Gpu code

Makefile.am

@@ -74,6 +74,23 @@ bin_PROGRAMS += fof_mpi
 endif
 endif
 
+# BUILD CUDA versions as well?
+if HAVECUDA
+bin_PROGRAMS += swift_cuda
+if HAVEMPI
+bin_PROGRAMS += swift_mpicuda
+endif
+endif
+
+
+# BUILD HIP versions as well?
+if HAVEHIP
+bin_PROGRAMS += swift_hip
+if HAVEMPI
+bin_PROGRAMS += swift_mpihip
+endif
+endif
+
 # engine_policy_setaffinity is available?
 if HAVESETAFFINITY
 ENGINE_POLICY_SETAFFINITY=| engine_policy_setaffinity
@@ -91,6 +108,28 @@ swift_mpi_SOURCES = swift.c
 swift_mpi_CFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(MPI_FLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)"
 swift_mpi_LDADD =  src/libswiftsim_mpi.la argparse/libargparse.la $(MPI_LIBS) $(VELOCIRAPTOR_MPI_LIBS) $(EXTRA_LIBS) $(LD_CSDS)
 
+# Sources for swift_cuda
+swift_cuda_SOURCES = swift.c dummy.C
+swift_cuda_CXXFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(CUDA_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_CUDA
+swift_cuda_LDADD =  src/.libs/libswiftsim_cuda.a src/cuda/.libs/libswiftCUDA.a $(EXTRA_LIBS) $(CUDA_LIBS) -lcudart argparse/.libs/libargparse.a src/.libs/libgrav.la
+
+# Sources for swift_hip
+swift_hip_SOURCES = swift.c dummy.C
+swift_hip_CXXFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(HIP_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_HIP
+swift_hip_LDADD =  src/.libs/libswiftsim_hip.a src/hip/.libs/libswiftHIP.a $(EXTRA_LIBS) $(HIP_LIBS) -lamdhip64 -L/opt/rocm-5.1.0/lib -lhsa-runtime64 -L/opt/rocm-5.1.0/lib64 -lamd_comgr argparse/.libs/libargparse.a src/.libs/libgrav.la
+
+# Sources for swift_mpicuda, do we need an affinity policy for MPI?
+swift_mpicuda_SOURCES = swift.c dummy.C
+swift_mpicuda_CXXFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(MPI_FLAGS) $(CUDA_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_CUDA
+swift_mpicuda_CFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(MPI_FLAGS) $(CUDA_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_CUDA
+swift_mpicuda_LDADD =  src/.libs/libswiftsim_mpicuda.a argparse/.libs/libargparse.a src/.libs/libgrav.la src/cuda/.libs/libswiftCUDA.a $(MPI_LIBS) $(EXTRA_LIBS) $(CUDA_LIBS) -lcudart
+
+# Sources for swift_mpihip, do we need an affinity policy for MPI?
+swift_mpihip_SOURCES = swift.c dummy.C
+swift_mpihip_CXXFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(MPI_FLAGS) $(HIP_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_HIP
+swift_mpihip_CFLAGS = $(MYFLAGS) $(AM_CFLAGS) $(MPI_FLAGS) $(HIP_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)" -DWITH_HIP
+swift_mpihip_LDADD =  src/.libs/libswiftsim_mpihip.a argparse/.libs/libargparse.a src/.libs/libgrav.la src/hip/.libs/libswiftHIP.a $(MPI_LIBS) $(EXTRA_LIBS) $(HIP_LIBS) -lamdhip64
+
 # Sources for fof
 fof_SOURCES = swift_fof.c
 fof_CFLAGS = $(MYFLAGS) $(AM_CFLAGS) -DENGINE_POLICY="engine_policy_keep $(ENGINE_POLICY_SETAFFINITY)"

configure.ac

@@ -41,6 +41,10 @@ AC_USE_SYSTEM_EXTENSIONS
 AC_PROG_CC
 AM_PROG_CC_C_O
 
+# Find and test the C++ compiler.
+AC_PROG_CXX
+AC_PROG_CXX_C_O
+
 # We need this for compilation hints and possibly FFTW.
 AX_OPENMP
 
