Merge branch 'master' into hines/nmodlrandom-simplified

CMakeLists.txt

@@ -74,6 +74,8 @@ option(NRN_ENABLE_RX3D "Enable rx3d support" ${NRN_ENABLE_RX3D_DEFAULT})
 option(NRN_ENABLE_CORENEURON "Enable CoreNEURON support" ${NRN_ENABLE_CORENEURON_DEFAULT})
 option(NRN_ENABLE_BACKTRACE "Enable pretty-printed backtraces" ${NRN_ENABLE_BACKTRACE_DEFAULT})
 option(NRN_ENABLE_TESTS "Enable unit tests" ${NRN_ENABLE_TESTS_DEFAULT})
+option(NRN_ENABLE_MATH_OPT "Enable extra math optimisations (to enable SIMD)"
+       ${NRN_ENABLE_MATH_OPT_DEFAULT})
 set(NRN_ENABLE_MODEL_TESTS
     "${NRN_ENABLE_MODEL_TESTS_DEFAULT}"
     CACHE STRING "Comma-separated list of detailed models to enable tests of.")
@@ -94,11 +96,11 @@ option(NRN_ENABLE_MOD_COMPATIBILITY "Enable CoreNEURON compatibility for MOD fil
        ${NRN_ENABLE_MOD_COMPATIBILITY_DEFAULT})
 option(NRN_ENABLE_REL_RPATH "Use relative RPATH in binaries. for relocatable installs/Python"
        ${NRN_ENABLE_REL_RPATH_DEFAULT})
-option(NRN_WHEEL_BUILD ${NRN_WHEEL_BUILD_DEFAULT})
+option(NRN_BINARY_DIST_BUILD ${NRN_BINARY_DIST_BUILD_DEFAULT})
 option(NRN_WHEEL_STATIC_READLINE "Use static readline libraries for the wheels."
        ${NRN_WHEEL_STATIC_READLINE_DEFAULT})
 mark_as_advanced(NRN_ENABLE_REL_RPATH)
-mark_as_advanced(NRN_WHEEL_BUILD)
+mark_as_advanced(NRN_BINARY_DIST_BUILD)
 
 # =============================================================================
 # Build options (string)
@@ -214,7 +216,7 @@ include(cmake/modules/FindPythonModule.cmake)
 include(cmake/Coverage.cmake)
 
 # set CMAKE_BUILD_TYPE and associated flags using allowableBuildTypes and CMAKE_BUILD_TYPE_DEFAULT
-set(allowableBuildTypes Custom Debug Release RelWithDebInfo Fast)
+set(allowableBuildTypes Custom Debug Release RelWithDebInfo Fast FastDebug)
 include(ReleaseDebugAutoFlags)
 
 # Try and emit an intelligent warning if the version number currently set in the CMake project(...)
@@ -929,7 +931,7 @@ if(BUILD_TYPE_UPPER MATCHES "CUSTOM")
 else()
   set(COMPILER_FLAGS "${CMAKE_CXX_FLAGS_${BUILD_TYPE_UPPER}}")
 endif()
-string(JOIN " " COMPILER_FLAGS "${COMPILER_FLAGS}" ${NRN_COMPILE_FLAGS})
+string(JOIN " " COMPILER_FLAGS "${COMPILER_FLAGS}" ${NRN_COMPILE_FLAGS} ${CMAKE_CXX_FLAGS})
 
 message(STATUS "")
 message(STATUS "Configured NEURON ${PROJECT_VERSION}")

ci/win_build_cmake.sh

@@ -29,6 +29,7 @@ cd $BUILD_SOURCESDIRECTORY/build
 	-DNRN_ENABLE_PYTHON=ON  \
 	-DNRN_ENABLE_RX3D=ON  \
 	-DNRN_RX3D_OPT_LEVEL=2 \
+	-DNRN_BINARY_DIST_BUILD=ON \
 	-DPYTHON_EXECUTABLE=/c/Python38/python.exe \
 	-DNRN_ENABLE_PYTHON_DYNAMIC=ON  \
 	-DNRN_PYTHON_DYNAMIC='c:/Python38/python.exe;c:/Python39/python.exe;c:/Python310/python.exe;c:/Python311/python.exe'  \

cmake/BuildOptionDefaults.cmake

@@ -28,6 +28,7 @@ set(NRN_ENABLE_REL_RPATH_DEFAULT OFF)
 set(NRN_AVOID_ABSOLUTE_PATHS_DEFAULT OFF)
 set(NRN_NMODL_CXX_FLAGS_DEFAULT "-O0")
 set(NRN_SANITIZERS_DEFAULT "")
+set(NRN_ENABLE_MATH_OPT_DEFAULT OFF)
 
