Iterator (const FlatTable & _tab, size_t _row) : tab(_tab), row(_row) { ; }
+      Iterator & operator++ () { ++row; return *this; }
+      FlatArray<T> operator* () const { return tab[row]; }
+      bool operator!= (const Iterator & it2) { return row != it2.row; }
+    };
+
+    Iterator begin() const { return Iterator(*this, BASE); }
+    Iterator end() const { return Iterator(*this, BASE+size); }
+  };
 
   NGCORE_API extern size_t * TablePrefixSum32 (FlatArray<unsigned int> entrysize);
   NGCORE_API extern size_t * TablePrefixSum64 (FlatArray<size_t> entrysize);
@@ -105,106 +105,119 @@ public:
   { return TablePrefixSum64 (entrysize); }
 
 
-/**
+  /**
-    A compact Table container.
+     A compact Table container.
-    A table contains size entries of variable size.
+     A table contains size entries of variable size.
-    The entry sizes must be known at construction.
+     The entry sizes must be known at construction.
-*/
+  */
   template <class T, class IndexType = size_t>
   class Table : public FlatTable<T, IndexType>
-{
-protected:
-
-  using FlatTable<T,IndexType>::size;
-  using FlatTable<T,IndexType>::index;
-  using FlatTable<T,IndexType>::data;
-
-public:
-  ///
-  NETGEN_INLINE Table () : FlatTable<T,IndexType> (0,nullptr,nullptr) { ; }
-  /// Construct table of uniform entrysize
-  NETGEN_INLINE Table (size_t asize, size_t entrysize)
-    : FlatTable<T,IndexType>( asize, new size_t[asize+1], new T[asize*entrysize] )
   {
-    for (size_t i : IntRange(size+1))
+  protected:
-      index[i] = i*entrysize;
-  }
 
-  /// Construct table of variable entrysize
+    using FlatTable<T,IndexType>::size;
-  template <typename TI>
+    using FlatTable<T,IndexType>::index;
-  NETGEN_INLINE Table (FlatArray<TI,IndexType> entrysize)
+    using FlatTable<T,IndexType>::data;
-    : FlatTable<T,IndexType> (0, nullptr, nullptr)
-  {
-    size  = entrysize.Size();
-    index = TablePrefixSum (FlatArray<TI> (entrysize.Size(), entrysize.Data()));
-    size_t cnt = index[size];
-    data = new T[cnt];
-  }
 
-  explicit NETGEN_INLINE Table (const Table & tab2)
+  public:
-    : FlatTable<T,IndexType>(0, nullptr, nullptr)
+    ///
-  {
+    NETGEN_INLINE Table () : FlatTable<T,IndexType> (0,nullptr,nullptr) { ; }
-    size = tab2.Size();
+    /// Construct table of uniform entrysize
+    NETGEN_INLINE Table (size_t asize, size_t entrysize)
+      : FlatTable<T,IndexType>( asize, new size_t[asize+1], new T[asize*entrysize] )
+    {
+      for (size_t i : IntRange(size+1))
+        index[i] = i*entrysize;
+    }
 
-    index = new size_t[size+1];
+    /// Construct table of variable entrysize
-    for (size_t i = 0; i <= size; i++)
+    template <typename TI>
-      index[i] = tab2.index[i];
+    NETGEN_INLINE Table (FlatArray<TI,IndexType> entrysize)
+      : FlatTable<T,IndexType> (0, nullptr, nullptr)
+    {
+      size  = entrysize.Size();
+      index = TablePrefixSum (FlatArray<TI> (entrysize.Size(), entrysize.Data()));
+      size_t cnt = index[size];
+      data = new T[cnt];
+    }
 
-    size_t cnt = index[size];
+    explicit NETGEN_INLINE Table (const Table & tab2)
-    data = new T[cnt];
+      : FlatTable<T,IndexType>(0, nullptr, nullptr)
-    for (size_t i = 0; i < cnt; i++)
+    {
-      data[i] = tab2.data[i];
+      size = tab2.Size();
-  }
 
-  NETGEN_INLINE Table (Table && tab2)
+      index = new size_t[size+1];
-    : FlatTable<T,IndexType>(0, nullptr, nullptr)
+      for (size_t i = 0; i <= size; i++)
-  {
+        index[i] = tab2.index[i];
-    Swap (size, tab2.size);
-    Swap (index, tab2.index);
-    Swap (data, tab2.data);
-  }
 
-  NETGEN_INLINE Table & operator= (Table && tab2)
+      size_t cnt = index[size];
-  {
+      data = new T[cnt];
-    Swap (size, tab2.size);
+      for (size_t i = 0; i < cnt; i++)
-    Swap (index, tab2.index);
+        data[i] = tab2.data[i];
-    Swap (data, tab2.data);
+    }
-    return *this;
+
-  }
+    NETGEN_INLINE Table (Table && tab2)
+      : FlatTable<T,IndexType>(0, nullptr, nullptr)
+    {
+      tab2.mt.Free(tab2.GetMemUsage());
+      Swap (size, tab2.size);
+      Swap (index, tab2.index);
+      Swap (data, tab2.data);
+    }
+
+    NETGEN_INLINE Table & operator= (Table && tab2)
+    {
+      mt.Swap(GetMemUsage(), tab2.mt, tab2.GetMemUsage());
+      Swap (size, tab2.size);
+      Swap (index, tab2.index);
+      Swap (data, tab2.data);
+      return *this;
+    }
 
 
 
-  /// Delete data
+    /// Delete data
-  NETGEN_INLINE ~Table ()
+    NETGEN_INLINE ~Table ()
-  {
+    {
-    delete [] data;
+      mt.Free(GetMemUsage());
-    delete [] index;
+      delete [] data;
-  }
+      delete [] index;
+    }
 
-  /// Size of table
+    /// Size of table
-  using FlatTable<T,IndexType>::Size;
+    using FlatTable<T,IndexType>::Size;
 
-  /// number of elements in all rows
+    /// number of elements in all rows
-  NETGEN_INLINE size_t NElements() const { return index[size]; }
+    NETGEN_INLINE size_t NElements() const { return index[size]; }
 
-  using FlatTable<T,IndexType>::operator[];
+    using FlatTable<T,IndexType>::operator[];
-};
+
+    NETGEN_INLINE void StartMemoryTracing (int /* mem_id */)
+    {
+      mt.Alloc(GetMemUsage());
+    }
+    const MemoryTracer& GetMemoryTracer() const { return mt; }
+
+  private:
+    size_t GetMemUsage() const { return size == 0 ? 0 : sizeof(T)*index[size] + sizeof(IndexType) * size+1; }
+    MemoryTracer mt;
+  };
 
 
-/// Print table
+  /// Print table
   template <class T, typename IndexType>
   inline ostream & operator<< (ostream & s, const Table<T,IndexType> & table)
-{
+  {
-  for (auto i : table.Range())
+    for (auto i : table.Range())
-    {
+      {
-      s << i << ":";
+        s << i << ":";
-      for (auto el : table[i])
+        for (auto el : table[i])
-	s << " " << el;
+          s << " " << el;
-      s << "\n";
+        s << "\n";
-    }
+      }
-  s << std::flush;
+    s << std::flush;
-  return s;
+    return s;
-}
+  }
 
 
 
@@ -349,142 +362,262 @@ public:
   };
 
 
-  /// Base class to generic DynamicTable.
-  class BaseDynamicTable
-  {
-  protected:
-
-    ///
-    struct linestruct
-    {
-      ///
-      int size;
-      ///
-      int maxsize;
-      ///
-      void * col;
-    };
-
-    ///
-    Array<linestruct> data;
-    ///
-    char * oneblock;
-
-  public:
-    ///
-    NGCORE_API BaseDynamicTable (int size);
-    ///
-    NGCORE_API BaseDynamicTable (const Array<int> & entrysizes, int elemsize);
-    ///
-    NGCORE_API ~BaseDynamicTable ();
-
-    /// Changes Size of table to size, deletes data
-    NGCORE_API void SetSize (int size);
-    ///
-    NGCORE_API void IncSize (int i, int elsize);
-
-    NGCORE_API void DecSize (int i);
-  };
-
-
 
   /**
-      A dynamic table class.
+     A dynamic table class.
 
-      A DynamicTable contains entries of variable size. Entry sizes can
+     A DynamicTable contains entries of variable size. Entry sizes can
-      be increased dynamically.
+     be increased dynamically.
   */
-  template <class T>
+  template <class T, class IndexType = size_t>
-  class DynamicTable : public BaseDynamicTable
+  class DynamicTable 
   {
+  protected:
+    static constexpr IndexType BASE = IndexBASE<IndexType>();
+
+    struct linestruct
+    {
+      int size;
+      int maxsize;
+      T * col;
+    };
+
+    Array<linestruct, IndexType> data;
+    T * oneblock = nullptr;
+  
   public:
     /// Creates table of size size
     DynamicTable (int size = 0)
-      : BaseDynamicTable (size) { ; }
+      : data(size)
-
-    /// Creates table with a priori fixed entry sizes.
-    DynamicTable (const Array<int> & entrysizes)
-      : BaseDynamicTable (entrysizes, sizeof(T)) { ; }
-
-    /// Inserts element acont into row i. Does not test if already used.
-    void Add (int i, const T & acont)
     {
-      if (data[i].size == data[i].maxsize)
+      for (auto & d : data)
-        IncSize (i, sizeof (T));
+        {
-      else
+          d.maxsize = 0;
-        data[i].size++;
+          d.size = 0;
-      static_cast<T*> (data[i].col) [data[i].size-1] = acont;
+          d.col = nullptr;
+        }
+      oneblock = nullptr;
     }
 
+    /// Creates table with a priori fixed entry sizes.
+    DynamicTable (const Array<int, IndexType> & entrysizes)
+      : data(entrysizes.Size())
+    {
+      size_t cnt = 0;
+      // size_t n = entrysizes.Size();
+      
+      for (auto es : entrysizes)
+        cnt += es;
+      oneblock = new T[cnt];
+      
+      cnt = 0;
+      for (auto i : data.Range())
+        {
+          data[i].maxsize = entrysizes[i];
+          data[i].size = 0;
+          data[i].col = &oneblock[cnt];
+          cnt += entrysizes[i];
+        }
+    }
+
+    DynamicTable (DynamicTable && tab2)
+    {
+      Swap (data, tab2.data);
+      Swap (oneblock, tab2.oneblock);
+    }
+
+    ~DynamicTable ()
+    {
+      if (oneblock)
+        delete [] oneblock;
+      else
+        for (auto & d : data)
+          delete [] d.col;
+    }
+    
+    DynamicTable & operator= (DynamicTable && tab2)
+    {
+      Swap (data, tab2.data);
+      Swap (oneblock, tab2.oneblock);
+      return *this;
+    }
+
+    /// Changes Size of table to size, deletes data
+    void SetSize (int size)
+    {
+      for (auto & d : data)
+        delete [] d.col;
+    
+      data.SetSize(size);
+      for (auto & d : data)
+        {
+          d.maxsize = 0;
+          d.size = 0;
+          d.col = nullptr;
+        }
+    }      
+
+    void ChangeSize (size_t size)
+    {
+      if (oneblock)
+        throw Exception ("cannot change size of oneblock dynamic table");
+      
+      size_t oldsize = data.Size();
+      if (size == oldsize) 
+        return;
+      
+      if (size < oldsize)
+        for (int i = size; i < oldsize; i++)
+          delete [] data[i+BASE].col;
+      
+      data.SetSize(size);
+
+      for (int i = oldsize; i < size; i++)
+        {
+          data[i+BASE].maxsize = 0;
+          data[i+BASE].size = 0;
+          data[i+BASE].col = nullptr;
+        }    
+    }
+    
+
+    
+    ///
+    void IncSize (IndexType i)
+    {
+      NETGEN_CHECK_RANGE(i,BASE,data.Size()+BASE);
+    
+      linestruct & line = data[i];
+    
+      if (line.size == line.maxsize)
+        {
+          T * p;
+          if constexpr (std::is_default_constructible<T>::value)
+            p = new T[(2*line.maxsize+5)];
+          else
+            p = reinterpret_cast<T*>(new char[(2*line.maxsize+5)*sizeof(T)]);
+          for (size_t i = 0; i < line.maxsize; i++)
+            p[i] = std::move(line.col[i]);
+          // memcpy (p, line.col, line.maxsize * sizeof(T));
+          delete [] line.col;
+          line.col = p;
+          line.maxsize = 2*line.maxsize+5;
+        }
+    
+      line.size++;
+    }
+  
+    void DecSize (IndexType i)
+    {
+      NETGEN_CHECK_RANGE(i,BASE,data.Size()+BASE);
+      linestruct & line = data[i];
+    
+#ifdef NETGEN_ENABLE_CHECK_RANGE
+      if (line.size == 0)
+        throw Exception ("BaseDynamicTable::Dec: EntrySize < 0");
+#endif
+    
+      line.size--;
+    }
+  
+  
+    /// Inserts element acont into row i. Does not test if already used.
+    void Add (IndexType i, const T & acont)
+    {
+      if (data[i].size == data[i].maxsize)
+        this->IncSize (i);
+      else
+        data[i].size++;
+      data[i].col[data[i].size-1] = acont;
+    }
+  
     /// Inserts element acont into row i, iff not yet exists.
-    void AddUnique (int i, const T & cont)
+    void AddUnique (IndexType i, const T & cont)
     {
       int es = EntrySize (i);
-      int * line = const_cast<int*> (GetLine (i));
+      T * line = data[i].col;
       for (int j = 0; j < es; j++)
         if (line[j] == cont)
           return;
       Add (i, cont);
     }
-
+  
 
     /// Inserts element acont into row i. Does not test if already used.
-    void AddEmpty (int i)
+    void AddEmpty (IndexType i)
     {
-      IncSize (i, sizeof (T));
+      IncSize (i);
     }
-
+  
     /** Set the nr-th element in the i-th row to acont.
         Does not check for overflow. */
-    void Set (int i, int nr, const T & acont)
+    void Set (IndexType i, int nr, const T & acont)
-    { static_cast<T*> (data[i].col)[nr] = acont; }
+    {
+      data[i].col[nr] = acont;
+    }
 
 
     /** Returns the nr-th element in the i-th row.
-      Does not check for overflow. */
+        Does not check for overflow. */
-    const T & Get (int i, int nr) const
+    const T & Get (IndexType i, int nr) const
-    { return static_cast<T*> (data[i].col)[nr]; }
+    {
-
+      return data[i].col[nr];
-
+    }
+  
+  
     /** Returns pointer to the first element in row i. */
-    const T * GetLine (int i) const
+    const T * GetLine (IndexType i) const
-    { return static_cast<T*> (data[i].col); }
+    {
-
+      return data[i].col;
-
+    }
+  
     /// Returns size of the table.
-    int Size () const
+    size_t Size () const
-    { return data.Size(); }
+    {
+      return data.Size();
+    }
+
+    auto Range () const
+    {
+      return data.Range();
+    }
 
     /// Returns size of the i-th row.
-    int EntrySize (int i) const
+    int EntrySize (IndexType i) const
-    { return data[i].size; }
+    {
+      return data[i].size;
+    }
 
     ///
-    void DecEntrySize (int i)
+    void DecEntrySize (IndexType i)
-    { DecSize(i); }
+    {
-
+      DecSize(i);
+    }
+  
     /// Access entry i
-    FlatArray<T> operator[] (int i)
+    FlatArray<T> operator[] (IndexType i)
-    { return FlatArray<T> (data[i].size, static_cast<T*> (data[i].col)); }
+    {
-
+      return FlatArray<T> (data[i].size, data[i].col);
+    }
+  
     /*
-    typedef const FlatArray<T> ConstFlatArray;
+      typedef const FlatArray<T> ConstFlatArray;
-    /// Access entry i
+      /// Access entry i
-    ConstFlatArray operator[] (int i) const
+      ConstFlatArray operator[] (int i) const
-    { return FlatArray<T> (data[i].size, static_cast<T*> (data[i].col)); }
+      { return FlatArray<T> (data[i].size, static_cast<T*> (data[i].col)); }
     */
-    FlatArray<T> operator[] (int i) const
+    FlatArray<T> operator[] (IndexType i) const
-    { return FlatArray<T> (data[i].size, static_cast<T*> (data[i].col)); }
+    {
+      return FlatArray<T> (data[i].size, data[i].col);
+    }
   };
 
 
-
-
   /// Print table
   template <class T>
   inline ostream & operator<< (ostream & s, const DynamicTable<T> & table)
   {
-    for (int i = 0; i < table.Size(); i++)
+    for (auto i : Range(table))
       {
         s << i << ":";
         for (int j = 0; j < table[i].Size(); j++)

libsrc/core/taskmanager.cpp

@@ -159,29 +159,18 @@ namespace ngcore
       active_workers = 0;
 
       static int cnt = 0;
-      char buf[100];
       if (use_paje_trace)
-        {
+          trace = new PajeTrace(num_threads, "ng" + ToString(cnt++));
-#ifdef PARALLEL
-          int is_init = -1;
-          MPI_Initialized(&is_init);
-          if (is_init)
-            sprintf(buf, "ng%d_rank%d.trace", cnt++, NgMPI_Comm(MPI_COMM_WORLD).Rank());
-          else
-#endif
-            sprintf(buf, "ng%d.trace", cnt++);
-        }
-      else
-        buf[0] = 0;
-      //sprintf(buf, "");
-      trace = new PajeTrace(num_threads, buf);
     }
 
 
   TaskManager :: ~TaskManager ()
   {
-    delete trace;
+    if (use_paje_trace)
-    trace = nullptr;
+      {
+        delete trace;
+        trace = nullptr;
+      }
     num_threads = 1;
   }
 
@@ -212,14 +201,14 @@ namespace ngcore
       ;
   }
 
-  static size_t calibrate_init_tsc = __rdtsc();
+  static size_t calibrate_init_tsc = GetTimeCounter();
   typedef std::chrono::system_clock TClock;
   static TClock::time_point calibrate_init_clock = TClock::now();
   
   void TaskManager :: StopWorkers()
   {
     done = true;
-    double delta_tsc = __rdtsc()-calibrate_init_tsc;
+    double delta_tsc = GetTimeCounter()-calibrate_init_tsc;
     double delta_sec = std::chrono::duration<double>(TClock::now()-calibrate_init_clock).count();
     double frequ = (delta_sec != 0) ? delta_tsc/delta_sec : 2.7e9;
     
@@ -348,8 +337,28 @@ namespace ngcore
         return;
       }
     
+    if (antasks == 1)
+      {
+        if (trace)
+          trace->StartJob(jobnr, afunc.target_type());
+        jobnr++;
+        if (startup_function) (*startup_function)();
+        TaskInfo ti;
+        ti.task_nr = 0;
+        ti.ntasks = 1;
+        ti.thread_nr = 0; ti.nthreads = 1;
+        {
+          RegionTracer t(ti.thread_nr, jobnr, RegionTracer::ID_JOB, ti.task_nr);
+          afunc(ti);
+        }
+        if (cleanup_function) (*cleanup_function)();
+        if (trace)
+          trace->StopJob();
+        return;
+      }
     
-    trace->StartJob(jobnr, afunc.target_type());
+    if (trace)
+        trace->StartJob(jobnr, afunc.target_type());
 
     func = &afunc;
 
@@ -412,14 +421,19 @@ namespace ngcore
       if (workers_on_node[j])
         {
           while (complete[j] != jobnr)
+          {
+#ifdef NETGEN_ARCH_AMD64
             _mm_pause();
+#endif // NETGEN_ARCH_AMD64
+          }
         }
 
     func = nullptr;
     if (ex)
       throw Exception (*ex);
 
-    trace->StopJob();
+    if (trace)
+        trace->StopJob();
   }
     
   void TaskManager :: Loop(int thd)

libsrc/core/taskmanager.hpp

@@ -17,6 +17,12 @@
 #include "paje_trace.hpp"
 #include "profiler.hpp"
 
+#ifdef USE_NUMA
+#include <numa.h>
+#include <sched.h>
+#endif
+
+
 namespace ngcore
 {
   using std::atomic;
@@ -1062,11 +1068,11 @@ public:
   {
     static Timer timer("ComputeColoring - "+Demangle(typeid(Tmask).name())); RegionTimer rt(timer);
     static_assert(sizeof(unsigned int)==4, "Adapt type of mask array");
-    auto n = colors.Size();
+    size_t n = colors.Size();
 
     Array<unsigned int> mask(ndofs);
 
-    int colored_blocks = 0;
+    size_t colored_blocks = 0;
 
     // We are coloring with 32 colors at once and use each bit to mask conflicts
     unsigned int check = 0;

libsrc/core/utils.cpp

@@ -13,7 +13,15 @@ namespace ngcore
 
 #ifdef WIN32
   // windows does demangling in typeid(T).name()
-  NGCORE_API std::string Demangle(const char* typeinfo) { return typeinfo; }
+  NGCORE_API std::string Demangle(const char* typeinfo) {
+      std::string name = typeinfo;
+      // remove "class " and "struct " at beginning of type names to be consistent with demangled names of gcc/clang
+      if(name.find("class ") == 0)
+          name.erase(0,6);
+      if(name.find("struct ") == 0)
+          name.erase(0,7);
+      return name;
+  }
 #else
   NGCORE_API std::string Demangle(const char* typeinfo)
   {

libsrc/core/utils.hpp

@@ -8,13 +8,19 @@
 #include <sstream>
 #include <string>
 
+#include "ngcore_api.hpp"       // for NGCORE_API and CPU arch macros
+
+#if defined(__APPLE__) && defined(NETGEN_ARCH_ARM64)
+#include <mach/mach_time.h>
+#endif
+
+#ifdef NETGEN_ARCH_AMD64
 #ifdef WIN32
 #include <intrin.h>   // for __rdtsc()  CPU time step counter
 #else
 #include <x86intrin.h>   // for __rdtsc()  CPU time step counter
 #endif // WIN32
-
+#endif // NETGEN_ARCH_AMD64
-#include "ngcore_api.hpp"       // for NGCORE_API
 
 namespace ngcore
 {
@@ -23,6 +29,10 @@ namespace ngcore
   
   NGCORE_API std::string Demangle(const char* typeinfo);
 
+  template<typename T>
+  std::string GetName(const T& obj)
+  { return Demangle(typeid(obj).name()); }
+
 #if defined(__GNUC__)
   inline bool likely (bool x) { return bool(__builtin_expect(long(x), 1L)); }
   inline bool unlikely (bool x) { return bool(__builtin_expect(long(x), 0L)); }
@@ -48,7 +58,19 @@ namespace ngcore
 
   inline TTimePoint GetTimeCounter() noexcept
   {
-      return TTimePoint(__rdtsc());
+#if defined(__APPLE__) && defined(NETGEN_ARCH_ARM64)
+    return mach_absolute_time();
+#elif defined(NETGEN_ARCH_AMD64)
+    return __rdtsc();
+#elif defined(NETGEN_ARCH_ARM64) && defined(__GNUC__)
+    // __GNUC__ is also defined by CLANG. Use inline asm to read Generic Timer
+    unsigned long long tics;
+    __asm __volatile("mrs %0, CNTVCT_EL0" : "=&r" (tics));
+    return tics;
+#else
+#warning "Unsupported CPU architecture"
+    return 0;
+#endif
   }
 
   template <class T>
@@ -157,7 +179,9 @@ namespace ngcore
       while (!m.compare_exchange_weak(should, true))
         {
           should = false;
+#ifdef NETGEN_ARCH_AMD64
           _mm_pause();
+#endif // NETGEN_ARCH_AMD64
         }
     }
     void unlock()

libsrc/core/version.cpp

@@ -0,0 +1,29 @@
+#include <map>
+
+#include <netgen_version.hpp>
+#include "exception.hpp"
+#include "version.hpp"
+
+namespace ngcore
+{
+  // clang-tidy should ignore this static object
+  static std::map<std::string, VersionInfo> library_versions;  // NOLINT
+
+  const VersionInfo& GetLibraryVersion(const std::string& library)
+  { return library_versions[library]; }
+
+  const std::map<std::string, VersionInfo>& GetLibraryVersions()
+  { return library_versions; }
+
+  void SetLibraryVersion(const std::string& library, const VersionInfo& version)
+  {
+    if(library_versions.count(library) && (library_versions[library] != version))
+      throw Exception("Failed to set library version for " + library + " to " + version.to_string() + ": version already set to " + library_versions[library].to_string());
+    library_versions[library] = version;
+  }
+
+  static bool dummy = [](){
+    SetLibraryVersion("netgen", NETGEN_VERSION);
+    return true;
+  }();
+} // namespace ngcore

libsrc/core/version.hpp

@@ -80,6 +80,10 @@ namespace ngcore
       return mayor_ == other.mayor_ && minor_ == other.minor_ && release == other.release
         && patch == other.patch;
     }
+    bool operator !=(const VersionInfo& other) const
+    {
+      return !(*this==other);
+    }
     bool operator >(const VersionInfo& other) const { return other < (*this); }
     bool operator <=(const VersionInfo& other) const { return !((*this) > other); }
     bool operator >=(const VersionInfo& other) const { return !((*this) < other); }
@@ -89,6 +93,10 @@ namespace ngcore
   {
     return ost << version.to_string();
   }
+
+  NGCORE_API const VersionInfo& GetLibraryVersion(const std::string& library);
+  NGCORE_API const std::map<std::string, VersionInfo>& GetLibraryVersions();
+  NGCORE_API void SetLibraryVersion(const std::string& library, const VersionInfo& version);
 } // namespace ngcore
 
 #endif // NETGEN_CORE_VERSION_HPP

libsrc/core/xbool.hpp

@@ -20,17 +20,27 @@ namespace ngcore
 
   public:
     xbool (bool b) : state(b ? 2 : 0) { ; }
-    xbool (TMAYBE x) : state(1) { ; }
+    xbool (TMAYBE /* x */) : state(1) { ; }
     xbool () = default;
     xbool (const xbool &) = default;
     
     xbool & operator= (bool b) { state = b ? 2 : 0; return *this; }
-    xbool & operator= (TMAYBE x) { state = 1; return *this; }
+    xbool & operator= (TMAYBE /* x */) { state = 1; return *this; }
     
     bool IsTrue () const { return state == 2; }
     bool IsMaybe () const { return state == 1; }
     bool IsFalse () const { return state == 0; }
+    bool IsMaybeTrue() const { return state >= 1; } 
+    bool IsMaybeFalse() const { return state <= 1; } 
+    friend ostream & operator<< (ostream & ost, xbool xb);
   };
+
+
+  static char output[] = "0?1";
+  inline ostream & operator<< (ostream & ost, xbool xb)
+  {
+    return ost << output[xb.state];
+  }
   
 } // namespace ngcore
 

libsrc/csg/CMakeLists.txt

@@ -11,7 +11,7 @@ if(APPLE)
     set_target_properties( csg PROPERTIES SUFFIX ".so")
 endif(APPLE)
 
-target_link_libraries(csg PUBLIC mesh PRIVATE netgen_python)
+target_link_libraries(csg PUBLIC mesh PRIVATE "$<BUILD_INTERFACE:netgen_python>")
 if(NOT WIN32)
   install( TARGETS csg ${NG_INSTALL_DIR})
 endif(NOT WIN32)
@@ -20,7 +20,7 @@ target_link_libraries(csg PUBLIC ngcore)
 
 if(USE_GUI)
     add_library(csgvis ${NG_LIB_TYPE} vscsg.cpp )
-    target_link_libraries(csgvis PRIVATE netgen_python PUBLIC ngcore)
+    target_link_libraries(csgvis PRIVATE "$<BUILD_INTERFACE:netgen_python>" PUBLIC ngcore)
     if(NOT WIN32)
         target_link_libraries(csgvis PUBLIC csg visual)
         if(APPLE)

libsrc/csg/csgeom.cpp

@@ -130,6 +130,7 @@ namespace netgen
                Point<3> & newp, EdgePointGeomInfo & newgi) const
   {
     Point<3> hnewp = p1+secpoint*(p2-p1);
+
     //(*testout) << "hnewp " << hnewp << " s1 " << surfi1 << " s2 " << surfi2 << endl;
     if (surfi1 != -1 && surfi2 != -1 && surfi1 != surfi2)
       {
@@ -175,9 +176,8 @@ namespace netgen
     
     solids.DeleteAll ();
 
-    for (int i = 0; i < splinecurves2d.Size(); i++)
-      delete splinecurves2d[i];
     splinecurves2d.DeleteAll();
+    splinecurves3d.DeleteAll();
     
     /*
     for (int i = 0; i < surfaces.Size(); i++)
@@ -711,24 +711,24 @@ namespace netgen
 
 
 
-  void CSGeometry :: SetSplineCurve (const char * name, SplineGeometry<2> * spl)
+  void CSGeometry :: SetSplineCurve (const char * name, shared_ptr<SplineGeometry<2>> spl)
   {
     splinecurves2d.Set(name,spl);
   }
-  void CSGeometry :: SetSplineCurve (const char * name, SplineGeometry<3> * spl)
+  void CSGeometry :: SetSplineCurve (const char * name, shared_ptr<SplineGeometry<3>> spl)
   {
     splinecurves3d.Set(name,spl);
   }
 
 
-  const SplineGeometry<2> * CSGeometry :: GetSplineCurve2d (const string & name) const
+  shared_ptr<SplineGeometry<2>> CSGeometry :: GetSplineCurve2d (const string & name) const
   {
     if (splinecurves2d.Used(name))
       return splinecurves2d[name];
     else
       return NULL;
   }
-  const SplineGeometry<3> * CSGeometry :: GetSplineCurve3d (const string & name) const
+  shared_ptr<SplineGeometry<3>> CSGeometry :: GetSplineCurve3d (const string & name) const
   {
     if (splinecurves3d.Used(name))
       return splinecurves3d[name];

libsrc/csg/csgeom.hpp

@@ -115,9 +115,9 @@ namespace netgen
     SymbolTable<Solid*> solids;
 
     /// all 2d splinecurves
-    SymbolTable< SplineGeometry<2>* > splinecurves2d;
+    SymbolTable<shared_ptr<SplineGeometry<2>>> splinecurves2d;
     /// all 3d splinecurves
-    SymbolTable< SplineGeometry<3>* > splinecurves3d;
+    SymbolTable<shared_ptr<SplineGeometry<3>>> splinecurves3d;
 
     /// all top level objects: solids and surfaces
     NgArray<TopLevelObject*> toplevelobjects;
@@ -232,10 +232,10 @@ namespace netgen
     const SymbolTable<Solid*> & GetSolids () const { return solids; }
 
 
-    void SetSplineCurve (const char * name, SplineGeometry<2> * spl);
+    void SetSplineCurve (const char * name, shared_ptr<SplineGeometry<2>> spl);
-    void SetSplineCurve (const char * name, SplineGeometry<3> * spl);
+    void SetSplineCurve (const char * name, shared_ptr<SplineGeometry<3>> spl);
-    const SplineGeometry<2> * GetSplineCurve2d (const string & name) const;
+    shared_ptr<SplineGeometry<2>> GetSplineCurve2d (const string & name) const;
-    const SplineGeometry<3> * GetSplineCurve3d (const string & name) const;
+    shared_ptr<SplineGeometry<3>> GetSplineCurve3d (const string & name) const;
 
     void DoArchive(Archive& archive) override;
     

libsrc/csg/csgparser.cpp

@@ -511,8 +511,8 @@ namespace netgen
 		  break;
 		}
 	      
-	      Primitive * nprim = new Extrusion(*(geom->GetSplineCurve3d(epath)),
+	      Primitive * nprim = new Extrusion(geom->GetSplineCurve3d(epath),
-						*(geom->GetSplineCurve2d(profile)),
+						geom->GetSplineCurve2d(profile),
 						z_dir);
 	      geom->AddSurfaces (nprim);
 	      return new Solid(nprim);
@@ -1186,7 +1186,7 @@ namespace netgen
 		ParseChar (scan, '=');
 		ParseChar (scan, '(');
 		
-		SplineGeometry<2> * newspline = new SplineGeometry<2>;
+		auto newspline = make_shared<SplineGeometry<2>>();
 		// newspline->CSGLoad(scan);
 		LoadSpline (*newspline, scan);
 
@@ -1212,7 +1212,7 @@ namespace netgen
 		ParseChar (scan, '=');
 		ParseChar (scan, '(');
 		
