New manual/gpu

configure

@@ -660,6 +660,8 @@ CPPFLAGS
 LDFLAGS
 CXXFLAGS
 CXX
+PLUMED_ACC_GPU
+PLUMED_ACC_TYPE
 MPIEXEC
 BASH_COMPLETION_DIR
 PYTHON_BIN
@@ -764,6 +766,8 @@ LDSHARED
 PYTHON_BIN
 BASH_COMPLETION_DIR
 MPIEXEC
+PLUMED_ACC_TYPE
+PLUMED_ACC_GPU
 CXX
 CXXFLAGS
 LDFLAGS
@@ -1455,7 +1459,9 @@ Optional Features:
   --enable-af_cpu         enable search for arrayfire_cpu, default: no
   --enable-libtorch       enable search for libtorch, default: no
   --enable-libmetatomic   enable search for libmetatomic, default: no
-  --enable-openacc        enable search for openacc, default: no
+  --enable-openacc        enable search for openacc (use PLUMED_ACC_TYPE and
+                          PLUMED_ACC_GPU for controlling the settings),
+                          default: no
   --disable-openmp        do not use OpenMP
 
 Some influential environment variables:
@@ -1472,6 +1478,12 @@ Some influential environment variables:
   MPIEXEC     command to run mpi programs in tests - default not specified,
               which means use PLUMED_MPIRUN env var at runtime for backward
               compatibility
+  PLUMED_ACC_TYPE
+              with --enable-openacc this is the setting ('host' or 'gpu') for
+              determining the target of the OPENACC pragmas - default: 'gpu'
+  PLUMED_ACC_GPU
+              if PLUMED_ACC_TYPE is 'gpu', this is read as a space separated
+              list of '-gpu' options for openacc (example "cc75 managed")
   CXX         C++ compiler command
   CXXFLAGS    C++ compiler flags
   LDFLAGS     linker flags, e.g. -L<lib dir> if you have libraries in a
@@ -3267,6 +3279,7 @@ fi
 
 
 
+
 # by default use -O flag
 # we override the autoconf default (-g) because in release build we do not want to
 # include symbol information (obj files are huge)
@@ -10169,7 +10182,9 @@ fi
 
 acc_found=ko
 if test $openacc = true; then
-  acc_found=ko
+  if test -z "$PLUMED_ACC_TYPE"; then
+    PLUMED_ACC_TYPE=gpu
+  fi
 # optional libraries follow
 
     found=ko
@@ -10358,9 +10373,9 @@ rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
 
 
   save_CXXFLAGS="$CXXFLAGS"
-  CXXFLAGS="$CXXFLAGS -acc=gpu"
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX accepts -acc=gpu" >&5
-$as_echo_n "checking whether $CXX accepts -acc=gpu... " >&6; }
+  CXXFLAGS="$CXXFLAGS -acc=$PLUMED_ACC_TYPE"
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX accepts -acc=$PLUMED_ACC_TYPE" >&5
+$as_echo_n "checking whether $CXX accepts -acc=$PLUMED_ACC_TYPE... " >&6; }
   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 
@@ -10402,18 +10417,61 @@ $as_echo "no" >&6; }; CXXFLAGS="$save_CXXFLAGS"
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
 