@@ -995,6 +999,78 @@ AH_VERBATIM([__STDC_FORMAT_MACROS],
 #define __STDC_FORMAT_MACROS 1
 #endif])
 
+
+
+# Check for CUDA
+have_cuda="no"
+AC_ARG_WITH([cuda],
+    [AS_HELP_STRING([--with-cuda=PATH],
+       [root directory where CUDA is installed @<:@yes/no@:>@]
+    )],
+    [],
+    [with_cuda="no"]
+)
+if test "x$with_cuda" != "xno"; then
+   if test "x$with_cuda" != "xyes"; then
+      CUDA_CFLAGS="-I$with_cuda/include"
+      CUDA_LIBS="-L$with_cuda/lib -L$with_cuda/lib64 -lcudart"
+      NVCC="$with_cuda/bin/nvcc"
+      have_cuda="yes"
+   else
+      AC_PATH_PROG([NVCC],[nvcc])
+      echo "Found nvcc = $NVCC"
+      if test -n "$NVCC"; then
+         CUDA_ROOT="`dirname $NVCC`/.."
+         CUDA_CFLAGS="-I${CUDA_ROOT}/include"
+         CUDA_LIBS="-L${CUDA_ROOT}/lib -L${CUDA_ROOT}/lib64 -lcudart"
+         have_cuda="yes"
+      fi
+   fi
+   if test "x$have_cuda" != "xno"; then
+      AC_DEFINE([HAVE_CUDA], 1, [The CUDA compiler is installed.])
+   fi
+   CFLAGS="${CFLAGS} "
+fi
+AC_SUBST(CUDA_CFLAGS)
+AC_SUBST(CUDA_LIBS)
+AC_SUBST(NVCC)
+AM_CONDITIONAL([HAVECUDA],[test -n "$NVCC"])
+
+# Check for HIP
+have_hip="no"
+AC_ARG_WITH([hip],
+    [AS_HELP_STRING([--with-hip=PATH],
+       [root directory where HIP is installed @<:@yes/no@:>@]
+    )],
+    [],
+    [with_hip="no"]
+)
+if test "x$with_hip" != "xno"; then
+   if test "x$with_hip" != "xyes"; then
+      HIP_CFLAGS="-I$with_hip/include"
+      HIP_LIBS="-L$with_hip/lib -L$with_hip/lib64"
+      HIPCC="$with_hip/bin/hipcc"
+      have_hip="yes"
+   else
+      AC_PATH_PROG([HIPCC],[hipcc])
+      echo "Found hipcc = $HIPCC"
+      if test -n "$HIPCC"; then
+         HIP_ROOT="`dirname $HIPCC`/.."
+         HIP_CFLAGS="-I${HIP_ROOT}/include"
+         HIP_LIBS="-L${HIP_ROOT}/lib -L${HIP_ROOT}/lib64"
+         have_hip="yes"
+      fi
+   fi
+   if test "x$have_hip" != "xno"; then
+      AC_DEFINE([HAVE_HIP], 1, [The HIP compiler is installed.])
+   fi
+   CFLAGS="${CFLAGS} "
+fi
+AC_SUBST(HIP_CFLAGS)
+AC_SUBST(HIP_LIBS)
+AC_SUBST(HIPCC)
+AM_CONDITIONAL([HAVEHIP],[test -n "$HIPCC"])
+
 # Check for FFTW. We test for this in the standard directories by default,
 # and only disable if using --with-fftw=no or --without-fftw. When a value
 # is given FFTW must be found.
@@ -3246,6 +3322,10 @@ AC_CONFIG_FILES([tests/testSelectOutput.sh], [chmod +x tests/testSelectOutput.sh
 AC_CONFIG_FILES([tests/testFormat.sh], [chmod +x tests/testFormat.sh])
 AC_CONFIG_FILES([tests/testNeutrinoCosmology.sh], [chmod +x tests/testNeutrinoCosmology.sh])
 AC_CONFIG_FILES([tests/output_list_params.yml])
+# cuda .in file
+AC_CONFIG_FILES([src/cuda/Makefile])
+# hip .in file
+AC_CONFIG_FILES([src/hip/Makefile])
 
