Merge pull request mfem#1485 from mfem/x86-updates

Some updates to the SIMD intrinsics branch

Author: v-dobrev

.travis.yml

@@ -11,8 +11,6 @@
 
 language: cpp
 
-sudo: false
-
 stages:
   - checks
   - tests
@@ -370,8 +368,10 @@ script:
    # Compiler
    - if [ $MPI == "YES" ]; then
         export MYCXX=mpic++;
+        export MAKE_CXX_FLAG=MPICXX=$MYCXX;
      else
         export MYCXX="$CXX";
+        export MAKE_CXX_FLAG=CXX=$MYCXX;
      fi
 
    # Print the compiler version
@@ -384,12 +384,9 @@ script:
      if [ "$CODECOV" == "YES" ]; then
         CPPFLAGS="--coverage -g";
      fi;
-     if [ "$CXX" == "clang++" ]; then
-        export MFEM_PERF_SW=clang;
-     fi
 
    # Configure the library
-   - make config MFEM_USE_MPI=$MPI MFEM_DEBUG=$DEBUG MFEM_CXX="$MYCXX"
+   - make config MFEM_USE_MPI=$MPI MFEM_DEBUG=$DEBUG $MAKE_CXX_FLAG
         MFEM_MPI_NP=$NPROCS CPPFLAGS="$CPPFLAGS"
    # Show the configuration
    - make info

INSTALL

@@ -396,9 +396,11 @@ MFEM_USE_SIDRE = YES/NO
    blueprint specification. When enabled, this option requires installation of
    HDF5 (see also MFEM_USE_NETCDF), Conduit and LLNL's axom project.
 
-MFEM_USE_SIMD = YES/NO 
-   Enables the high performance templated classes to use specific intrinsics 
-   instead of the AutoSIMD (config/simd/auto.hpp) classe.
+MFEM_USE_SIMD = YES/NO
+   Enables the high performance templated classes to use architecture dependent
+   SIMD intrinsics instead of the generic implementation of class AutoSIMD in
+   linalg/simd/auto.hpp. This option should be combined with suitable
+   compiler options, such as -march=native, to enable optimal vectorization.
 
 MFEM_USE_CONDUIT = YES/NO
    Enables support for converting MFEM Mesh and Grid Function objects to and

config/cmake/MFEMConfig.cmake.in

@@ -47,6 +47,7 @@ set(MFEM_USE_OCCA @MFEM_USE_OCCA@)
 set(MFEM_USE_RAJA @MFEM_USE_RAJA@)
 set(MFEM_USE_CEED @MFEM_USE_CEED@)
 set(MFEM_USE_UMPIRE @MFEM_USE_UMPIRE@)
+set(MFEM_USE_SIMD @MFEM_USE_SIMD@)
 set(MFEM_USE_ADIOS2 @MFEM_USE_ADIOS2@)
 
 set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")

config/cmake/config.hpp.in

@@ -107,6 +107,9 @@
 // Enable MFEM functionality based on the Sidre library
 #cmakedefine MFEM_USE_SIDRE
 
+// Enable the high performance templated classes to use SIMD
+#cmakedefine MFEM_USE_SIMD
+
 // Enable MFEM functionality based on Conduit
 #cmakedefine MFEM_USE_CONDUIT
 

config/cmake/modules/MfemCmakeUtilities.cmake

@@ -733,7 +733,7 @@ function(mfem_export_mk_files)
       MFEM_USE_SUPERLU MFEM_USE_STRUMPACK MFEM_USE_GNUTLS
       MFEM_USE_GSLIB MFEM_USE_NETCDF MFEM_USE_PETSC MFEM_USE_MPFR MFEM_USE_SIDRE
       MFEM_USE_CONDUIT MFEM_USE_PUMI MFEM_USE_CUDA MFEM_USE_OCCA MFEM_USE_RAJA
-      MFEM_USE_UMPIRE)
+      MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADIOS2)
   foreach(var ${CONFIG_MK_BOOL_VARS})
     if (${var})
       set(${var} YES)
@@ -743,6 +743,7 @@ function(mfem_export_mk_files)
   endforeach()
   # TODO: Add support for MFEM_USE_CUDA=YES
   set(MFEM_CXX ${CMAKE_CXX_COMPILER})
+  set(MFEM_HOST_CXX ${MFEM_CXX})
   set(MFEM_CPPFLAGS "")
   string(STRIP "${CMAKE_CXX_FLAGS_${BUILD_TYPE}} ${CMAKE_CXX_FLAGS}"
          MFEM_CXXFLAGS)

config/config.mk.in

@@ -54,6 +54,7 @@ MFEM_USE_ADIOS2        = @MFEM_USE_ADIOS2@
 
 # Compiler, compile options, and link options
 MFEM_CXX       = @MFEM_CXX@
+MFEM_HOST_CXX  = @MFEM_HOST_CXX@
 MFEM_CPPFLAGS  = @MFEM_CPPFLAGS@
 MFEM_CXXFLAGS  = @MFEM_CXXFLAGS@
 MFEM_TPLFLAGS  = @MFEM_TPLFLAGS@

config/defaults.cmake

@@ -49,6 +49,7 @@ option(MFEM_USE_OCCA "Enable OCCA" OFF)
 option(MFEM_USE_RAJA "Enable RAJA" OFF)
 option(MFEM_USE_CEED "Enable CEED" OFF)
 option(MFEM_USE_UMPIRE "Enable Umpire" OFF)
+option(MFEM_USE_SIMD "Enable use of SIMD intrinsics" ON)
 option(MFEM_USE_ADIOS2 "Enable ADIOS2" OFF)
 
 set(MFEM_MPI_NP 4 CACHE STRING "Number of processes used for MPI tests")