-    #the cc must be an user input!!!!!!!!!!!!!!!!!!!!!
-    #use CXXFLAGS="-O3 -gpu=cc80" at configure time
-    #PLUMED_CHECK_CXXFLAG([-gpu=cc75])
-    #PLUMED_CHECK_CXXFLAG([-gpu=ccall])
-    #PLUMED_CHECK_CXXFLAG([-gpu=rdc])
-    #PLUMED_CHECK_CXXFLAG([-stdpar])
-    #PLUMED_CHECK_CXXFLAG([-gpu=managed])
-    #PLUMED_CHECK_LDFLAGS([-static-nvidia])
-    #PLUMED_CHECK_LDFLAGS([-acc])
+    if test "$PLUMED_ACC_TYPE" = "gpu"; then
+      if test -n "$PLUMED_ACC_GPU"; then
+        for flag in $PLUMED_ACC_GPU; do
+
+  save_CXXFLAGS="$CXXFLAGS"
+  CXXFLAGS="$CXXFLAGS -gpu=$flag"
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX accepts -gpu=$flag" >&5
+$as_echo_n "checking whether $CXX accepts -gpu=$flag... " >&6; }
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_compile "$LINENO"; then :
+
+      cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_link "$LINENO"; then :
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: not linking" >&5
+$as_echo "not linking" >&6; }; CXXFLAGS="$save_CXXFLAGS"
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+
+else
+  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }; CXXFLAGS="$save_CXXFLAGS"
+
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+
+        done
+      fi
+    fi
     LDSHARED="$LDSHARED -acc -static-nvidia"
-    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Please remember to add \"CXXFLAGS\" the compute capability flag of your device, for example \"-gpu=cc80\"" >&5
-$as_echo "$as_me: WARNING: Please remember to add \"CXXFLAGS\" the compute capability flag of your device, for example \"-gpu=cc80\"" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: Please remember to personalize the openacc settings with  \"PLUMED_ACC_GPU\"" >&5
+$as_echo "$as_me: WARNING: Please remember to personalize the openacc settings with  \"PLUMED_ACC_GPU\"" >&2;}
   fi
 else
   acc_found=ko
@@ -11313,8 +11371,8 @@ if test $openacc = true; then
     { $as_echo "$as_me:${as_lineno-$LINENO}: **** PLUMED will be compiled using OPENACC" >&5
 $as_echo "$as_me: **** PLUMED will be compiled using OPENACC" >&6;}
   else
-    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: **** PLUMED will NOT be compiled using MPI because OPENACC have not been found!" >&5
-$as_echo "$as_me: WARNING: **** PLUMED will NOT be compiled using MPI because OPENACC have not been found!" >&2;}
+    { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: **** PLUMED will NOT be compiled using OPENACC because OPENACC have not been found!" >&5
+$as_echo "$as_me: WARNING: **** PLUMED will NOT be compiled using OPENACC because OPENACC have not been found!" >&2;}
   fi
 else
     { $as_echo "$as_me:${as_lineno-$LINENO}: **** PLUMED will be compiled without OPENACC" >&5

configure.ac

@@ -324,15 +324,16 @@ PLUMED_CONFIG_ENABLE([af_cuda],[search for arrayfire_cuda],[no])
 PLUMED_CONFIG_ENABLE([af_cpu],[search for arrayfire_cpu],[no])
 PLUMED_CONFIG_ENABLE([libtorch],[search for libtorch],[no]) #added by luigibonati
 PLUMED_CONFIG_ENABLE([libmetatomic],[search for libmetatomic],[no])
-PLUMED_CONFIG_ENABLE([openacc],[search for openacc],[no])
+PLUMED_CONFIG_ENABLE([openacc],[search for openacc (use PLUMED_ACC_TYPE and PLUMED_ACC_GPU for controlling the settings)],[no])
 
 AC_ARG_VAR(SOEXT,[extension of dynamic libraries (so/dylib)])
 AC_ARG_VAR(STATIC_LIBS,[variables that should be linked statically directly to MD code - configure will add here -ldl if necessary ])
 AC_ARG_VAR(LDSHARED,[command for linking shared library - configure will use CXX plus the proper flags ])
 AC_ARG_VAR(PYTHON_BIN,[python executable (e.g. python2.7 or /opt/local/bin/python2.7) - default: search for a python executable])
 AC_ARG_VAR(BASH_COMPLETION_DIR,[path where bash completion will be installed - default: search with pkg-config])
 AC_ARG_VAR(MPIEXEC,[command to run mpi programs in tests - default not specified, which means use PLUMED_MPIRUN env var at runtime for backward compatibility])
-
+AC_ARG_VAR(PLUMED_ACC_TYPE,[with --enable-openacc this is the setting  ('host' or 'gpu') for determining the target of the OPENACC pragmas - default: 'gpu'])
+AC_ARG_VAR(PLUMED_ACC_GPU,[if PLUMED_ACC_TYPE is 'gpu', this is read as a space separated list of '-gpu' options for openacc (example "cc75 managed")])
 # by default use -O flag
 # we override the autoconf default (-g) because in release build we do not want to
 # include symbol information (obj files are huge)
@@ -1004,25 +1005,25 @@ fi
 