 # Save the compilation options
 AC_DEFINE_UNQUOTED([SWIFT_CONFIG_FLAGS],["$swift_config_flags"],[Flags passed to configure])
@@ -3276,6 +3356,8 @@ AC_MSG_RESULT([
    HDF5 enabled         : $with_hdf5
     - parallel          : $have_parallel_hdf5
    METIS/ParMETIS       : $have_metis / $have_parmetis
+   CUDA enabled         : $have_cuda
+   HIP enabled          : $have_hip
    FFTW3 enabled        : $have_fftw
     - threaded/openmp   : $have_threaded_fftw / $have_openmp_fftw
     - MPI               : $have_mpi_fftw

cudalt.py

@@ -0,0 +1,80 @@
+#!/usr/bin/python3
+# libtoolish hack: compile a .cu file like libtool does
+import sys
+import os
+
+lo_filepath = sys.argv[1]
+o_filepath = lo_filepath.replace(".lo", ".o")
+
+try:
+   i = o_filepath.rindex("/")
+   lo_dir = o_filepath[0:i+1]
+   o_filename = o_filepath[i+1:]
+
+except ValueError:
+   lo_dir = ""
+   o_filename = o_filepath
+
+local_pic_dir = ".libs/"
+local_npic_dir = ""
+pic_dir = lo_dir + local_pic_dir
+npic_dir = lo_dir + local_npic_dir
+
+pic_filepath = pic_dir + o_filename
+npic_filepath = npic_dir + o_filename
+local_pic_filepath = local_pic_dir + o_filename
+local_npic_filepath = local_npic_dir + o_filename
+
+# Make lib dir
+try:
+   os.mkdir(pic_dir)
+except OSError:
+   pass
+
+# generate the command to compile the .cu for shared library
+args = sys.argv[2:]
+args.extend(["-Xcompiler","-fPIC"]) 
+# position indep code
+args.append("-o")
+args.append(pic_filepath)
+command = " ".join(args)
+print (command)
+
+# compile the .cu
+rv = os.system(command)
+if rv != 0:
+    sys.exit(1)
+
+# generate the command to compile the .cu for static library
+args = sys.argv[2:]
+args.append("-o")
+args.append(npic_filepath)
+command = " ".join(args)
+print (command)
+
+# compile the .cu
+rv = os.system(command)
+if rv != 0:
+    sys.exit(1)
+
+# get libtool version
+fd = os.popen("libtool --version")
+libtool_version = fd.readline()
+fd.close()
+
+# generate the .lo file
+f = open(lo_filepath, "w")
+f.write("# " +  lo_filepath + " - a libtool object file\n")
+f.write("# Generated by " + libtool_version + "\n")
+f.write("#\n")
+f.write("# Please DO NOT delete this file!\n")
+f.write("# It is necessary for linking the library.\n\n")
+
+f.write("# Name of the PIC object.\n")
+f.write("pic_object='" + local_pic_filepath + "'\n\n")
+
+f.write("# Name of the non-PIC object.\n")
+f.write("non_pic_object='" + local_npic_filepath + "'\n")
+f.close()
+
+sys.exit(0)

dummy.C

@@ -0,0 +1,3 @@
+void dummy(){
+
+}

examples/HydroTests/GreshoVortex_3D/getGlass.sh

@@ -1,2 +1,2 @@
 #!/bin/bash
-wget http://virgodb.cosma.dur.ac.uk/swift-webstorage/ICs/glassCube_64.hdf5
+wget http://virgodb.cosma.dur.ac.uk/swift-webstorage/ICs/glassCube_128.hdf5

examples/HydroTests/GreshoVortex_3D/gresho.yml

@@ -7,20 +7,24 @@ InternalUnitSystem:
   UnitTemp_in_cgs:     1   # Kelvin
 
 Scheduler:
-  max_top_level_cells: 15
-
+  max_top_level_cells: 8
+  tasks_per_cell: 200
+  deadlock_waiting_time_s:   10
+  cell_split_size: 80
+  cell_sub_size_pair_hydro:  50 # (Optional) Maximal number of hydro-hydro interactions per sub-pair hydro/star task (this is the default value).
+  cell_sub_size_self_hydro:  50     # (Optional) Maximal number of hydro-hydro interactions per sub-self hydro/star task. Set to how many cells are targeted for GPU tasks  
 # Parameters governing the time integration
 TimeIntegration:
   time_begin: 0.    # The starting time of the simulation (in internal units).
   time_end:   1.    # The end time of the simulation (in internal units).
   dt_min:     1e-6  # The minimal time-step size of the simulation (in internal units).
-  dt_max:     1e-2  # The maximal time-step size of the simulation (in internal units).
+  dt_max:     1e-4  # The maximal time-step size of the simulation (in internal units).
 