 # Some distributions may set the prefix. To avoid errors, unset it
 set(NRN_PYTHON_DYNAMIC_DEFAULT "")
@@ -43,7 +44,7 @@ set(PYTHON_EXECUTABLE_DEFAULT "")
 set(IV_LIB_DEFAULT "")
 
 # For wheel deployment
-set(NRN_WHEEL_BUILD_DEFAULT OFF)
+set(NRN_BINARY_DIST_BUILD_DEFAULT OFF)
 set(NRN_WHEEL_STATIC_READLINE_DEFAULT OFF)
 
 # we add some coreneuron options in order to check support like GPU

cmake/CompilerFlagsHelpers.cmake

@@ -14,7 +14,8 @@ set(SUPPORTED_COMPILER_LANGUAGE_LIST "C;CXX")
 foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
   if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "XL")
     set(CMAKE_${COMPILER_LANGUAGE}_COMPILER_IS_XLC ON)
-  elseif(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "Intel")
+  elseif(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "Intel"
+         OR CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "IntelLLVM")
     set(CMAKE_${COMPILER_LANGUAGE}_COMPILER_IS_ICC ON)
   elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
     set(CMAKE_${COMPILER_LANGUAGE}_COMPILER_IS_MSVC)
@@ -27,6 +28,10 @@ foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
   endif()
 endforeach()
 
+set(UNSAFE_MATH_FLAG
+    "-ffinite-math-only -fno-math-errno -funsafe-math-optimizations -fno-associative-math")
+set(FASTDEBUG_FLAG "-g -O1")
+
 # Set optimization flags for each compiler
 foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
 
@@ -39,6 +44,7 @@ foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NONE "-O0")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NORMAL "-O2")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_FAST "-O3")
+    set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG ${FASTDEBUG_FLAG})
     set(CMAKE_${COMPILER_LANGUAGE}_STACK_PROTECTION "-qstackprotect")
     set(CMAKE_${COMPILER_LANGUAGE}_POSITION_INDEPENDENT "-qpic=small")
     set(CMAKE_${COMPILER_LANGUAGE}_VECTORIZE "-qhot")
@@ -59,9 +65,11 @@ foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NONE "-O0")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NORMAL "-O2")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_FAST "-O3")
+    set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG ${FASTDEBUG_FLAG})
     set(CMAKE_${COMPILER_LANGUAGE}_STACK_PROTECTION "-fstack-protector")
     set(CMAKE_${COMPILER_LANGUAGE}_POSITION_INDEPENDENT "-fPIC")
     set(CMAKE_${COMPILER_LANGUAGE}_VECTORIZE "-ftree-vectorize")
+    set(CMAKE_${COMPILER_LANGUAGE}_UNSAFE_MATH ${UNSAFE_MATH_FLAG})
     set(IGNORE_UNKNOWN_PRAGMA_FLAGS "-Wno-unknown-pragmas")
 
     if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_VERSION VERSION_GREATER "4.7.0")
@@ -82,6 +90,11 @@ foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NONE "-O0")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NORMAL "-O2")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_FAST "-O3")
+    if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "IntelLLVM")
+      set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG "${FASTDEBUG_FLAG} -fp-model precise")
+    else()
+      set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG "${FASTDEBUG_FLAG} -fp-model consistent")
+    endif()
     set(CMAKE_${COMPILER_LANGUAGE}_STACK_PROTECTION "-fstack-protector")
     set(CMAKE_${COMPILER_LANGUAGE}_POSITION_INDEPENDENT "-fpic")
     set(CMAKE_${COMPILER_LANGUAGE}_VECTORIZE "")
@@ -94,11 +107,18 @@ foreach(COMPILER_LANGUAGE ${SUPPORTED_COMPILER_LANGUAGE_LIST})
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NONE "-O0")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_NORMAL "-O2")
     set(CMAKE_${COMPILER_LANGUAGE}_OPT_FAST "-O3")
+    set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG "-g -O1")
     set(CMAKE_${COMPILER_LANGUAGE}_STACK_PROTECTION "")
     set(CMAKE_${COMPILER_LANGUAGE}_POSITION_INDEPENDENT "-fPIC")
     set(CMAKE_${COMPILER_LANGUAGE}_VECTORIZE "")
     if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "PGI")
       set(CMAKE_${COMPILER_LANGUAGE}_WARNING_ALL "")
     endif()
+    if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "Clang")
+      set(CMAKE_${COMPILER_LANGUAGE}_UNSAFE_MATH ${UNSAFE_MATH_FLAG})
+    endif()
+    if(CMAKE_${COMPILER_LANGUAGE}_COMPILER_ID STREQUAL "NVHPC")
+      set(CMAKE_${COMPILER_LANGUAGE}_OPT_FASTDEBUG "${FASTDEBUG_FLAG} -fno-omit-frame-pointer")
+    endif()
   endif()
 endforeach()