 acc_found=ko
 if test $openacc = true; then
-  acc_found=ko
+  if test -z "$PLUMED_ACC_TYPE"; then
+    PLUMED_ACC_TYPE=gpu
+  fi
 # optional libraries follow
   PLUMED_CHECK_PACKAGE([openacc.h],[acc_get_device_type],[__PLUMED_HAS_OPENACC])
   if test "$__PLUMED_HAS_OPENACC" = yes; then
     acc_found=ok
     #test if we can compile with openacc
     PLUMED_CHECK_CXXFLAG([-acc])
-    PLUMED_CHECK_CXXFLAG([-acc=gpu])
-    #the cc must be an user input!!!!!!!!!!!!!!!!!!!!!
-    #use CXXFLAGS="-O3 -gpu=cc80" at configure time
-    #PLUMED_CHECK_CXXFLAG([-gpu=cc75])
-    #PLUMED_CHECK_CXXFLAG([-gpu=ccall])
-    #PLUMED_CHECK_CXXFLAG([-gpu=rdc])
-    #PLUMED_CHECK_CXXFLAG([-stdpar])
-    #PLUMED_CHECK_CXXFLAG([-gpu=managed])
-    #PLUMED_CHECK_LDFLAGS([-static-nvidia])
-    #PLUMED_CHECK_LDFLAGS([-acc])
+    PLUMED_CHECK_CXXFLAG([-acc=$PLUMED_ACC_TYPE])
+    if test "$PLUMED_ACC_TYPE" = "gpu"; then
+      if test -n "$PLUMED_ACC_GPU"; then
+        for flag in $PLUMED_ACC_GPU; do
+          PLUMED_CHECK_CXXFLAG([-gpu=$flag])
+        done
+      fi
+    fi
     LDSHARED="$LDSHARED -acc -static-nvidia"
-    AC_MSG_WARN([Please remember to add "CXXFLAGS" the compute capability flag of your device, for example "-gpu=cc80"])
+    AC_MSG_WARN([Please remember to personalize the openacc settings with  "PLUMED_ACC_GPU"])
   fi
 else
   acc_found=ko
@@ -1584,7 +1585,7 @@ if test $openacc = true; then
   if test $acc_found = ok; then
     AC_MSG_NOTICE([**** PLUMED will be compiled using OPENACC])
   else
-    AC_MSG_WARN([**** PLUMED will NOT be compiled using MPI because OPENACC have not been found!])
+    AC_MSG_WARN([**** PLUMED will NOT be compiled using OPENACC because OPENACC have not been found!])
   fi
 else
     AC_MSG_NOTICE([**** PLUMED will be compiled without OPENACC])

new-manual/gpu.md

@@ -5,3 +5,48 @@ command. GPU parallelism in PLUMED has been implemented using [openACC](https://
 on these features at the moment. __There is thus no guarantee that the GPU accelerated versions of actions are any faster than 
 the CPU versions.__ If you have experimented with these features on your own calculations we would love to hear from you (even 
 if your experience was negative.)
+
+## [Experimental] Compiling plumed with openacc
+
+_This section will likely be moved  in the proper installation page_
+
+To compile PLUMED with openacc enabled you will need to have the [NVIDIA HPC SDK](https://developer.nvidia.com/hpc-sdk) avaiable on your path.
+
+Plumed is tested with the [24.3](https://developer.nvidia.com/nvidia-hpc-sdk-243-downloads) version.
+
+To prepare the compilation add `--enable-opeacc` to the `./configure` options.
+It is also possible to pass some extra options by exporting or specifying the the following variables:
+ - **PLUMED_ACC_TYPE**: if omitted defaults to `gpu`, can be changed to `host` or `multicore` to try a compilation that targets the CPU also for the openacc accellerate part of the code.
+ - **PLUMED_ACC_GPU**: if `PLUMED_ACC_TYPE` is set to `gpu`, it can be used to specify a range of `-gpu` optios to pass to the nvhpc compiler (for example the target gpu, see the [compiler manual](https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html) for the options) options can be comma separated or space separated
+
+
+!!! warning
+    (Currently) modules that use a custom openmp reduction can be compiled with nvhpc.
+    Currently `membranefusion` is not compatible and should be excluded from the compilation
+
+
+## List of actions that can be called with the USEGPU option:
+
+ - module:
+   - ACTION
+
+ - colvar:
+   - [ANGLE](ANGLE.md)
+   - [DIPOLE](DIPOLE.md)
+   - [DISTANCE](DISTANCE.md)
+   - [PLANE](PLANE.md)
+   - [POSITION](POSITION.md)
+   - [TORSION](TORSION.md)
+ - crystdistrib:
+   - ~~[QUATERNION_BOND_PRODUCT_MATRIX](QUATERNION_BOND_PRODUCT_MATRIX.md)~~ setup, but deactivated
+ - secondarystructure:
+   - [SECONDARY_STRUCTURE_DRMSD](SECONDARY_STRUCTURE_DRMSD.md), and in particular:
+     - [ALPHARMSD](ALPHARMSD.md) only with **TYPE=DRMSD**
+     - [ANTIBETARMSD](ANTIBETARMSD.md) only with **TYPE=DRMSD**
+     - [PARABETARMSD](PARABETARMSD.md) only with **TYPE=DRMSD**
+ - volumes:
+   - [AROUND](AROUND.md)
+   - [INCYLINDER](INCYLINDER.md)
+   - [INSPHERE](INSPHERE.md)
+- adjmat
+   - [CONTACT_MATRIX](CONTACT_MATRIX.md)