 # Parameters governing the snapshots
 Snapshots:
   basename:            gresho # Common part of the name of output files
   time_first:          0.     # Time of the first output (in internal units)
-  delta_time:          1e-1   # Time difference between consecutive outputs (in internal units)
+  delta_time:          1e-3   # Time difference between consecutive outputs (in internal units)
   compression:         1
 
 # Parameters governing the conserved quantities statistics
@@ -29,10 +33,11 @@ Statistics:
 
 # Parameters for the hydrodynamics scheme
 SPH:
-  resolution_eta:        1.2348   # Target smoothing length in units of the mean inter-particle separation (1.2348 == 48Ngbs with the cubic spline kernel).
+  resolution_eta:        1.9   # Target smoothing length in units of the mean inter-particle separation (1.2348 == 48Ngbs with the cubic spline kernel).
   CFL_condition:         0.1      # Courant-Friedrich-Levy condition for time integration.
 
 # Parameters related to the initial conditions
 InitialConditions:
   file_name:  ./greshoVortex.hdf5     # The file to read
-  periodic:   1
+  periodic:   1
+#  replicate:  2

src/Makefile.am

@@ -16,7 +16,10 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 # Add the non-standard paths to the included library headers
-AM_CFLAGS = $(HDF5_CPPFLAGS) $(GSL_INCS) $(FFTW_INCS) $(NUMA_INCS) $(GRACKLE_INCS)  $(SUNDIALS_INCS) $(CHEALPIX_CFLAGS)
+AM_CFLAGS = $(HDF5_CPPFLAGS) $(GSL_INCS) $(FFTW_INCS) $(NUMA_INCS) $(GRACKLE_INCS)  $(SUNDIALS_INCS) $(CHEALPIX_CFLAGS) -O0
+
+# Add HIP Path
+AM_CFLAGS += -D__HIP_PLATFORM_AMD__
 
 # Assign a "safe" version number
 AM_LDFLAGS = $(HDF5_LDFLAGS) $(FFTW_LIBS)
@@ -40,6 +43,22 @@ lib_LTLIBRARIES += libswiftsim_mpi.la
 noinst_LTLIBRARIES += libgrav_mpi.la
 endif
 
+# Build a cuda version too?
+if HAVECUDA
+lib_LTLIBRARIES += libswiftsim_cuda.la
+if HAVEMPI
+lib_LTLIBRARIES += libswiftsim_mpicuda.la
+endif
+endif
+
+# Build a hip version too?
+if HAVEHIP
+lib_LTLIBRARIES += libswiftsim_hip.la
+if HAVEMPI
+lib_LTLIBRARIES += libswiftsim_mpihip.la
+endif
+endif
+
 # List required headers
 include_HEADERS = space.h runner.h queue.h task.h lock.h cell.h part.h const.h 
 include_HEADERS += cell_hydro.h cell_stars.h cell_grav.h cell_sinks.h cell_black_holes.h cell_rt.h cell_grid.h
@@ -161,7 +180,7 @@ endif
 AM_SOURCES = space.c space_rebuild.c space_regrid.c space_unique_id.c 
 AM_SOURCES += space_sort.c space_split.c space_extras.c space_first_init.c space_init.c 
 AM_SOURCES += space_cell_index.c space_recycle.c 
-AM_SOURCES += runner_main.c runner_doiact_hydro.c runner_doiact_limiter.c
+AM_SOURCES += runner_main.c runner_doiact_hydro.c runner_doiact_limiter.c runner_gpu_pack_functions.c
 AM_SOURCES += runner_doiact_stars.c runner_doiact_black_holes.c runner_ghost.c
 AM_SOURCES += runner_recv.c runner_pack.c
 AM_SOURCES += runner_sort.c runner_drift.c runner_black_holes.c runner_time_integration.c 
@@ -208,7 +227,7 @@ AM_SOURCES += $(SPHM1RT_RT_SOURCES)
 AM_SOURCES += $(GEAR_RT_SOURCES)
 