-		SplineGeometry<3> * newspline = new SplineGeometry<3>;
+		auto newspline = make_shared<SplineGeometry<3>>();
 		// newspline->CSGLoad(scan);
 		LoadSpline (*newspline, scan);
 

libsrc/csg/edgeflw.cpp

@@ -970,7 +970,7 @@ namespace netgen
   
     for (int i = 0; i < geometry.GetNTopLevelObjects(); i++)
       {
-	Solid * locsol;
+	// Solid * locsol;
 
 	if (geometry.GetTopLevelObject(i)->GetLayer() != layer) 
 	  continue;
@@ -978,7 +978,8 @@ namespace netgen
 	const Solid * sol = geometry.GetTopLevelObject(i)->GetSolid();
 	const Surface * surf = geometry.GetTopLevelObject(i)->GetSurface();
 
-	sol -> TangentialSolid (hp, locsol, locsurfind, size*ideps);
+	// sol -> TangentialSolid (hp, locsol, locsurfind, size*ideps);
+        auto locsol = sol -> TangentialSolid (hp, locsurfind, size*ideps);
 
 	//*testout << "hp = " << hp << endl;
 	//(*testout) << "locsol: " << endl;
@@ -995,7 +996,8 @@ namespace netgen
 	ReducePrimitiveIterator rpi(boxp);
 	UnReducePrimitiveIterator urpi;
       
-	((Solid*)locsol) -> IterateSolid (rpi);
+	// ((Solid*)locsol) -> IterateSolid (rpi);
+        locsol -> IterateSolid (rpi);
 
 	locsol -> CalcSurfaceInverse ();
       
@@ -1020,7 +1022,8 @@ namespace netgen
 		}
 	  }
 
-	((Solid*)locsol) -> IterateSolid (urpi);
+	// ((Solid*)locsol) -> IterateSolid (urpi);
+        locsol -> IterateSolid (urpi);
 
       
 	if (debug)
@@ -1085,23 +1088,33 @@ namespace netgen
 	    //int k;
 	    double eps = 1e-8*size;
 
-	    NgArray<bool> pre_ok(2);
+            ArrayMem<bool,2> pre_ok(2);
+            bool flip = false;
 
  	    do
  	      {
  		eps *= 0.5;
- 		pre_ok[0] = (locsol -> VectorIn2 (hp, m, n, eps) == IS_OUTSIDE &&
+                auto in00 = locsol -> VectorIn2 (hp, m, n, eps);
-			     locsol -> VectorIn2 (hp, m, -1. * n, eps) == IS_INSIDE);
+                auto in01 = locsol -> VectorIn2 (hp, m, -1. * n, eps);
- 		pre_ok[1] = (locsol -> VectorIn2 (hp, -1.*m, n, eps) == IS_OUTSIDE &&
+                pre_ok[0] = in00 == IS_OUTSIDE && in01 == IS_INSIDE;
-			     locsol -> VectorIn2 (hp, -1.*m, -1. * n, eps) == IS_INSIDE);
+
+                if(in00 == IS_INSIDE && in01 == IS_OUTSIDE)
+                  pre_ok[0] = flip = true;
+
+                auto in10 = locsol -> VectorIn2 (hp, -1.*m, n, eps);
+                auto in11 = locsol -> VectorIn2 (hp, -1.*m, -1. * n, eps);
+                pre_ok[1] = (in10 == IS_OUTSIDE && in11 == IS_INSIDE);
+
+                if(in10 == IS_INSIDE && in11 == IS_OUTSIDE)
+                  pre_ok[1] = flip = true;
 		
 		if (debug)
 		  {
 		    *testout << "eps = " << eps << endl;
-		    *testout << "in,1 = " << locsol -> VectorIn2 (hp, m, n, eps) << endl;
+                    *testout << "in,1 = " << in00 << endl;
-		    *testout << "in,1 = " << locsol -> VectorIn2 (hp, m, -1. * n, eps) << endl;
+                    *testout << "in,1 = " << in01 << endl;
-		    *testout << "in,1 = " << locsol -> VectorIn2 (hp, -1.*m, n, eps) << endl;
+                    *testout << "in,1 = " << in10 << endl;
-		    *testout << "in,1 = " << locsol -> VectorIn2 (hp, -1.*m, -1. * n, eps) << endl;
+                    *testout << "in,1 = " << in11 << endl;
 		  }
  	      }
  	    while(pre_ok[0] && pre_ok[1] && eps > 1e-16*size);
@@ -1150,10 +1163,10 @@ namespace netgen
 		m2 = fac * grad;
 		// (*testout) << "hp = " << hp << ", m = " << m << ", m2 = " << m2 << endl;
 
-		Solid * locsol2;
+		// Solid * locsol2;
-		locsol -> TangentialSolid3 (hp, m, m2, locsol2, locsurfind2, ideps*size);
+		auto locsol2 = locsol -> TangentialSolid3 (hp, m, m2, locsurfind2, ideps*size);
 		if (!locsol2) ok = 0;
-		delete locsol2;
+		// delete locsol2;
 
 
 		if (ok)
@@ -1197,7 +1210,10 @@ namespace netgen
 		  
 		    if (!surf)
 		      {
-			if (sameasref)
+                        bool inside = sameasref;
+                        if(flip)
+                          inside = !inside;
+                        if (inside)
 			  refedges.Elem(hi).domin = i;
 			else 
 			  refedges.Elem(hi).domout = i;
@@ -1246,7 +1262,7 @@ namespace netgen
 		m *= -1;
 	      } 
 	  }
-	delete locsol;          
+	// delete locsol;          
       }
 
    
@@ -1255,6 +1271,9 @@ namespace netgen
 	*testout << "Refsegments, before delete: " << endl << refedges << endl;
 	*testout << "inv: " << endl << refedgesinv << endl;
       }
+
+    if(refedges.Size() == 0)
+      throw Exception("No edges found, something wrong.");
     
     NgBitArray todelete(refedges.Size());
     todelete.Clear();
@@ -1764,7 +1783,7 @@ namespace netgen
     int nsurf = geometry.GetNSurf();
     int layer = 0;
 
-    Solid * tansol;
+    // Solid * tansol;
     NgArray<int> tansurfind;
 
     double size = geometry.MaxSize();
@@ -1822,7 +1841,8 @@ namespace netgen
 		  continue;
 		  
 		const Solid * sol = geometry.GetTopLevelObject(j)->GetSolid();
-		sol -> TangentialSolid (p1, tansol, tansurfind, ideps*size);
+		// sol -> TangentialSolid (p1, tansol, tansurfind, ideps*size);
+                auto tansol = sol -> TangentialSolid (p1, tansurfind, ideps*size);
 		layer = geometry.GetTopLevelObject(j)->GetLayer();
 
 		
@@ -1852,7 +1872,7 @@ namespace netgen
 			//        seg.invs1 = surfaces[i] -> Inverse();
 			//        seg.invs2 = ! (surfaces[i] -> Inverse());
 		      }
-		    delete tansol;
+		    // delete tansol;
 		  }
 	      }
 

libsrc/csg/extrusion.cpp

@@ -41,6 +41,18 @@ namespace netgen
 	    loc_z_dir[i] = glob_z_direction;
 	  }
       }
+
+    double cum_angle = 0.;
+    for(auto i : Range(path->GetSplines()))
+      {
+        const auto& sp = path->GetSpline(i);
+        auto t1 = sp.GetTangent(0.);
+        t1.Normalize();
+        auto t2 = sp.GetTangent(1.);
+        t2.Normalize();
+        cum_angle += acos(t1 * t2);
+        angles.Append(cum_angle);
+      }
     
     profile->GetCoeff(profile_spline_coeff);
     latest_point3d = -1.111e30;
@@ -415,6 +427,14 @@ namespace netgen
   }
 
 
+  bool ExtrusionFace :: PointInFace (const Point<3> & p, const double eps) const
+  {
+    Point<3> hp = p;
+    Project(hp);
+    return Dist2(p,hp) < sqr(eps);
+  }
+  
+
   void ExtrusionFace :: LineIntersections ( const Point<3> & p,
 					    const Vec<3> & v,
 					    const double eps,
@@ -648,20 +668,35 @@ namespace netgen
     dez /= lenz;
     dez -= (dez * ez) * ez;
   }
-  
 
-  Extrusion :: Extrusion(const SplineGeometry<3> & path_in,
+  void ExtrusionFace :: DefineTangentialPlane(const Point<3>& ap1,
-			 const SplineGeometry<2> & profile_in,
+                                              const Point<3>& ap2)
+  {
+    Surface::DefineTangentialPlane(ap1, ap2);
+    tangential_plane_seg = latest_seg;
+  }
+
+  void ExtrusionFace :: ToPlane(const Point<3>& p3d, Point<2>& p2d,
+                                double h, int& zone) const
+  {
+    Surface::ToPlane(p3d, p2d, h, zone);
+    double angle = angles[tangential_plane_seg] - angles[latest_seg];
+    if(fabs(angle) > 3.14/2.)
+      zone = -1;
+  }
+
+  Extrusion :: Extrusion(shared_ptr<SplineGeometry<3>> path_in,
+			 shared_ptr<SplineGeometry<2>> profile_in,
 			 const Vec<3> & z_dir) :
-    path(&path_in), profile(&profile_in), z_direction(z_dir)
+    path(path_in), profile(profile_in), z_direction(z_dir)
   {
     surfaceactive.SetSize(0);
     surfaceids.SetSize(0);
 
     for(int j=0; j<profile->GetNSplines(); j++)
       {
-	ExtrusionFace * face = new ExtrusionFace(&((*profile).GetSpline(j)),
+	ExtrusionFace * face = new ExtrusionFace(&(profile->GetSpline(j)),
-						 path,
+						 path.get(),
 						 z_direction);
 	faces.Append(face);
 	surfaceactive.Append(true);
@@ -737,6 +772,16 @@ namespace netgen
     return PointInSolid(p,eps,NULL);    
   }
 
+  void Extrusion :: GetTangentialSurfaceIndices (const Point<3> & p, 
+                                                 NgArray<int> & surfind, double eps) const
+  {
+    for (int j = 0; j < faces.Size(); j++)
+      if (faces[j] -> PointInFace(p, eps))
+        if (!surfind.Contains (GetSurfaceId(j)))
+          surfind.Append (GetSurfaceId(j));
+  }
+
+  
   INSOLID_TYPE Extrusion :: VecInSolid (const Point<3> & p,
 					const Vec<3> & v,
 					double eps) const
@@ -838,7 +883,7 @@ namespace netgen
       return retval;
 
     if(latestfacenum >= 0)
-      return faces[latestfacenum]->VecInFace(p,v2,0);
+      return faces[latestfacenum]->VecInFace(p,v2,eps);
     else
       return VecInSolid(p,v2,eps);
   }

libsrc/csg/extrusion.hpp

@@ -12,8 +12,10 @@ namespace netgen
     const SplineSeg<2> * profile;
     const SplineGeometry<3> * path;
     Vec<3> glob_z_direction;
+    Array<double> angles;
 
     bool deletable;
+    int tangential_plane_seg;
   
     NgArray< const SplineSeg3<3> * > spline3_path;
     NgArray< const LineSeg<3> * > line_path;
@@ -54,7 +56,7 @@ namespace netgen
 
     ~ExtrusionFace();
 
-    virtual void DoArchive(Archive& ar)
+    void DoArchive(Archive& ar) override
     {
       Surface::DoArchive(ar);
       ar & profile & path & glob_z_direction & deletable & spline3_path & line_path &
@@ -62,25 +64,25 @@ namespace netgen
         profile_spline_coeff & latest_seg & latest_t & latest_point2d & latest_point3d;
     }
   
-    virtual int IsIdentic (const Surface & s2, int & inv, double eps) const; 
+    int IsIdentic (const Surface & s2, int & inv, double eps) const override;
   
-    virtual double CalcFunctionValue (const Point<3> & point) const;
+    double CalcFunctionValue (const Point<3> & point) const override;
-    virtual void CalcGradient (const Point<3> & point, Vec<3> & grad) const;
+    void CalcGradient (const Point<3> & point, Vec<3> & grad) const override;
-    virtual void CalcHesse (const Point<3> & point, Mat<3> & hesse) const;
+    void CalcHesse (const Point<3> & point, Mat<3> & hesse) const override;
-    virtual double HesseNorm () const;
+    double HesseNorm () const override;
 
-    virtual double MaxCurvature () const;
+    double MaxCurvature () const override;
     //virtual double MaxCurvatureLoc (const Point<3> & /* c */ , 
     //				  double /* rad */) const;
 
-    virtual void Project (Point<3> & p) const;
+    void Project (Point<3> & p) const override;
 
-    virtual Point<3> GetSurfacePoint () const;
+    Point<3> GetSurfacePoint () const override;
-    virtual void Print (ostream & str) const;
+    void Print (ostream & str) const override;
   
-    virtual void GetTriangleApproximation (TriangleApproximation & tas, 
+    void GetTriangleApproximation (TriangleApproximation & tas, 
 					   const Box<3> & boundingbox, 
-					   double facets) const;
+					   double facets) const override;
 
     const SplineGeometry<3> & GetPath(void) const {return *path;}
     const SplineSeg<2> & GetProfile(void) const {return *profile;}
@@ -94,6 +96,9 @@ namespace netgen
 			     int & after,
 			     bool & intersecting ) const;
 
+
+    bool PointInFace (const Point<3> & p, const double eps) const;
+    
     INSOLID_TYPE VecInFace ( const Point<3> & p,
 			     const Vec<3> & v,
 			     const double eps ) const;
@@ -111,6 +116,11 @@ namespace netgen
 				    Vec<3> & ex, Vec<3> & ey, Vec<3> & ez,
 				    Vec<3> & dex, Vec<3> & dey, Vec<3> & dez) const;
 
+    void DefineTangentialPlane(const Point<3>& ap1,
+                               const Point<3>& ap2) override;
+    void ToPlane(const Point<3>& p3d, Point<2>& p2d,
+                 double h, int& zone) const override;
+
   };
 
 
@@ -118,8 +128,8 @@ namespace netgen
   class Extrusion : public Primitive
   {
   private:
-    const SplineGeometry<3>* path;
+    shared_ptr<SplineGeometry<3>> path;
-    const SplineGeometry<2>* profile; // closed, clockwise oriented curve
+    shared_ptr<SplineGeometry<2>> profile; // closed, clockwise oriented curve
 
     Vec<3> z_direction;
 
@@ -128,43 +138,46 @@ namespace netgen
     mutable int latestfacenum;
 
   public:
-    Extrusion(const SplineGeometry<3> & path_in,
+    Extrusion(shared_ptr<SplineGeometry<3>> path_in,
-	      const SplineGeometry<2> & profile_in,
+	      shared_ptr<SplineGeometry<2>> profile_in,
 	      const Vec<3> & z_dir);
     // default constructor for archive
     Extrusion() {}
     ~Extrusion();
 
-    virtual void DoArchive(Archive& ar)
+    void DoArchive(Archive& ar) override
     {
       Primitive::DoArchive(ar);
       ar & path & profile & z_direction & faces & latestfacenum;
     }
-    virtual INSOLID_TYPE BoxInSolid (const BoxSphere<3> & box) const;
+    INSOLID_TYPE BoxInSolid (const BoxSphere<3> & box) const override;
-    virtual INSOLID_TYPE PointInSolid (const Point<3> & p,
+    INSOLID_TYPE PointInSolid (const Point<3> & p,
-				       double eps) const;
+                               double eps) const override;
     INSOLID_TYPE PointInSolid (const Point<3> & p,
 			       double eps,
 			       NgArray<int> * const facenums) const;
-    virtual INSOLID_TYPE VecInSolid (const Point<3> & p,
+
-				     const Vec<3> & v,
+    void GetTangentialSurfaceIndices (const Point<3> & p, 
-				     double eps) const;
+                                      NgArray<int> & surfind, double eps) const override;
+
+    INSOLID_TYPE VecInSolid (const Point<3> & p,
+                             const Vec<3> & v,
+                             double eps) const override;
 
     // checks if lim s->0 lim t->0  p + t(v1 + s v2) in solid
-    virtual INSOLID_TYPE VecInSolid2 (const Point<3> & p,
+    INSOLID_TYPE VecInSolid2 (const Point<3> & p,
-				      const Vec<3> & v1,
+                              const Vec<3> & v1,
-				      const Vec<3> & v2,
+                              const Vec<3> & v2,
-				      double eps) const;
+                              double eps) const override;
 
   
-    virtual int GetNSurfaces() const;
+    int GetNSurfaces() const override;
-    virtual Surface & GetSurface (int i = 0);
+    Surface & GetSurface (int i = 0) override;
-    virtual const Surface & GetSurface (int i = 0) const;
+    const Surface & GetSurface (int i = 0) const override;
 
 
-    virtual void Reduce (const BoxSphere<3> & box);
+    void Reduce (const BoxSphere<3> & box) override;
-    virtual void UnReduce ();
+    void UnReduce () override;
-
   };
 
 }

libsrc/csg/identify.cpp

@@ -318,6 +318,10 @@ GetIdentifiedPoint (class Mesh & mesh,  int pi)
 
 void PeriodicIdentification :: IdentifyPoints (class Mesh & mesh)
 {
+  Point3d p1, p2;
+  mesh.GetBox(p1, p2);
+  auto eps = 1e-6 * (p2-p1).Length();
+
   for (int i = 1; i <= mesh.GetNP(); i++)
     {
       Point<3> p = mesh.Point(i);
@@ -327,7 +331,7 @@ void PeriodicIdentification :: IdentifyPoints (class Mesh & mesh)
           pp = trafo(pp);
 	  s2->Project (pp);
 	  for (int j = 1; j <= mesh.GetNP(); j++)
-	    if (Dist2(mesh.Point(j), pp) < 1e-6)
+	    if (Dist2(mesh.Point(j), pp) < eps)
 	      {
 		mesh.GetIdentifications().Add (i, j, nr);
 		/*

libsrc/csg/polyhedra.hpp

@@ -48,46 +48,63 @@ namespace netgen
 
   public:
     Polyhedra ();
-    virtual ~Polyhedra ();
+    virtual ~Polyhedra () override;
     static Primitive * CreateDefault ();
 
-    virtual INSOLID_TYPE BoxInSolid (const BoxSphere<3> & box) const;
+    virtual INSOLID_TYPE BoxInSolid (const BoxSphere<3> & box) const override;
     virtual INSOLID_TYPE PointInSolid (const Point<3> & p,
-				       double eps) const;
+				       double eps) const override;
-    virtual INSOLID_TYPE VecInSolid (const Point<3> & p,
+    virtual INSOLID_TYPE VecInSolidNew (const Point<3> & p,
+                                        const Vec<3> & v,
+                                        double eps, bool printing = false) const;
+    virtual INSOLID_TYPE VecInSolidOld (const Point<3> & p,
 				     const Vec<3> & v,
 				     double eps) const;
+    
+    virtual INSOLID_TYPE VecInSolid (const Point<3> & p,
+				     const Vec<3> & v,
+				     double eps) const override;
 
-    // checks if lim s->0 lim t->0  p + t(v1 + s v2) in solid
     virtual INSOLID_TYPE VecInSolid2 (const Point<3> & p,
 				      const Vec<3> & v1,
 				      const Vec<3> & v2,
-				      double eps) const;
+				      double eps) const override;
+    
+    virtual INSOLID_TYPE VecInSolid3 (const Point<3> & p,
+				      const Vec<3> & v1,
+				      const Vec<3> & v2,
+				      double eps) const override;
 
+    virtual INSOLID_TYPE VecInSolid4 (const Point<3> & p,
+				      const Vec<3> & v,
+				      const Vec<3> & v2,
+				      const Vec<3> & m,
+				      double eps) const override;
+    
     virtual void GetTangentialSurfaceIndices (const Point<3> & p, 
-					      NgArray<int> & surfind, double eps) const;
+					      NgArray<int> & surfind, double eps) const override;
 
 
     virtual void GetTangentialVecSurfaceIndices2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
-						  NgArray<int> & surfind, double eps) const;
+						  NgArray<int> & surfind, double eps) const override;
 
-    virtual void CalcSpecialPoints (NgArray<Point<3> > & pts) const;
+    virtual void CalcSpecialPoints (NgArray<Point<3> > & pts) const override;
     virtual void AnalyzeSpecialPoint (const Point<3> & pt, 
-				      NgArray<Point<3> > & specpts) const;
+				      NgArray<Point<3> > & specpts) const override;
-    virtual Vec<3> SpecialPointTangentialVector (const Point<3> & p, int s1, int s2) const;
+    virtual Vec<3> SpecialPointTangentialVector (const Point<3> & p, int s1, int s2) const override;
 
-    virtual int GetNSurfaces() const 
+    virtual int GetNSurfaces() const override
     { return planes.Size(); }
-    virtual Surface & GetSurface (int i) 
+    virtual Surface & GetSurface (int i) override
     { return *planes[i]; }
-    virtual const Surface & GetSurface (int i) const
+    virtual const Surface & GetSurface (int i) const override
     { return *planes[i]; }
 
-    virtual void GetPrimitiveData (const char *& classname, NgArray<double> & coeffs) const;
+    virtual void GetPrimitiveData (const char *& classname, NgArray<double> & coeffs) const override;
-    virtual void SetPrimitiveData (NgArray<double> & coeffs);
+    virtual void SetPrimitiveData (NgArray<double> & coeffs) override;
 
-    virtual void Reduce (const BoxSphere<3> & box);
+    virtual void Reduce (const BoxSphere<3> & box) override;
-    virtual void UnReduce ();
+    virtual void UnReduce () override;
 
     int AddPoint (const Point<3> & p);
     int AddFace (int pi1, int pi2, int pi3, int inputnum);

libsrc/csg/python_csg.cpp

@@ -170,7 +170,8 @@ namespace netgen
 
 DLL_HEADER void ExportCSG(py::module &m)
 {
-  py::class_<SplineGeometry<2>> (m, "SplineCurve2d")
+  py::class_<SplineGeometry<2>, shared_ptr<SplineGeometry<2>>>
+    (m, "SplineCurve2d")
     .def(py::init<>())
     .def ("AddPoint", FunctionPointer
           ([] (SplineGeometry<2> & self, double x, double y)
@@ -329,14 +330,31 @@ DLL_HEADER void ExportCSG(py::module &m)
                                            Solid * sol = new Solid(rev);
                                            return make_shared<SPSolid> (sol);
                                          }));
-  m.def ("Extrusion", FunctionPointer([](const SplineGeometry<3> & path,
+  m.def ("Extrusion", [](shared_ptr<SplineGeometry<3>> path,
-					 const SplineGeometry<2> & profile,
+                         shared_ptr<SplineGeometry<2>> profile,
-					 Vec<3> n)
+                         Vec<3> d)
-                                         {
+  {
-                                           Extrusion * extr = new Extrusion (path,profile,n);
+    Extrusion * extr = new Extrusion (path,profile,d);
-                                           Solid * sol = new Solid(extr);
+    Solid * sol = new Solid(extr);
-                                           return make_shared<SPSolid> (sol);
+    return make_shared<SPSolid> (sol);
-                                         }));
+  }, py::arg("path"), py::arg("profile"), py::arg("d"),
+     R"delimiter(A body of extrusion is defined by its profile
+(which has to be a closed, clockwiseoriented 2D curve),
+ by a path (a 3D curve) and a vector d. It is constructed
+ as follows: Take a point p on the path and denote the
+ (unit-)tangent of the path in this point by t. If we cut
+ the body by the plane given by p and t as normal vector,
+ the cut is the profile. The profile is oriented by the
+ (local) y-direction `y:=d−(d·t)t` and the (local) x-direction
+ `x:=t \times y`.
+The following points have to be noticed:
+ * If the path is not closed, then also the body is NOT closed.
+   In this case e.g. planes or orthobricks have to be used to
+   construct a closed body.
+ * The path has to be smooth, i.e. the tangents at the end- resp.
+   start-point of two consecutive spline or line patches have to
+   have the same directions.
+)delimiter");
   m.def("EllipticCone", [](const Point<3>& a, const Vec<3>& v, const Vec<3>& w,
                             double h, double r)
         {
@@ -354,6 +372,35 @@ When r =1, the truncated elliptic cone becomes an elliptic cylinder.
 When r tends to zero, the truncated elliptic cone tends to a full elliptic cone.
 However, when r = 0, the top part becomes a point(tip) and meshing fails!
 )raw_string");
+
+  m.def("Polyhedron", [](py::list points, py::list faces)
+  {
+    auto poly = new Polyhedra();
+    for(auto p : points)
+      poly->AddPoint(py::cast<Point<3>>(p));
+    int fnr = 0;
+    for(auto face : faces)
+      {
+        auto lface = py::cast<py::list>(face);
+        if(py::len(lface) == 3)
+          poly->AddFace(py::cast<int>(lface[0]),
+                        py::cast<int>(lface[1]),
+                        py::cast<int>(lface[2]),
+                        fnr++);
+        else if(py::len(lface) == 4)
+          {
+            poly->AddFace(py::cast<int>(lface[0]),
+                          py::cast<int>(lface[1]),
+                          py::cast<int>(lface[2]),
+                          fnr);
+            poly->AddFace(py::cast<int>(lface[0]),
+                          py::cast<int>(lface[2]),
+                          py::cast<int>(lface[3]),
+                          fnr++);
+          }
+      }
+    return make_shared<SPSolid>(new Solid(poly));
+  });
   
   m.def ("Or", FunctionPointer([](shared_ptr<SPSolid> s1, shared_ptr<SPSolid> s2)
                                  {

libsrc/csg/revolution.cpp

@@ -670,6 +670,23 @@ namespace netgen
 	surfaceactive.Append(1);
 	surfaceids.Append(0);
       }
+
+    // checking
+    if (type == 2)
+      {
+        auto t0 = spline_in.GetSpline(0).GetTangent(0);
+        cout << "tstart (must be vertically): " << t0 << endl;
+
+        auto tn = spline_in.GetSpline(nsplines-1).GetTangent(1);
+        cout << "tend (must be vertically): " << tn << endl;
+
+        for (int i = 0; i < nsplines-1; i++)
+          {
+            auto ta = spline_in.GetSpline(i).GetTangent(1);
+            auto tb = spline_in.GetSpline(i+1).GetTangent(0);
+            cout << "sin (must not be 0) = " << abs(ta(0)*tb(1)-ta(1)*tb(0)) / (Abs(ta)*Abs(tb)); 
+          }
+      }
   }
   
   Revolution::~Revolution()
@@ -764,8 +781,9 @@ namespace netgen
     int intersections_before(0), intersections_after(0);
     double randomx = 7.42357;
     double randomy = 1.814756;
-    randomx *= 1./sqrt(randomx*randomx+randomy*randomy);
+    double randomlen = sqrt(randomx*randomx+randomy*randomy);
-    randomy *= 1./sqrt(randomx*randomx+randomy*randomy);
+    randomx *= 1./randomlen;
+    randomy *= 1./randomlen;
     
 
     const double a = randomy;
@@ -930,6 +948,67 @@ namespace netgen
 
     return VecInSolid(p,v1+0.01*v2,eps);
   }
+
+  void Revolution ::
+  GetTangentialVecSurfaceIndices2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
+                                   NgArray<int> & surfind, double eps) const
+  {
+    *testout << "tangentialvecsurfind2, p = " << p << endl;
+    for (int i = 0; i < faces.Size(); i++)
+      if (faces[i]->PointInFace (p, eps))
+        {
+          *testout << "check face " << i << endl;
+          Point<2> p2d;
+          Vec<2> v12d;
+          faces[i]->CalcProj(p,p2d,v1,v12d);
+          *testout << "v12d = " << v12d << endl;
+          auto & spline = faces[i]->GetSpline();
+          if (Dist2 (spline.StartPI(), p2d) < sqr(eps))
+            {
+              *testout << "start pi" << endl;
+              Vec<2> tang = spline.GetTangent(0);
+              double ip = tang*v12d;
+              *testout << "ip = " << ip << endl;
+              if (ip > eps)
+                surfind.Append(GetSurfaceId(i));
+              else if (ip > -eps)
+                {
+                  Vec<2> v22d;
+                  faces[i]->CalcProj(p,p2d,v2,v22d);
+                  double ip2 = tang*v22d;
+                  *testout << "ip2 = " << ip2 << endl;
+                  if (ip2 > -eps)
+                    surfind.Append(GetSurfaceId(i));                    
+                }
+            }
+          else if (Dist2 (faces[i]->GetSpline().EndPI(), p2d) < sqr(eps))
+            {
+              *testout << "end pi" << endl;
+              
+              Vec<2> tang = spline.GetTangent(1);
+              double ip = tang*v12d;
+              *testout << "ip = " << ip << endl;
+              if (ip < -eps)
+                surfind.Append(GetSurfaceId(i));
+              else if (ip < eps)
+                {
+                  Vec<2> v22d;
+                  faces[i]->CalcProj(p,p2d,v2,v22d);
+                  double ip2 = tang*v22d;
+                  *testout << "ip2 = " << ip2 << endl;                  
+                  if (ip2 < eps)
+                    surfind.Append(GetSurfaceId(i));                    
+                }
+            }
+          else
+            {
+              *testout << "inner point" << endl;
+              surfind.Append(GetSurfaceId(i));
+            }
+        }
+  }
+
+
   
   int Revolution :: GetNSurfaces() const
   {

libsrc/csg/revolution.hpp

@@ -67,7 +67,10 @@ namespace netgen
     virtual double MaxCurvature () const;
     //virtual double MaxCurvatureLoc (const Point<3> & /* c */ , 
     //				  double /* rad */) const;
-
+    
+    Point<3> P0() const { return p0; }
+    Vec<3> Axis() const { return v_axis; }
+    
     virtual void Project (Point<3> & p) const;
 
     virtual Point<3> GetSurfacePoint () const;
@@ -155,7 +158,10 @@ namespace netgen
 				      const Vec<3> & v2,
 				      double eps) const;
 
-  
+    virtual void GetTangentialVecSurfaceIndices2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
+						  NgArray<int> & surfind, double eps) const;
+
+    
     virtual int GetNSurfaces() const;
     virtual Surface & GetSurface (int i = 0);
     virtual const Surface & GetSurface (int i = 0) const;

libsrc/csg/singularref.cpp

@@ -168,7 +168,7 @@ void SingularPoint :: FindPoints (class Mesh & mesh)
 	  for (int k = 1; k <= 3; k++)
 	    {
 	      const Solid * solk(NULL);
-	      Solid *tansol;
+	      // Solid *tansol;
 	      switch (k)
 		{
 		case 1:  solk = sol1; break;
@@ -176,7 +176,7 @@ void SingularPoint :: FindPoints (class Mesh & mesh)
 		case 3:  solk = sol3; break;
 		}
 
-	      solk -> TangentialSolid (p, tansol, surfk, 1e-3);
+	      auto tansol = solk -> TangentialSolid (p, surfk, 1e-3);
 	      (*testout) << "Tansol = " << *tansol << endl;
 
 	      if (!tansol) continue;
@@ -195,7 +195,7 @@ void SingularPoint :: FindPoints (class Mesh & mesh)
 		if (!surf.Contains (surfk[i]))
 		  surf.Append (surfk[i]);
 	      
-	      delete tansol;
+	      // delete tansol;
 	    }
 
 	  if (surf.Size() < 3) continue;

libsrc/csg/solid.cpp

@@ -6,21 +6,6 @@
 
 namespace netgen
 {
-  //using namespace netgen;
-
-
-  /*
-  SolidIterator :: SolidIterator ()
-  {
-    ;
-  }
-
-  SolidIterator :: ~SolidIterator ()
-  {
-    ;
-  }
-  */
-
 
 
   // int Solid :: cntnames = 0;
@@ -193,9 +178,70 @@ namespace netgen
 
 
 
