Add PME for GIST

.github/workflows/merge-gate.yml

@@ -41,19 +41,13 @@ jobs:
     steps:
       - name: Install prerequisite packages
         run: |
-          sudo apt-get install gfortran \
-                               libbz2-dev \
-                               libblas-dev \
-                               liblapack-dev \
-                               libnetcdf-dev \
-                               libfftw3-dev \
-                               netcdf-bin \
-                               clang \
-                               openmpi-bin \
-                               openmpi-common \
-                               libopenmpi-dev \
-                               cmake-data \
-                               cmake
+          sudo apt-get install gfortran
+          sudo apt-get install libbz2-dev
+          sudo apt-get install libblas-dev liblapack-dev
+          sudo apt-get install libfftw3-dev
+          sudo apt-get install clang
+          sudo apt-get install openmpi-bin openmpi-common libopenmpi-dev
+          sudo apt-get install cmake-data cmake
 
       - name: Checkout source code
         uses: actions/checkout@v2
@@ -68,6 +62,15 @@ jobs:
           mkdir -p include && mv AmberTools/src/sander/sander.h include
           mv lib include $HOME
 
+          curl -OL ftp://ftp.unidata.ucar.edu/pub/netcdf/netcdf-4.6.1.tar.gz
+          tar -zxf netcdf-4.6.1.tar.gz
+          cd netcdf-4.6.1
+          ./configure --disable-netcdf-4 --disable-dap --disable-doxygen --prefix=$HOME
+          make -j2
+          make install
+          cd ..
+          export PATH=$HOME/bin:$PATH
+
           if [ $USE_OPENMP = "yes" ]; then
             export OPT="openmp"
             export OMP_NUM_THREADS=4
@@ -92,13 +95,14 @@ jobs:
             cd build
             cmake .. $BUILD_FLAGS -DCOMPILER=${COMPILER^^} -DINSTALL_HEADERS=FALSE \
                      -DCMAKE_INSTALL_PREFIX=$installdir -DCMAKE_LIBRARY_PATH=$HOME/lib \
-                     -DPRINT_PACKAGING_REPORT=TRUE
+                     -DPRINT_PACKAGING_REPORT=TRUE -DNetCDF_LIBRARIES_C=$HOME/lib/libnetcdf.so \
+                     -DNetCDF_INCLUDES=$HOME/include
             make -j2 install
             cd ..
             export PATH=$installdir/bin:$PATH
           else
             export LD_LIBRARY_PATH=$HOME/lib:${LD_LIBRARY_PATH}
-            ./configure ${BUILD_FLAGS} ${COMPILER}
+            ./configure --with-netcdf=$HOME ${BUILD_FLAGS} ${COMPILER}
             make -j2 install
           fi
           cd test && make $TEST_TYPE