 # Include files for distribution, not installation.
-nobase_noinst_HEADERS = align.h approx_math.h atomic.h barrier.h cycle.h error.h inline.h kernel_hydro.h kernel_gravity.h 
+nobase_noinst_HEADERS = align.h approx_math.h atomic.h barrier.h cycle.h error.h inline.h kernel_hydro.h kernel_gravity.h runner_gpu_pack_functions.h
 nobase_noinst_HEADERS += gravity_iact.h kernel_long_gravity.h vector.h accumulate.h cache.h exp.h log.h
 nobase_noinst_HEADERS += runner_doiact_nosort.h runner_doiact_hydro.h runner_doiact_stars.h runner_doiact_black_holes.h runner_doiact_grav.h
 nobase_noinst_HEADERS += runner_doiact_functions_hydro.h runner_doiact_functions_stars.h runner_doiact_functions_black_holes.h 
@@ -526,6 +545,33 @@ libswiftsim_mpi_la_LDFLAGS = $(AM_LDFLAGS) $(MPI_LIBS) $(EXTRA_LIBS) -version-in
 libswiftsim_mpi_la_SHORTNAME = mpi
 libswiftsim_mpi_la_LIBADD = $(GRACKLE_LIBS) $(VELOCIRAPTOR_LIBS) $(MPI_LIBS) libgrav_mpi.la
 
+# Sources and flags for regular CUDA library
+libswiftsim_cuda_la_SOURCES = $(AM_SOURCES)
+libswiftsim_cuda_la_CFLAGS = $(AM_CFLAGS) $(CUDA_CFLAGS) -DWITH_CUDA
+libswiftsim_cuda_la_CXXFLAGS = $(AM_CFLAGS) $(CUDA_CFLAGS) -DWITH_CUDA
+libswiftsim_cuda_la_LDFLAGS = $(AM_LDFLAGS) $(EXTRA_LIBS) $(CUDA_LIBS)
+libswiftsim_cuda_la_SHORTNAME = cuda
+libswiftsim_cuda_la_LIBADD = $(GRACKLE_LIBS) $(VELOCIRAPTOR_LIBS) $(MPI_LIBS) libgrav.la
+
+# Sources and flags for regular HIP library
+libswiftsim_hip_la_SOURCES = $(AM_SOURCES)
+libswiftsim_hip_la_CFLAGS = $(AM_CFLAGS) $(HIP_CFLAGS) -DWITH_HIP
+libswiftsim_hip_la_LDFLAGS = $(AM_LDFLAGS) $(EXTRA_LIBS) $(HIP_LIBS) -lamdhip64
+libswiftsim_hip_la_SHORTNAME = hip
+libswiftsim_hip_la_LIBADD = $(GRACKLE_LIBS) $(VELOCIRAPTOR_LIBS) $(MPI_LIBS) libgrav.la
+
+# Sources and flags for MPI CUDA library
+libswiftsim_mpicuda_la_SOURCES = $(AM_SOURCES)
+libswiftsim_mpicuda_la_CFLAGS = $(AM_CFLAGS) $(MPI_FLAGS) $(CUDA_CFLAGS) -DWITH_CUDA
+libswiftsim_mpicuda_la_CXXFLAGS = $(AM_CFLAGS) $(MPI_FLAGS) $(CUDA_CFLAGS) -DWITH_CUDA
+libswiftsim_mpicuda_la_LDFLAGS = $(AM_LDFLAGS) $(MPI_LIBS) $(EXTRA_LIBS) $(CUDA_LIBS)
+libswiftsim_mpicuda_la_SHORTNAME = mpicuda
+libswiftsim_mpicuda_la_LIBADD = $(GRACKLE_LIBS) $(VELOCIRAPTOR_LIBS) $(MPI_LIBS) libgrav_mpi.la
+
+#subdir
+SUBDIRS = . cuda
+SUBDIRS += . hip
+
 # Versioning. If any sources change then update the version_string.h file with
 # the current git revision and package version.
 # May have a checkout without a version_string.h file and no git command (tar/zip