+  INSOLID_TYPE Solid ::
+  PointInSolid (const Point<3> & p, double eps) const
+  {
+    switch (op)
+      {
+      case TERM: case TERM_REF:
+        return prim->PointInSolid (p, eps);
+      case SECTION:
+        return Intersection (s1->PointInSolid (p, eps), s2->PointInSolid (p, eps));
+      case UNION:
+        return Union (s1->PointInSolid (p, eps), s2->PointInSolid (p, eps));          
+      case SUB:
+        return Complement (s1->PointInSolid (p, eps));
+      case ROOT:
+	return s1->PointInSolid (p, eps);
+      }
+  }
+
+  
+  INSOLID_TYPE Solid ::
+  VecInSolid (const Point<3> & p, const Vec<3> & v, double eps) const
+  {
+    switch (op)
+      {
+      case TERM: case TERM_REF:
+        return prim->VecInSolid (p, v, eps);
+      case SECTION:
+        return Intersection (s1->VecInSolid (p, v, eps), s2->VecInSolid (p, v, eps));        
+      case UNION:
+        return Union (s1->VecInSolid (p, v, eps), s2->VecInSolid (p, v, eps));                
+      case SUB:
+        return Complement (s1->VecInSolid (p, v, eps));
+      case ROOT:
+	return s1->VecInSolid (p, v, eps);
+      }
+  }
+  
+  // checks if lim s->0 lim t->0  p + t(v1 + s v2) in solid
+  INSOLID_TYPE Solid ::
+  VecInSolid2 (const Point<3> & p, const Vec<3> & v1,
+               const Vec<3> & v2, double eps) const
+  {
+    switch (op)
+      {
+      case TERM: case TERM_REF:
+        return prim->VecInSolid2 (p, v1, v2, eps);
+      case SECTION:
+        return Intersection (s1->VecInSolid2 (p, v1, v2, eps), s2->VecInSolid2 (p, v1, v2, eps));
+      case UNION:
+        return Union (s1->VecInSolid2 (p, v1, v2, eps), s2->VecInSolid2 (p, v1, v2, eps));          
+      case SUB:
+        return Complement (s1->VecInSolid2 (p, v1, v2, eps));
+      case ROOT:
+	return s1->VecInSolid2 (p, v1, v2, eps);
+      }
+  }
+
+  
+
 
   bool Solid :: IsIn (const Point<3> & p, double eps) const
   {
+    return PointInSolid (p,eps) != IS_OUTSIDE;
+      /*
     switch (op)
       {
       case TERM: case TERM_REF:
@@ -213,10 +259,13 @@ namespace netgen
 	return s1->IsIn (p, eps);
       }
     return 0;
+      */
   }
 
   bool Solid :: IsStrictIn (const Point<3> & p, double eps) const
   {
+    return PointInSolid (p,eps) == IS_INSIDE;
+    /*
     switch (op)
       {
       case TERM: case TERM_REF:
@@ -234,11 +283,14 @@ namespace netgen
 	return s1->IsStrictIn (p, eps);
       }
     return 0;
+    */
   }
 
   bool Solid :: VectorIn (const Point<3> & p, const Vec<3> & v, 
 			 double eps) const
   {
+    return VecInSolid (p,v,eps) != IS_OUTSIDE;
+    /*
     Vec<3> hv;
     switch (op)
       {
@@ -257,11 +309,14 @@ namespace netgen
 	return s1->VectorIn(p, v, eps);
       }
     return 0;
+    */
   }
 
   bool Solid :: VectorStrictIn (const Point<3> & p, const Vec<3> & v,
 			       double eps) const
   {
+    return VecInSolid (p,v,eps) == IS_INSIDE;
+    /*
     Vec<3> hv;
     switch (op)
       {
@@ -282,9 +337,11 @@ namespace netgen
 	return s1->VectorStrictIn(p, v, eps);
       }
     return 0;
+    */
   }
 
 
+  /*
   bool Solid::VectorIn2 (const Point<3> & p, const Vec<3> & v1, 
 			const Vec<3> & v2, double eps) const
   {
@@ -317,10 +374,57 @@ namespace netgen
       }
     return 0;  
   }
+  */
 
 
+  bool Solid::VectorIn2 (const Point<3> & p, const Vec<3> & v1, 
+                         const Vec<3> & v2, double eps) const
+  {
+    return VecInSolid2 (p,v1,v2,eps) != IS_OUTSIDE;
+    /*
+    switch (op)
+      {
+      case TERM: case TERM_REF:
+        {
+          auto res = prim->VecInSolid2 (p, v1, v2, eps);
+          return res != IS_OUTSIDE;
+        }
+      case SECTION:
+	return s1->VectorIn2 (p, v1, v2, eps) && s2->VectorIn2 (p, v1, v2, eps);
+      case UNION:
+	return s1->VectorIn2 (p, v1, v2, eps) || s2->VectorIn2 (p, v1, v2, eps);
+      case SUB:
+	return !s1->VectorStrictIn2 (p, v1, v2, eps);
+      case ROOT:
+	return s1->VectorIn2 (p, v1, v2, eps);
+      }
+    // return 0;  
+    */
+  }
 
-
+  bool Solid :: VectorStrictIn2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
+                                 double eps) const
+  {
+    return VecInSolid2 (p,v1,v2,eps) == IS_INSIDE;
+    /*
+    switch (op)
+      {
+      case TERM: case TERM_REF:
+        {
+          auto res = prim->VecInSolid2 (p, v1, v2, eps);
+          return (res == IS_INSIDE);
+        }
+      case SECTION:
+	return s1->VectorStrictIn2 (p, v1, v2, eps) && s2->VectorStrictIn2 (p, v1, v2, eps);
+      case UNION:
+	return s1->VectorStrictIn2 (p, v1, v2, eps) || s2->VectorStrictIn2 (p, v1, v2, eps);
+      case SUB:
+	return !s1->VectorIn2 (p, v1, v2, eps);
+      case ROOT:
+	return s1->VectorStrictIn2 (p, v1, v2, eps);
+      }
+    */
+  }
 
 
   void Solid :: Print (ostream & str) const
@@ -638,18 +742,20 @@ namespace netgen
   }
 
 
-  void Solid :: TangentialSolid (const Point<3> & p, Solid *& tansol, NgArray<int> & surfids, double eps) const
+  unique_ptr<Solid> Solid :: TangentialSolid (const Point<3> & p, NgArray<int> & surfids, double eps) const
   {
-    int in, strin;
+    bool in, strin;
+    Solid * tansol = nullptr;
     RecTangentialSolid (p, tansol, surfids, in, strin, eps);
     surfids.SetSize (0);
     if (tansol)
       tansol -> GetTangentialSurfaceIndices (p, surfids, eps);
+    return unique_ptr<Solid> (tansol);
   }
 
 
   void Solid :: RecTangentialSolid (const Point<3> & p, Solid *& tansol, NgArray<int> & surfids,
-				    int & in, int & strin, double eps) const
+				    bool & in, bool & strin, double eps) const
   {
     tansol = NULL;
 
@@ -671,7 +777,7 @@ namespace netgen
 	}
       case SECTION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialSolid (p, tansol1, surfids, in1, strin1, eps);
@@ -686,13 +792,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 && in2);
+	  in = in1 && in2;
-	  strin = (strin1 && strin2);
+	  strin = strin1 && strin2;
 	  break;
 	}
       case UNION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1 = 0, * tansol2 = 0;
 
 	  s1 -> RecTangentialSolid (p, tansol1, surfids, in1, strin1, eps);
@@ -712,13 +818,13 @@ namespace netgen
 	      delete tansol1;
 	      delete tansol2;
 	    }
-	  in = (in1 || in2);
+	  in = in1 || in2;
-	  strin = (strin1 || strin2);
+	  strin = strin1 || strin2;
 	  break;
 	}
       case SUB:
 	{
-	  int hin, hstrin;
+	  bool hin, hstrin;
 	  Solid * tansol1;
 
 	  s1 -> RecTangentialSolid (p, tansol1, surfids, hin, hstrin, eps);
@@ -740,22 +846,24 @@ namespace netgen
 
 
 
-  void Solid :: TangentialSolid2 (const Point<3> & p, 
+  unique_ptr<Solid> Solid :: TangentialSolid2 (const Point<3> & p, 
-				  const Vec<3> & t,
+                                               const Vec<3> & t,
-				  Solid *& tansol, NgArray<int> & surfids, double eps) const
+                                               NgArray<int> & surfids, double eps) const
   {
-    int in, strin;
+    Solid * tansol = nullptr;
+    bool in, strin;
     surfids.SetSize (0);
     RecTangentialSolid2 (p, t, tansol, surfids, in, strin, eps);
     if (tansol)
       tansol -> GetTangentialSurfaceIndices2 (p, t, surfids, eps);
+    return unique_ptr<Solid> (tansol);
   }
 
   void Solid :: RecTangentialSolid2 (const Point<3> & p, const Vec<3> & t,
 				     Solid *& tansol, NgArray<int> & surfids, 
-				     int & in, int & strin, double eps) const
+				     bool & in, bool & strin, double eps) const
   {
-    tansol = NULL;
+    tansol = nullptr;
 
     switch (op)
       {
@@ -774,8 +882,8 @@ namespace netgen
 	  if (ist == DOES_INTERSECT)
 	    ist = prim->VecInSolid (p, t, eps);
 
-	  in = (ist == IS_INSIDE || ist == DOES_INTERSECT);
+	  in = (ist == IS_INSIDE) || (ist == DOES_INTERSECT);
-	  strin = (ist == IS_INSIDE);
+	  strin = ist == IS_INSIDE;
 
 	  if (ist == DOES_INTERSECT)
 	    {
@@ -786,7 +894,7 @@ namespace netgen
 	}
       case SECTION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialSolid2 (p, t, tansol1, surfids, in1, strin1, eps);
@@ -801,13 +909,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 && in2);
+	  in = in1 && in2;
-	  strin = (strin1 && strin2);
+	  strin = strin1 && strin2;
 	  break;
 	}
       case UNION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialSolid2 (p, t, tansol1, surfids, in1, strin1, eps);
@@ -822,13 +930,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 || in2);
+	  in = in1 || in2;
-	  strin = (strin1 || strin2);
+	  strin = strin1 || strin2;
 	  break;
 	}
       case SUB:
 	{
-	  int hin, hstrin;
+	  bool hin, hstrin;
 	  Solid * tansol1;
 
 	  s1 -> RecTangentialSolid2 (p, t, tansol1, surfids, hin, hstrin, eps);
@@ -854,25 +962,28 @@ namespace netgen
 
 
 
-  void Solid :: TangentialSolid3 (const Point<3> & p, 
+  unique_ptr<Solid> Solid :: TangentialSolid3 (const Point<3> & p, 
-				  const Vec<3> & t, const Vec<3> & t2,
+                                               const Vec<3> & t, const Vec<3> & t2,
-				  Solid *& tansol, NgArray<int> & surfids, 
+                                               NgArray<int> & surfids, 
-				  double eps) const
+                                               double eps) const
   {
-    int in, strin;
+    bool in, strin;
+    Solid * tansol = nullptr;
     surfids.SetSize (0);
     RecTangentialSolid3 (p, t, t2, tansol, surfids, in, strin, eps);
 
     if (tansol)
       tansol -> GetTangentialSurfaceIndices3 (p, t, t2, surfids, eps);
+
+    return unique_ptr<Solid>(tansol);
   }
 
   void Solid :: RecTangentialSolid3 (const Point<3> & p, 
 				     const Vec<3> & t, const Vec<3> & t2,
 				     Solid *& tansol, NgArray<int> & surfids, 
-				     int & in, int & strin, double eps) const
+				     bool & in, bool & strin, double eps) const
   {
-    tansol = NULL;
+    tansol = nullptr;
 
     switch (op)
       {
@@ -882,8 +993,8 @@ namespace netgen
 
 	  if (ist == DOES_INTERSECT)
 	    ist = prim->VecInSolid3 (p, t, t2, eps);
-	  in = (ist == IS_INSIDE || ist == DOES_INTERSECT);
+	  in = (ist == IS_INSIDE) || (ist == DOES_INTERSECT);
-	  strin = (ist == IS_INSIDE);
+	  strin = ist == IS_INSIDE;
 
 	  if (ist == DOES_INTERSECT)
 	    {
@@ -894,7 +1005,7 @@ namespace netgen
 	}
       case SECTION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialSolid3 (p, t, t2, tansol1, surfids, in1, strin1, eps);
@@ -909,13 +1020,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 && in2);
+	  in = in1 && in2;
-	  strin = (strin1 && strin2);
+	  strin = strin1 && strin2;
 	  break;
 	}
       case UNION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialSolid3 (p, t, t2, tansol1, surfids, in1, strin1, eps);
@@ -930,13 +1041,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 || in2);
+	  in = in1 || in2;
-	  strin = (strin1 || strin2);
+	  strin = strin1 || strin2;
 	  break;
 	}
       case SUB:
 	{
-	  int hin, hstrin;
+	  bool hin, hstrin;
 	  Solid * tansol1;
 
 	  s1 -> RecTangentialSolid3 (p, t, t2, tansol1, surfids, hin, hstrin, eps);
@@ -965,12 +1076,13 @@ namespace netgen
 
 
 
-  void Solid :: TangentialEdgeSolid (const Point<3> & p, 
+  unique_ptr<Solid> Solid :: TangentialEdgeSolid (const Point<3> & p, 
-				     const Vec<3> & t, const Vec<3> & t2, const Vec<3> & m, 
+                                                  const Vec<3> & t, const Vec<3> & t2, const Vec<3> & m, 
-				     Solid *& tansol, NgArray<int> & surfids, 
+                                                  NgArray<int> & surfids, 
-				     double eps) const
+                                                  double eps) const
   {
-    int in, strin;
+    Solid * tansol = nullptr;
+    bool in, strin;
     surfids.SetSize (0);
 
     // *testout << "tangentialedgesolid,sol = " << (*this) << endl;
@@ -978,12 +1090,14 @@ namespace netgen
 
     if (tansol)
       tansol -> RecGetTangentialEdgeSurfaceIndices (p, t, t2, m, surfids, eps);
+
+    return unique_ptr<Solid> (tansol);
   }
 
   void Solid :: RecTangentialEdgeSolid (const Point<3> & p, 
 					const Vec<3> & t, const Vec<3> & t2, const Vec<3> & m,
 					Solid *& tansol, NgArray<int> & surfids, 
-					int & in, int & strin, double eps) const
+					bool & in, bool & strin, double eps) const
   {
     tansol = NULL;
 
@@ -1005,8 +1119,8 @@ namespace netgen
 
 	  // (*testout) << "ist2 = " << ist << endl;
 
-	  in = (ist == IS_INSIDE || ist == DOES_INTERSECT);
+	  in = (ist == IS_INSIDE) || (ist == DOES_INTERSECT);
-	  strin = (ist == IS_INSIDE);
+	  strin = ist == IS_INSIDE;
 
 	  if (ist == DOES_INTERSECT)
 	    {
@@ -1017,7 +1131,7 @@ namespace netgen
 	}
       case SECTION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialEdgeSolid (p, t, t2, m, tansol1, surfids, in1, strin1, eps);
@@ -1032,13 +1146,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 && in2);
+	  in = in1 && in2;
-	  strin = (strin1 && strin2);
+	  strin = strin1 && strin2;
 	  break;
 	}
       case UNION:
 	{
-	  int in1, in2, strin1, strin2;
+	  bool in1, in2, strin1, strin2;
 	  Solid * tansol1, * tansol2;
 
 	  s1 -> RecTangentialEdgeSolid (p, t, t2, m, tansol1, surfids, in1, strin1, eps);
@@ -1053,13 +1167,13 @@ namespace netgen
 	      else if (tansol2)
 		tansol = tansol2;
 	    }
-	  in = (in1 || in2);
+	  in = in1 || in2;
-	  strin = (strin1 || strin2);
+	  strin = strin1 || strin2;
 	  break;
 	}
       case SUB:
 	{
-	  int hin, hstrin;
+	  bool hin, hstrin;
 	  Solid * tansol1;
 
 	  s1 -> RecTangentialEdgeSolid (p, t, t2, m, tansol1, surfids, hin, hstrin, eps);
@@ -1095,29 +1209,31 @@ namespace netgen
 
   int Solid :: Edge (const Point<3> & p, const Vec<3> & v, double eps) const
   {
-    int in, strin, faces;
+    bool in, strin;
+    int faces;
     RecEdge (p, v, in, strin, faces, eps);
     return faces >= 2;
   }
 
   int Solid :: OnFace (const Point<3> & p, const Vec<3> & v, double eps) const
   {
-    int in, strin, faces;
+    bool in, strin;
+    int faces;
     RecEdge (p, v, in, strin, faces, eps);
     return faces >= 1;
   }
 
 
   void Solid :: RecEdge (const Point<3> & p, const Vec<3> & v,
-			 int & in, int & strin, int & faces, double eps) const
+			 bool & in, bool & strin, int & faces, double eps) const
   {
     switch (op)
       {
       case TERM: case TERM_REF:
 	{
 	  INSOLID_TYPE ist = prim->VecInSolid (p, v, eps);
-	  in = (ist == IS_INSIDE || ist == DOES_INTERSECT);
+	  in = (ist == IS_INSIDE) || (ist == DOES_INTERSECT);
-	  strin = (ist == IS_INSIDE);
+	  strin = ist == IS_INSIDE;
 	  /*
 	    in = VectorIn (p, v);
 	    strin = VectorStrictIn (p, v);
@@ -1143,7 +1259,8 @@ namespace netgen
 	}
       case SECTION:
 	{
-	  int in1, in2, strin1, strin2, faces1, faces2;
+	  bool in1, in2, strin1, strin2;
+          int faces1, faces2;
 
 	  s1 -> RecEdge (p, v, in1, strin1, faces1, eps);
 	  s2 -> RecEdge (p, v, in2, strin2, faces2, eps);
@@ -1157,7 +1274,8 @@ namespace netgen
 	}
       case UNION:
 	{
-	  int in1, in2, strin1, strin2, faces1, faces2;
+	  bool in1, in2, strin1, strin2;
+          int faces1, faces2;
 
 	  s1 -> RecEdge (p, v, in1, strin1, faces1, eps);
 	  s2 -> RecEdge (p, v, in2, strin2, faces2, eps);
@@ -1171,7 +1289,7 @@ namespace netgen
 	}
       case SUB:
 	{
-	  int in1, strin1;
+	  bool in1, strin1;
 	  s1 -> RecEdge (p, v, in1, strin1, faces, eps);
 	  in = !strin1;
 	  strin = !in1;

libsrc/csg/solid.hpp

@@ -33,6 +33,26 @@ namespace netgen
   };
 
 
+  inline INSOLID_TYPE Intersection (INSOLID_TYPE ina, INSOLID_TYPE inb)
+  {
+    if (ina == IS_INSIDE && inb == IS_INSIDE) return IS_INSIDE;
+    if (ina == IS_OUTSIDE || inb == IS_OUTSIDE) return IS_OUTSIDE;
+    return DOES_INTERSECT;
+  }
+  
+  inline INSOLID_TYPE Union (INSOLID_TYPE ina, INSOLID_TYPE inb)
+  {
+    if (ina == IS_INSIDE || inb == IS_INSIDE) return IS_INSIDE;
+    if (ina == IS_OUTSIDE && inb == IS_OUTSIDE) return IS_OUTSIDE;
+    return DOES_INTERSECT;
+  }
+
+  inline INSOLID_TYPE Complement (INSOLID_TYPE in)
+  {
+    if (in == IS_INSIDE) return IS_OUTSIDE;
+    if (in == IS_OUTSIDE) return IS_INSIDE;
+    return DOES_INTERSECT;
+  }
 
   class Solid
   {
@@ -102,6 +122,14 @@ namespace netgen
 
     // geometric tests
 
+    INSOLID_TYPE PointInSolid (const Point<3> & p, double eps) const;
+    INSOLID_TYPE VecInSolid (const Point<3> & p, const Vec<3> & v, double eps) const;
+
+    // checks if lim s->0 lim t->0  p + t(v1 + s v2) in solid
+    INSOLID_TYPE VecInSolid2 (const Point<3> & p, const Vec<3> & v1,
+                              const Vec<3> & v2, double eps) const;
+
+    
     bool IsIn (const Point<3> & p, double eps = 1e-6) const;
     bool IsStrictIn (const Point<3> & p, double eps = 1e-6) const;
     bool VectorIn (const Point<3> & p, const Vec<3> & v, double eps = 1e-6) const;
@@ -109,21 +137,24 @@ namespace netgen
   
     bool VectorIn2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
 		    double eps) const;
+    /*
     bool VectorIn2Rec (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
 		       double eps) const;
-
+    */
+    bool VectorStrictIn2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
+                          double eps) const;
 
     /// compute localization in point p
-    void TangentialSolid (const Point<3> & p, Solid *& tansol, NgArray<int> & surfids, double eps) const;
+    unique_ptr<Solid> TangentialSolid (const Point<3> & p, NgArray<int> & surfids, double eps) const;
 
     /// compute localization in point p tangential to vector t
-    void TangentialSolid2 (const Point<3> & p, const Vec<3> & t,
+    unique_ptr<Solid> TangentialSolid2 (const Point<3> & p, const Vec<3> & t,
-			   Solid *& tansol, NgArray<int> & surfids, double eps) const;
+                                        NgArray<int> & surfids, double eps) const;
 
     /** compute localization in point p, with second order approximation to edge
 	p + s t + s*s/2 t2 **/
-    void TangentialSolid3 (const Point<3> & p, const Vec<3> & t, const Vec<3> & t2, 
+    unique_ptr<Solid> TangentialSolid3 (const Point<3> & p, const Vec<3> & t, const Vec<3> & t2, 
-			   Solid *& tansol, NgArray<int> & surfids, double eps) const;
+                                        NgArray<int> & surfids, double eps) const;
 
 
 
@@ -133,9 +164,9 @@ namespace netgen
 	p + s t + s*s/2 t2 + r m
 	with first order
     **/
-    void TangentialEdgeSolid (const Point<3> & p, const Vec<3> & t, const Vec<3> & t2, 
+    unique_ptr<Solid> TangentialEdgeSolid (const Point<3> & p, const Vec<3> & t, const Vec<3> & t2, 
-			      const Vec<3> & m, 
+                                           const Vec<3> & m, 
-			      Solid *& tansol, NgArray<int> & surfids, double eps) const;
+                                           NgArray<int> & surfids, double eps) const;
 
 
     void CalcOnePrimitiveSpecialPoints (const Box<3> & box, NgArray<Point<3> > & pts) const;
@@ -180,24 +211,24 @@ namespace netgen
 			int & in, int & strin) const;
     ///
     void RecTangentialSolid (const Point<3> & p, Solid *& tansol, NgArray<int> & surfids, 
-			     int & in, int & strin, double eps) const;
+                             bool & in, bool & strin, double eps) const;
 
     void RecTangentialSolid2 (const Point<3> & p, const Vec<3> & vec, 
 			      Solid *& tansol, NgArray<int> & surfids, 
-			      int & in, int & strin, double eps) const;
+			      bool & in, bool & strin, double eps) const;
     ///
     void RecTangentialSolid3 (const Point<3> & p, const Vec<3> & vec,const Vec<3> & vec2, 
 			      Solid *& tansol, NgArray<int> & surfids, 
-			      int & in, int & strin, double eps) const;
+			      bool & in, bool & strin, double eps) const;
     ///
     void RecTangentialEdgeSolid (const Point<3> & p, const Vec<3> & t, const Vec<3> & t2, 
 				 const Vec<3> & m, 
 				 Solid *& tansol, NgArray<int> & surfids, 
-				 int & in, int & strin, double eps) const;
+				 bool & in, bool & strin, double eps) const;
 
     ///
     void RecEdge (const Point<3> & p, const Vec<3> & v,
-		  int & in, int & strin, int & faces, double eps) const;
+		  bool & in, bool & strin, int & faces, double eps) const;
     ///
     void CalcSurfaceInverseRec (int inv);
     ///

libsrc/csg/specpoin.cpp

@@ -286,36 +286,30 @@ namespace netgen
 					  dynamic_cast<const Plane*> (geometry->GetSurface(locsurf[k3])),
 					  pts);
 
-		      for (int j = 0; j < pts.Size(); j++)
+                      for (auto pnt : pts)
-			if (Dist (pts[j], box.Center()) < box.Diam()/2)
+                        if (Dist (pnt, box.Center()) < box.Diam()/2)                        
 			  {
-			    Solid * tansol;
+			    auto tansol = sol -> TangentialSolid (pnt, surfids, 1e-9*size);
-			    sol -> TangentialSolid (pts[j], tansol, surfids, 1e-9*size);
+                            if (tansol)
-
+                              {
-			    if(!tansol)
+                                bool ok1 = false, ok2 = false, ok3 = false;
-			      continue;
+                                int rep1 = geometry->GetSurfaceClassRepresentant(locsurf[k1]);
-
+                                int rep2 = geometry->GetSurfaceClassRepresentant(locsurf[k2]);
-			    bool ok1 = false, ok2 = false, ok3 = false;
+                                int rep3 = geometry->GetSurfaceClassRepresentant(locsurf[k3]);
-			    int rep1 = geometry->GetSurfaceClassRepresentant(locsurf[k1]);
+                                
-			    int rep2 = geometry->GetSurfaceClassRepresentant(locsurf[k2]);
+                                for (auto surfid : surfids)
-			    int rep3 = geometry->GetSurfaceClassRepresentant(locsurf[k3]);
+                                  {
-			    for(int jj=0; jj<surfids.Size(); jj++)
+                                    int actrep = geometry->GetSurfaceClassRepresentant(surfid);
-			      {
+                                    if (actrep == rep1) ok1 = true;
-				int actrep = geometry->GetSurfaceClassRepresentant(surfids[jj]);
+                                    if (actrep == rep2) ok2 = true;
-				if(actrep == rep1) ok1 = true;
+                                    if (actrep == rep3) ok3 = true;				  
-				if(actrep == rep2) ok2 = true;
+                                  }
-				if(actrep == rep3) ok3 = true;				  
+                                
-			      }
+                                if (ok1 && ok2 && ok3)
-
+                                  if (AddPoint (pnt, layer))
-			    
+                                    (*testout) << "cross point found, 1: " << pnt << endl;
-			    if (tansol && ok1 && ok2 && ok3)
+                              }
-			    // if (sol -> IsIn (pts[j], 1e-6*size) && !sol->IsStrictIn (pts[j], 1e-6*size))
+                          }
-			      {
-				if (AddPoint (pts[j], layer))
-				  (*testout) << "cross point found, 1: " << pts[j] << endl;
-			      }  
-			    delete tansol;
-			  }
 		    }
 
 
@@ -330,39 +324,30 @@ namespace netgen
 					    qsurf, pts);
 			//(*testout) << "checking pot. crosspoints: " << pts << endl;
 
-			for (int j = 0; j < pts.Size(); j++)
+                        for (auto pnt : pts)
-			  if (Dist (pts[j], box.Center()) < box.Diam()/2)
+                          if (Dist (pnt, box.Center()) < box.Diam()/2)
 			    {
-			      Solid * tansol;
+			      auto tansol = sol -> TangentialSolid (pnt, surfids, 1e-9*size);
-			      sol -> TangentialSolid (pts[j], tansol, surfids, 1e-9*size);
+                              if (tansol)
-
+                                {
-			      if(!tansol)
+                                  bool ok1 = false, ok2 = false, ok3 = true;//false;
-				continue;
+                                  int rep1 = geometry->GetSurfaceClassRepresentant(locsurf[k1]);
-			      			      
+                                  int rep2 = geometry->GetSurfaceClassRepresentant(locsurf[k2]);
-			      bool ok1 = false, ok2 = false, ok3 = true;//false;
+                                  
-			      int rep1 = geometry->GetSurfaceClassRepresentant(locsurf[k1]);
+                                  for (auto surfid : surfids)
-			      int rep2 = geometry->GetSurfaceClassRepresentant(locsurf[k2]);
+                                    {
-			      //int rep3 = geometry->GetSurfaceClassRepresentant(quadi);
+                                      int actrep = geometry->GetSurfaceClassRepresentant(surfid);
-			      for(int jj=0; jj<surfids.Size(); jj++)
+                                      if (actrep == rep1) ok1 = true;
-				{
+                                      if (actrep == rep2) ok2 = true;
-				  int actrep = geometry->GetSurfaceClassRepresentant(surfids[jj]);
+                                    }
-				  if(actrep == rep1) ok1 = true;
+                                  
-				  if(actrep == rep2) ok2 = true;
+                                  if (ok1 && ok2 && ok3)
-				  //if(actrep == rep3) ok3 = true;				  
+                                    if (AddPoint (pnt, layer))
-				}
+                                      (*testout) << "cross point found, 2: " << pnt << endl;
-
+                                }
-
-			      if (tansol && ok1 && ok2 && ok3)
-				//if (sol -> IsIn (pts[j], 1e-6*size) && !sol->IsStrictIn (pts[j], 1e-6*size) )
-				{
-				  if (AddPoint (pts[j], layer))
-				    (*testout) << "cross point found, 2: " << pts[j] << endl;
-				}  
-			      delete tansol;
 			    }
 		      }
 
-
 		for (int k1 = 0; k1 < numprim; k1++)
 		  if (k1 != quadi)
 		    {
@@ -372,15 +357,10 @@ namespace netgen
 		      for (int j = 0; j < pts.Size(); j++)
 			if (Dist (pts[j], box.Center()) < box.Diam()/2)
 			  {
-			    Solid * tansol;
+			    auto tansol = sol -> TangentialSolid (pts[j], surfids, 1e-9*size);
-			    sol -> TangentialSolid (pts[j], tansol, surfids, 1e-9*size);
 			    if (tansol)
-			      // sol -> IsIn (pts[j], 1e-6*size) && !sol->IsStrictIn (pts[j], 1e-6*size) )
+                              if (AddPoint (pts[j], layer))
-			      {
+                                (*testout) << "extremal point found, 1: " << pts[j] << endl;
-				if (AddPoint (pts[j], layer))
-				  (*testout) << "extremal point found, 1: " << pts[j] << endl;
-			      }  
-			    delete tansol;
 			  }
 		    }
 	      }
@@ -395,8 +375,6 @@ namespace netgen
 	    NgArray<Point<3> > pts;
 	    NgArray<int> surfids;
 
-
-	    
 	    for (int k1 = 0; k1 < numprim; k1++)
 	      for (int k2 = 0; k2 < k1; k2++)
 		for (int k3 = 0; k3 < k2; k3++)
@@ -409,16 +387,14 @@ namespace netgen
 		    for (int j = 0; j < pts.Size(); j++)
 		      if (Dist (pts[j], box.Center()) < box.Diam()/2)
 			{
-			  Solid * tansol;
+			  auto tansol = sol -> TangentialSolid (pts[j], surfids, 1e-9*size);
-			  sol -> TangentialSolid (pts[j], tansol, surfids, 1e-9*size);
+			  if (!tansol) continue;
-			  
-			  if(!tansol)
-			    continue;
 			  
 			  bool ok1 = false, ok2 = false, ok3 = false;
 			  int rep1 = geometry->GetSurfaceClassRepresentant(locsurf[k1]);
 			  int rep2 = geometry->GetSurfaceClassRepresentant(locsurf[k2]);
 			  int rep3 = geometry->GetSurfaceClassRepresentant(locsurf[k3]);
+                          
 			  for(int jj=0; jj<surfids.Size(); jj++)
 			    {
 			      int actrep = geometry->GetSurfaceClassRepresentant(surfids[jj]);
@@ -427,14 +403,9 @@ namespace netgen
 			      if(actrep == rep3) ok3 = true;				  
 			    }
 			  
-			  
+			  if (ok1 && ok2 && ok3)
-			  if (tansol && ok1 && ok2 && ok3)
+                            if (AddPoint (pts[j], layer))
-			    // if (sol -> IsIn (pts[j], 1e-6*size) && !sol->IsStrictIn (pts[j], 1e-6*size))
+                              (*testout) << "cross point found, 1: " << pts[j] << endl;
-			    {
-			      if (AddPoint (pts[j], layer))
-				(*testout) << "cross point found, 1: " << pts[j] << endl;
-			    }  
-			  delete tansol;
 			}
 		  }
 	    
@@ -449,23 +420,37 @@ namespace netgen
 		  for (int j = 0; j < pts.Size(); j++)
 		    if (Dist (pts[j], box.Center()) < box.Diam()/2)
 		      {
-			Solid * tansol;
+			auto tansol = sol -> TangentialSolid (pts[j], surfids, 1e-9*size);
-			sol -> TangentialSolid (pts[j], tansol, surfids, 1e-9*size);
 			if (tansol)
-			  // sol -> IsIn (pts[j], 1e-6*size) && !sol->IsStrictIn (pts[j], 1e-6*size) )
+                          if (AddPoint (pts[j], layer))
-			  {
+                            (*testout) << "extremal point found, spheres: " << pts[j] << endl;
-			    if (AddPoint (pts[j], layer))
-			      (*testout) << "extremal point found, spheres: " << pts[j] << endl;
-			  }  
-			delete tansol;
 		      }
 		}
 	    
 	    return;
 	  }
-      }
 
 
+        if (numprim == 2)
+          {
+            auto rev0 = dynamic_cast<const RevolutionFace*> (geometry->GetSurface(locsurf[0]));
+            auto rev1 = dynamic_cast<const RevolutionFace*> (geometry->GetSurface(locsurf[1]));
+            if (rev0 && rev1)
+              {
+                NgArray<Point<3>> pts;
+                bool check = ComputeExtremalPoints (rev0, rev1, pts);
+                if (check)
+                  {
+                    for (auto p : pts)
+                      if (box.IsIn(p))
+                        AddPoint (p, layer);
+                    return;
+                  }
+              }
+          }
+      } // end if (numprim <= check_crosspoint)
+    
+
     
     possiblecrossp = (numprim >= 3) && calccp;
     surecrossp = 0;
@@ -494,7 +479,6 @@ namespace netgen
 		(*testout) << "k1,2,3 = " << k1 << "," << k2 << "," << k3 << ", nc = " << nc << ", deg = " << deg << endl;
 #endif
 