configure

@@ -75,16 +75,21 @@ UsageFull() {
   echo "    NVCCFLAGS    : Flags to pass to the nvcc compiler."
   echo "    DBGFLAGS     : Any additional flags to pass to all compilers."
   echo "    SHADER_MODEL : (-cuda) Should be set to 'sm_XX', where XX is CUDA compute architecture."
-  echo "                   SM6.2 = GP10B"
-  echo "                   SM6.1 = GP106 = GTX-1070, GP104 = GTX-1080, GP102 = Titan-X[P]"
-  echo "                   SM6.0 = GP100 / P100 = DGX-1"
-  echo "                   SM5.3 = GM200 [Grid] = M60, M40?"
-  echo "                   SM5.2 = GM200 = GTX-Titan-X, M6000 etc."
-  echo "                   SM5.0 = GM204 = GTX980, 970 etc"
-  echo "                   SM3.7 = GK210 = K80"
-  echo "                   SM3.5 = GK110 = K20[x], K40, GTX780, GTX-Titan, GTX-Titan-Black, GTX-Titan-Z"
-  echo "                   SM3.0 = GK104 = K10, GTX680, 690 etc."
-  echo "                   SM2.0 = All GF variants = C2050, 2075, M2090, GTX480, GTX580 etc."
+  echo "                   sm_86 = GA102, 104, 106, 107"
+  echo "                   sm_80 = GA100"
+  echo "                   sm_75 = Turing"
+  echo "                   sm_72 = GV10B"
+  echo "                   sm_70 = GV100"
+  echo "                   sm_62 = GP10B"
+  echo "                   sm_61 = GP106 = GTX-1070, GP104 = GTX-1080, GP102 = Titan-X[P]"
+  echo "                   sm_60 = GP100 / P100 = DGX-1"
+  echo "                   sm_53 = GM200 [Grid] = M60, M40?"
+  echo "                   sm_52 = GM200 = GTX-Titan-X, M6000 etc."
+  echo "                   sm_50 = GM204 = GTX980, 970 etc"
+  echo "                   sm_37 = GK210 = K80"
+  echo "                   sm_35 = GK110 = K20[x], K40, GTX780, GTX-Titan, GTX-Titan-Black, GTX-Titan-Z"
+  echo "                   sm_30 = GK104 = K10, GTX680, 690 etc."
+  echo "                   sm_20 = All GF variants = C2050, 2075, M2090, GTX480, GTX580 etc."
   echo "  EXPERIMENTAL OPTIONS:"
   echo "    --compile-verbose : Turn on compile details."
   echo "    -profile          : Use Gnu compiler profiling (>= V4.5)*"
@@ -1204,8 +1209,18 @@ SetupLibraries() {
           fi
           lflag=${LIB_FLAG[$i]}
         else
-          # Lib home specified
-          linc="-I$lhome/include"
+          # Lib home specified.
+          # Determine include directory.
+          incdir="$lhome/include"
+          linc="-I$incdir"
+          if [ ! -d "$incdir" ] ; then
+            # include dir is not in the usual place, happens with e.g. some CUDA installs.
+            if [ -d "$lhome/targets/x86_64-linux/include" ] ; then
+              linc="-I$lhome/targets/x86_64-linux/include"
+            else
+              WrnMsg "Include dir $incdir not found. Linking ${LIB_CKEY[$i]} may fail."
+            fi
+          fi
           # Check if architecture-specific lib dir exists. Use that if so.
           lhdir="$lhome/lib"
           ladir="$lhome/lib$NBITS"
@@ -1688,6 +1703,65 @@ SetupMKL() {
     fi
   fi
 }
+# ------------------------------------------------------------------------------
+# Define different shader models/compute architectures and their CUDA limits
+# CUDA version
+#  3.0-3.1    ...
+#  3.2        .........
+#  4.0-4.2    .........
+#  5.X        ...........................
+#  6.0        ...........................
+#  6.5        ...................................................
+#  7.X        ...................................................
+#  8.X        .....................................................................
+#  9.X                    .....................................................................
+# 10.X                    ...........................................................................
+# 11.X                                ...........................................................................
+CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53 sm_60 sm_61 sm_62 sm_70 sm_72 sm_75 sm_80 sm_86'
+
+# SetSupportedSM <major v> <minor v>
+# Set Shader models supported by current cuda version
+SetSupportedSM() {
+  if [ $1 -lt 3 ] ; then
+    Err "CUDA < 3 not supported."
+  fi
+  if [ $1 -eq 3 ] ; then
+    if [ $2 -ge 2 ] ; then
+      CUDA_SM_LIST='sm_20 sm_21'
+    else
+      CUDA_SM_LIST='sm_20'
+    fi
+  elif [ $1 -eq 4 ] ; then
+    CUDA_SM_LIST='sm_20 sm_21'
+  elif [ $1 -eq 5 ] ; then
+    CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35'
+  elif [ $1 -eq 6 ] ; then
+    if [ $2 -ge 5 ] ; then
+      CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53'
+    else
+      CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35'
+    fi
+  elif [ $1 -eq 7 ] ; then
+    CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53'
+  elif [ $1 -eq 8 ] ; then
+    CUDA_SM_LIST='sm_20 sm_21 sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53 sm_60 sm_61 sm_62'
+  elif [ $1 -eq 9 ] ; then
+    CUDA_SM_LIST='sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53 sm_60 sm_61 sm_62 sm_70 sm_72'
+  elif [ $1 -eq 10 ] ; then
+    CUDA_SM_LIST='sm_30 sm_32 sm_35 sm_37 sm_50 sm_52 sm_53 sm_60 sm_61 sm_62 sm_70 sm_72 sm_75'
+  else # >= 11
+    CUDA_SM_LIST='sm_35 sm_37 sm_50 sm_52 sm_53 sm_60 sm_61 sm_62 sm_70 sm_72 sm_75 sm_80 sm_86'
+  fi
+  echo "  Supported shader models: $CUDA_SM_LIST"
+}
+
+# SetCudaArch <sm>
+# Set CUDA_ARCH variable with compute_XX value for given SM
+SetCudaArch() {
+  smversion=${1#sm_}
+  CUDA_ARCH="compute_$smversion"
+  #echo "$1 $CUDA_ARCH"
+}
 