cmake/ReleaseDebugAutoFlags.cmake

@@ -21,6 +21,9 @@ endif()
 # Release : Release mode, no debuginfo
 # RelWithDebInfo : Distribution mode, basic optimizations for portable code with debuginfos
 # Fast : Maximum level of optimization. Target native architecture, not portable code
+# FastDebug: Similar to Debug with a bit higher level optimisations (-O1) and other compiler
+#            flags so that it's faster than -O0 but still produces consistent results for
+#            testing and debugging purposes.
 # ~~~
 
 include(CompilerFlagsHelpers)
@@ -33,17 +36,56 @@ set(CMAKE_CXX_FLAGS_DEBUG
     "${CMAKE_CXX_DEBUGINFO_FLAGS}  ${CMAKE_CXX_OPT_NONE} ${CMAKE_CXX_STACK_PROTECTION} ${CMAKE_CXX_IGNORE_WARNINGS}"
 )
 
-set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_OPT_NORMAL} ${CMAKE_C_IGNORE_WARNINGS}")
-set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_OPT_NORMAL} ${CMAKE_CXX_IGNORE_WARNINGS}")
+set(C_UNSAFE_MATH_FLAGS "")
+set(CXX_UNSAFE_MATH_FLAGS "")
+if(NRN_ENABLE_MATH_OPT)
+  set(C_UNSAFE_MATH_FLAGS ${CMAKE_C_UNSAFE_MATH})
+  set(CXX_UNSAFE_MATH_FLAGS ${CMAKE_CXX_UNSAFE_MATH})
+endif()
+
+set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_OPT_NORMAL} ${CMAKE_C_IGNORE_WARNINGS} ${C_UNSAFE_MATH_FLAGS}")
+set(CMAKE_CXX_FLAGS_RELEASE
+    "${CMAKE_CXX_OPT_NORMAL} ${CMAKE_CXX_IGNORE_WARNINGS} ${CXX_UNSAFE_MATH_FLAGS}")
 
 set(CMAKE_C_FLAGS_RELWITHDEBINFO
-    "${CMAKE_C_DEBUGINFO_FLAGS}  ${CMAKE_C_OPT_NORMAL} ${CMAKE_C_IGNORE_WARNINGS}")
+    "${CMAKE_C_DEBUGINFO_FLAGS}  ${CMAKE_C_OPT_NORMAL} ${CMAKE_C_IGNORE_WARNINGS} ${C_UNSAFE_MATH_FLAGS}"
+)
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO
-    "${CMAKE_CXX_DEBUGINFO_FLAGS}  ${CMAKE_CXX_OPT_NORMAL} ${CMAKE_CXX_IGNORE_WARNINGS}")
+    "${CMAKE_CXX_DEBUGINFO_FLAGS}  ${CMAKE_CXX_OPT_NORMAL} ${CMAKE_CXX_IGNORE_WARNINGS} ${CXX_UNSAFE_MATH_FLAGS}"
+)
 
 set(CMAKE_C_FLAGS_FAST
     "${CMAKE_C_OPT_FAST} ${CMAKE_C_LINK_TIME_OPT} ${CMAKE_C_GEN_NATIVE} ${CMAKE_C_IGNORE_WARNINGS}")
 set(CMAKE_CXX_FLAGS_FAST
     "${CMAKE_CXX_OPT_FAST} ${CMAKE_CXX_LINK_TIME_OPT} ${CMAKE_CXX_GEN_NATIVE} ${CMAKE_CXX_IGNORE_WARNINGS}"
 )
+
+set(CMAKE_C_FLAGS_FASTDEBUG "${CMAKE_C_OPT_FASTDEBUG} ${CMAKE_C_IGNORE_WARNINGS}")
+set(CMAKE_CXX_FLAGS_FASTDEBUG "${CMAKE_CXX_OPT_FASTDEBUG} ${CMAKE_CXX_IGNORE_WARNINGS}")
 # ~~~
+
+# for binary distributions, avoid addition of OpenMP specific flag as compiler on end-user machine
+# may not support it.
+if(NOT DEFINED NRN_BINARY_DIST_BUILD OR NOT NRN_BINARY_DIST_BUILD)
+  include(CheckCXXCompilerFlag)
+  # Check support for OpenMP SIMD constructs
+  set(SIMD_FLAGS "")
+
+  if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Intel")
+    set(SIMD_FLAGS "-qopenmp-simd")
+  elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
+    set(SIMD_FLAGS "-openmp:experimental")
+  elseif(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+    set(SIMD_FLAGS "-openmp-simd")
+  else() # not ICC, MSVC, or Clang => GCC and others
+    set(SIMD_FLAGS "-fopenmp-simd")
+  endif()
+
+  check_cxx_compiler_flag("${SIMD_FLAGS}" COMPILER_SUPPORT_SIMD)
+  if(COMPILER_SUPPORT_SIMD)
+    set(CMAKE_C_FLAGS "${CMAKE_CXX_FLAGS} ${SIMD_FLAGS}")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SIMD_FLAGS}")
+  else()
+    message(STATUS "The compiler ${CMAKE_CXX_COMPILER} has no support for OpenMP SIMD construct")
+  endif()
+endif()