-
 		if (!nc && !deg) decision = 0;
 		if (nc) surecrossp = 1;
 	      }
@@ -608,9 +592,9 @@ namespace netgen
 		    decision = 0;
 		}
 	    }
-
+#ifdef DEVELOP
-	// (*testout) << "l = " << level << " dec/sureexp = " << decision << sureexp << endl;
+        (*testout) << "edgepnt decision = " << decision << " sure = " << sureexp << endl;
-
+#endif
 	if (decision && sureexp)
 	  {
 	    for (int k1 = 0; k1 < locsurf.Size() - 1; k1++)
@@ -890,9 +874,18 @@ namespace netgen
 	f1->CalcGradient (p, g1);
 	f2->CalcGradient (p, g2);
 
-	if ( sqr (g1 * g2) > (1 - 1e-10) * Abs2 (g1) * Abs2 (g2))
+	// if ( sqr (g1 * g2) > (1 - 1e-10) * Abs2 (g1) * Abs2 (g2))
-	  return 1;
+        // return 1;
+        if ( Abs2 (Cross(g1,g2)) < 1e-10 * Abs2 (g1) * Abs2 (g2))  // same, but stable
+          {
+            if (Abs2(vrs) < 1e-12*sqr(size))  // degenerate only if on both surfaces
+              return 1;
+            else
+              return 0;
+          }
 
+
+        
 	for (int j = 0; j < 3; j++)
 	  {
 	    mat(0,j) = g1(j);
@@ -1534,7 +1527,61 @@ namespace netgen
   }
 
 
+  bool SpecialPointCalculation :: 
+  ComputeExtremalPoints (const RevolutionFace * rev1, 
+			 const RevolutionFace * rev2, 
+			 NgArray<Point<3> > & pts)
+  {
+    // if (rev1 -> P0() != rev2 -> P0()) return false; // missing ????
+    if (Dist2 (rev1 -> P0(), rev2 -> P0()) > 1e-20*sqr(size)) return false;
+    // if (rev1 -> Axis() != rev2 -> Axis()) return false;
+    if ( (rev1 -> Axis()-rev2 -> Axis()).Length2() > 1e-16) return false;
 
+    Point<2> p1s = rev1->GetSpline().StartPI();
+    Point<2> p1e = rev1->GetSpline().EndPI();
+    Point<2> p2s = rev2->GetSpline().StartPI();
+    Point<2> p2e = rev2->GetSpline().EndPI();
+
+    Point<2> p2d;
+    if (Dist2(p1s,p2e) < 1e-20*sqr(size))
+      p2d = p1s;
+    else if (Dist2(p1e, p2s) < 1e-20*sqr(size))
+      p2d = p1e;
+    else
+      return false;
+    *testout << "Norm axis = " << rev1->Axis().Length() << endl;
+    Point<3> center = rev1->P0() + p2d(0)*rev1->Axis();
+    Vec<3> n = rev1->Axis();
+    // extremal points of circle, center, normal axis, radius p2d(1)
+    // Lagrange:
+    // L(x, lam1, lam2) = x_i + lam1 * (x-c)*v + lam2 * ( |x-c|^2 - r^2 )
+    for (double i = 0; i < 3; i++)
+      {
+        double lam1 = -n(i) / n.Length2();
+        Vec<3> ei(0,0,0); ei(i) = 1;
+        // double lam2 = 1/(2*p2d(1)) * sqrt(1 - sqr(n(i))/n.Length2());
+        double fac = 1-sqr(n(i))/n.Length2();
+        // if (fabs(lam2) > 1e-10)
+        if (fac > 1e-10)
+          {
+            double lam2 = 1/(2*p2d(1)) * sqrt(fac);
+            Point<3> x = center - 1.0/(2*lam2) * (ei + lam1*n);
+            pts.Append (x);
+            x = center + 1.0/(2*lam2) * (ei + lam1*n);
+            pts.Append (x);
+
+            /*
+            // check:
+            Point<2> p2d;
+            rev1 -> CalcProj (x, p2d);
+            *testout << "special solution, p2d = " << p2d << endl;
+            rev2 -> CalcProj (x, p2d);
+            *testout << "special solution, p2d = " << p2d << endl;
+            */
+          }
+      }
+    return true;
+  }
 
 
 
@@ -1739,9 +1786,7 @@ namespace netgen
 	      continue;
 	    
 
-	    Solid * locsol;
+	    auto locsol = sol -> TangentialSolid (p, surfind, ideps*geomsize);
-	    sol -> TangentialSolid (p, locsol, surfind, ideps*geomsize);
-
 
 	    rep_surfind.SetSize (surfind.Size());
 	    int num_indep_surfs = 0;
@@ -1769,10 +1814,10 @@ namespace netgen
 	    if (surf)
 	      {
 		// locsol -> GetSurfaceIndices (surfind);
-		bool hassurf = 0;
+		bool hassurf = false;
 		for (int m = 0; m < surfind.Size(); m++)
 		  if (ageometry.GetSurface(surfind[m]) == surf)
-		    hassurf = 1;
+		    hassurf = true;
 
 		if (!hassurf)
 		  continue;
@@ -1818,8 +1863,11 @@ namespace netgen
 		    }
 
 
-		  if (Abs2 (t) < 1e-8)
+		  if (Abs2 (t) < 1e-16)
-		    continue;
+                    {
+                      // cerr << "normal vectors degenerated" << endl;
+                      continue;
+                    }
 
 #ifdef DEVELOP
 		  *testout << "           tangential vector " << t << endl;
@@ -1860,17 +1908,17 @@ namespace netgen
 		      rhs(0) = -t * (hessej * t);
 		      rhs(1) = -t * (hessek * t);
 
-		      CalcInverse (mat, inv);
+                      CalcInverse (mat, inv);
-		      t2 = inv * rhs;
+                      t2 = inv * rhs;
-
+#ifdef DEVELOP
-		      
+                      *testout << "t = " << t << ", t2 = " << t2 << endl;
+#endif
 		      /*
 		      ageometry.GetIndependentSurfaceIndices 
 			(locsol, p, t, surfind2);
 		      */
 
-		      Solid * locsol2;
+		      auto locsol2 = locsol -> TangentialSolid3 (p, t, t2, surfind2, ideps*geomsize); 
-		      locsol -> TangentialSolid3 (p, t, t2, locsol2, surfind2, ideps*geomsize); 
 		      if (!locsol2) continue;
 		      
 		      // locsol2 -> GetTangentialSurfaceIndices3 (p, t, t2, surfind2, 1e-9*geomsize);
@@ -1911,31 +1959,31 @@ namespace netgen
 			  Vec<3> nv =
 			    ageometry.GetSurface(surfind2[l]) -> GetNormalVector(p);
 
-			 
 			  Vec<3> m1 = Cross (t, nv);
 			  Vec<3> m2 = -m1;
 			  bool isface1 = 0, isface2 = 0;
 			  
-			  Solid * locsol3;
-
 			  // locsol2 -> TangentialSolid2 (p, m1, locsol3, surfind3, 1e-9*geomsize);
-			  locsol -> TangentialEdgeSolid (p, t, t2, m1, locsol3, surfind3, ideps*geomsize);
+			  auto locsol3 = locsol -> TangentialEdgeSolid (p, t, t2, m1, surfind3, ideps*geomsize);
-
+#ifdef DEVELOP
+                          (*testout) << "m1 = " << m1 << ", surfind3 = " << surfind3 << endl;
+#endif
 			  //ageometry.GetIndependentSurfaceIndices (surfind3);
 
 			  if (surfind3.Contains(surfind2[l]))
 			    isface1 = 1;
-			  delete locsol3;
 			  
 			  // locsol2 -> TangentialSolid2 (p, m2, locsol3, surfind3, 1e-9*geomsize);
-			  locsol -> TangentialEdgeSolid (p, t, t2, m2, locsol3, surfind3, ideps*geomsize); 
+			  locsol3 = locsol -> TangentialEdgeSolid (p, t, t2, m2, surfind3, ideps*geomsize); 
+#ifdef DEVELOP
+                          (*testout) << "m2 = " << m2 << ", surfind3 = " << surfind3 << endl;
+#endif
 
 			  // ageometry.GetIndependentSurfaceIndices (surfind3);
 
 			  
 			  if (surfind3.Contains(surfind2[l]))
 			    isface2 = 1;
-			  delete locsol3;
 
 			  if (isface1 != isface2)
 			    cnt_tang_faces++;
@@ -1948,7 +1996,6 @@ namespace netgen
 		      if (cnt_tang_faces < 1)
 			ok = false;
 
-		      delete locsol2;
 		      if (!ok) continue;
 		    }
 
@@ -1973,16 +2020,47 @@ namespace netgen
 			    continue;
 			  
 			  Vec<3> s = Cross (normalvecs[m], t);
+
 			  Vec<3> t2a = t + 0.01 *s;
 			  Vec<3> t2b = t - 0.01 *s;
-			  
+
-			  bool isface =
+			  bool isfaceold =
 			    (locsol->VectorIn (p, t2a, 1e-6*geomsize) &&
 			     !locsol->VectorStrictIn (p, t2a, 1e-6*geomsize))
 			    ||
 			    (locsol->VectorIn (p, t2b, 1e-6*geomsize) &&
 			     !locsol->VectorStrictIn (p, t2b, 1e-6*geomsize));
-			  
+
+                          bool isfacenew =
+                            locsol -> VecInSolid2(p, t, s, 1e-6*geomsize) == DOES_INTERSECT ||
+                            locsol -> VecInSolid2(p, t, -s, 1e-6*geomsize) == DOES_INTERSECT;
+                            /*
+                            (locsol->VectorIn2 (p, t, s, 1e-6*geomsize) && !locsol->VectorStrictIn2 (p, t, s, 1e-6*geomsize)) || 
+                            (locsol->VectorIn2 (p, t, -s, 1e-6*geomsize) && !locsol->VectorStrictIn2 (p, t, -s, 1e-6*geomsize));
+                            */
+
+                          bool isface = isfacenew;
+                          
+                          if (isfaceold != isfacenew)
+                            {
+                              *testout << "different, p = " << p << ", t = " << t << ", s = " << s << endl;
+                              *testout << "tlo = " << si << endl;
+                              *testout << "isface, old = " << isface << ", isfacenew = " << isfacenew << endl;
+                              
+                              *testout << "t2a = " << t2a << endl;
+                              *testout << "vecin(p,t2a) = " << locsol->VectorIn (p, t2a, 1e-6*geomsize) << endl;
+                              *testout << "vecstrictin(p,t2a) = " << locsol->VectorStrictIn (p, t2a, 1e-6*geomsize) << endl;
+                              *testout << "vectorin2 = " << locsol->VectorIn2 (p, t, s, 1e-6*geomsize)  << endl;
+                              *testout << "vectorstrictin2 = " << locsol->VectorStrictIn2 (p, t, s, 1e-6*geomsize)  << endl;
+                              
+                              *testout << "t2b = " << t2b << endl;
+                              *testout << "vecin(p,t2b) = " << locsol->VectorIn (p, t2b, 1e-6*geomsize) << endl;
+                              *testout << "vecstrictin(p,t2b) = " << locsol->VectorStrictIn (p, t2b, 1e-6*geomsize) << endl;
+                              *testout << "vectorin2- = " << locsol->VectorIn2 (p, t, -s, 1e-6*geomsize)  << endl;
+                              *testout << "vectorstrictin2- = " << locsol->VectorStrictIn2 (p, t, -s, 1e-6*geomsize)  << endl;
+                            }
+                          
+                          
 			  /*
 			  bool isface =
 			    (locsol->VectorIn (p, t2a) &&
@@ -1992,10 +2070,8 @@ namespace netgen
 			     !locsol->VectorStrictIn (p, t2b));
 			  */
 
-			  if (isface)
+                          if (isface)
-			    {
+                            cnts++;
-			      cnts++;
-			    }
 			}
 		      if (cnts < 2) isedge = 0;
 		    }
@@ -2054,54 +2130,22 @@ namespace netgen
 		    }
 		  
 		}
-
-	    delete locsol;
 	  }
       }
 
-    /*
-    NgBitArray testuncond (specpoints.Size());
-    testuncond.Clear();
-    for(int i = 0; i<specpoints.Size(); i++)
-      {
-	if(testuncond.Test(i))
-	  continue;
-	
-	NgArray<int> same;
-	same.Append(i);
-	
-	for(int j = i+1; j<specpoints.Size(); j++)
-	  {
-	    if(Dist(specpoints[i].p,specpoints[j].p) < 1e-20)
-	      {
-		same.Append(j);
-		testuncond.Set(j);
-	      }
-	  }
-	
-	if(same.Size() < 3)
-	  for(int j=0; j<same.Size(); j++)
-	    {
-	      (*testout) << "setting " << specpoints[same[j]].p << "; " << specpoints[same[j]].v << "; " 
-			 <<specpoints[same[j]].unconditional << " to conditional" << endl;
-	      specpoints[same[j]].unconditional=0;
-	    }
-      }
-    */
-
 
+    
     // if special point is unconditional on some solid,
     // it must be unconditional everywhere:
 
-    NgBitArray uncond (apoints.Size());
+    BitArray uncond (apoints.Size());
     uncond.Clear();
 
     for (int i = 0; i < specpoints.Size(); i++)
       if (specpoints[i].unconditional)
-	uncond.Set (specpoint2point[i]);
+	uncond.SetBit (specpoint2point[i]);
   
     for (int i = 0; i < specpoints.Size(); i++)
-      specpoints[i].unconditional = 
+      specpoints[i].unconditional = uncond.Test (specpoint2point[i]);
-	uncond.Test (specpoint2point[i]) ? 1 : 0;
   }
 }

libsrc/csg/specpoin.hpp

@@ -167,6 +167,9 @@ namespace netgen
 				const Sphere * sphere2, 
 				NgArray<Point<3> > & pts);
 
+    bool ComputeExtremalPoints (const RevolutionFace * rev1, 
+				const RevolutionFace * rev2, 
+				NgArray<Point<3> > & pts);
 
     void ComputeCrossPoints (const Plane * plane1, 
 			     const Plane * plane2, 

libsrc/csg/surface.cpp

@@ -16,6 +16,7 @@ Surface :: Surface ()
   strcpy (name, "noname");
   bcprop = -1;
   bcname = "default";
+  inverse = false;
 }
 
 Surface :: ~Surface()
@@ -427,11 +428,20 @@ VecInSolid2 (const Point<3> & p,
   if (hv1 >= eps)
     return IS_OUTSIDE;
 
+  double hv2 = v2 * hv;
+  if (hv2 <= -eps)
+    return IS_INSIDE;
+  if (hv2 >= eps)
+    return IS_OUTSIDE;
+  return DOES_INTERSECT;
+  
+  /*
   double hv2 = v2 * hv;
   if (hv2 <= 0)
     return IS_INSIDE;
   else
     return IS_OUTSIDE;
+  */
 }
   
 

libsrc/csg/surface.hpp

@@ -213,8 +213,6 @@ namespace netgen
     INSOLID_TYPE;
 
 
-
-
   class DummySurface : public Surface
   {
     virtual double CalcFunctionValue (const Point<3> & /* point */) const
@@ -293,9 +291,13 @@ namespace netgen
 				      const Vec<3> & m,
 				      double eps) const;
 
+    // for a point p in the surface, into which (closed) surfaces does v point into ? 
     virtual void GetTangentialVecSurfaceIndices (const Point<3> & p, const Vec<3> & v,
 						 NgArray<int> & surfind, double eps) const;
 
+    // a point p in the surface, and v a tangential vector
+    // for arbitrary small, but positive t consider q := Project(p+t*v)
+    // into which (closed) surfaces does v2 point into, when starting from q ?     
     virtual void GetTangentialVecSurfaceIndices2 (const Point<3> & p, const Vec<3> & v1, const Vec<3> & v2,
 						  NgArray<int> & surfind, double eps) const;
 

libsrc/general/CMakeLists.txt

@@ -11,7 +11,7 @@ target_sources(gen INTERFACE
 install(FILES
   ngarray.hpp autodiff.hpp autoptr.hpp ngbitarray.hpp
   dynamicmem.hpp hashtabl.hpp mpi_interface.hpp myadt.hpp
-  ngsimd.hpp mystring.hpp netgenout.hpp ngpython.hpp
+  mystring.hpp netgenout.hpp ngpython.hpp
   optmem.hpp parthreads.hpp seti.hpp sort.hpp
   spbita2d.hpp stack.hpp table.hpp template.hpp
   gzstream.h

libsrc/general/mpi_interface.hpp

@@ -14,65 +14,8 @@
 
 namespace netgen
 {
-  // using ngcore::id;
-  // using ngcore::ntasks;
-
-#ifndef PARALLEL
-  /** without MPI, we need a dummy typedef **/
-  // typedef int MPI_Comm;
-#endif
-
-  /** This is the "standard" communicator that will be used for netgen-objects. **/
-  // extern DLL_HEADER NgMPI_Comm ng_comm;
 
 #ifdef OLD
-#ifdef PARALLEL
-  inline int MyMPI_GetNTasks (MPI_Comm comm /* = ng_comm */)
-  {
-    int ntasks;
-    MPI_Comm_size(comm, &ntasks);
-    return ntasks;
-  }
-  inline int MyMPI_GetId (MPI_Comm comm /* = ng_comm */)
-  {
-    int id;
-    MPI_Comm_rank(comm, &id);
-    return id;
-  }
-#else
-  // enum { MPI_COMM_WORLD = 12345, MPI_COMM_NULL = 0};
-  inline int MyMPI_GetNTasks (MPI_Comm comm /* = ng_comm */) { return 1; }
-  inline int MyMPI_GetId (MPI_Comm comm /* = ng_comm */) { return 0; }
-#endif
-#endif
-  
-  /*
-#ifdef PARALLEL
-  // For python wrapping of communicators
-  struct PyMPI_Comm {
-    MPI_Comm comm;
-    bool owns_comm;
-    PyMPI_Comm (MPI_Comm _comm, bool _owns_comm = false) : comm(_comm), owns_comm(_owns_comm) { }
-    PyMPI_Comm (const PyMPI_Comm & c) = delete;
-    ~PyMPI_Comm () {
-      if (owns_comm)
-	MPI_Comm_free(&comm);
-    }
-    inline int Rank() const { return MyMPI_GetId(comm); }
-    inline int Size() const { return MyMPI_GetNTasks(comm); }
-  };
-#else
-  // dummy without MPI
-  struct PyMPI_Comm {
-    MPI_Comm comm = 0;
-    PyMPI_Comm (MPI_Comm _comm, bool _owns_comm = false) { }
-    ~PyMPI_Comm () { }
-    inline int Rank() const { return 0; }
-    inline int Size() const { return 1; }
-  };
-#endif
-  */
-  
 #ifdef PARALLEL
   template <class T>
   inline MPI_Datatype MyGetMPIType ( ) 
@@ -93,32 +36,37 @@ namespace netgen
   typedef int MPI_Datatype;
   template <class T> inline MPI_Datatype MyGetMPIType ( ) { return 0; }
 #endif
+#endif
 
+  
 #ifdef PARALLEL
   enum { MPI_TAG_CMD = 110 };
   enum { MPI_TAG_MESH = 210 };
   enum { MPI_TAG_VIS = 310 };
 
-  inline void MyMPI_Send (int i, int dest, int tag, MPI_Comm comm /* = ng_comm */)
+
+  [[deprecated("mympi_send int, use comm.Send instead")]]            
+  inline void MyMPI_Send (int i, int dest, int tag, MPI_Comm comm)
   {
     int hi = i;
     MPI_Send( &hi, 1, MPI_INT, dest, tag, comm);
   }
-
+  
-  inline void MyMPI_Recv (int & i, int src, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_revc int, use comm.Recv instead")]]            
+  inline void MyMPI_Recv (int & i, int src, int tag, MPI_Comm comm)
   {
     MPI_Status status;
     MPI_Recv( &i, 1, MPI_INT, src, tag, comm, &status);
   }
 
-
+  [[deprecated("mympi_send string, use comm.Send instead")]]              
-
+  inline void MyMPI_Send (const string & s, int dest, int tag, MPI_Comm comm)
-  inline void MyMPI_Send (const string & s, int dest, int tag, MPI_Comm comm /* = ng_comm */)
   {
     MPI_Send( const_cast<char*> (s.c_str()), s.length(), MPI_CHAR, dest, tag, comm);
   }
 
-  inline void MyMPI_Recv (string & s, int src, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_revc string, use comm.Recv instead")]]              
+  inline void MyMPI_Recv (string & s, int src, int tag, MPI_Comm comm)
   {
     MPI_Status status;
     int len;
@@ -130,34 +78,35 @@ namespace netgen
 
  
 
-
   template <class T, int BASE>
-  inline void MyMPI_Send (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_send ngflatarray, use comm.send instead")]]              
+  inline void MyMPI_Send (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm)
   {
-    MPI_Send( &s.First(), s.Size(), MyGetMPIType<T>(), dest, tag, comm);
+    MPI_Send( &s.First(), s.Size(), GetMPIType<T>(), dest, tag, comm);
   }
 
   template <class T, int BASE>
-  inline void MyMPI_Recv ( NgFlatArray<T, BASE> s, int src, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_recv ngflatarray, use comm.Recv instead")]]                
+  inline void MyMPI_Recv ( NgFlatArray<T, BASE> s, int src, int tag, MPI_Comm comm)
   {
     MPI_Status status;
-    MPI_Recv( &s.First(), s.Size(), MyGetMPIType<T>(), src, tag, comm, &status);
+    MPI_Recv( &s.First(), s.Size(), GetMPIType<T>(), src, tag, comm, &status);
   }
 
   template <class T, int BASE>
-  inline void MyMPI_Recv ( NgArray <T, BASE> & s, int src, int tag, MPI_Comm comm /* = ng_comm */)
+  inline void MyMPI_Recv ( NgArray <T, BASE> & s, int src, int tag, MPI_Comm comm)
   {
     MPI_Status status;
     int len;
     MPI_Probe (src, tag, comm, &status);
-    MPI_Get_count (&status, MyGetMPIType<T>(), &len);
+    MPI_Get_count (&status, GetMPIType<T>(), &len);
 
     s.SetSize (len);
-    MPI_Recv( &s.First(), len, MyGetMPIType<T>(), src, tag, comm, &status);
+    MPI_Recv( &s.First(), len, GetMPIType<T>(), src, tag, comm, &status);
   }
 
   template <class T, int BASE>
-  inline int MyMPI_Recv ( NgArray <T, BASE> & s, int tag, MPI_Comm comm /* = ng_comm */)
+  inline int MyMPI_Recv ( NgArray <T, BASE> & s, int tag, MPI_Comm comm)
   {
     MPI_Status status;
     int len;
@@ -165,10 +114,10 @@ namespace netgen
 
     int src = status.MPI_SOURCE;
 
-    MPI_Get_count (&status, MyGetMPIType<T>(), &len);
+    MPI_Get_count (&status, GetMPIType<T>(), &len);
 
     s.SetSize (len);
-    MPI_Recv( &s.First(), len, MyGetMPIType<T>(), src, tag, comm, &status);
+    MPI_Recv( &s.First(), len, GetMPIType<T>(), src, tag, comm, &status);
 
     return src;
   }
@@ -190,22 +139,23 @@ namespace netgen
   */
 
   template <class T, int BASE>
-  inline MPI_Request MyMPI_ISend (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_isend ngflatarray, use comm.send instead")]]                
+  inline MPI_Request MyMPI_ISend (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm)
   {
     MPI_Request request;
-    MPI_Isend( &s.First(), s.Size(), MyGetMPIType<T>(), dest, tag, comm, &request);
+    MPI_Isend( &s.First(), s.Size(), GetMPIType<T>(), dest, tag, comm, &request);
     return request;
   }
 
-
   template <class T, int BASE>
-  inline MPI_Request MyMPI_IRecv (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_irecv ngflatarray, use comm.recv instead")]]                
+  inline MPI_Request MyMPI_IRecv (NgFlatArray<T, BASE> s, int dest, int tag, MPI_Comm comm)
   {
     MPI_Request request;
-    MPI_Irecv( &s.First(), s.Size(), MyGetMPIType<T>(), dest, tag, comm, &request);
+    MPI_Irecv( &s.First(), s.Size(), GetMPIType<T>(), dest, tag, comm, &request);
     return request;
   }
-
+  
   /*
   template <class T, int BASE>
   inline void MyMPI_ISend (NgFlatArray<T, BASE> s, int dest, int tag)
@@ -232,106 +182,59 @@ namespace netgen
     receive-table entries will be set
    */
 
-  /*
   template <typename T>
   inline void MyMPI_ExchangeTable (TABLE<T> & send_data, 
 				   TABLE<T> & recv_data, int tag,
-				   MPI_Comm comm = MPI_COMM_WORLD)
+				   const NgMPI_Comm & comm)
   {
-    int ntasks, rank;
-    MPI_Comm_size(comm, &ntasks);
-    MPI_Comm_rank(comm, &rank);
-
-    NgArray<MPI_Request> requests;
-    for (int dest = 0; dest < ntasks; dest++)
-      if (dest != rank)
-	requests.Append (MyMPI_ISend (send_data[dest], dest, tag, comm));
-
-    for (int i = 0; i < ntasks-1; i++)
-      {
-	MPI_Status status;
-	MPI_Probe (MPI_ANY_SOURCE, tag, comm, &status);
-	int size, src = status.MPI_SOURCE;
-	MPI_Get_count (&status, MPI_INT, &size);
-	recv_data.SetEntrySize (src, size, sizeof(T));
-	requests.Append (MyMPI_IRecv (recv_data[src], src, tag, comm));
-      }
-    MPI_Barrier (comm);
-    MPI_Waitall (requests.Size(), &requests[0], MPI_STATUS_IGNORE);
-  }
-  */
-
-  template <typename T>
-  inline void MyMPI_ExchangeTable (TABLE<T> & send_data, 
-				   TABLE<T> & recv_data, int tag,
-				   const NgMPI_Comm & comm /* = ng_comm */)
-  {
-    /*
-    int rank = MyMPI_GetId(comm);
-    int ntasks = MyMPI_GetNTasks(comm);
-    */
     int rank = comm.Rank();
     int ntasks = comm.Size();
     
-    NgArray<int> send_sizes(ntasks);
+    Array<int> send_sizes(ntasks);
-    NgArray<int> recv_sizes(ntasks);
+    Array<int> recv_sizes(ntasks);
     for (int i = 0; i < ntasks; i++)
       send_sizes[i] = send_data[i].Size();
+
+    comm.AllToAll (send_sizes, recv_sizes);
     
-    MPI_Alltoall (&send_sizes[0], 1, MPI_INT, 
-		  &recv_sizes[0], 1, MPI_INT, comm);
-
-      // in-place is buggy !
-//    MPI_Alltoall (MPI_IN_PLACE, 1, MPI_INT, 
-//		  &recv_sizes[0], 1, MPI_INT, comm);
-
-
     for (int i = 0; i < ntasks; i++)
       recv_data.SetEntrySize (i, recv_sizes[i], sizeof(T));
     
-    NgArray<MPI_Request> requests;
+    Array<MPI_Request> requests;
     for (int dest = 0; dest < ntasks; dest++)
       if (dest != rank && send_data[dest].Size())
-	requests.Append (MyMPI_ISend (send_data[dest], dest, tag, comm));
+        requests.Append (comm.ISend (FlatArray<T>(send_data[dest]), dest, tag));
 
     for (int dest = 0; dest < ntasks; dest++)
       if (dest != rank && recv_data[dest].Size())
-	requests.Append (MyMPI_IRecv (recv_data[dest], dest, tag, comm));
+        requests.Append (comm.IRecv (FlatArray<T>(recv_data[dest]), dest, tag));
 
-    // MPI_Barrier (comm);
+    MyMPI_WaitAll (requests);
-    MPI_Waitall (requests.Size(), &requests[0], MPI_STATUS_IGNORE);
   }
 
 
-
-
-
-
-
   extern void MyMPI_SendCmd (const char * cmd);
   extern string MyMPI_RecvCmd ();
 
 
-
-
   template <class T>
-  inline void MyMPI_Bcast (T & s, MPI_Comm comm /* = ng_comm */)
+  inline void MyMPI_Bcast (T & s, MPI_Comm comm)
   {
-    MPI_Bcast (&s, 1, MyGetMPIType<T>(), 0, comm);
+    MPI_Bcast (&s, 1, GetMPIType<T>(), 0, comm);
   }
 
   template <class T>
-  inline void MyMPI_Bcast (NgArray<T, 0> & s, NgMPI_Comm comm /* = ng_comm */)
+  inline void MyMPI_Bcast (NgArray<T, 0> & s, NgMPI_Comm comm)
   {
     int size = s.Size();
     MyMPI_Bcast (size, comm);
     // if (MyMPI_GetId(comm) != 0) s.SetSize (size);
     if (comm.Rank() != 0) s.SetSize (size);
-    MPI_Bcast (&s[0], size, MyGetMPIType<T>(), 0, comm);
+    MPI_Bcast (&s[0], size, GetMPIType<T>(), 0, comm);
   }
 
   template <class T>
-  inline void MyMPI_Bcast (NgArray<T, 0> & s, int root, MPI_Comm comm /* = ng_comm */)
+  inline void MyMPI_Bcast (NgArray<T, 0> & s, int root, MPI_Comm comm)
   {
     int id;
     MPI_Comm_rank(comm, &id);
@@ -340,67 +243,23 @@ namespace netgen
     MPI_Bcast (&size, 1, MPI_INT, root, comm);
     if (id != root) s.SetSize (size);
     if ( !size ) return;
-    MPI_Bcast (&s[0], size, MyGetMPIType<T>(), root, comm);
+    MPI_Bcast (&s[0], size, GetMPIType<T>(), root, comm);
   }
 
   template <class T, class T2>
-  inline void MyMPI_Allgather (const T & send, NgFlatArray<T2> recv, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_allgather deprecated, use comm.allgather")]]                    
+  inline void MyMPI_Allgather (const T & send, NgFlatArray<T2> recv, MPI_Comm comm)
   {
-    MPI_Allgather( const_cast<T*> (&send), 1, MyGetMPIType<T>(), &recv[0], 1, MyGetMPIType<T2>(), comm);
+    MPI_Allgather( const_cast<T*> (&send), 1, GetMPIType<T>(), &recv[0], 1, GetMPIType<T2>(), comm);
   }
 
   template <class T, class T2>
-  inline void MyMPI_Alltoall (NgFlatArray<T> send, NgFlatArray<T2> recv, MPI_Comm comm /* = ng_comm */)
+  [[deprecated("mympi_alltoall deprecated, use comm.alltoall")]]                  
+  inline void MyMPI_Alltoall (NgFlatArray<T> send, NgFlatArray<T2> recv, MPI_Comm comm)
   {
-    MPI_Alltoall( &send[0], 1, MyGetMPIType<T>(), &recv[0], 1, MyGetMPIType<T2>(), comm);
+    MPI_Alltoall( &send[0], 1, GetMPIType<T>(), &recv[0], 1, GetMPIType<T2>(), comm);
   }
 
-//   template <class T, class T2>
-//   inline void MyMPI_Alltoall_Block (NgFlatArray<T> send, NgFlatArray<T2> recv, int blocklen, MPI_Comm comm = ng_comm)
-//   {
-//     MPI_Alltoall( &send[0], blocklen, MyGetMPIType<T>(), &recv[0], blocklen, MyGetMPIType<T2>(), comm);
-//   }
-
-
-
-  /*
-  inline void MyMPI_Send (  int *& s, int len,  int dest, int tag)
-  {
-    int hlen = len;
-    MPI_Send( &hlen, 1, MPI_INT, dest, tag, MPI_COMM_WORLD);
-    MPI_Send( s, len, MPI_INT, dest, tag, MPI_COMM_WORLD);
-  }
-
-
-  inline void MyMPI_Recv ( int *& s, int & len, int src, int tag)
-  {
-    MPI_Status status;
-    MPI_Recv( &len, 1, MPI_INT, src, tag, MPI_COMM_WORLD, &status);
-    if ( s ) 
-      delete [] s;
-    s = new int [len];
-    MPI_Recv( s, len, MPI_INT, src, tag, MPI_COMM_WORLD, &status);
-  }
-
-
-
-  inline void MyMPI_Send ( double * s, int len,  int dest, int tag)
-  {
-     MPI_Send( &len, 1, MPI_INT, dest, tag, MPI_COMM_WORLD);
-     MPI_Send( s, len, MPI_DOUBLE, dest, tag, MPI_COMM_WORLD);
-  }
-
-
-  inline void MyMPI_Recv ( double *& s, int & len, int src, int tag)
-  {
-    MPI_Status status;
-    MPI_Recv( &len, 1, MPI_INT, src, tag, MPI_COMM_WORLD, &status);
-    if ( s )
-      delete [] s;
-    s = new double [len];
-    MPI_Recv( s, len, MPI_DOUBLE, src, tag, MPI_COMM_WORLD, &status);
-  }
-  */
 