 # ------------------------------------------------------------------------------
 # Check that CUDA_HOME is defined and set up flags for nvcc
@@ -1696,44 +1770,52 @@ SetupCUDA() {
     Err "CUDA_HOME not set. Set CUDA_HOME to point to your NVIDIA tools installation."
   fi
   if [ ! -x "$CUDA_HOME/bin/nvcc" ]; then
-    Err "Error: nvcc cuda compiler not found in $CUDA_HOME/bin"
+    Err "nvcc cuda compiler not found in $CUDA_HOME/bin"
   fi
   if [ -z "$NVCC" ]; then NVCC="$CUDA_HOME/bin/nvcc"; fi
   cuda_version=`$NVCC --version | grep 'release' | cut -d' ' -f5 | cut -d',' -f1`
+  cuda_major_version=`echo "$cuda_version" | awk 'BEGIN{FS=".";}{printf("%i", $1);}'`
+  cuda_minor_version=`echo "$cuda_version" | awk 'BEGIN{FS=".";}{printf("%i", $2);}'`
   echo "  CUDA version $cuda_version detected."
-  SM_CONFIG="Configuring for $SHADER_MODEL"
+  # A zero version indicates version detection failed.
+  if [ $cuda_major_version -lt 1 ] ; then
+    Err "CUDA version detection failed."
+  fi
+  SetSupportedSM $cuda_major_version $cuda_minor_version
+
   if [ -z "$NVCCFLAGS" -a -z "$SHADER_MODEL" ] ; then
+    # Compile for multiple shader models
     WrnMsg "SHADER_MODEL not set. Compiling for multiple architectures."
     WrnMsg "To compile for a specific architecture set SHADER_MODEL"
     WrnMsg "to 'sm_XX', where XX is the shader model version."
-    # NOTE: From AmberTools configure2
-    #Note at present we do not include SM3.5 or SM3.7 since they sometimes show performance
-    #regressions over just using SM3.0.
-    # TODO fix for volta?
-    sm70flags='-gencode arch=compute_60,code=sm_70'
-    sm62flags='-gencode arch=compute_62,code=sm_62'
-    sm61flags='-gencode arch=compute_61,code=sm_61'
-    sm60flags='-gencode arch=compute_60,code=sm_60'
-    sm53flags='-gencode arch=compute_53,code=sm_53'
-    sm52flags='-gencode arch=compute_52,code=sm_52'
-    sm50flags='-gencode arch=compute_50,code=sm_50'
-    sm37flags='-gencode arch=compute_37,code=sm_37'
-    sm35flags='-gencode arch=compute_35,code=sm_35'
-    sm30flags='-gencode arch=compute_30,code=sm_30'
-    sm20flags='-gencode arch=compute_20,code=sm_20'
-    if [ "$cuda_version" = '9.0' -o "$cuda_version" = '9.1' -o "$cuda_version" = '9.2' -o "$cuda_version" = "10.0" -o "$cuda_version" = "10.1" ] ; then
-      SM_CONFIG="Configuring for SM3.0, SM5.0, SM5.2, SM5.3, SM6.0, SM6.1, and SM7.0"
-      NVCCFLAGS="$sm30flags $sm50flags $sm52flags $sm53flags $sm60flags $sm61flags $sm70flags"
-    elif [ "$cuda_version" = '8.0' ] ; then
-      SM_CONFIG="Configuring for SM2.0, SM3.0, SM5.0, SM5.2, SM5.3, SM6.0 and SM6.1"
-      NVCCFLAGS="$sm20flags $sm30flags $sm50flags $sm52flags $sm53flags $sm60flags $sm61flags"
-    else
-      SM_CONFIG="Configuring for SM2.0, SM3.0, SM5.0, SM5.2 and SM5.3"
-      echo "BE AWARE: CUDA < 8.0 does not support GTX-1080, Titan-XP, DGX-1 or other Pascal based GPUs."
-      NVCCFLAGS="$sm20flags $sm30flags $sm50flags $sm52flags $sm53flags"
+    # TODO determine why Amber has arch=compute_60 for 70 and 75
+    SM_CONFIG="Configuring for"
+    NVCCFLAGS="$DBFLAG"
+    # Loop over supported shader models for this CUDA
+    for sm in $CUDA_SM_LIST ; do
+      SetCudaArch $sm
+      SM_CONFIG="$SM_CONFIG $sm"
+      NVCCFLAGS="$NVCCFLAGS -gencode arch=$CUDA_ARCH,code=$sm"
+    done
+  elif [ -z "$NVCCFLAGS" -a ! -z "$SHADER_MODEL" ] ; then
+    # Compile for single shader model
+    SM_CONFIG="Configuring for $SHADER_MODEL"
+    # See if it is supported.
+    sm_is_supported=0
+    for sm in $CUDA_SM_LIST ; do
+      if [ "$sm" = "$SHADER_MODEL" ] ; then
+        sm_is_supported=1
+        break
+      fi
+    done
+    if [ $sm_is_supported -eq 0 ] ; then
+      Err "Shader model $SHADER_MODEL is not supported by CUDA $cuda_version"
     fi
+    NVCCFLAGS="$DBFLAG -arch=$SHADER_MODEL"
+  else
+    # Use specified NVCC flags
+    SM_CONFIG="Using NVCCFLAGS: $NVCCFLAGS"
   fi
-  if [ -z "$NVCCFLAGS" ]; then NVCCFLAGS="$DBFLAG -arch=$SHADER_MODEL"; fi
   if [ ! -z "$picflag" ] ; then
     NVCCFLAGS="--compiler-options $picflag $NVCCFLAGS"
   fi