docs/cmake_doc/options.rst

@@ -12,14 +12,14 @@ together using the following instructions:
 .. code-block:: shell
 
    git clone https://github.com/neuronsimulator/nrn           # latest development branch
-   git clone https://github.com/neuronsimulator/nrn -b 8.0.0  # specific release version 8.0.0
+   git clone https://github.com/neuronsimulator/nrn -b 8.2.3  # specific release version 8.2.3
    cd nrn
 
 ..
 
    .. warning:: To build NEURON from source you either need to clone the
    NEURON Git repository or download a source code archive that includes
-   Git submodules, such as the ``full-src-package-X.Y.Z.tar.gz`` file in
+   Git submodules, such as the ``nrn-full-src-package-X.Y.Z.tar.gz`` file in
    the `NEURON
    releases <https://github.com/neuronsimulator/nrn/releases>`__ on
    GitHub. The tarballs like ``Source code (tar.gz)`` or
@@ -662,3 +662,15 @@ NRN_PYTHON_EXTRA_FOR_TESTS:STRING=
   built with support for, for use in tests of error messages and reporting.
   For these purposes, minor versions (3.X and 3.Y) are considered different
   and patch versions (3.8.X and 3.8.Y) are considered to be the same.
+
+NRN_ENABLE_MATH_OPT:BOOL=OFF
+-------------------------------------
+  Enable extra math optimisations.
+
+  When using compilers like GCC and Clang, one needs to explicitly use compiler
+  flags like `-funsafe-math-optimizations` in order to generate SIMD/vectorised
+  code using vector math library. This flag adds these extra compiler flags
+  to enable SIMD code.
+
+  Note: Compilers like Intel, NVHPC, Cray etc enable such optimisations
+  by default.

packaging/python/build_wheels.bash

@@ -91,7 +91,7 @@ build_wheel_linux() {
 
     CMAKE_DEFS="NRN_MPI_DYNAMIC=$3"
     if [ "$USE_STATIC_READLINE" == "1" ]; then
-      CMAKE_DEFS="$CMAKE_DEFS,NRN_WHEEL_BUILD=ON,NRN_WHEEL_STATIC_READLINE=ON"
+      CMAKE_DEFS="$CMAKE_DEFS,NRN_BINARY_DIST_BUILD=ON,NRN_WHEEL_STATIC_READLINE=ON"
     fi
 
     if [ "$2" == "coreneuron" ]; then
@@ -148,7 +148,7 @@ build_wheel_osx() {
 
     CMAKE_DEFS="NRN_MPI_DYNAMIC=$3"
     if [ "$USE_STATIC_READLINE" == "1" ]; then
-      CMAKE_DEFS="$CMAKE_DEFS,NRN_WHEEL_BUILD=ON,NRN_WHEEL_STATIC_READLINE=ON"
+      CMAKE_DEFS="$CMAKE_DEFS,NRN_BINARY_DIST_BUILD=ON,NRN_WHEEL_STATIC_READLINE=ON"
     fi
 
     # We need to "fix" the platform tag if the Python installer is universal2

src/coreneuron/CMakeLists.txt

@@ -318,7 +318,7 @@ endif()
 # Do not set this when building wheels. The nrnivmodl workflow means that we do not know what
 # compiler will be invoked with these flags, so we have to use flags that are as generic as
 # possible.
-if(NOT DEFINED NRN_WHEEL_BUILD OR NOT NRN_WHEEL_BUILD)
+if(NOT DEFINED NRN_BINARY_DIST_BUILD OR NOT NRN_BINARY_DIST_BUILD)
   list(APPEND CORENRN_EXTRA_CXX_FLAGS "${IGNORE_UNKNOWN_PRAGMA_FLAGS}")
 endif()