 #endif // PARALLEL
 

libsrc/general/myadt.hpp

@@ -47,7 +47,5 @@ namespace netgen
 #include "netgenout.hpp"
 #include "gzstream.h"
 
-#include "ngsimd.hpp"
-
 
 #endif

libsrc/general/ngarray.hpp

@@ -205,6 +205,12 @@ namespace netgen
     {
       return ( Pos(elem) >= 0 );
     }
+
+    operator FlatArray<T> () const
+    {
+      static_assert (BASE==0);
+      return FlatArray<T>(size, data);
+    }
   };
 
 
@@ -730,7 +736,7 @@ namespace netgen
 
   /// bubble sort array
   template <class T, class S>
-  inline void BubbleSort (NgFlatArray<T> & data, NgFlatArray<S> & slave)
+  inline void BubbleSort (NgFlatArray<T> & data, NgFlatArray<S> & index)
   {
     for (int i = 0; i < data.Size(); i++)
       for (int j = i+1; j < data.Size(); j++)
@@ -740,16 +746,16 @@ namespace netgen
 	    data[i] = data[j];
 	    data[j] = hv;
 	    
-	    S hvs = slave[i];
+	    S hvs = index[i];
-	    slave[i] = slave[j];
+	    index[i] = index[j];
-	    slave[j] = hvs;
+	    index[j] = hvs;
 	  }
   }
 
 
   template <class T, class S>
   void QuickSortRec (NgFlatArray<T> & data,
-		     NgFlatArray<S> & slave,
+		     NgFlatArray<S> & index,
 		     int left, int right)
   {
     int i = left;
@@ -764,20 +770,20 @@ namespace netgen
 	if (i <= j)
 	  {
             ngcore::Swap (data[i], data[j]);
-            ngcore::Swap (slave[i], slave[j]);
+            ngcore::Swap (index[i], index[j]);
 	    i++; j--;
 	  }
       }
     while (i <= j);
-    if (left < j) QuickSortRec (data, slave, left, j);
+    if (left < j) QuickSortRec (data, index, left, j);
-    if (i < right) QuickSortRec (data, slave, i, right);
+    if (i < right) QuickSortRec (data, index, i, right);
   }
 
   template <class T, class S>
-  void QuickSort (NgFlatArray<T> & data, NgFlatArray<S> & slave)
+  void QuickSort (NgFlatArray<T> & data, NgFlatArray<S> & index)
   {
     if (data.Size() > 1)
-      QuickSortRec (data, slave, 0, data.Size()-1);
+      QuickSortRec (data, index, 0, data.Size()-1);
   }
 
 

libsrc/general/ngsimd.hpp

@@ -1,672 +0,0 @@
-#ifndef FILE_NGSIMD
-#define FILE_NGSIMD
-/**************************************************************************/
-/* File:   ngsimd.hpp                                                     */
-/* Author: Joachim Schoeberl                                              */
-/* Date:   25. Mar. 16                                                    */
-/**************************************************************************/
-
-#include <immintrin.h>
-#include <tuple>
-#include <ostream>
-#include <stdexcept>
-#include <string>
-#include <type_traits>
-
-#include <core/utils.hpp>
-
-#ifdef WIN32
-#ifndef AVX_OPERATORS_DEFINED
-#define AVX_OPERATORS_DEFINED
-NG_INLINE __m128d operator- (__m128d a) { return _mm_xor_pd(a, _mm_set1_pd(-0.0)); }
-NG_INLINE __m128d operator+ (__m128d a, __m128d b) { return _mm_add_pd(a,b); }
-NG_INLINE __m128d operator- (__m128d a, __m128d b) { return _mm_sub_pd(a,b); }
-NG_INLINE __m128d operator* (__m128d a, __m128d b) { return _mm_mul_pd(a,b); }
-NG_INLINE __m128d operator/ (__m128d a, __m128d b) { return _mm_div_pd(a,b); }
-NG_INLINE __m128d operator* (double a, __m128d b) { return _mm_set1_pd(a)*b; }
-NG_INLINE __m128d operator* (__m128d b, double a) { return _mm_set1_pd(a)*b; }
-
-NG_INLINE __m128d operator+= (__m128d &a, __m128d b) { return a = a+b; }
-NG_INLINE __m128d operator-= (__m128d &a, __m128d b) { return a = a-b; }
-NG_INLINE __m128d operator*= (__m128d &a, __m128d b) { return a = a*b; }
-NG_INLINE __m128d operator/= (__m128d &a, __m128d b) { return a = a/b; }
-
-NG_INLINE __m256d operator- (__m256d a) { return _mm256_xor_pd(a, _mm256_set1_pd(-0.0)); }
-NG_INLINE __m256d operator+ (__m256d a, __m256d b) { return _mm256_add_pd(a,b); }
-NG_INLINE __m256d operator- (__m256d a, __m256d b) { return _mm256_sub_pd(a,b); }
-NG_INLINE __m256d operator* (__m256d a, __m256d b) { return _mm256_mul_pd(a,b); }
-NG_INLINE __m256d operator/ (__m256d a, __m256d b) { return _mm256_div_pd(a,b); }
-NG_INLINE __m256d operator* (double a, __m256d b) { return _mm256_set1_pd(a)*b; }
-NG_INLINE __m256d operator* (__m256d b, double a) { return _mm256_set1_pd(a)*b; }
-
-NG_INLINE __m256d operator+= (__m256d &a, __m256d b) { return a = a+b; }
-NG_INLINE __m256d operator-= (__m256d &a, __m256d b) { return a = a-b; }
-NG_INLINE __m256d operator*= (__m256d &a, __m256d b) { return a = a*b; }
-NG_INLINE __m256d operator/= (__m256d &a, __m256d b) { return a = a/b; }
-#endif
-#endif
-
-
-
-namespace ngsimd
-{
-  // MSVC does not define SSE. It's always present on 64bit cpus
-#if (defined(_M_AMD64) || defined(_M_X64) || defined(__AVX__))
-#ifndef __SSE__
-#define __SSE__
-#endif
-#ifndef __SSE2__
-#define __SSE2__
-#endif
-#endif
-
-  
-  constexpr int GetDefaultSIMDSize() {
-#if defined __AVX512F__
-    return 8;
-#elif defined __AVX__
-    return 4;
-#elif defined __SSE__
-    return 2;
-#else
-    return 1;
-#endif
-  }
-
-#if defined __AVX512F__
-    typedef __m512 tAVX;
-    typedef __m512d tAVXd;
-#elif defined __AVX__
-    typedef __m256 tAVX;
-    typedef __m256d tAVXd;
-#elif defined __SSE__
-    typedef __m128 tAVX;
-    typedef __m128d tAVXd;
-#endif
-
-
-  
-  template <typename T, int N=GetDefaultSIMDSize()> class SIMD;
-
-  template <typename T>
-  struct has_call_operator
-  {
-      template <typename C> static std::true_type check( decltype( sizeof(&C::operator() )) ) { return std::true_type(); }
-      template <typename> static std::false_type check(...) { return std::false_type(); }
-      typedef decltype( check<T>(sizeof(char)) ) type;
-      static constexpr type value = type();
-  };
-
-  template <typename T1, typename T2, typename T3>
-  // a*b+c
-  NG_INLINE auto FMA(T1 a, T2 b, T3 c)
-  {
-    return a*b+c;
-  }
-
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator+ (T a, SIMD<double,N> b) { return SIMD<double,N>(a) + b; }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator- (T a, SIMD<double,N> b) { return SIMD<double,N>(a) - b; }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator* (T a, SIMD<double,N> b) { return SIMD<double,N>(a) * b; }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator/ (T a, SIMD<double,N> b) { return SIMD<double,N>(a) / b; }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator+ (SIMD<double,N> a, T b) { return a + SIMD<double,N>(b); }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator- (SIMD<double,N> a, T b) { return a - SIMD<double,N>(b); }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator* (SIMD<double,N> a, T b) { return a * SIMD<double,N>(b); }
-  template<int N, typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    NG_INLINE SIMD<double,N> operator/ (SIMD<double,N> a, T b) { return a / SIMD<double,N>(b); }
-
-
-using std::sqrt;
-using std::fabs;
-
-  class ExceptionNOSIMD : public std::runtime_error
-  {
-  public:
-    using std::runtime_error::runtime_error;
-    std::string What() { return what(); }
-  };
-
-  using std::exp;
-  template<int N> NG_INLINE SIMD<double,N> exp (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return exp(a[i]); } );
-  }
-
-  using std::log;
-  template<int N> NG_INLINE SIMD<double,N> log (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return log(a[i]); } );
-  }
-
-  using std::pow;
-  template<int N> NG_INLINE SIMD<double,N> pow (SIMD<double,N> a, double x)
-  {
-    return SIMD<double,N>([&](int i)->double { return pow(a[i],x); } );
-  }
-
-  using std::sin;
-  template<int N> NG_INLINE SIMD<double,N> sin (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return sin(a[i]); } );
-  }
-
-  using std::cos;
-  template<int N> NG_INLINE SIMD<double,N> cos (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return cos(a[i]); } );
-  }
-
-  using std::tan;
-  template<int N> NG_INLINE SIMD<double,N> tan (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return tan(a[i]); } );
-  }
-
-  using std::atan;
-  template<int N> NG_INLINE SIMD<double,N> atan (SIMD<double,N> a)
-  {
-    return SIMD<double,N>([&](int i)->double { return atan(a[i]); } );
-  }
-
-
-/////////////////////////////////////////////////////////////////////////////
-// SIMD width 1 (in case no AVX support is available)
-/////////////////////////////////////////////////////////////////////////////
-  template<>
-  class SIMD<double,1>
-  {
-    double data;
-
-  public:
-    static constexpr int Size() { return 1; }
-    SIMD () = default;
-    SIMD (const SIMD &) = default;
-    SIMD & operator= (const SIMD &) = default;
-
-    // only called if T has a call operator of appropriate type
-    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
-    SIMD (const T & func)
-    {
-      data = func(0);
-    }
-
-    // only called if T is arithmetic (integral or floating point types)
-    template<typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    SIMD (const T & val)
-    {
-      data = val;
-    }
-
-    SIMD (double const * p)
-    {
-      data = *p;
-    }
-
-    NG_INLINE operator double() const { return data; }
-    NG_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
-    NG_INLINE double Data() const { return data; }
-    NG_INLINE double & Data() { return data; }
-
-    NG_INLINE SIMD<double,1> &operator+= (SIMD<double,1> b) { data+=b.Data(); return *this; }
-    NG_INLINE SIMD<double,1> &operator-= (SIMD<double,1> b) { data-=b.Data(); return *this; }
-    NG_INLINE SIMD<double,1> &operator*= (SIMD<double,1> b) { data*=b.Data(); return *this; }
-    NG_INLINE SIMD<double,1> &operator/= (SIMD<double,1> b) { data/=b.Data(); return *this; }
-
-  };
-
-  NG_INLINE SIMD<double,1> operator+ (SIMD<double,1> a, SIMD<double,1> b) { return a.Data()+b.Data(); }
-  NG_INLINE SIMD<double,1> operator- (SIMD<double,1> a, SIMD<double,1> b) { return a.Data()-b.Data(); }
-  NG_INLINE SIMD<double,1> operator- (SIMD<double,1> a) { return -a.Data(); }
-  NG_INLINE SIMD<double,1> operator* (SIMD<double,1> a, SIMD<double,1> b) { return a.Data()*b.Data(); }
-  NG_INLINE SIMD<double,1> operator/ (SIMD<double,1> a, SIMD<double,1> b) { return a.Data()/b.Data(); }
-
-  NG_INLINE SIMD<double,1> sqrt (SIMD<double,1> a) { return std::sqrt(a.Data()); }
-  NG_INLINE SIMD<double,1> floor (SIMD<double,1> a) { return std::floor(a.Data()); }
-  NG_INLINE SIMD<double,1> ceil (SIMD<double,1> a) { return std::ceil(a.Data()); }
-  NG_INLINE SIMD<double,1> fabs (SIMD<double,1> a) { return std::fabs(a.Data()); }
-  NG_INLINE SIMD<double,1> L2Norm2 (SIMD<double,1> a) { return a.Data()*a.Data(); }
-  NG_INLINE SIMD<double,1> Trans (SIMD<double,1> a) { return a; }
-  NG_INLINE SIMD<double,1> IfPos (SIMD<double,1> a, SIMD<double,1> b, SIMD<double,1> c)
-  {
-    return (a.Data() > 0) ? b : c;
-  }
-
-  NG_INLINE double HSum (SIMD<double,1> sd)
-  {
-    return sd.Data();
-  }
-
-  NG_INLINE auto HSum (SIMD<double,1> sd1, SIMD<double,1> sd2)
-  {
-    return std::make_tuple(sd1.Data(), sd2.Data());
-  }
-
-  NG_INLINE auto HSum (SIMD<double,1> sd1, SIMD<double,1> sd2, SIMD<double,1> sd3, SIMD<double,1> sd4)
-  {
-    return std::make_tuple(sd1.Data(), sd2.Data(), sd3.Data(), sd4.Data());
-  }
-
-
-/////////////////////////////////////////////////////////////////////////////
-// SSE - Simd width 2
-/////////////////////////////////////////////////////////////////////////////
-#ifdef __SSE__
-  template<>
-  class alignas(16) SIMD<double,2>
-  {
-    __m128d data;
-
-  public:
-    static constexpr int Size() { return 2; }
-    SIMD () = default;
-    SIMD (const SIMD &) = default;
-    SIMD & operator= (const SIMD &) = default;
-
-    SIMD (__m128d adata)
-      : data(adata)
-      { ; }
-
-    // only called if T has a call operator of appropriate type
-    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
-    SIMD (const T & func)
-    {
-      data = _mm_set_pd(func(1), func(0));
-    }
-
-    // only called if T is arithmetic (integral or floating point types)
-    template<typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    SIMD (const T & val)
-    {
-      data = _mm_set1_pd(val);
-    }
-
-    SIMD (double const * p)
-    {
-      data = _mm_loadu_pd(p);
-    }
-
-    NG_INLINE operator __m128d() const { return data; }
-    NG_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
-    NG_INLINE double& operator[] (int i) { return ((double*)(&data))[i]; }
-    NG_INLINE __m128d Data() const { return data; }
-    NG_INLINE __m128d & Data() { return data; }
-
-    // NG_INLINE operator std::tuple<double&,double&,double&,double&> ()
-    // { return std::tuple<double&,double&,double&,double&>((*this)[0], (*this)[1], (*this)[2], (*this)[3]); }
-
-
-    NG_INLINE SIMD<double,2> &operator+= (SIMD<double,2> b) { data+=b.Data(); return *this; }
-    NG_INLINE SIMD<double,2> &operator-= (SIMD<double,2> b) { data-=b.Data(); return *this; }
-    NG_INLINE SIMD<double,2> &operator*= (SIMD<double,2> b) { data*=b.Data(); return *this; }
-    NG_INLINE SIMD<double,2> &operator/= (SIMD<double,2> b) { data/=b.Data(); return *this; }
-
-  };
-
-  NG_INLINE SIMD<double,2> operator+ (SIMD<double,2> a, SIMD<double,2> b) { return a.Data()+b.Data(); }
-  NG_INLINE SIMD<double,2> operator- (SIMD<double,2> a, SIMD<double,2> b) { return a.Data()-b.Data(); }
-  NG_INLINE SIMD<double,2> operator- (SIMD<double,2> a) { return -a.Data(); }
-  NG_INLINE SIMD<double,2> operator* (SIMD<double,2> a, SIMD<double,2> b) { return a.Data()*b.Data(); }
-  NG_INLINE SIMD<double,2> operator/ (SIMD<double,2> a, SIMD<double,2> b) { return a.Data()/b.Data(); }
-
-  /*
-  NG_INLINE SIMD<double,4> sqrt (SIMD<double,4> a) { return _mm256_sqrt_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> floor (SIMD<double,4> a) { return _mm256_floor_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> ceil (SIMD<double,4> a) { return _mm256_ceil_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> fabs (SIMD<double,4> a) { return _mm256_max_pd(a.Data(), -a.Data()); }
-  NG_INLINE SIMD<double,4> L2Norm2 (SIMD<double,4> a) { return a.Data()*a.Data(); }
-  NG_INLINE SIMD<double,4> Trans (SIMD<double,4> a) { return a; }
-  NG_INLINE SIMD<double,4> IfPos (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
-  {
-    auto cp = _mm256_cmp_pd (a.Data(), _mm256_setzero_pd(), _CMP_GT_OS);
-    return _mm256_blendv_pd(c.Data(), b.Data(), cp);
-  }
-
-  NG_INLINE double HSum (SIMD<double,4> sd)
-  {
-    __m128d hv = _mm_add_pd (_mm256_extractf128_pd(sd.Data(),0), _mm256_extractf128_pd(sd.Data(),1));
-    return _mm_cvtsd_f64 (_mm_hadd_pd (hv, hv));
-  }
-
-  NG_INLINE auto HSum (SIMD<double,4> sd1, SIMD<double,4> sd2)
-  {
-    __m256d hv = _mm256_hadd_pd(sd1.Data(), sd2.Data());
-    __m128d hv2 = _mm_add_pd (_mm256_extractf128_pd(hv,0), _mm256_extractf128_pd(hv,1));
-    return std::make_tuple(_mm_cvtsd_f64 (hv2),  _mm_cvtsd_f64(_mm_shuffle_pd (hv2, hv2, 3)));
-  }
-
-  NG_INLINE auto HSum (SIMD<double,4> v1, SIMD<double,4> v2, SIMD<double,4> v3, SIMD<double,4> v4)
-  {
-    __m256d hsum1 = _mm256_hadd_pd (v1.Data(), v2.Data());
-    __m256d hsum2 = _mm256_hadd_pd (v3.Data(), v4.Data());
-    __m256d hsum = _mm256_add_pd (_mm256_permute2f128_pd (hsum1, hsum2, 1+2*16),
-                                  _mm256_blend_pd (hsum1, hsum2, 12));
-    return SIMD<double,4>(hsum);
-  }
-  */
-#endif // __SSE__
-
-  
-
-/////////////////////////////////////////////////////////////////////////////
-// AVX - Simd width 4
-/////////////////////////////////////////////////////////////////////////////
-#ifdef __AVX__
-  template<>
-  class alignas(32) SIMD<double,4>
-  {
-    __m256d data;
-
-  public:
-    static constexpr int Size() { return 4; }
-    SIMD () = default;
-    SIMD (const SIMD &) = default;
-    SIMD & operator= (const SIMD &) = default;
-
-    SIMD (__m256d adata)
-      : data(adata)
-      { ; }
-
-    // only called if T has a call operator of appropriate type
-    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
-    SIMD (const T & func)
-    {
-      data = _mm256_set_pd(func(3), func(2), func(1), func(0));
-    }
-
-    // only called if T is arithmetic (integral or floating point types)
-    template<typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    SIMD (const T & val)
-    {
-      data = _mm256_set1_pd(val);
-    }
-
-    SIMD (double const * p)
-    {
-      data = _mm256_loadu_pd(p);
-    }
-
-    NG_INLINE operator __m256d() const { return data; }
-    NG_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
-    NG_INLINE double& operator[] (int i) { return ((double*)(&data))[i]; }
-    NG_INLINE __m256d Data() const { return data; }
-    NG_INLINE __m256d & Data() { return data; }
-
-    NG_INLINE operator std::tuple<double&,double&,double&,double&> ()
-    { return std::tuple<double&,double&,double&,double&>((*this)[0], (*this)[1], (*this)[2], (*this)[3]); }
-
-
-    NG_INLINE SIMD<double,4> &operator+= (SIMD<double,4> b) { data+=b.Data(); return *this; }
-    NG_INLINE SIMD<double,4> &operator-= (SIMD<double,4> b) { data-=b.Data(); return *this; }
-    NG_INLINE SIMD<double,4> &operator*= (SIMD<double,4> b) { data*=b.Data(); return *this; }
-    NG_INLINE SIMD<double,4> &operator/= (SIMD<double,4> b) { data/=b.Data(); return *this; }
-
-  };
-
-  NG_INLINE SIMD<double,4> operator+ (SIMD<double,4> a, SIMD<double,4> b) { return a.Data()+b.Data(); }
-  NG_INLINE SIMD<double,4> operator- (SIMD<double,4> a, SIMD<double,4> b) { return a.Data()-b.Data(); }
-  NG_INLINE SIMD<double,4> operator- (SIMD<double,4> a) { return -a.Data(); }
-  NG_INLINE SIMD<double,4> operator* (SIMD<double,4> a, SIMD<double,4> b) { return a.Data()*b.Data(); }
-  NG_INLINE SIMD<double,4> operator/ (SIMD<double,4> a, SIMD<double,4> b) { return a.Data()/b.Data(); }
-
-  NG_INLINE SIMD<double,4> sqrt (SIMD<double,4> a) { return _mm256_sqrt_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> floor (SIMD<double,4> a) { return _mm256_floor_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> ceil (SIMD<double,4> a) { return _mm256_ceil_pd(a.Data()); }
-  NG_INLINE SIMD<double,4> fabs (SIMD<double,4> a) { return _mm256_max_pd(a.Data(), -a.Data()); }
-  NG_INLINE SIMD<double,4> L2Norm2 (SIMD<double,4> a) { return a.Data()*a.Data(); }
-  NG_INLINE SIMD<double,4> Trans (SIMD<double,4> a) { return a; }
-  NG_INLINE SIMD<double,4> IfPos (SIMD<double,4> a, SIMD<double,4> b, SIMD<double,4> c)
-  {
-    auto cp = _mm256_cmp_pd (a.Data(), _mm256_setzero_pd(), _CMP_GT_OS);
-    return _mm256_blendv_pd(c.Data(), b.Data(), cp);
-  }
-
-  NG_INLINE double HSum (SIMD<double,4> sd)
-  {
-    __m128d hv = _mm_add_pd (_mm256_extractf128_pd(sd.Data(),0), _mm256_extractf128_pd(sd.Data(),1));
-    return _mm_cvtsd_f64 (_mm_hadd_pd (hv, hv));
-  }
-
-  NG_INLINE auto HSum (SIMD<double,4> sd1, SIMD<double,4> sd2)
-  {
-    __m256d hv = _mm256_hadd_pd(sd1.Data(), sd2.Data());
-    __m128d hv2 = _mm_add_pd (_mm256_extractf128_pd(hv,0), _mm256_extractf128_pd(hv,1));
-    return std::make_tuple(_mm_cvtsd_f64 (hv2),  _mm_cvtsd_f64(_mm_shuffle_pd (hv2, hv2, 3)));
-  }
-
-  NG_INLINE auto HSum (SIMD<double,4> v1, SIMD<double,4> v2, SIMD<double,4> v3, SIMD<double,4> v4)
-  {
-    __m256d hsum1 = _mm256_hadd_pd (v1.Data(), v2.Data());
-    __m256d hsum2 = _mm256_hadd_pd (v3.Data(), v4.Data());
-    __m256d hsum = _mm256_add_pd (_mm256_permute2f128_pd (hsum1, hsum2, 1+2*16),
-                                  _mm256_blend_pd (hsum1, hsum2, 12));
-    return SIMD<double,4>(hsum);
-  }
-
-#endif // __AVX__
-
-/////////////////////////////////////////////////////////////////////////////
-// AVX512 - Simd width 8
-/////////////////////////////////////////////////////////////////////////////
-#ifdef __AVX512F__
-  template<>
-  class alignas(64) SIMD<double,8>
-  {
-    __m512d data;
-
-  public:
-    static constexpr int Size() { return 8; }
-    SIMD () = default;
-    SIMD (const SIMD &) = default;
-    SIMD & operator= (const SIMD &) = default;
-
-    SIMD (__m512d adata)
-      : data(adata)
-      { ; }
-
-    // only called if T has a call operator of appropriate type
-    template<typename T, typename std::enable_if<std::is_convertible<T, std::function<double(int)>>::value, int>::type = 0>
-    SIMD (const T & func)
-    {
-      data = _mm512_set_pd(func(7), func(6), func(5), func(4),
-                           func(3), func(2), func(1), func(0));
-    }
-
-    // only called if T is arithmetic (integral or floating point types)
-    template<typename T, typename std::enable_if<std::is_arithmetic<T>::value, int>::type = 0>
-    SIMD (const T & val)
-    {
-      data = _mm512_set1_pd(val);
-    }
-
-    SIMD (double const * p)
-    {
-      data = _mm512_loadu_pd(p);
-    }
-
-    NG_INLINE operator __m512d() const { return data; }
-    NG_INLINE double operator[] (int i) const { return ((double*)(&data))[i]; }
-    NG_INLINE __m512d Data() const { return data; }
-    NG_INLINE __m512d & Data() { return data; }
-
-    NG_INLINE SIMD<double,8> &operator+= (SIMD<double,8> b) { data+=b.Data(); return *this; }
-    NG_INLINE SIMD<double,8> &operator-= (SIMD<double,8> b) { data-=b.Data(); return *this; }
-    NG_INLINE SIMD<double,8> &operator*= (SIMD<double,8> b) { data*=b.Data(); return *this; }
-    NG_INLINE SIMD<double,8> &operator/= (SIMD<double,8> b) { data/=b.Data(); return *this; }
-
-  };
-
-  NG_INLINE SIMD<double,8> operator- (SIMD<double,8> a) { return _mm512_sub_pd(_mm512_setzero_pd(), a.Data()); }
-
-  NG_INLINE SIMD<double,8> operator+ (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_add_pd(a.Data(),b.Data()); }
-  NG_INLINE SIMD<double,8> operator- (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_sub_pd(a.Data(),b.Data()); }
-  NG_INLINE SIMD<double,8> operator* (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_mul_pd(a.Data(),b.Data()); }
-  NG_INLINE SIMD<double,8> operator/ (SIMD<double,8> a, SIMD<double,8> b) { return _mm512_div_pd(a.Data(),b.Data()); }
-
-  NG_INLINE SIMD<double,8> sqrt (SIMD<double,8> a) { return _mm512_sqrt_pd(a.Data()); }
-  NG_INLINE SIMD<double,8> floor (SIMD<double,8> a) { return _mm512_floor_pd(a.Data()); }
-  NG_INLINE SIMD<double,8> ceil (SIMD<double,8> a) { return _mm512_ceil_pd(a.Data()); }
-  NG_INLINE SIMD<double,8> fabs (SIMD<double,8> a) { return _mm512_max_pd(a.Data(), -a.Data()); }
-  NG_INLINE SIMD<double,8> L2Norm2 (SIMD<double,8> a) { return a.Data()*a.Data(); }
-  NG_INLINE SIMD<double,8> Trans (SIMD<double,8> a) { return a; }
-  NG_INLINE SIMD<double,8> IfPos (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
-  {
-    auto cp = _mm512_cmp_pd_mask (a.Data(), _mm512_setzero_pd(), _MM_CMPINT_GT);
-    return _mm512_mask_blend_pd(cp, c.Data(), b.Data());
-  }
-
-
-  template<> NG_INLINE auto FMA (SIMD<double,8> a, SIMD<double,8> b, SIMD<double,8> c)
-  {
-    return _mm512_fmadd_pd (a.Data(), b.Data(), c.Data());
-  }
-
-  NG_INLINE double HSum (SIMD<double,8> sd)
-  {
-    SIMD<double,4> low = _mm512_extractf64x4_pd(sd.Data(),0);
-    SIMD<double,4> high = _mm512_extractf64x4_pd(sd.Data(),1);
-    return HSum(low)+HSum(high);
-  }
-
-  NG_INLINE auto HSum (SIMD<double,8> sd1, SIMD<double,8> sd2)
-  {
-    return std::make_tuple(HSum(sd1), HSum(sd2));
-  }
-
-  NG_INLINE SIMD<double,4> HSum (SIMD<double,8> v1, SIMD<double,8> v2, SIMD<double,8> v3, SIMD<double,8> v4)
-  {
-    SIMD<double,4> high1 = _mm512_extractf64x4_pd(v1.Data(),1);
-    SIMD<double,4> high2 = _mm512_extractf64x4_pd(v2.Data(),1);
-    SIMD<double,4> high3 = _mm512_extractf64x4_pd(v3.Data(),1);
-    SIMD<double,4> high4 = _mm512_extractf64x4_pd(v4.Data(),1);
-    SIMD<double,4> low1 = _mm512_extractf64x4_pd(v1.Data(),0);
-    SIMD<double,4> low2 = _mm512_extractf64x4_pd(v2.Data(),0);
-    SIMD<double,4> low3 = _mm512_extractf64x4_pd(v3.Data(),0);
-    SIMD<double,4> low4 = _mm512_extractf64x4_pd(v4.Data(),0);
-    return HSum(low1,low2,low3,low4) + HSum(high1,high2,high3,high4);
-  }
-#endif // __AVX512F__
-
-
-////////////////////////////////////////////////////////////////////////////////
-// MultiSIMD - Multiple SIMD values in one struct using head-tail implementation
-////////////////////////////////////////////////////////////////////////////////
-  template <int D, typename T>
-  class MultiSIMD
-  {
-    SIMD<T> head;
-    MultiSIMD<D-1,T> tail;
-  public:
-    MultiSIMD () = default;
-    MultiSIMD (const MultiSIMD & ) = default;
-    MultiSIMD (T v) : head(v), tail(v) { ; }
-    MultiSIMD (SIMD<T> _head, MultiSIMD<D-1,T> _tail)
-      : head(_head), tail(_tail) { ; }
-    SIMD<T> Head() const { return head; }
-    MultiSIMD<D-1,T> Tail() const { return tail; }
-    SIMD<T> & Head() { return head; }
-    MultiSIMD<D-1,T> & Tail() { return tail; }
-
-    template <int NR>
-    SIMD<T> Get() const { return NR==0 ? head : tail.template Get<NR-1>(); }
-    template <int NR>
-    SIMD<T> & Get() { return NR==0 ? head : tail.template Get<NR-1>(); }
-    auto MakeTuple() { return std::tuple_cat(std::tuple<SIMD<T>&> (head), tail.MakeTuple()); }
-    // not yet possible for MSVC
-    // operator auto () { return MakeTuple(); }
-  };
-
-  template <typename T>
-  class MultiSIMD<2,T>
-  {
-    SIMD<T> v0, v1;
-  public:
-    MultiSIMD () = default;
-    MultiSIMD (const MultiSIMD & ) = default;
-    MultiSIMD (T v) : v0(v), v1(v) { ; }
-    MultiSIMD (SIMD<T> _v0, SIMD<T> _v1) : v0(_v0), v1(_v1) { ; }
-
-    SIMD<T> Head() const { return v0; }
-    SIMD<T> Tail() const { return v1; }
-    SIMD<T> & Head() { return v0; }
-    SIMD<T> & Tail() { return v1; }
-
-    template <int NR>
-    SIMD<T> Get() const { return NR==0 ? v0 : v1; }
-    template <int NR>
-    SIMD<T> & Get() { return NR==0 ? v0 : v1; }
-    auto MakeTuple() { return std::tuple<SIMD<T>&, SIMD<T>&> (v0, v1); }
-    operator std::tuple<SIMD<T>&, SIMD<T>&>() { return MakeTuple(); }
-  };
-
-  template <int D> NG_INLINE MultiSIMD<D,double> operator+ (MultiSIMD<D,double> a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a.Head()+b.Head(), a.Tail()+b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator+ (double a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a+b.Head(), a+b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator+ (MultiSIMD<D,double> b, double a)
-  { return MultiSIMD<D,double> (a+b.Head(), a+b.Tail()); }
-
-  template <int D> NG_INLINE MultiSIMD<D,double> operator- (MultiSIMD<D,double> a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a.Head()-b.Head(), a.Tail()-b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator- (double a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a-b.Head(), a-b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator- (MultiSIMD<D,double> b, double a)
-  { return MultiSIMD<D,double> (b.Head()-a, b.Tail()-a); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator- (MultiSIMD<D,double> a)
-  { return MultiSIMD<D,double> (-a.Head(), -a.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator* (MultiSIMD<D,double> a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a.Head()*b.Head(), a.Tail()*b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator/ (MultiSIMD<D,double> a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> (a.Head()/b.Head(), a.Tail()/b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator* (double a, MultiSIMD<D,double> b)
-  { return MultiSIMD<D,double> ( a*b.Head(), a*b.Tail()); }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator* (MultiSIMD<D,double> b, double a)
-  { return MultiSIMD<D,double> ( a*b.Head(), a*b.Tail()); }
-
-  template <int D> NG_INLINE MultiSIMD<D,double> & operator+= (MultiSIMD<D,double> & a, MultiSIMD<D,double> b)
-  { a.Head()+=b.Head(); a.Tail()+=b.Tail(); return a; }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator-= (MultiSIMD<D,double> & a, double b)
-  { a.Head()-=b; a.Tail()-=b; return a; }
-  template <int D> NG_INLINE MultiSIMD<D,double> operator-= (MultiSIMD<D,double> & a, MultiSIMD<D,double> b)
-  { a.Head()-=b.Head(); a.Tail()-=b.Tail(); return a; }
-  template <int D> NG_INLINE MultiSIMD<D,double> & operator*= (MultiSIMD<D,double> & a, MultiSIMD<D,double> b)
-  { a.Head()*=b.Head(); a.Tail()*=b.Tail(); return a; }
-  template <int D> NG_INLINE MultiSIMD<D,double> & operator*= (MultiSIMD<D,double> & a, double b)
-  { a.Head()*=b; a.Tail()*=b; return a; }
-  // NG_INLINE MultiSIMD<double> operator/= (MultiSIMD<double> & a, MultiSIMD<double> b) { return a.Data()/=b.Data(); }
-
-  NG_INLINE SIMD<double> HVSum (SIMD<double> a) { return a; }
-  template <int D>
-  NG_INLINE SIMD<double> HVSum (MultiSIMD<D,double> a) { return a.Head() + HVSum(a.Tail()); }
-
-  template <int D> NG_INLINE double HSum (MultiSIMD<D,double> a) { return HSum(HVSum(a)); }
-  template <int D> NG_INLINE auto HSum (MultiSIMD<D,double> a, MultiSIMD<D,double> b)
-  { return HSum(HVSum(a), HVSum(b)); }
-
-  template <int D, typename T>
-  std::ostream & operator<< (std::ostream & ost, MultiSIMD<D,T> multi)
-  {
-    ost << multi.Head() << " " << multi.Tail();
-    return ost;
-  }
-
-  template <typename T>
-  std::ostream & operator<< (std::ostream & ost, SIMD<T> simd)
-  {
-    ost << simd[0];
-    for (int i = 1; i < simd.Size(); i++)
-      ost << " " << simd[i];
-    return ost;
-  }
-}
-
-namespace netgen
-{
-  using namespace ngsimd;
-}
-#endif

libsrc/general/optmem.hpp

@@ -26,39 +26,13 @@ private:
   mutex block_allocator_mutex;
 public:
   ///
-  BlockAllocator (unsigned asize, unsigned ablocks = 100);
+  DLL_HEADER BlockAllocator (unsigned asize, unsigned ablocks = 100);
   ///
-  ~BlockAllocator ();
+  DLL_HEADER ~BlockAllocator ();
   ///
-
+  DLL_HEADER void * Alloc ();
-  void * Alloc ();
-  /*
-  {
-    if (!freelist)
-      Alloc2();
-
-    void * p = freelist;
-    // freelist = *(void**)freelist;
-    freelist = *static_cast<void**> (freelist);
-
-    return p;
-  }
-  */
-
-
   ///
-  void Free (void * p);
+  DLL_HEADER void Free (void * p);
-  /*
-  {
-    if (!bablocks.Size()) return;
-    *(void**)p = freelist;
-    freelist = p;
-  }
-  */
-
-
-private:
-  //  void Alloc2 ();
 };
 
