GENESIS 1.6.0 with GPU support

ウェブページ

https://www.r-ccs.riken.jp/labs/cbrt/

バージョン

1.6.0

ビルド環境

  • Intel Parallel Studio XE 2018 Update 4
  • CUDA 11.1 Update 1

ビルドに必要なファイル

  • genesis-1.6.0.tar.bz2
  • tests-1.6.0.tar.bz2
  • fortdep.py.patch

--- fortdep.py.org      2020-12-28 14:05:33.000000000 +0900
+++ fortdep.py  2020-12-28 14:39:43.000000000 +0900
@@ -24,6 +24,7 @@
   re_fcomment = re.compile( "(^[^!]*)!(.*$)" )
   re_module   = re.compile( "^(.*;+|\s*)\s*module\s*([^\s,]*)\s*", re.I )
   re_use      = re.compile( "^(.*;+\s*|\s*)use\s+([^\s,]*)\s*", re.I )
+  re_include  = re.compile( "^(\s*)#include\s+['\"]([^'\"]+)['\"]\s*" )
   mod_ext     = ".mod"
 
   def __init__( self, fname = "", ext = ".o" ):
@@ -31,6 +32,7 @@
     self.filename = fname
     self.modules  = []
     self.depmods  = []
+    self.includes = []
 
   def setFilename( self, fname, ext = ".o" ):
     if len(fname) == 0:
@@ -54,6 +56,14 @@
       if ( ma ):
         modname = ma.group(2).lower()
         self.appendDependMods( modname )
+      mi = FortranFile.re_include.search( line )
+      if ( mi ):
+        srcname = mi.group(2)
+        if re.search( r".h$", srcname, re.I ):
+          #self.includes.append( srcname )
+          pass
+        else:
+          self.includes.append( re.sub( r'\.[a-zA-Z0-9]+$', ".o", srcname ) )
     myf.close()
     self.makeUniqList()
 
@@ -64,8 +74,9 @@
     self.depmods.append( modname + FortranFile.mod_ext )
 
   def makeUniqList( self ):
-    self.modules = list(set(self.modules))
-    self.depmods = list(set(self.depmods))
+    self.modules  = list(set(self.modules))
+    self.depmods  = list(set(self.depmods))
+    self.includes = list(set(self.includes))
 
   def getMyModuleFilenames( self ):
     return self.modules
@@ -87,7 +98,8 @@
       if m.lower() in mods_avail or not mods_avail:
         depmods.append(m)
     ret = ""
-    ret += self.objname + ": " + self.filename + " " + " ".join(depmods) + " " + static_deps
+    ret += self.objname + ": " + self.filename + " " + " ".join(depmods) + \
+                          " " + " ".join(self.includes) + " " + static_deps
     if len(self.modules) > 0:
       ret += "\n"
       ret += " ".join(self.modules) + ": " + self.filename + " " + self.objname

ビルド手順

#!/bin/sh

VERSION=1.6.0
BASEDIR=/home/users/${USER}/Software/GENESIS/${VERSION}
SRC_TARBALL=${BASEDIR}/genesis-${VERSION}.tar.bz2
TESTS_TARBALL=${BASEDIR}/tests-${VERSION}.tar.bz2
FORTDEP_PATCH=${BASEDIR}/fortdep.py.patch

INSTALLDIR=/local/apl/lx/genesis160-CUDA

WORKDIR=/work/users/${USER}
BUILDDIR=${WORKDIR}/genesis-${VERSION}
TESTSDIR=${WORKDIR}/tests-${VERSION}

PARALLEL=12
PARALLEL_TESTS=8

# ----------------------------------------------------------------------------
umask 0022

module purge
module load intel_parallelstudio/2018update4
module load cuda/11.1

export LANG=C
export LC_ALL=C
export OMP_NUM_THREADS=1

cd ${WORKDIR}
if [ -d genesis-${VERSION} ]; then
  mv genesis-${VERSION} genesis-erase
  rm -rf genesis-erase &
fi

if [ -d tests-${VERSION} ]; then
  mv tests-${VERSION} tests-erase
  rm -rf tests-erase &
fi

tar jxf ${SRC_TARBALL}
tar jxf ${TESTS_TARBALL}

cd ${BUILDDIR}
patch -p0 < ${FORTDEP_PATCH}
FC=mpiifort CC=mpiicc ./configure --prefix=${INSTALLDIR} \
                                  --enable-gpu \
                                  --enable-single \
                                  --with-cuda=/local/apl/lx/cuda-11.1
make depend

make -j ${PARALLEL} && make install

SPDYN=${INSTALLDIR}/bin/spdyn

cd ${TESTSDIR}/regression_test

# spdyn tests
./test.py       "mpirun -np ${PARALLEL_TESTS} $SPDYN" gpu
./test_remd.py  "mpirun -np ${PARALLEL_TESTS} $SPDYN" gpu
./test_rpath.py "mpirun -np ${PARALLEL_TESTS} $SPDYN" gpu
./test_gamd.py  "mpirun -np ${PARALLEL_TESTS} $SPDYN" gpu
./test_fep.py   "mpirun -np ${PARALLEL_TESTS} $SPDYN" gpu

メモ

  • テストは全て通過(仕様上動作しない test_remd の 9 個のテストを除く)
  • バイナリ(spdyn)は P100, V100 のどちらでも動作します(ccgpup, ccgpuv で確認)。ただし、ネイティブサポートにはなっていない可能性があります。
View PDF