 }

libsrc/general/parthreads.hpp

@@ -96,8 +96,8 @@ void ParallelFor( int first, int next, const TFunc & f )
 
 
   
-  typedef void (*TaskManager)(std::function<void(int,int)>);
+  typedef void (*NgTaskManager)(std::function<void(int,int)>);
-  typedef void (*Tracer)(string, bool);  // false .. start, true .. stop
+  typedef void (*NgTracer)(string, bool);  // false .. start, true .. stop
 
   inline void DummyTaskManager (std::function<void(int,int)> func)
   {
@@ -108,7 +108,7 @@ void ParallelFor( int first, int next, const TFunc & f )
   inline void DummyTracer (string, bool) { ; }
   
   template <typename FUNC>
-  inline void ParallelFor (TaskManager tm, size_t n, FUNC func)
+  inline void ParallelFor (NgTaskManager tm, size_t n, FUNC func)
   {
     (*tm) ([n,func] (size_t nr, size_t nums)
            {
@@ -121,7 +121,7 @@ void ParallelFor( int first, int next, const TFunc & f )
   }
   
   template <typename FUNC>
-  inline void ParallelForRange (TaskManager tm, size_t n, FUNC func)
+  inline void ParallelForRange (NgTaskManager tm, size_t n, FUNC func)
   {
     (*tm) ([n,func] (size_t nr, size_t nums)
            {

libsrc/general/table.hpp

@@ -114,11 +114,22 @@ public:
   /// Creates table of size size
   inline TABLE (int size) : BASE_TABLE (size) { ; }
 
+  TABLE (TABLE && tab2)
+    : BASE_TABLE(move(tab2))
+  { }
+  
   /// Creates fixed maximal element size table
   inline TABLE (const NgFlatArray<int,BASE> & entrysizes)
     : BASE_TABLE (NgFlatArray<int> (entrysizes.Size(), const_cast<int*>(&entrysizes[BASE])), 
 		  sizeof(T))
   { ; }
+
+  TABLE & operator= (TABLE && tab2)
+  {
+    BASE_TABLE::operator=(move(tab2));
+    return *this;
+  }
+
   
   /// Changes Size of table to size, deletes data
   inline void SetSize (int size)

libsrc/general/template.hpp

@@ -14,6 +14,8 @@ namespace netgen
    templates, global types, defines and variables
 */
 
+DLL_HEADER extern const string netgen_version;
+
 ///	The following value may be adapted to the hardware !
 #ifndef CLOCKS_PER_SEC
 #define CLOCKS_PER_SEC 1000000

libsrc/geom2d/CMakeLists.txt

@@ -1,10 +1,10 @@
-add_definitions(-DNGLIB_EXPORTS)
+add_definitions(-DNGINTERFACE_EXPORTS)
-add_library(geom2d ${NG_LIB_TYPE} genmesh2d.cpp geometry2d.cpp python_geom2d.cpp )
+add_library(geom2d ${NG_LIB_TYPE} csg2d.cpp genmesh2d.cpp geometry2d.cpp python_geom2d.cpp )
 if(APPLE)
     set_target_properties( geom2d PROPERTIES SUFFIX ".so")
 endif(APPLE)
 
-target_link_libraries(geom2d PUBLIC ngcore mesh PRIVATE netgen_python)
+target_link_libraries(geom2d PUBLIC ngcore mesh PRIVATE "$<BUILD_INTERFACE:netgen_python>")
 if(NOT WIN32)
   install( TARGETS geom2d ${NG_INSTALL_DIR})
 endif(NOT WIN32)
@@ -20,6 +20,6 @@ endif(USE_GUI)
 
 install(FILES
   geometry2d.hpp spline2d.hpp
-  vsgeom2d.hpp
+  vsgeom2d.hpp csg2d.hpp
   DESTINATION ${NG_INSTALL_DIR_INCLUDE}/geom2d COMPONENT netgen_devel
 )

libsrc/geom2d/csg2d.hpp

@@ -0,0 +1,741 @@
+#ifndef NETGEN_CSG2D_HPP_INCLUDED
+#define NETGEN_CSG2D_HPP_INCLUDED
+
+#include <variant>
+
+#include "geometry2d.hpp"
+
+namespace netgen
+{
+
+using namespace ngcore;
+using netgen::Point;
+using netgen::Vec;
+using Spline = SplineSeg3<2>;
+using netgen::Box;
+
+inline double Area(const Point<2>& P, const Point<2>& Q, const Point<2>& R)
+{
+  return (Q[0]-P[0]) * (R[1]-P[1]) - (Q[1]-P[1]) * (R[0]-P[0]);
+}
+
+// compute weight of spline such that p lies on it
+void ComputeWeight( Spline & s, Point<2> p );
+
+enum IntersectionType
+{                 // types of intersection (detected in the first phase)
+  NO_INTERSECTION = 0,
+  X_INTERSECTION,
+  T_INTERSECTION_Q,
+  T_INTERSECTION_P,
+  V_INTERSECTION,
+  X_OVERLAP,      // Q0 -- P1 -- Q1 -- P0   (different direction)
+  T_OVERLAP_Q,    // same direction or P inside Q
+  T_OVERLAP_P,    // same direction or Q inside P
+  V_OVERLAP       // one common point
+};
+
+enum IntersectionLabel
+{      // for the classification of intersection vertices in the second phase
+  NONE,
+  CROSSING,
+  BOUNCING,
+  LEFT_ON,
+  RIGHT_ON,
+  ON_ON,
+  ON_LEFT,
+  ON_RIGHT,
+  DELAYED_CROSSING,
+  DELAYED_BOUNCING
+};
+
+enum EntryExitLabel
+{         // for marking intersection vertices as "entry" or "exit"
+  EXIT,
+  ENTRY,
+  NEITHER
+};
+
+enum IteratorType
+{
+  SOURCE,
+  INTERSECTION,
+  CROSSING_INTERSECTION,
+  ALL
+};
+
+inline constexpr const double MAXH_DEFAULT{1e99};
+inline const string POINT_NAME_DEFAULT{""};
+inline const string BC_DEFAULT{""};
+inline const string MAT_DEFAULT{""};
+
+struct EdgeInfo
+{
+  optional<Point<2>> control_point = nullopt; // for spline segments
+  double maxh = MAXH_DEFAULT;
+  string bc = BC_DEFAULT;
+
+  EdgeInfo() = default;
+  EdgeInfo(Point<2> p) : control_point(p) {}
+  EdgeInfo(double h) : maxh(h) {}
+  EdgeInfo(string s) : bc(s) {}
+  EdgeInfo(optional<Point<2>> p, double h, string s)
+      : control_point(p), maxh(h), bc(s)
+    {}
+
+  void Assign( EdgeInfo other )
+    {
+      if(other.control_point != nullopt)
+          control_point = other.control_point;
+      if(other.bc != BC_DEFAULT)
+          bc = other.bc;
+      if(other.maxh != MAXH_DEFAULT)
+          maxh = min(maxh, other.maxh);
+    }
+};
+
+struct PointInfo
+{
+  double maxh = MAXH_DEFAULT;
+  string name = POINT_NAME_DEFAULT;
+  PointInfo() = default;
+  PointInfo(const PointInfo& other) = default;
+  PointInfo(double amaxh) : maxh(amaxh) {}
+  PointInfo(string aname) : name(aname) {}
+  PointInfo(double amaxh, string aname) : maxh(amaxh), name(aname) {}
+
+  void Assign(const PointInfo& other)
+  {
+    maxh = min(maxh, other.maxh);
+    if(other.name != POINT_NAME_DEFAULT)
+      name = other.name;
+  }
+};
+
+struct Vertex : Point<2>
+{
+  Vertex (Point<2> p) : Point<2>(p) {}
+  Vertex (const Vertex & v) : Point<2>(v)
+  {
+    spline = v.spline;
+    info = v.info;
+    pinfo = v.pinfo;
+    is_source = true;
+  }
+
+  Vertex * prev = nullptr;
+  Vertex * next = nullptr;
+  unique_ptr<Vertex> pnext = nullptr;
+  Vertex * neighbour = nullptr; // same vertex in other polygon (at intersections)
+  double lam = -1.0;
+  bool is_intersection = false;
+  bool is_source = false;
+
+  IntersectionLabel label = NONE;    // type of intersection vertex
+  EntryExitLabel enex = NEITHER;        // entry/exit "flag"
+
+  // In case the edge this - next is curved, store the spline information here
+  optional<Spline> spline = nullopt;
+  EdgeInfo info;
+  PointInfo pinfo;
+
+  DLL_HEADER Vertex * Insert(Point<2> p, double lam = -1.0);
+
+  void Link( Vertex * v )
+  {
+    neighbour = v;
+    v->neighbour = this;
+    is_intersection = true;
+    v->is_intersection = true;
+  }
+};
+
+struct VertexIterator
+{
+  struct iterator
+  {
+    iterator(Vertex* root, IteratorType IterType) :
+      root(root), V(NULL), iterType(IterType)
+    {
+      if (root == NULL)
+        return;
+
+      if (nextVertex() == NULL)         // no (source/intersection) vertex found
+        root = V = NULL;                // -> mark iterator as "end"
+    }
+
+    const iterator& operator++()
+    {
+      nextVertex();
+      return *this;
+    }
+
+    Vertex* operator*()
+    {
+      return V;
+    }
+
+    bool operator!=(const iterator& other) const
+    {
+      return (root != other.root) || (V != other.V);
+    }
+
+    private:
+    Vertex* root;
+    Vertex* V;
+    IteratorType iterType;
+
+    //
+    // find the next vertex
+    // if iterType is ALL, then it is just the next vertex
+    // if iterType is SOURCE, then it is the next source vertex
+    // if iterType is INTERSECTION, then it is the next intersection vertex
+    // if iterType is CROSSING_INTERSECTION, then it is the next intersection vertex with CROSSING label
+    //
+    Vertex* nextVertex()
+    {
+      bool nextFound = false;
+
+      if (V == NULL)
+      {                  // find first (source/intersection) vertex
+        V = root;
+        switch(iterType)
+        {
+          case ALL:
+            nextFound = true;
+            break;
+          case SOURCE:
+            if (V->is_source)
+              nextFound = true;
+            break;
+          case INTERSECTION:
+            if (V->is_intersection)
+              nextFound = true;
+            break;
+          case CROSSING_INTERSECTION:
+            if (V->is_intersection && (V->label == CROSSING))
+              nextFound = true;
+            break;
+        }
+      }
+
+      while (!nextFound)
+      {              // find next (source/intersection) vertex
+        switch(iterType)
+        {
+          case ALL:
+            V = V->next;
+            break;
+          case SOURCE:
+            do {
+              V = V->next;
+            } while (!V->is_source && V != root);
+            break;
+          case INTERSECTION:
+            do {
+              V = V->next;
+            } while (!V->is_intersection && V != root);
+            break;
+          case CROSSING_INTERSECTION:
+            do {
+              V = V->next;
+            } while ( ( !V->is_intersection || (V->label != CROSSING) ) && V != root);
+            break;
+        }
+
+        if (V == root)
+        {                // back at the root vertex?
+          root = V = NULL;              // -> mark iterator as "end"
+          return(V);
+        }
+
+        switch(iterType)
+        {
+          case ALL:
+            nextFound = true;
+            break;
+          case SOURCE:
+            if (V->is_source)
+              nextFound = true;
+            break;
+          case INTERSECTION:
+            if (V->is_intersection)
+              nextFound = true;
+            break;
+          case CROSSING_INTERSECTION:
+            if (V->is_intersection && (V->label == CROSSING))
+              nextFound = true;
+            break;
+        }
+      }
+      return(V);
+    }
+  };
+
+  public:
+  VertexIterator() : root(NULL) {};
+
+  iterator begin() { return iterator(root, iterType); }
+  iterator end()   { return iterator(NULL, iterType); }
+
+  Vertex* root;
+  IteratorType iterType;
+};
+
+
+struct Edge
+{
+  Vertex * v0 = nullptr;
+  Vertex * v1 = nullptr;
+
+  Edge (Vertex* v, Vertex* w) : v0(v), v1(w) { };
+};
+
+struct EdgeIterator
+{
+  struct iterator
+  {
+    iterator(Vertex* root, IteratorType IterType) :
+      root(root), one(NULL), two(NULL), iterType(IterType)
+    {
+      if (root == NULL)
+        return;
+
+      if (nextEdge() == NULL)           // no source edge found
+        root = one = two = NULL;        // -> mark iterator as "end"
+    }
+
+    const iterator& operator++() { nextEdge(); return *this; }
+
+    Edge operator*()
+    {
+      return Edge(one,two);
+    }
+
+    bool operator!=(const iterator& other) const
+    {
+      return (root != other.root) || (one != other.one) || (two != other.two);
+    }
+
+    private:
+    Vertex* root;
+    Vertex* one;
+    Vertex* two;
+    IteratorType iterType;
+
+    //
+    // find the next vertex, starting at curr
+    // if iterType is ALL, then it is just the next vertex
+    // if iterType is SOURCE, then it is the next source vertex
+    //
+    Vertex* nextVertex(Vertex* curr)
+    {
+      if (curr == NULL)
+        return(NULL);
+
+      switch(iterType)
+      {
+        case ALL:
+          curr = curr->next;
+          break;
+
+        case SOURCE:
+          do {
+            curr = curr->next;
+          } while (!curr->is_source);
+          break;
+        default:
+          ;
+      }
+
+      return(curr);
+    }
+
+    //
+    // find the next edge
+    //
+    Vertex* nextEdge()
+    {
+      if (root == NULL)                 // empty polygon?
+        return (NULL);
+
+      if (one == NULL)
+      {                // find one (source) vertex
+        one = root;                     // note: root is always a (source) vertex
+        two = nextVertex(one);
+        if (two == one)                 // just one (source) vertex
+          return(NULL);                 // -> no (source) edges
+        return(one);
+      }
+
+      if (two == root)
+      {                // back at the root vertex?
+        root = one = two = NULL;        // -> mark iterator as "end"
+        return(NULL);
+      }
+
+      one = two;
+      two = nextVertex(one);
+
+      return (one);
+    }
+  };
+
+  public:
+  EdgeIterator() : root(NULL) {};
+
+  iterator begin() { return iterator(root, iterType); }
+  iterator end()   { return iterator(NULL, iterType); }
+
+  Vertex* root;
+  IteratorType iterType;
+};
+
+
+inline int CalcSide( const Point<2> & p0, const Point<2> & p1, const Point<2> & r )
+{
+  if ( (p0[1] < r[1]) != (p1[1] < r[1]) )
+  {
+    if (p0[0] >= r[0])
+    {
+      if (p1[0] > r[0])
+        return 2 * (p1[1] > p0[1]) - 1;
+      else
+        if ( (Area(p0,p1,r) > 0) == (p1[1] > p0[1]) )
+          return 2 * (p1[1] > p0[1]) - 1;
+    }
+    else
+    {
+      if (p1[0] > r[0])
+        if ( (Area(p0,p1,r) > 0) == (p1[1] > p0[1]) )
+          return 2 * (p1[1] > p0[1]) - 1;
+    }
+  }
+  return 0;
+}
+
+struct Loop
+{
+  unique_ptr<Vertex> first = nullptr;
+  unique_ptr<Box<2>> bbox = nullptr;
+
+  Loop() = default;
+
+  Loop(const Loop & p)
+    : first(nullptr)
+  {
+    for(auto v : p.Vertices(ALL))
+      AppendVertex(*v);
+  }
+
+  Loop(Loop && p) = default;
+
+  Loop & operator=(Loop && p) = default;
+
+  Loop & operator=(const Loop & p)
+  {
+    // static Timer t("Loop::operator="); RegionTimer rt(t);
+    first = nullptr;
+    if(p.first)
+    {
+      size_t n = p.Size();
+      Array<unique_ptr<Vertex>> new_verts(n);
+      {
+        size_t i = 0;
+        for(const auto v : p.Vertices(ALL))
+          new_verts[i++] = make_unique<Vertex>(*v);
+      }
+
+      for(auto i : IntRange(n-1))
+      {
+        Vertex * v  = new_verts[i].get();
+        Vertex * vn = new_verts[i+1].get();
+        v->next = vn;
+        vn->prev = v;
+      }
+      Vertex * vfirst = new_verts[0].get();
+      Vertex * vlast = new_verts[n-1].get();
+      vfirst->prev = vlast;
+      vlast->next = vfirst;
+
+      for(auto i : IntRange(1,n))
+        new_verts[n-1-i]->pnext = std::move(new_verts[n-i]);
+
+      first = std::move(new_verts[0]);
+    }
+    bbox = nullptr;
+    return *this;
+  }
+
+  void Clear()
+  {
+    first = nullptr;
+  }
+
+  Vertex & AppendVertex(const Vertex & v)
+  {
+    auto & vnew = Append( static_cast<Point<2>>(v), true );
+    vnew.info = v.info;
+    vnew.pinfo = v.pinfo;
+    if(v.spline)
+      vnew.spline = *v.spline;
+    if(bbox)
+      bbox->Add(v);
+    return vnew;
+  }
+
+  Vertex & Append(Point<2> p, bool source = false)
+  {
+    Vertex * vnew;
+    if(first==nullptr)
+    {
+      first = make_unique<Vertex>(p);
+      first->next = first.get();
+      first->prev = first.get();
+      vnew = first.get();
+    }
+    else
+    {
+      vnew = first->prev->Insert(p);
+    }
+
+    vnew->is_source = source;
+    //     cout << "size after " << Size() << endl;
+    if(bbox)
+      bbox->Add(p);
+    return *vnew;
+  }
+
+  void Remove (Vertex* v)
+  {
+    v->prev->next = v->next;
+    v->next->prev = v->prev;
+    if(first.get() == v)
+      first = std::move(v->pnext);
+    else
+      v->prev->pnext = std::move(v->pnext);
+    bbox.reset();
+  }
+
+  bool IsInside( Point<2> r ) const;
+  bool IsLeftInside( const Vertex & p0 );
+  bool IsRightInside( const Vertex & p0 );
+
+  EdgeIterator Edges(IteratorType iterType) const
+  {
+    EdgeIterator it;
+    it.iterType = iterType;
+    it.root = first.get();
+    return it;
+  }
+
+  VertexIterator Vertices(IteratorType iterType, Vertex* first_ = nullptr) const
+  {
+    VertexIterator it;
+    it.iterType = iterType;
+    it.root = (first_ == nullptr) ? first.get() : first_;
+    return it;
+  }
+
+  //
+  // check, if all vertices have the ON_ON label
+  //
+  bool allOnOn()
+  {
+    for (Vertex* v : Vertices(ALL))
+      if (v->label != ON_ON)
+        return(false);
+    return(true);
+  }
+
+  //
+  // check, if the polygon does not contain any crossing intersection vertex
+  // or crossing intersection chain or (if we want to compute the union instead
+  // of the intersection) a bouncing vertex or a bouncing intersection chain
+  //
+  bool noCrossingVertex(bool union_case = false)
+  {
+    for (Vertex* v : Vertices(ALL))
+      if (v->is_intersection)
+      {
+        if ( (v->label == CROSSING) || (v->label == DELAYED_CROSSING) )
+          return(false);
+
+        if (union_case && ( (v->label == BOUNCING) || (v->label == DELAYED_BOUNCING) ) )
+          return(false);
+      }
+    return(true);
+  }
+
+  //
+  // return a non-intersection point
+  //
+  Point<2> getNonIntersectionPoint()
+  {
+    for (Vertex* v : Vertices(ALL))
+      if (!v->is_intersection)
+        return *v;
+
+    // no non-intersection vertex found -> find suitable edge midpoint
+    for (Vertex* v : Vertices(ALL))
+      // make sure that edge from V to V->next is not collinear with other polygon
+      if ( (v->next->neighbour != v->neighbour->prev) && (v->next->neighbour != v->neighbour->next) )
+        // return edge midpoint
+        return Center(*v, *v->next);
+    throw Exception("no point found");
+  }
+
+  //
+  // return and insert a non-intersection vertex
+  //
+  Vertex* getNonIntersectionVertex();
+
+  void SetBC(string bc)
+  {
+    for(auto v : Vertices(ALL))
+      v->info.bc = bc;
+  }
+
+  size_t Size() const
+  {
+    if(first==nullptr) return 0;
+
+    size_t cnt = 0;
+
+    for(auto v : Vertices(ALL))
+      cnt++;
+
+    return cnt;
+  }
+
+  const Box<2> & GetBoundingBox()
+  {
+    if(bbox==nullptr)
+    {
+      static Timer tall("Loop::GetBoundingBox"); RegionTimer rt(tall);
+      bbox = make_unique<Box<2>>(Box<2>::EMPTY_BOX);
+      for(auto v : Vertices(ALL))
+      {
+        bbox->Add(*v);
+        if(v->spline)
+          bbox->Add(v->spline->TangentPoint());
+      }
+    }
+    return *bbox;
+  }
+};
+
+
+struct Solid2d
+{
+  Array<Loop> polys;
+
+  string name = MAT_DEFAULT;
+
+  Solid2d() = default;
+  Solid2d(string name_) : name(name_) {}
+  DLL_HEADER Solid2d(const Array<std::variant<Point<2>, EdgeInfo, PointInfo>> & points, string name_=MAT_DEFAULT, string bc_=BC_DEFAULT);
+  Solid2d(Solid2d && other) = default;
+  Solid2d(const Solid2d & other) = default;
+
+  DLL_HEADER Solid2d operator+(const Solid2d & other) const;
+  DLL_HEADER Solid2d operator*(const Solid2d & other) const;
+  DLL_HEADER Solid2d operator-(const Solid2d & other) const;
+
+  Solid2d& operator=(Solid2d && other) = default;
+  Solid2d& operator=(const Solid2d & other) = default;
+  DLL_HEADER Solid2d& operator+=(const Solid2d & other);
+  DLL_HEADER Solid2d& operator*=(const Solid2d & other);
+  DLL_HEADER Solid2d& operator-=(const Solid2d & other);
+
+  void Append( const Loop & poly )
+  {
+    polys.Append(poly);
+  }
+
+  bool IsInside( Point<2> r ) const;
+  bool IsLeftInside( const Vertex & p0 );
+  bool IsRightInside( const Vertex & p0 );
+
+  template<typename TFunc>
+  Solid2d & Transform( const TFunc & func )
+    {
+      for(auto & poly : polys)
+          for(auto v : poly.Vertices(ALL))
+            {
+              auto p = func(*v);
+              (*v)[0] = p[0];
+              (*v)[1] = p[1];
+              if(v->spline)
+                {
+                  auto &s = *v->spline;
+                  auto pmid = func(s.GetPoint(0.5));
+                  s = Spline(func(s.StartPI()), func(s.TangentPoint()), func(s.EndPI()));
+                  ComputeWeight(s, pmid);
+                }
+            }
+      return *this;
+    }
+
+  Solid2d & Move( Vec<2> v );
+  Solid2d & Scale( double s );
+  Solid2d & Scale( Vec<2> s );
+  Solid2d & RotateRad( double ang, Point<2> center = {0,0} );
+  Solid2d & RotateDeg( double ang, Point<2> center = {0,0} )
+    {
+      return RotateRad( ang/180.*M_PI, center );
+    }
+
+  Solid2d & BC(string bc)
+  {
+    for(auto & p : polys)
+      for(auto v : p.Vertices(ALL))
+        v->info.bc = bc;
+    return *this;
+  }
+
+  Solid2d & Maxh(double maxh)
+  {
+    for(auto & p : polys)
+      for(auto v : p.Vertices(ALL))
+        v->info.maxh = maxh;
+    return *this;
+  }
+
+  Solid2d & Mat(string mat)
+  {
+    name = mat;
+    return *this;
+  }
+
+  Box<2> GetBoundingBox() const;
+};
+
+
+class CSG2d
+{
+  public:
+    Array<Solid2d> solids;
+
+    void Add ( Solid2d s )
+    {
+      solids.Append(s);
+    }
+
+    DLL_HEADER shared_ptr<netgen::SplineGeometry2d> GenerateSplineGeometry();
+    DLL_HEADER shared_ptr<netgen::Mesh> GenerateMesh(MeshingParameters & mp);
+};
+
+DLL_HEADER Solid2d Circle( Point<2> center, double r, string name="", string bc="");
+DLL_HEADER Solid2d Rectangle( Point<2> p0, Point<2> p1, string mat=MAT_DEFAULT, string bc=BC_DEFAULT );
+
+DLL_HEADER void AddIntersectionPoints ( Solid2d & s1, Solid2d & s2 );
+DLL_HEADER Solid2d ClipSolids ( const Solid2d & s1, const Solid2d & s2, char op);
+DLL_HEADER Solid2d ClipSolids ( const Solid2d & s1, Solid2d && s2, char op);
+DLL_HEADER Solid2d ClipSolids ( Solid2d && s1, const Solid2d & s2, char op);
+DLL_HEADER Solid2d ClipSolids ( Solid2d && s1, Solid2d && s2, char op);
+
+DLL_HEADER IntersectionType intersect(const Point<2> P1, const Point<2> P2, const Point<2> Q1, const Point<2> Q2, double& alpha, double& beta);
+
+}
+#endif // NETGEN_CSG2D_HPP_INCLUDED

libsrc/geom2d/genmesh2d.cpp

@@ -18,29 +18,44 @@ namespace netgen
   {
     double fperel, oldf, f;
 
-    int n = 10000;
+    int n = 1;
-    
+    NgArray<Point<2> > xi;
-    NgArray<Point<2> > xi(n);
+    NgArray<double> hi;
-    NgArray<double> hi(n);
+
-    
+    // do one extra step
-    for (int i = 0; i < n; i++)
+    int not_fine_enough = 2;
+
+    while(not_fine_enough && n < 10000)
+    {
+      not_fine_enough--;
+      n*=4;
+
+      xi.SetSize(n);
+      hi.SetSize(n);
+
+      for (int i = 0; i < n; i++)
       {
-	xi[i] = spline.GetPoint ( (i+0.5) / n );
+        xi[i] = spline.GetPoint ( (i+0.5) / n );
-	hi[i] = mesh.GetH (Point<3> (xi[i](0), xi[i](1), 0));
+        hi[i] = mesh.GetH (Point<3> (xi[i](0), xi[i](1), 0));
       }
 
-    // limit slope
+      // limit slope
-    double gradh = min(1/elto0,mp.grading);
+      double gradh = min(1/elto0,mp.grading);
-    for (int i = 0; i < n-1; i++)
+      for (int i = 0; i < n-1; i++)
       {
-	double hnext = hi[i] + gradh * (xi[i+1]-xi[i]).Length();
+        double hnext = hi[i] + gradh * (xi[i+1]-xi[i]).Length();
-	hi[i+1] = min(hi[i+1], hnext);
+        if(hnext > 2*hi[i])
+          not_fine_enough = 2;
+        hi[i+1] = min(hi[i+1], hnext);
       } 
-    for (int i = n-1; i > 1; i--)
+      for (int i = n-1; i > 1; i--)
       {
-	double hnext = hi[i] + gradh * (xi[i-1]-xi[i]).Length();
+        double hnext = hi[i] + gradh * (xi[i-1]-xi[i]).Length();
-	hi[i-1] = min(hi[i-1], hnext);
+        if(hnext > 2*hi[i])
+          not_fine_enough = 2;
+        hi[i-1] = min(hi[i-1], hnext);
       }
+    }
 
     points.SetSize (0);
 
@@ -233,13 +248,22 @@ namespace netgen
 	double hr = GetDomainMaxh (spline.rightdom);
 	if (hr > 0) hcurve = min2 (hcurve, hr);
 
-	int np = 1000;
+        // skip curvature restrictions for straight lines
-	for (double t = 0.5/np; t < 1; t += 1.0/np)
+        if(spline.MaxCurvature()==0)
-	  {
+        {
-	    Point<2> x = spline.GetPoint(t);
+          mesh2d.RestrictLocalHLine (Point<3>(p1(0),p1(1),0), 
-	    double hc = 1.0/mp.curvaturesafety / (1e-99+spline.CalcCurvature (t));
+                                     Point<3>(p2(0),p2(1),0), hcurve);
-	    mesh2d.RestrictLocalH (Point<3> (x(0), x(1), 0), min2(hc, hcurve));
+        }
-	  }
+        else
+        {
+          int np = 1000;
+          for (double t = 0.5/np; t < 1; t += 1.0/np)
+            {
+              Point<2> x = spline.GetPoint(t);
+              double hc = 1.0/mp.curvaturesafety / (1e-99+spline.CalcCurvature (t));
+              mesh2d.RestrictLocalH (Point<3> (x(0), x(1), 0), min2(hc, hcurve));
+            }
+        }
       }
     
     for (auto mspnt : mp.meshsize_points)
@@ -278,6 +302,11 @@ namespace netgen
 	      {
 		npi = mesh2d.AddPoint (newp, layer);
 		searchtree.Insert (newp, npi);
+                mesh2d.AddLockedPoint(npi);
+                Element0d el(npi, npi);
+                el.name = "";
+                mesh2d.SetCD2Name(npi, "");
+                mesh2d.pointelements.Append (el);
 	      }
           }
     }
@@ -392,10 +421,16 @@ namespace netgen
 			 shared_ptr<Mesh> & mesh, 
 			 MeshingParameters & mp)
   {
+    static Timer tall("MeshFromSpline2D"); RegionTimer rtall(tall);
+    static Timer t_h("SetH");
+    static Timer t_tensor("tensor domain meshing");
+    static Timer t_part_boundary("PartitionBoundary");
+    static Timer t_hpref("mark hpref points");
     PrintMessage (1, "Generate Mesh from spline geometry");
 
     Box<2> bbox = geometry.GetBoundingBox ();
 
+    t_h.Start();
     if (bbox.Diam() < mp.maxh) 
       mp.maxh = bbox.Diam();
 
@@ -412,11 +447,14 @@ namespace netgen
 
     
 
+    t_part_boundary.Start();
     geometry.PartitionBoundary (mp, mp.maxh, *mesh);
+    t_part_boundary.Stop();
     
     PrintMessage (3, "Boundary mesh done, np = ", mesh->GetNP());
 
 
+    t_hpref.Start();
     // marks mesh points for hp-refinement
     for (int i = 0; i < geometry.GetNP(); i++)
       if (geometry.GetPoint(i).hpref)
@@ -434,6 +472,7 @@ namespace netgen
 	      }
 	  (*mesh)[mpi].Singularity(geometry.GetPoint(i).hpref);
 	}
+    t_hpref.Stop();
 
 
     int maxdomnr = 0;
@@ -443,6 +482,16 @@ namespace netgen
 	if ( (*mesh)[si].domout > maxdomnr) maxdomnr = (*mesh)[si].domout;
       }
 
+    TableCreator<const Segment*> dom2seg_creator(maxdomnr+1);
+    for ( ; !dom2seg_creator.Done(); dom2seg_creator++)
+      for (const Segment & seg : mesh->LineSegments())
+        {
+          dom2seg_creator.Add (seg.domin, &seg);
+          if (seg.domin != seg.domout)
+            dom2seg_creator.Add (seg.domout, &seg);
+        }
+    auto dom2seg = dom2seg_creator.MoveTable();
+    
     mesh->ClearFaceDescriptors();
     for (int i = 1; i <= maxdomnr; i++)
       mesh->AddFaceDescriptor (FaceDescriptor (i, 0, 0, i));
@@ -458,10 +507,8 @@ namespace netgen
     for ( int sindex = 0; sindex < maxsegmentindex; sindex++ )
       mesh->SetBCName ( sindex, geometry.GetBCName( sindex+1 ) );
 
-    for (SegmentIndex si = 0; si < mesh->GetNSeg(); si++)
-      (*mesh)[si].SetBCName ( (*mesh).GetBCNamePtr( (*mesh)[si].si-1 ) );
-  
     mesh->CalcLocalH(mp.grading);
+    t_h.Stop();
 
     int bnp = mesh->GetNP(); // boundary points
     auto BndPntRange = mesh->Points().Range();
@@ -472,6 +519,7 @@ namespace netgen
     for (int domnr = 1; domnr <= maxdomnr; domnr++)
       if (geometry.GetDomainTensorMeshing (domnr))
         { // tensor product mesh
+          RegionTimer rt(t_tensor);
           
           NgArray<PointIndex, PointIndex::BASE> nextpi(bnp);
           NgArray<int, PointIndex::BASE> si1(bnp), si2(bnp);
@@ -527,11 +575,16 @@ namespace netgen
 
 
           for (PointIndex pix = nextpi[c1], ix = 0; pix != c2; pix = nextpi[pix], ix++)
+          {
+            Point<3> px = (*mesh)[pix];
             for (PointIndex piy = nextpi[c2], iy = 0; piy != c3; piy = nextpi[piy], iy++)
               {
-                Point<3> p = (*mesh)[pix] + ( (*mesh)[piy] - (*mesh)[c2] );
+                double lam = Dist((*mesh)[piy],(*mesh)[c2]) / Dist((*mesh)[c3],(*mesh)[c2]);
-                pts[(nex+1)*(iy+1) + ix+1] = mesh -> AddPoint (p , 1, FIXEDPOINT);
+                auto pix1 = pts[(nex+1)*ney+ix+1];
+                auto pnew = px + lam*((*mesh)[pix1]-px);
+                pts[(nex+1)*(iy+1) + ix+1] = mesh -> AddPoint (pnew, 1, FIXEDPOINT);
               }
+          }
 
           for (int i = 0; i < ney; i++)
             for (int j = 0; j < nex; j++)
@@ -545,13 +598,20 @@ namespace netgen
 
                 mesh -> AddSurfaceElement (el);
               }
+          char* material;
+          geometry.GetMaterial(domnr, material);
+          if(material)
+            mesh->SetMaterial(domnr, material);
         }
 
 
 
 
+    static Timer t_domain("Mesh domain");
+    static Timer t_points("Mesh domain - find points");
     for (int domnr = 1; domnr <= maxdomnr; domnr++)
       {
+        RegionTimer rt(t_domain);
         if (geometry.GetDomainTensorMeshing (domnr)) continue;
         
         double h = mp.maxh;
@@ -567,19 +627,43 @@ namespace netgen
 
 	Meshing2 meshing (geometry, mp, Box<3> (pmin, pmax));
 
-	NgArray<int, PointIndex::BASE> compress(bnp);
+	NgArray<int, PointIndex::BASE> compress(mesh->GetNP());
 	compress = -1;
 	int cnt = 0;
-        for (PointIndex pi : BndPntRange)
+
-	  if ( (*mesh)[pi].GetLayer() == geometry.GetDomainLayer(domnr))
+        t_points.Start();
-	    {
+        /*
-	      meshing.AddPoint ((*mesh)[pi], pi);
+        for (SegmentIndex si = 0; si < mesh->GetNSeg(); si++)
-	      cnt++;
+        {
-	      compress[pi] = cnt;
+          const auto & s = (*mesh)[si];
-	    }
+          if ( s.domin==domnr || s.domout==domnr )
+          {
+            for (auto pi : {s[0], s[1]})
+            {
+              if(compress[pi]==-1)
+              {
+                meshing.AddPoint((*mesh)[pi], pi);
+                cnt++;
+                compress[pi] = cnt;
+              }
+            }
+          }
+        }
+        */
+        for (const Segment * seg : dom2seg[domnr])
+          if (seg->domin==domnr || seg->domout==domnr )
+            for (auto pi : {(*seg)[0], (*seg)[1]})
+              if (compress[pi]==-1)
+                {
+                  meshing.AddPoint((*mesh)[pi], pi);
+                  cnt++;
+                  compress[pi] = cnt;
+                }
+
 
 	PointGeomInfo gi;
 	gi.trignum = 1;
+        /*
 	for (SegmentIndex si = 0; si < mesh->GetNSeg(); si++)
 	  {
 	    if ( (*mesh)[si].domin == domnr)
@@ -593,13 +677,31 @@ namespace netgen
                                             compress[(*mesh)[si][0]], gi, gi);
 	      }
 	  }
+        */
 
-        // not complete, use at own risk ...
+	for (const Segment * seg : dom2seg[domnr])
-        // meshing.Delaunay(*mesh, domnr, mp);
+	  {
-        mp.checkoverlap = 0;
+	    if (seg->domin == domnr)
-        auto res = meshing.GenerateMesh (*mesh, mp, h, domnr);
+              meshing.AddBoundaryElement (compress[(*seg)[0]], 
-        if (res != 0)
+                                          compress[(*seg)[1]], gi, gi);
-          throw NgException("meshing failed");
+            
+	    if (seg->domout == domnr)
+              meshing.AddBoundaryElement (compress[(*seg)[1]],
+                                          compress[(*seg)[0]], gi, gi);
+	  }
+
+        
+        t_points.Stop();
+
+        if(mp.delaunay2d && cnt>1)
+          meshing.Delaunay(*mesh, domnr, mp);
+        else
+        {
+          // mp.checkoverlap = 0;
+          auto res = meshing.GenerateMesh (*mesh, mp, h, domnr);
+          if (res != 0)
+            throw NgException("meshing failed");
+        }
         
 	for (SurfaceElementIndex sei = oldnf; sei < mesh->GetNSE(); sei++)
 	  (*mesh)[sei].SetIndex (domnr);
@@ -611,6 +713,8 @@ namespace netgen
           mesh->SetMaterial (domnr, material);
       }
 
+    mesh->Compress();
+
     mp.quad = hquad;
 
 

libsrc/geom2d/geometry2d.hpp

@@ -9,6 +9,7 @@
 
 #include <myadt.hpp>
 #include <gprim.hpp>
+#include <meshing.hpp>
 
 
 // #include "../gprim/spline.hpp"
@@ -133,7 +134,7 @@ namespace netgen
     NgArray<char*> materials;
     NgArray<double> maxh;
     NgArray<bool> quadmeshing;
-    NgArray<bool> tensormeshing;
+    Array<bool> tensormeshing;
     NgArray<int> layer;
     NgArray<string*> bcnames;
     double elto0 = 1.0;
@@ -203,8 +204,8 @@ namespace netgen
 
 
     size_t GetNDomains() const { return materials.Size(); }
-    void GetMaterial (int  domnr, char* & material );
+    DLL_HEADER void GetMaterial (int  domnr, char* & material );
-    void SetMaterial (int  domnr, const string & material);
+    DLL_HEADER void SetMaterial (int  domnr, const string & material);
 
     double GetDomainMaxh ( const int domnr );
     void SetDomainMaxh ( const int domnr, double maxh );
@@ -214,11 +215,36 @@ namespace netgen
       if ( quadmeshing.Size() ) return quadmeshing[domnr-1]; 
       else return false;
     }
+    void SetDomainQuadMeshing ( int domnr, bool quad_meshing )
+    {
+      auto oldsize = quadmeshing.Size();
+
+      if ( oldsize<domnr )
+        {
+          quadmeshing.SetSize(domnr);
+          for(auto dom : IntRange(oldsize, domnr-1))
+              quadmeshing[dom] = false;
+        }
+
+      quadmeshing[domnr-1] = quad_meshing;
+    }
+
     bool GetDomainTensorMeshing ( int domnr ) 
     { 
-      if ( tensormeshing.Size() ) return tensormeshing[domnr-1]; 
+      if ( tensormeshing.Size()>=domnr ) return tensormeshing[domnr-1];
       else return false;
     }
+    void SetDomainTensorMeshing ( int domnr, bool tm )
+    {
+      if ( tensormeshing.Size()<domnr )
+      {
+        auto oldsize = tensormeshing.Size();
+        tensormeshing.SetSize(domnr);
+        for(auto i : IntRange(oldsize, domnr-1))
+          tensormeshing[i] = false;
+      }
+      tensormeshing[domnr-1] = tm;
+    }
     int GetDomainLayer ( int domnr ) 
     { 
       if ( layer.Size() ) return layer[domnr-1]; 

libsrc/geom2d/python_geom2d.cpp

@@ -6,8 +6,10 @@
 
 #include <meshing.hpp>
 #include <geometry2d.hpp>
+#include <csg2d.hpp>
 
 using namespace netgen;
+using namespace pybind11::literals;
 
 namespace netgen
 {
@@ -15,7 +17,7 @@ namespace netgen
 }
 
 
-DLL_HEADER void ExportGeom2d(py::module &m) 
+NGCORE_API_EXPORT void ExportGeom2d(py::module &m) 
 {
   py::class_<SplineSegExt, shared_ptr<SplineSegExt>>
     (m, "Spline", "Spline of a SplineGeometry object")
@@ -65,24 +67,45 @@ DLL_HEADER void ExportGeom2d(py::module &m)
                                       optional<variant<int, string>> bc, optional<int> copy, double maxh,
                                       double hpref, double hprefleft, double hprefright)
 	  {
-            auto segtype = py::cast<std::string>(segment[0]);
-            
             SplineSegExt * seg;
-            if (segtype == "line")
+            if(py::isinstance<py::str>(segment[0]))
               {
-                LineSeg<2> * l = new LineSeg<2>(self.GetPoint(py::cast<int>(segment[1])),
+                auto segtype = py::cast<std::string>(segment[0]);
-                                                self.GetPoint(py::cast<int>(segment[2])));
+            
-                seg = new SplineSegExt(*l);
+                if (segtype == "line")
-              }
+                  {
-            else if (segtype == "spline3")
+                    LineSeg<2> * l = new LineSeg<2>(self.GetPoint(py::cast<int>(segment[1])),
-              {
+                                                    self.GetPoint(py::cast<int>(segment[2])));
-                SplineSeg3<2> * seg3 = new SplineSeg3<2>(self.GetPoint(py::cast<int>(segment[1])),
+                    seg = new SplineSegExt(*l);
-                                                         self.GetPoint(py::cast<int>(segment[2])),
+                  }
-                                                         self.GetPoint(py::cast<int>(segment[3])));
+                else if (segtype == "spline3")
-                seg = new SplineSegExt(*seg3);
+                  {
+                    SplineSeg3<2> * seg3 = new SplineSeg3<2>(self.GetPoint(py::cast<int>(segment[1])),
+                                                             self.GetPoint(py::cast<int>(segment[2])),
+                                                             self.GetPoint(py::cast<int>(segment[3])));
+                    seg = new SplineSegExt(*seg3);
+                  }
+                else
+                  throw Exception("Appended segment is not a line or a spline3");
               }
             else
-                throw Exception("Appended segment is not a line or a spline3");
+              {
+                if(py::len(segment) == 2)
+                  {
+                    auto l = new LineSeg<2>(self.GetPoint(py::cast<int>(segment[0])),
+                                            self.GetPoint(py::cast<int>(segment[1])));
+                    seg = new SplineSegExt(*l);
+                  }
+                else if(py::len(segment) == 3)
+                  {
+                    SplineSeg3<2> * seg3 = new SplineSeg3<2>(self.GetPoint(py::cast<int>(segment[0])),
+                                                             self.GetPoint(py::cast<int>(segment[1])),
+                                                             self.GetPoint(py::cast<int>(segment[2])));
+                    seg = new SplineSegExt(*seg3);
+                  }
+                else
+                  throw Exception("Appended segment must either have 2 or 3 points");
+              }
             seg->leftdom = leftdomain;
             seg->rightdom = rightdomain;
             seg->hmax = maxh;
@@ -265,6 +288,7 @@ DLL_HEADER void ExportGeom2d(py::module &m)
                  {
                    double len = self.splines[i]->Length();
                    int n = floor(len/(0.05*min(xdist,ydist)));
+                   n = max(3, n);
                    lst.push_back(self.splines[i]->StartPI());
                    for (int j = 1; j < n; j++){
                      lst.push_back(self.splines[i]->GetPoint(j*1./n));
@@ -371,9 +395,8 @@ DLL_HEADER void ExportGeom2d(py::module &m)
           })
          )
     
-    // If we change to c++17 this can become optional<MeshingParameters>
     .def("GenerateMesh", [](shared_ptr<SplineGeometry2d> self,
-                            MeshingParameters* pars, py::kwargs kwargs)
+                            optional<MeshingParameters> pars, py::kwargs kwargs)
 		{
                   MeshingParameters mp;
                   if(pars) mp = *pars;
@@ -389,11 +412,89 @@ DLL_HEADER void ExportGeom2d(py::module &m)
                   if(result != 0)
                     throw Exception("Meshing failed!");
 		  return mesh;
-                }, py::arg("mp") = nullptr,
+                }, py::arg("mp") = nullopt,
+      py::call_guard<py::gil_scoped_release>(),
+      meshingparameter_description.c_str())
+    .def("_SetDomainTensorMeshing", &SplineGeometry2d::SetDomainTensorMeshing)
+    ;
+  
+  py::class_<Solid2d>(m, "Solid2d")
+    .def(py::init<>())
+    .def(py::init<Array<std::variant<Point<2>, EdgeInfo, PointInfo>>, std::string, std::string>(), py::arg("points"), py::arg("mat")=MAT_DEFAULT, py::arg("bc")=BC_DEFAULT)
+
+    .def(py::self+py::self)
+    .def(py::self-py::self)
+    .def(py::self*py::self)
+    .def(py::self+=py::self)
+    .def(py::self-=py::self)
+    .def(py::self*=py::self)
+
+    .def("Mat", &Solid2d::Mat)
+    .def("BC", &Solid2d::BC)
+    .def("Maxh", &Solid2d::Maxh)
+
+    .def("Copy", [](Solid2d & self) -> Solid2d { return self; })
+    .def("Move", &Solid2d::Move)
+    .def("Scale", static_cast<Solid2d& (Solid2d::*)(double)>(&Solid2d::Scale))
+    .def("Scale", static_cast<Solid2d& (Solid2d::*)(Vec<2>)>(&Solid2d::Scale))
+    .def("Rotate", &Solid2d::RotateDeg, py::arg("angle"), py::arg("center")=Point<2>{0,0})
+    ;
+  
+
+  m.def("Rectangle", [](Point<2> p0, Point<2> p1, string mat, string bc, optional<string> bottom, optional<string> right, optional<string> top, optional<string> left) -> Solid2d
+		  {
+                      using P = Point<2>;
+                      return { {
+                              p0,    EdgeInfo{bottom ? *bottom : bc},
+                              P{p1[0],p0[1]}, EdgeInfo {right  ? *right  : bc},
+                              p1,             EdgeInfo {top    ? *top    : bc},
+                              P{p0[0],p1[1]}, EdgeInfo {left   ? *left   : bc},
+                             }, mat};
+                  },
+		  "pmin"_a, "pmax"_a, "mat"_a=MAT_DEFAULT, "bc"_a=BC_DEFAULT,
+                  "bottom"_a=nullopt, "right"_a=nullopt, "top"_a=nullopt, "left"_a=nullopt
+       );
+  m.def("Circle", Circle, py::arg("center"), py::arg("radius"), py::arg("mat")=MAT_DEFAULT, py::arg("bc")=BC_DEFAULT);
+
+  py::class_<CSG2d>(m, "CSG2d")
+    .def(py::init<>())
+    .def("GenerateSplineGeometry", &CSG2d::GenerateSplineGeometry)
+    .def("Add", &CSG2d::Add)
+    .def("GenerateMesh", [](CSG2d & self, optional<MeshingParameters> pars, py::kwargs kwargs)
+		{
+                  MeshingParameters mp;
+                  if(pars) mp = *pars;
+                  {
+                    py::gil_scoped_acquire aq;
+                    CreateMPfromKwargs(mp, kwargs);
+                  }
+		  auto mesh = make_shared<Mesh>();
+                  auto geo = self.GenerateSplineGeometry();
+                  mesh->SetGeometry(geo);
+                  SetGlobalMesh (mesh);
+                  ng_geometry = geo;
+		  auto result = geo->GenerateMesh(mesh, mp);
+                  if(result != 0)
+                    throw Exception("Meshing failed!");
+		  return mesh;
+                }, py::arg("mp") = nullopt,
       py::call_guard<py::gil_scoped_release>(),
       meshingparameter_description.c_str())
     ;
-  
+
+  py::class_<EdgeInfo>(m, "EdgeInfo")
+    .def(py::init<>())
+    .def(py::init<const Point<2>&>(), py::arg("control_point"))
+    .def(py::init<double>(), py::arg("maxh"))
+    .def(py::init<string>(), py::arg("bc"))
+    .def(py::init<optional<Point<2>>, double, string>(), py::arg("control_point")=nullopt, py::arg("maxh")=MAXH_DEFAULT, py::arg("bc")=BC_DEFAULT)
+    ;
+  py::class_<PointInfo>(m, "PointInfo")
+    .def(py::init<>())
+    .def(py::init<double>(), "maxh"_a)
+    .def(py::init<string>(), "name"_a)
+    .def(py::init<double, string>(), "maxh"_a, "name"_a)
+    ;
 }
 
 PYBIND11_MODULE(libgeom2d, m) {

libsrc/gprim/adtree.hpp

@@ -1099,6 +1099,284 @@ public:
 //     auto & Tree() { return *tree; };
 //   };
 
+  template<int dim, typename T=INDEX, typename TSCAL=double>
+  class DelaunayTree
+  {
+  public:
+    // Number of entries per leaf
+    static constexpr int N = 100;
+
+    struct Node;
+
+    struct Leaf
+    {
+      Point<2*dim, TSCAL> p[N];
+      T index[N];
+      int n_elements;
+      int nr;
+
+      Leaf() : n_elements(0)
+      { }
+
+
+      void Add( Array<Leaf*> &leaves, Array<T> &leaf_index, const Point<2*dim> &ap, T aindex )
+        {
+          p[n_elements] = ap;
+          index[n_elements] = aindex;
+          n_elements++;
+          if(leaf_index.Size()<aindex+1)
+              leaf_index.SetSize(aindex+1);
+          leaf_index[aindex] = nr;
+        }
+    };
+
+    struct Node
+    {
+      union
+      {
+        Node *children[2];
+        Leaf *leaf;
+      };
+      double sep;
+      int level;
+
+      Node()
+          : children{nullptr,nullptr}
+      { }
+
+      ~Node()
+       { }
+
+      Leaf *GetLeaf() const
+        {
+          return children[1] ? nullptr : leaf;
+        }
+    };
+
+  private:
+    Node root;
+
+    Array<Leaf*> leaves;
+    Array<T> leaf_index;
+
+    Point<dim> global_min, global_max;
+    double tol;
+    size_t n_leaves;
+    size_t n_nodes;
+    BlockAllocator ball_nodes;
+    BlockAllocator ball_leaves;
+
+  public:
+
+    DelaunayTree (const Point<dim> & pmin, const Point<dim> & pmax)
+        : global_min(pmin), global_max(pmax), n_leaves(1), n_nodes(1), ball_nodes(sizeof(Node)), ball_leaves(sizeof(Leaf))
+      {
+        root.leaf = (Leaf*) ball_leaves.Alloc(); new (root.leaf) Leaf();
+        root.leaf->nr = 0;
+        leaves.Append(root.leaf);
+        root.level = 0;
+        tol = 1e-7 * Dist(pmax, pmin);
+      }
+
+    DelaunayTree (const Box<dim> & box)
+        : DelaunayTree(box.PMin(), box.PMax())
+      { }
+
+    size_t GetNLeaves()
+      {
+        return n_leaves;
+      }
+
+    size_t GetNNodes()
+      {
+        return n_nodes;
+      }
+
+    template<typename TFunc>
+    void GetFirstIntersecting (const Point<dim> & pmin, const Point<dim> & pmax,
+            TFunc func=[](auto pi){return false;}) const
+      {
+        // static Timer timer("DelaunayTree::GetIntersecting"); RegionTimer rt(timer);
+        // static Timer timer1("DelaunayTree::GetIntersecting-LinearSearch");
+        ArrayMem<const Node*, 100> stack;
+        ArrayMem<int, 100> dir_stack;
+
+
+        Point<2*dim> tpmin, tpmax;
+
+        for (size_t i : IntRange(dim))
+          {
+            tpmin(i) = global_min(i);
+            tpmax(i) = pmax(i)+tol;
+
+            tpmin(i+dim) = pmin(i)-tol;
+            tpmax(i+dim) = global_max(i);
+          }
+
+        stack.SetSize(0);
+        stack.Append(&root);
+        dir_stack.SetSize(0);
+        dir_stack.Append(0);
+
+        while(stack.Size())
+          {
+            const Node *node = stack.Last();
+            stack.DeleteLast();
+
+            int dir = dir_stack.Last();
+            dir_stack.DeleteLast();
+
+            if(Leaf *leaf = node->GetLeaf())
+              {
+                //               RegionTimer rt1(timer1);
+                for (auto i : IntRange(leaf->n_elements))
+                  {
+                    bool intersect = true;
+                    const auto p = leaf->p[i];
+
+                    for (int d = 0; d < dim; d++)
+                        if (p[d] > tpmax[d])
+                            intersect = false;
+                    for (int d = dim; d < 2*dim; d++)
+                        if (p[d] < tpmin[d])
+                            intersect = false;
+                    if(intersect)
+                        if(func(leaf->index[i])) return;
+                  }
+              }
+            else
+              {
+                int newdir = dir+1;
+                if(newdir==2*dim) newdir = 0;
+                if (tpmin[dir] <= node->sep)
+                  {
+                    stack.Append(node->children[0]);
+                    dir_stack.Append(newdir);
+                  }
+                if (tpmax[dir] >= node->sep)
+                  {
+                    stack.Append(node->children[1]);
+                    dir_stack.Append(newdir);
+                  }
+              }
+          }
+      }
+
+    void GetIntersecting (const Point<dim> & pmin, const Point<dim> & pmax,
+            NgArray<T> & pis) const
+      {
+        pis.SetSize(0);
+        GetFirstIntersecting(pmin, pmax, [&pis](auto pi) { pis.Append(pi); return false;});
+      }
+
+    void Insert (const Box<dim> & box, T pi)
+      {
+        Insert (box.PMin(), box.PMax(), pi);
+      }
+
+    void Insert (const Point<dim> & pmin, const Point<dim> & pmax, T pi)
+      {
+        // static Timer timer("DelaunayTree::Insert"); RegionTimer rt(timer);
+        int dir = 0;
+        Point<2*dim> p;
+        for (auto i : IntRange(dim))
+          {
+            p(i) = pmin[i];
+            p(i+dim) = pmax[i];
+          }
+
+        Node * node = &root;
+        Leaf * leaf = node->GetLeaf();
+
+        // search correct leaf to add point
+        while(!leaf)
+          {
+            node = p[dir] < node->sep ? node->children[0] : node->children[1];
+            dir++;
+            if(dir==2*dim) dir = 0;
+            leaf = node->GetLeaf();
+          }
+
+        // add point to leaf
+        if(leaf->n_elements < N)
+            leaf->Add(leaves, leaf_index, p,pi);
+        else // assume leaf->n_elements == N
+          {
+            // add two new nodes and one new leaf
+            int n_elements = leaf->n_elements;
+            ArrayMem<TSCAL, N> coords(n_elements);
+            ArrayMem<int, N> order(n_elements);
+
+            // separate points in two halves, first sort all coordinates in direction dir
+            for (auto i : IntRange(n_elements))
+              {
+                order[i] = i;
+                coords[i] = leaf->p[i][dir];
+              }
+
+            QuickSortI(coords, order);
+            int isplit = N/2;
+            Leaf *leaf1 = (Leaf*) ball_leaves.Alloc(); new (leaf1) Leaf();
+            Leaf *leaf2 = (Leaf*) ball_leaves.Alloc(); new (leaf2) Leaf();
+
+            leaf1->nr = leaf->nr;
+            leaf2->nr = leaves.Size();
+            leaves.Append(leaf2);
+            leaves[leaf1->nr] = leaf1;
+
+            for (auto i : order.Range(isplit))
+                leaf1->Add(leaves, leaf_index, leaf->p[i], leaf->index[i] );
+            for (auto i : order.Range(isplit, N))
+                leaf2->Add(leaves, leaf_index, leaf->p[i], leaf->index[i] );
+
+            Node *node1 = (Node*) ball_nodes.Alloc(); new (node1) Node();
+            node1->leaf = leaf1;
+            node1->level = node->level+1;
+
+            Node *node2 = (Node*) ball_nodes.Alloc(); new (node2) Node();
+            node2->leaf = leaf2;
+            node2->level = node->level+1;
+
+            node->children[0] = node1;
+            node->children[1] = node2;
+            node->sep = 0.5 * (leaf->p[order[isplit-1]][dir] + leaf->p[order[isplit]][dir]);
+
+            // add new point to one of the new leaves
+            if (p[dir] < node->sep)
+                leaf1->Add( leaves, leaf_index, p, pi );
+            else
+                leaf2->Add( leaves, leaf_index, p, pi );
+
+            ball_leaves.Free(leaf);
+            n_leaves++;
+            n_nodes+=2;
+          }
+      }
+
+    void DeleteElement (T pi)
+      {
+        // static Timer timer("DelaunayTree::DeleteElement"); RegionTimer rt(timer);
+        Leaf *leaf = leaves[leaf_index[pi]];
+        leaf_index[pi] = -1;
+        auto & n_elements = leaf->n_elements;
+        auto & index = leaf->index;
+        auto & p = leaf->p;
+
+        for (auto i : IntRange(n_elements))
+          {
+            if(index[i] == pi)
+              {
+                n_elements--;
+                if(i!=n_elements)
+                  {
+                    index[i] = index[n_elements];
+                    p[i] = p[n_elements];
+                  }
+                return;
+              }
+          }
+      }
+  };
 }
 
 #endif

libsrc/gprim/geom2d.cpp

@@ -3,10 +3,6 @@
 #include <myadt.hpp>
 #include <gprim.hpp>
 
-#ifndef M_PI
-#define M_PI        3.14159265358979323846
-#endif
-
 namespace netgen
 {
 
@@ -251,239 +247,4 @@ int PTRIANGLE2D :: IsIn (const Point2d & p) const
 }
 
 #endif
-
-
-
-
-
-
-Polygon2d :: Polygon2d ()
-{
-  ;
-}
-
-Polygon2d :: ~Polygon2d ()
-{
-  ;
-}
-
-void Polygon2d :: AddPoint (const Point2d & p)
-{ 
-  points.Append(p); 
-}
-
-
-double Polygon2d :: HArea () const
-{
-  int i;
-  double ar = 0;
-  for (i = 1; i <= points.Size(); i++)
-    {
-      const Point2d & p1 = points.Get(i);
-      const Point2d & p2 = points.Get(i%points.Size()+1);
-      ar += 
-	(p2.X()-p1.X()) * p1.Y() -
-	(p2.Y()-p1.Y()) * p1.X();
-    }
-  return ar/2;
-  /*
-  CURSOR c;
-  double ar = 0;
-  Point2d * p1, * p2, p0 = Point2d(0, 0);
-  Vec2d v1, v2 = Vec2d(1, 0);
-
-  p2 = points[points.Last()];
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    p1 = p2;
-    p2 = points[c];
-    ar += Cross ( (*p2-*p1), (*p1 - p0));
-    }
-  return ar / 2;
-  */
-}
-
-
-int Polygon2d :: IsOn (const Point2d & p) const
-{
-  int i;
-  for (i = 1; i <= points.Size(); i++)
-    {
-      const Point2d & p1 = points.Get(i);
-      const Point2d & p2 = points.Get(i%points.Size()+1);
-      if (IsOnLine (Line2d(p1, p2), p)) return 1;
-    }
-  return 0;
-  /*
-  CURSOR c;
-  Point2d * p1, * p2;
-  
-  p2 = points[points.Last()];
-  for (c = points.First(); c != points.Head(); c++)
-    {
-      p1 = p2;
-      p2 = points[c];
-      if (IsOnLine (Line2d(*p1, *p2), p)) return 1;
-    }
-  return 0;
-  */
-}
-
-
-int Polygon2d :: IsIn (const Point2d & p) const
-{
-  int i;
-  double sum = 0, ang;
-  for (i = 1; i <= points.Size(); i++)
-    {
-      const Point2d & p1 = points.Get(i);
-      const Point2d & p2 = points.Get(i%points.Size()+1);
-      ang = Angle ( (p1 - p), (p2 - p) );
-      if (ang > M_PI) ang -= 2 * M_PI;
-      sum += ang;
-    }
-  return fabs(sum) > M_PI;
-  /*
-  CURSOR c;
-  Point2d * p1, * p2;
-  double sum = 0, ang;
-
-  p2 = points[points.Last()];
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    p1 = p2;
-    p2 = points[c];
-    ang = Angle ( (*p1 - p), (*p2 - p) );
-    if (ang > M_PI) ang -= 2 * M_PI;
-    sum += ang;
-    }
-
-  return fabs(sum) > M_PI;
-  */
-}
-
-int Polygon2d :: IsConvex () const
-  {
-    /*
-  Point2d *p, *pold, *pnew;
-  char cw;
-  CURSOR c;
-
-  if (points.Length() < 3) return 0;
-
-  c = points.Last();
-  p = points[c];
-  c--;
-  pold = points[c];
-  pnew = points[points.First()];
-  cw = ::CW (*pold, *p, *pnew);
-
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    pnew = points[c];
-    if (cw != ::CW (*pold, *p, *pnew))
-      return 0;
-    pold = p;
-    p = pnew;
-    }
-    */
-    return 0;
-  }
-
-
-int Polygon2d :: IsStarPoint (const Point2d & p) const
-  {
-    /*
-  Point2d *pnew, *pold;
-  char cw;
-  CURSOR c;
-
-  if (points.Length() < 3) return 0;
-
-  pold = points[points.Last()];
-  pnew = points[points.First()];
-
-  cw = ::CW (p, *pold, *pnew);
-
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    pnew = points[c];
-    if (cw != ::CW (p, *pold, *pnew))
-      return 0;
-    pold = pnew;
-    }
-  return 1;
-    */
-    return 0;
-  }
-
-
-Point2d Polygon2d :: Center () const
-  {
-    /*
-  double ai, a = 0, x = 0, y = 0;
-  Point2d * p, *p2;
-  Point2d p0 = Point2d(0, 0);
-  CURSOR c;
-
-  p2 = points[points.Last()];
-
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    p = points[c];
-    ai = Cross (*p2 - p0, *p - p0);
-    x += ai / 3 * (p2->X() + p->X());
-    y += ai / 3 * (p2->Y() + p->Y());
-    a+= ai;
-    p2 = p;
-    }
-  if (a != 0)
-    return Point2d (x / a, y / a);
-  else
-    return Point2d (0, 0);
-    */
-    return Point2d (0, 0);
-  }
-
-
-
-Point2d Polygon2d :: EqualAreaPoint () const
-  {
-    /*
-  double a11 = 0, a12 = 0, a21= 0, a22 = 0;
-  double b1 = 0, b2 = 0, dx, dy;
-  double det;
-  Point2d * p, *p2;
-  CURSOR c;
-
-  p = points[points.Last()];
-
-  for (c = points.First(); c != points.Head(); c++)
-    {
-    p2 = p;
-    p = points[c];
-
-    dx = p->X() - p2->X();
-    dy = p->Y() - p2->Y();
-
-    a11 += sqr (dy);
-    a12 -= dx * dy;
-    a21 -= dx * dy;
-    a22 += sqr (dx);
-    b1 -= dy * (p->X() * p2->Y() - p2->X() * p->Y());
-    b2 -= dx * (p->Y() * p2->X() - p2->Y() * p->X());
-    }
-
-  det = a11 * a22 - a21 * a12;
-
-  if (det != 0)
-    return Point2d ( (b1 * a22 - b2 * a12) / det,
-                     (a11 * b2 - a21 * b1) / det);
-  else
-    return Point2d (0, 0);
-*/
-    return Point2d (0, 0);
-  }
-
-
 }

libsrc/gprim/geom2d.hpp

@@ -609,40 +609,6 @@ namespace netgen
 #endif
 
 
-
-  class Polygon2d
-  {
-  protected:
-    NgArray<Point2d> points;
-  
-  public:
-    Polygon2d ();
-    ~Polygon2d ();
-  
-    void AddPoint (const Point2d & p);
-    int GetNP() const { return points.Size(); }
-    void GetPoint (int i, Point2d & p) const
-    { p = points.Get(i); }
-    void GetLine (int i, Point2d & p1, Point2d & p2) const
-    { p1 = points.Get(i); p2 = points.Get(i%points.Size()+1); }
-
-    double Area () const { return fabs (HArea()); }
-    int CW () const { return HArea() > 0; }
-    int CCW () const { return HArea() < 0; }
-  
-    int IsOn (const Point2d & p) const;
-    int IsIn (const Point2d & p) const;
-
-    int IsConvex () const;
-
-    int IsStarPoint (const Point2d & p) const;
-    Point2d Center() const;
-    Point2d EqualAreaPoint () const;
-  private:
-    double HArea () const;
-  };
-
-
   /** Cheap approximation to atan2.
       A monotone function of atan2(x,y) is computed.
   */

libsrc/gprim/geomfuncs.cpp

@@ -83,7 +83,7 @@ double Det (const Mat<3,3> & m)
 
 void EigenValues (const Mat<3,3> & m, Vec<3> & ev)
 {
-  const double pi = 3.141592;
+  const double pi = M_PI;
   double a, b, c, d;
   double p, q;
   double arg;

libsrc/gprim/geomfuncs.hpp

@@ -142,10 +142,24 @@ namespace netgen
 
   inline void CalcInverse (const Mat<2,3> & m, Mat<3,2> & inv)
   {
+    Vec<3> a0 = m.Row(0);
+    Vec<3> a1 = m.Row(1);
+    Vec<3> n = Cross(a0, a1);
+    Vec<3> d0 = Cross(a1, n);
+    Vec<3> d1 = Cross(a0, n);
+    double s0 = 1.0/(a0*d0);
+    double s1 = 1.0/(a1*d1);
+    for (int i = 0; i < 3; i++)
+      {
+        inv(i,0) = s0*d0(i);
+        inv(i,1) = s1*d1(i);
+      }
+    /*
     Mat<2,2> a = m * Trans (m);
     Mat<2,2> ainv;
     CalcInverse (a, ainv);
     inv = Trans (m) * ainv;
+    */
   }
 
   void CalcInverse (const Mat<3,2> & m, Mat<2,3> & inv);
@@ -166,6 +180,31 @@ namespace netgen
   void EigenValues (const Mat<3,3> & m, Vec<3> & ev);
   void EigenValues (const Mat<2,2> & m, Vec<3> & ev);
 
+
+  template <typename T> 
+  Vec<3,T> StableSolve (Mat<2,3,T> mat, Vec<2,T> rhs)
+  {
+    Vec<3> a0 = mat.Row(0);
+    Vec<3> a1 = mat.Row(1);
+    /*
+    Vec<3> d = Cross ( Cross (a0, a1), a0);
+
+    double alpha = rhs(0) / a0.Length2();
+    double beta = (rhs(1)-alpha* (a0*a1)) / (d*a1);
+    return alpha * a0 + beta * d;
+    */
+    Vec<3> n = Cross(a0, a1);
+    Vec<3> d0 = Cross(a1, n);
+    Vec<3> d1 = Cross(a0, n);
+    double alpha = rhs(0) / (a0*d0);
+    double beta = rhs(1) / (a1*d1);
+    return alpha * d0 + beta * d1;
+  }
+
+
+
+
+  
 }
 
 #endif

libsrc/gprim/geomobjects.hpp

@@ -384,8 +384,16 @@ namespace netgen
     bool IsIn (const Point<D> & p) const
     {
       for (int i = 0; i < D; i++)
-	if (p(i) < pmin(i) || p(i) > pmax(i)) return 0;
+	if (p(i) < pmin(i) || p(i) > pmax(i)) return false;
-      return 1;
+      return true;
+    }
+
+    // is point in eps-increased box
+    bool IsIn (const Point<D> & p, double eps) const
+    {
+      for (int i = 0; i < D; i++)
+	if (p(i) < pmin(i)-eps || p(i) > pmax(i)+eps) return false;
+      return true;
     }
 
 
@@ -478,7 +486,7 @@ namespace netgen
   };
 
 
-#ifdef PARALLEL
+#ifdef PARALLEL_OLD
   template <> 
   inline MPI_Datatype MyGetMPIType<Vec<3, double> > ()
   {

libsrc/gprim/geomtest3d.cpp

@@ -1035,6 +1035,39 @@ double MinDistLP2 (const Point3d & lp1, const Point3d & lp2, const Point3d & p)
 }
 
 
+double MinDistLP2 (const Point3d & lp1, const Point3d & lp2, const Point3d & p, double & lam)
+{
+  Vec3d v(lp1, lp2);
+  Vec3d vlp(lp1, p);
+
+  // dist(lam) = \| vlp \|^2 - 2 lam (v1p, v) + lam^2 \| v \|^2
+
+  // lam = (v * vlp) / (v * v);
+  // if (lam < 0) lam = 0;
+  // if (lam > 1) lam = 1;
+
+  double num = v*vlp;
+  double den = v*v;
+
+  if (num <= 0) 
+  {
+    lam = 0.0;
+    return Dist2 (lp1, p);
+  }
+
+  if (num >= den) 
+  {
+    lam = 1.0;
+    return Dist2 (lp2, p);
+  }
+  
+  lam = num/den;
+  if (den > 0)
+      return vlp.Length2() - num * num /den;
+  else
+    return vlp.Length2();
+}
+
 
 double MinDistTP2 (const Point3d & tp1, const Point3d & tp2, 
 		   const Point3d & tp3, const Point3d & p)
@@ -1102,7 +1135,7 @@ double MinDistTP2 (const Point3d & tp1, const Point3d & tp2,
 
 // 0 checks !!!
 double MinDistLL2 (const Point3d & l1p1, const Point3d & l1p2,
-		  const Point3d & l2p1, const Point3d & l2p2)
+		  const Point3d & l2p1, const Point3d & l2p2, double & lam1, double & lam2 )
 {
   // dist(lam1,lam2) = \| l2p1+lam2v2 - (l1p1+lam1 v1) \|
   // min !
@@ -1112,7 +1145,7 @@ double MinDistLL2 (const Point3d & l1p1, const Point3d & l1p2,
   Vec3d v2 (l2p1, l2p2);
 
   double a11, a12, a22, rs1, rs2;
-  double lam1, lam2, det;
+  double det;
 
   a11 = v1*v1;
   a12 = -(v1*v2);
@@ -1138,14 +1171,27 @@ double MinDistLL2 (const Point3d & l1p1, const Point3d & l1p2,
     }
 
   double minv, hv;
-  minv = MinDistLP2 (l1p1, l1p2, l2p1);
+  minv = MinDistLP2 (l1p1, l1p2, l2p1, lam1);
-  hv =  MinDistLP2 (l1p1, l1p2, l2p2);
+  lam2 = 0.;
-  if (hv < minv) minv = hv;
+  hv =  MinDistLP2 (l1p1, l1p2, l2p2, lam1);
+  if (hv < minv)
+  {
+    lam2 = 1.;
+    minv = hv;
+  }
 
-  hv =  MinDistLP2 (l2p1, l2p2, l1p1);
+  hv =  MinDistLP2 (l2p1, l2p2, l1p1, lam2);
-  if (hv < minv) minv = hv;
+  if (hv < minv)
-  hv =  MinDistLP2 (l2p1, l2p2, l1p2);
+  {
-  if (hv < minv) minv = hv;
+    lam1 = 0.;
+    minv = hv;
+  }
+  hv =  MinDistLP2 (l2p1, l2p2, l1p2, lam2);
+  if (hv < minv)
+  {
+    lam1 = 1.;
+    minv = hv;
+  }
 
   return minv;
 }

libsrc/gprim/geomtest3d.hpp

@@ -75,7 +75,7 @@ extern double MinDistLP2 (const Point2d & lp1, const Point2d & lp2, const Point2
 extern double MinDistLP2 (const Point3d & lp1, const Point3d & lp2, const Point3d & p);
 
 /// Minimal distance of point p to the triangle segment [tp1,tp2,pt3]
-extern double MinDistTP2 (const Point3d & tp1, const Point3d & tp2, 
+DLL_HEADER double MinDistTP2 (const Point3d & tp1, const Point3d & tp2, 
 			  const Point3d & tp3, const Point3d & p);
 
   inline double MinDistTP2 (const Point<2> & tp1, const Point<2> & tp2, 
@@ -91,6 +91,9 @@ extern double MinDistTP2 (const Point3d & tp1, const Point3d & tp2,
 extern double MinDistLL2 (const Point3d & l1p1, const Point3d & l1p2,
 			  const Point3d & l2p1, const Point3d & l2p2);
 
+extern double MinDistLL2 (const Point3d & l1p1, const Point3d & l1p2,
+		  const Point3d & l2p1, const Point3d & l2p2, double & lam1, double & lam2 );
+
 }
 
 #endif

libsrc/gprim/spline.cpp

@@ -98,6 +98,16 @@ namespace netgen
     proj_latest_t = 0.5;
   }
 
+  template<int D>
+  SplineSeg3<D> :: SplineSeg3 (const GeomPoint<D> & ap1,
+			       const GeomPoint<D> & ap2,
+			       const GeomPoint<D> & ap3,
+                               double aweight)
+    : p1(ap1), p2(ap2), p3(ap3), weight(aweight)
+  {
+    proj_latest_t = 0.5;
+  }
+
   template<int D>
   Point<D> SplineSeg3<D> :: GetPoint (double t) const
   {

libsrc/gprim/spline.hpp

@@ -185,6 +185,10 @@ namespace netgen
     SplineSeg3 (const GeomPoint<D> & ap1, 
 		const GeomPoint<D> & ap2, 
 		const GeomPoint<D> & ap3);
+    SplineSeg3 (const GeomPoint<D> & ap1,
+		const GeomPoint<D> & ap2,
+		const GeomPoint<D> & ap3,
+                double aweight);
     // default constructor for archive
     SplineSeg3() {}
     ///
@@ -192,9 +196,13 @@ namespace netgen
     {
       ar & p1 & p2 & p3 & weight & proj_latest_t;
     }
-    virtual Point<D> GetPoint (double t) const;
     ///
-    virtual Vec<D> GetTangent (const double t) const;
+    double GetWeight () const { return weight; }
+    void SetWeight (double w) { weight = w; }
+    ///
+    DLL_HEADER virtual Point<D> GetPoint (double t) const;
+    ///
+    DLL_HEADER virtual Vec<D> GetTangent (const double t) const;
 
   
     DLL_HEADER virtual void GetDerivatives (const double t, 
@@ -202,12 +210,12 @@ namespace netgen
 				 Vec<D> & first,
 				 Vec<D> & second) const;
     ///
-    virtual const GeomPoint<D> & StartPI () const { return p1; };
+    DLL_HEADER virtual const GeomPoint<D> & StartPI () const { return p1; };
     ///
-    virtual const GeomPoint<D> & EndPI () const { return p3; }
+    DLL_HEADER virtual const GeomPoint<D> & EndPI () const { return p3; }
     ///
-    virtual void GetCoeff (Vector & coeffs) const;
+    DLL_HEADER virtual void GetCoeff (Vector & coeffs) const;
-    virtual void GetCoeff (Vector & coeffs, Point<D> p0) const;
+    DLL_HEADER virtual void GetCoeff (Vector & coeffs, Point<D> p0) const;
     
     virtual string GetType(void) const {return "spline3";}
 

libsrc/include/incopengl.hpp

@@ -2,9 +2,10 @@
 #define INCOPENGL_HPP___
 #define GL_GLEXT_PROTOTYPES
 
+#include <mystdlib.h>
 #include <mydefs.hpp>
 
-#  if defined(TOGL_AGL) || defined(TOGL_AGL_CLASSIC) || defined(TOGL_NSOPENGL)
+#  ifdef __APPLE__
 #define GL_SILENCE_DEPRECATION
 #define GL_DO_NOT_WARN_IF_MULTI_GL_VERSION_HEADERS_INCLUDED
 #    include <OpenGL/gl3.h>

libsrc/include/mydefs.hpp

@@ -11,22 +11,15 @@
   defines for graphics, testmodes, ...
 */
 
+#include <core/ngcore.hpp>
 #define PACKAGE_VERSION "6.2-dev"
 
 // #define DEBUG
 
-#ifdef WIN32
+#if defined(NGINTERFACE_EXPORTS) || ( defined(WIN32) && (defined(NGLIB_EXPORTS) || defined(nglib_EXPORTS)) )
-   #if NGINTERFACE_EXPORTS || NGLIB_EXPORTS || nglib_EXPORTS
+   #define DLL_HEADER   NGCORE_API_EXPORT
-      #define DLL_HEADER   __declspec(dllexport)
-   #else
-      #define DLL_HEADER   __declspec(dllimport)
-   #endif
 #else
-   #if __GNUC__ >= 4
+   #define DLL_HEADER   NGCORE_API_IMPORT
-      #define DLL_HEADER __attribute__ ((visibility ("default")))
-   #else
-      #define DLL_HEADER
-   #endif
 #endif
 
 

libsrc/include/nginterface.h

@@ -11,22 +11,13 @@
 /* Date:   20. Nov. 99                                                    */
 /**************************************************************************/
 
-#include <core/ngcore.hpp>
+#include "mydefs.hpp"
 
 /*
   Application program interface to Netgen
 
 */
 
-#ifndef DLL_HEADER
-  #if NGINTERFACE_EXPORTS || NGLIB_EXPORTS || nglib_EXPORTS
-    #define DLL_HEADER NGCORE_API_EXPORT
-  #else
-    #define DLL_HEADER NGCORE_API_IMPORT
-  #endif
-#endif
-
-
 // max number of nodes per element
 #define NG_ELEMENT_MAXPOINTS 20
 

libsrc/include/nginterface_v2.hpp

@@ -8,7 +8,7 @@
 /* Date:   May  09                                                        */
 /**************************************************************************/
 
-#include <core/ngcore.hpp>
+#include "mydefs.hpp"
 
 /*
   C++ interface to Netgen
@@ -276,7 +276,7 @@ namespace netgen
     void UpdateTopology ();
     void DoArchive (Archive & archive);
 
-    NgMPI_Comm GetCommunicator() const;
+    const NgMPI_Comm & GetCommunicator() const;
     
     virtual ~Ngx_Mesh();
 
@@ -343,8 +343,11 @@ namespace netgen
       void SetRefinementFlag (size_t elnr, bool flag);
     
     void Curve (int order);
+    int GetCurveOrder ();
 
-    void Refine (NG_REFINEMENT_TYPE reftype,
+    void EnableTable (string name, bool set);
+
+    void Refine (NG_REFINEMENT_TYPE reftype, bool onlyonce,
                  void (*taskmanager)(function<void(int,int)>) = &DummyTaskManager2,
                  void (*tracer)(string, bool) = &DummyTracer2);
 
@@ -354,6 +357,10 @@ namespace netgen
     int GetParentElement (int ei) const;
     int GetParentSElement (int ei) const;
 
+    bool HasParentEdges() const;
+    tuple<int, std::array<int,3>> GetParentEdges (int enr) const;
+    tuple<int, std::array<int,4>> GetParentFaces (int fnr) const;
+    
     int GetNIdentifications() const;
     int GetIdentificationType(int idnr) const;
     Ng_Buffer<int[2]> GetPeriodicVertices(int idnr) const;
@@ -367,8 +374,9 @@ namespace netgen
     
 
     // for MPI-parallel
-    std::tuple<int,int*> GetDistantProcs (int nodetype, int locnum) const;
+    FlatArray<int> GetDistantProcs (int nodetype, int locnum) const;
-
+    size_t GetGlobalVertexNum (int locnum) const;
+                               
     shared_ptr<Mesh> GetMesh () const { return mesh; } 
     shared_ptr<Mesh> SelectMesh () const;
     inline auto GetTimeStamp() const;

libsrc/include/nginterface_v2_impl.hpp

@@ -111,7 +111,13 @@ NGX_INLINE DLL_HEADER Ng_Element Ngx_Mesh :: GetElement<1> (size_t nr) const
   ret.faces.num = 0;
   ret.faces.ptr = NULL;
 
-  if (mesh->GetDimension() == 2)
+  if (mesh->GetDimension() == 3)
+    {
+      ret.facets.num = 0;
+      ret.facets.base = 0;
+      ret.facets.ptr = nullptr;
+    }
+  else if (mesh->GetDimension() == 2)
     {
       ret.facets.num = 1;
       ret.facets.base = 0;
@@ -330,6 +336,20 @@ NGX_INLINE void Ngx_Mesh :: GetParentNodes (int ni, int * parents) const
     parents[0] = parents[1] = -1;
 }
 
+inline bool Ngx_Mesh :: HasParentEdges() const
+{
+  return mesh->GetTopology().HasParentEdges();
+}
+
+inline tuple<int, std::array<int,3>> Ngx_Mesh :: GetParentEdges (int enr) const
+{
+  return mesh->GetTopology().GetParentEdges(enr);
+}
+
+inline tuple<int, std::array<int,4>> Ngx_Mesh :: GetParentFaces (int fnr) const
+{
+  return mesh->GetTopology().GetParentFaces(fnr);
+}
 
 
 inline auto Ngx_Mesh :: GetTimeStamp() const { return mesh->GetTimeStamp(); }

libsrc/interface/CMakeLists.txt

@@ -4,10 +4,10 @@ add_library(interface ${NG_LIB_TYPE}
         read_fnf_mesh.cpp readtetmesh.cpp readuser.cpp writeabaqus.cpp writediffpack.cpp 
         writedolfin.cpp writeelmer.cpp writefeap.cpp writefluent.cpp writegmsh.cpp writejcm.cpp 
         writepermas.cpp writetecplot.cpp writetet.cpp writetochnog.cpp writeuser.cpp
-        wuchemnitz.cpp writegmsh2.cpp writeOpenFOAM15x.cpp 
+        wuchemnitz.cpp writegmsh2.cpp writeOpenFOAM15x.cpp rw_cgns.cpp
         )
 
-target_link_libraries(interface mesh csg geom2d stl visual)
+target_link_libraries(interface PUBLIC mesh csg geom2d stl visual PRIVATE netgen_cgns)
 
 if(NOT WIN32)
   install( TARGETS interface ${NG_INSTALL_DIR})

libsrc/interface/nginterface.cpp

@@ -560,7 +560,7 @@ char * Ng_GetSurfaceElementBCName (int ei)
   if ( mesh->GetDimension() == 3 )
     return const_cast<char *>(mesh->GetFaceDescriptor(mesh->SurfaceElement(ei).GetIndex()).GetBCName().c_str());
   else
-    return const_cast<char *>(mesh->LineSegment(ei).GetBCName().c_str());
+    return const_cast<char *>(mesh->GetBCName(mesh->LineSegment(ei).si).c_str());
 }
 
 
@@ -905,6 +905,7 @@ void Ng_GetSurfaceElementNeighbouringDomains(const int selnr, int & in, int & ou
 
 // gibt anzahl an distant pnums zurueck
 // * pnums entspricht ARRAY<int[2] >
+[[deprecated("Use GetDistantNodeNums(locnum) -> FlatArray instead!")]]                    
 int NgPar_GetDistantNodeNums ( int nodetype, int locnum, int * distnums )
 {
   int size = NgPar_GetNDistantNodeNums (nodetype, locnum);
@@ -931,6 +932,7 @@ int NgPar_GetDistantNodeNums ( int nodetype, int locnum, int * distnums )
   return size;
 }
 
+[[deprecated("Use GetDistantNodeNums(locnum) -> FlatArray instead!")]]                    
 int NgPar_GetNDistantNodeNums ( int nodetype, int locnum )
 {
   locnum++;
@@ -944,6 +946,7 @@ int NgPar_GetNDistantNodeNums ( int nodetype, int locnum )
   return -1;
 }
 
+[[deprecated("Use GetDistantNodeNums(locnum) -> FlatArray instead!")]]                    
 int NgPar_GetGlobalNodeNum (int nodetype, int locnum)
 {
   locnum++;

libsrc/interface/nginterface_v2.cpp

@@ -61,8 +61,12 @@ namespace netgen
     SetGlobalMesh (mesh);
   }
 
-  NgMPI_Comm Ngx_Mesh :: GetCommunicator() const
+  const NgMPI_Comm & Ngx_Mesh :: GetCommunicator() const
-  { return Valid() ? mesh->GetCommunicator() : NgMPI_Comm{}; }
+  {
+    // return Valid() ? mesh->GetCommunicator() : NgMPI_Comm{};
+    if (!Valid()) throw Exception("Ngx_mesh::GetCommunicator: don't have a valid mesh");
+    return mesh->GetCommunicator();
+  }
 
   void Ngx_Mesh :: SaveMesh (ostream & ost) const
   {
@@ -775,29 +779,23 @@ namespace netgen
 
 
 
-#ifdef __SSE__
-#include <immintrin.h>
-  
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<1,1> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     cout << "multi-eltrafo simd called, 1,1,simd" << endl;
   }
 
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<2,2> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     mesh->GetCurvedElements().CalcMultiPointSurfaceTransformation<2>
-      (elnr, npts,
+      (elnr, npts, xi, sxi, x, sx, dxdxi, sdxdxi);
-       reinterpret_cast<const SIMD<double>*> (xi), sxi,
-       reinterpret_cast<SIMD<double>*> (x), sx,
-       reinterpret_cast<SIMD<double>*> (dxdxi), sdxdxi);
     /*
     for (int i = 0; i < npts; i++)
       {
@@ -824,15 +822,15 @@ namespace netgen
 
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<3,3> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     mesh->GetCurvedElements().CalcMultiPointElementTransformation
       (elnr, npts,
-       reinterpret_cast<const SIMD<double>*> (xi), sxi,
+       xi, sxi,
-       reinterpret_cast<SIMD<double>*> (x), sx,
+       x, sx,
-       reinterpret_cast<SIMD<double>*> (dxdxi), sdxdxi);
+       dxdxi, sdxdxi);
     /*
     for (int i = 0; i < npts; i++)
       {
@@ -859,33 +857,43 @@ namespace netgen
 
   template<> DLL_HEADER void Ngx_Mesh ::
   MultiElementTransformation<0,2> (int elnr, int npts,
-				   const tAVXd *xi, size_t sxi,
+				   const SIMD<double> *xi, size_t sxi,
-				   tAVXd * x, size_t sx,
+				   SIMD<double> * x, size_t sx,
-				   tAVXd * dxdxi, size_t sdxdxi) const
+				   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
-    cout << "MultiElementtransformation<0,2> simd not implemented" << endl;
+    //cout << "MultiElementtransformation<0,2> simd not implemented" << endl;
+
+    PointIndex pi = mesh->pointelements[elnr].pnum;
+    Point<3> xg = mesh->Point(pi);
+    if (x)
+      for (int j = 0; j < npts; j++)
+	for (int i = 0; i < 2; i++)
+	  x[j*sx+i] = xg(i);
   }
   
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<0,1> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
-    cout << "multi-eltrafo simd called, 0,1,simd" << endl;
+    //cout << "multi-eltrafo simd called, 0,1,simd" << endl;
+    PointIndex pi = mesh->pointelements[elnr].pnum;
+    Point<3> xg = mesh->Point(pi);
+    if (x)
+      for (int j = 0; j < npts; j++)
+	for (int i = 0; i < 1; i++)
+	  x[j*sx+i] = xg(i);
   }
 
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<1,3> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     mesh->GetCurvedElements().CalcMultiPointSegmentTransformation<3>
-      (elnr, npts,
+      (elnr, npts, xi, sxi, x, sx, dxdxi, sdxdxi);
-       reinterpret_cast<const SIMD<double>*> (xi), sxi,
-       reinterpret_cast<SIMD<double>*> (x), sx,
-       reinterpret_cast<SIMD<double>*> (dxdxi), sdxdxi);
     /*
     double hxi[4][1];
     double hx[4][3];
@@ -908,15 +916,12 @@ namespace netgen
   
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<1,2> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     mesh->GetCurvedElements().CalcMultiPointSegmentTransformation<2>
-      (elnr, npts,
+      (elnr, npts, xi, sxi, x, sx, dxdxi, sdxdxi);
-       reinterpret_cast<const SIMD<double>*> (xi), sxi,
-       reinterpret_cast<SIMD<double>*> (x), sx,
-       reinterpret_cast<SIMD<double>*> (dxdxi), sdxdxi);
     /*
     for (int i = 0; i < npts; i++)
       {
@@ -943,15 +948,12 @@ namespace netgen
 
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<2,3> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     mesh->GetCurvedElements().CalcMultiPointSurfaceTransformation<3>
-      (elnr, npts,
+      (elnr, npts, xi, sxi, x, sx, dxdxi, sdxdxi);
-       reinterpret_cast<const SIMD<double>*> (xi), sxi,
-       reinterpret_cast<SIMD<double>*> (x), sx,
-       reinterpret_cast<SIMD<double>*> (dxdxi), sdxdxi);
     /*
     for (int i = 0; i < npts; i++)
       {
@@ -978,9 +980,9 @@ namespace netgen
 
   template<> DLL_HEADER void Ngx_Mesh :: 
   MultiElementTransformation<0,3> (int elnr, int npts,
-                                   const tAVXd * xi, size_t sxi,
+                                   const SIMD<double> * xi, size_t sxi,
-                                   tAVXd * x, size_t sx,
+                                   SIMD<double> * x, size_t sx,
-                                   tAVXd * dxdxi, size_t sdxdxi) const
+                                   SIMD<double> * dxdxi, size_t sdxdxi) const
   {
     for (int i = 0; i < npts; i++)
       {
@@ -999,10 +1001,6 @@ namespace netgen
       }
   }
 
-  
-#endif
-  
-
 
 
   
@@ -1037,6 +1035,14 @@ namespace netgen
       case 2:
         {
           Point<3> p(hp[0], hp[1],0);
+          try
+            {
+              auto ind = mesh->GetSurfaceElementOfPoint(p, lami, nullptr,
+                                                        build_searchtree);
+              return ind - 1;
+            }
+          catch(NgException e) // quads not implemented curved yet
+            {
           for (SegmentIndex si = 0; si < mesh->GetNSeg(); si++)
             {
               auto & seg = (*mesh)[si];
@@ -1060,6 +1066,7 @@ namespace netgen
                   return si;
                 }
             }
+            }
         }
         break;
       case 3:
@@ -1134,7 +1141,18 @@ namespace netgen
     mesh->BuildCurvedElements(order);
   }
 
+  int Ngx_Mesh :: GetCurveOrder ()
+  {
+    return mesh->GetCurvedElements().GetOrder();
+  }
 
+  void Ngx_Mesh :: EnableTable (string name, bool set)
+  {
+    mesh->GetTopology().EnableTable (name, set);
+  }
+
+
+  
   template <>
   DLL_HEADER void Ngx_Mesh :: SetRefinementFlag<2> (size_t elnr, bool flag)
   {
@@ -1147,9 +1165,9 @@ namespace netgen
     mesh->VolumeElement(elnr+1).SetRefinementFlag(flag);    
   }
   
-  void Ngx_Mesh :: Refine (NG_REFINEMENT_TYPE reftype,
+  void Ngx_Mesh :: Refine (NG_REFINEMENT_TYPE reftype, bool onlyonce,
                            void (*task_manager)(function<void(int,int)>),
-                           Tracer tracer)
+                           NgTracer tracer)
   {
     NgLock meshlock (mesh->MajorMutex(), 1);
     
@@ -1157,6 +1175,7 @@ namespace netgen
     biopt.usemarkedelements = 1;
     biopt.refine_p = 0;
     biopt.refine_hp = 0;
+    biopt.onlyonce = onlyonce;
     if (reftype == NG_REFINE_P)
       biopt.refine_p = 1;
     if (reftype == NG_REFINE_HP)
@@ -1288,33 +1307,35 @@ void Ngx_Mesh::SetSurfaceElementOrders (int enr, int ox, int oy)
 
   
 
+size_t Ngx_Mesh :: GetGlobalVertexNum (int locnum) const
+{
+#ifdef PARALLEL  
+  return mesh->GetParallelTopology().GetGlobalPNum (locnum+1)-1;
+#else
+  return locnum;
+#endif
+}
 
   
-  std::tuple<int,int*>  Ngx_Mesh :: GetDistantProcs (int nodetype, int locnum) const
+FlatArray<int>  Ngx_Mesh :: GetDistantProcs (int nodetype, int locnum) const
   {
 #ifdef PARALLEL
+    if (mesh->GetCommunicator().Size() == 1)
+      return FlatArray<int>(0,nullptr);
+    
     switch (nodetype)
       {
       case 0:
-	{
+        return mesh->GetParallelTopology().GetDistantPNums(locnum);
-	  NgFlatArray<int> dn = mesh->GetParallelTopology().GetDistantPNums(locnum);
-	  return std::tuple<int,int*>(dn.Size(), &dn[0]);
-	}
       case 1:
-	{
+        return mesh->GetParallelTopology().GetDistantEdgeNums(locnum);
-	  NgFlatArray<int> dn = mesh->GetParallelTopology().GetDistantEdgeNums(locnum);
-	  return std::tuple<int,int*>(dn.Size(), &dn[0]);
-	}
       case 2:
-	{
+        return mesh->GetParallelTopology().GetDistantFaceNums(locnum);
-	  NgFlatArray<int> dn = mesh->GetParallelTopology().GetDistantFaceNums(locnum);
-	  return std::tuple<int,int*>(dn.Size(), &dn[0]);
-	}
       default:
-	return std::tuple<int,int*>(0,nullptr);
+	return FlatArray<int>(0, nullptr);
       }
 #else
-    return std::tuple<int,int*>(0,nullptr);
+    return FlatArray<int>(0,nullptr);
 #endif
   }
 }