Merge branch 'Crusher' of github.com:JamesEMcClure/LBPM-WIA into Crusher
This commit is contained in:
commit
736ab1c101
47
sample_scripts/configure_crusher_cpu
Executable file
47
sample_scripts/configure_crusher_cpu
Executable file
@ -0,0 +1,47 @@
|
||||
#module load cmake/3.21.3
|
||||
#module load PrgEnv-gnu
|
||||
module load PrgEnv-amd
|
||||
module load rocm/4.5.0
|
||||
module load cray-mpich
|
||||
module load cray-hdf5-parallel
|
||||
#module load craype-accel-amd-gfx908
|
||||
|
||||
## These must be set before compiling so the executable picks up GTL
|
||||
export PE_MPICH_GTL_DIR_amd_gfx90a="-L${CRAY_MPICH_ROOTDIR}/gtl/lib"
|
||||
export PE_MPICH_GTL_LIBS_amd_gfx90a="-lmpi_gtl_hsa"
|
||||
|
||||
export LD_LIBRARY_PATH=${CRAY_LD_LIBRARY_PATH}:${LD_LIBRARY_PATH}
|
||||
|
||||
# Need a new version of cmake
|
||||
export CMAKE_DIR=/gpfs/alpine/csc380/proj-shared/LBPM/cmake-3.21.0/bin
|
||||
|
||||
#-I${MPICH_DIR}/include
|
||||
#-L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa
|
||||
|
||||
HIPFLAGS = --amdgpu-target=gfx90a
|
||||
|
||||
# configure
|
||||
rm -rf CMake*
|
||||
${CMAKE_DIR}/cmake \
|
||||
-D CMAKE_BUILD_TYPE:STRING=Release \
|
||||
-D CMAKE_C_COMPILER:PATH=cc \
|
||||
-D CMAKE_CXX_COMPILER:PATH=CC \
|
||||
-D CMAKE_CXX_STANDARD=14 \
|
||||
-D DISABLE_GOLD:BOOL=TRUE \
|
||||
-D DISABLE_LTO:BOOL=TRUE \
|
||||
-D CMAKE_C_FLAGS="-L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -I${HDF5_DIR}/include" \
|
||||
-D CMAKE_CXX_FLAGS="-L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -I${HDF5_DIR}/include" \
|
||||
-D LINK_LIBRARIES="${ROCM_PATH}/lib/libamdhip64.so;${CRAY_MPICH_ROOTDIR}/gtl/lib/libmpi_gtl_hsa.so" \
|
||||
-D USE_HIP=0 \
|
||||
-D CMAKE_HIP_COMPILER_TOOLKIT_ROOT=$ROCM_PATH/hip \
|
||||
-D USE_MPI=1 \
|
||||
-D MPI_SKIP_SEARCH=1 \
|
||||
-D MPIEXEC="srun" \
|
||||
-D USE_HDF5=1 \
|
||||
-D HDF5_DIRECTORY="${HDF5_DIR}" \
|
||||
-D USE_SILO=0 \
|
||||
-D USE_TIMER=0 \
|
||||
-D USE_DOXYGEN:BOOL=false \
|
||||
~/LBPM-WIA
|
||||
|
||||
|
@ -18,7 +18,11 @@ export CMAKE_DIR=/gpfs/alpine/csc380/proj-shared/LBPM/cmake-3.21.0/bin
|
||||
#-I${MPICH_DIR}/include
|
||||
#-L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa
|
||||
|
||||
HIPFLAGS = --amdgpu-target=gfx90a
|
||||
#export HIPFLAGS="--amdgpu-target=gfx90a --save-temps"
|
||||
|
||||
|
||||
#THIS IS HOW TO CHECK FOR SPILLS (example)
|
||||
# hipcc -c -g -ggdb --save-temps Color.hip
|
||||
|
||||
# configure
|
||||
rm -rf CMake*
|
||||
|
@ -270,17 +270,19 @@ int main(int argc, char **argv)
|
||||
//.......................................................................
|
||||
|
||||
//...........................................................................
|
||||
comm.barrier();
|
||||
//comm.barrier();
|
||||
if (rank == 0) cout << "Domain set." << endl;
|
||||
//...........................................................................
|
||||
|
||||
cout << flush;
|
||||
//...........................................................................
|
||||
if (rank==0) printf ("Create ScaLBL_Communicator \n");
|
||||
cout << flush;
|
||||
// Create a communicator for the device (will use optimized layout)
|
||||
ScaLBL_Communicator ScaLBL_Comm(Dm);
|
||||
|
||||
int Npad=(Np/16 + 2)*16;
|
||||
if (rank==0) printf ("Set up memory efficient layout, %i | %i | %i \n", Np, Npad, N);
|
||||
cout << flush;
|
||||
auto neighborList= new int[18*Npad];
|
||||
IntArray Map(Nx,Ny,Nz);
|
||||
Map.fill(-2);
|
||||
@ -291,7 +293,8 @@ int main(int argc, char **argv)
|
||||
//......................device distributions.................................
|
||||
dist_mem_size = Np*sizeof(double);
|
||||
if (rank==0) printf ("Allocating distributions \n");
|
||||
|
||||
cout << flush;
|
||||
|
||||
int *NeighborList;
|
||||
int *dvcMap;
|
||||
double *fq;
|
||||
@ -321,6 +324,9 @@ int main(int argc, char **argv)
|
||||
ScaLBL_DeviceBarrier();
|
||||
delete [] TmpMap;
|
||||
|
||||
if (rank==0) printf("Map is copied to GPU \n");
|
||||
cout << flush;
|
||||
|
||||
//...........................................................................
|
||||
|
||||
/* // Write the communcation structure into a file for debugging
|
||||
@ -352,11 +358,13 @@ int main(int argc, char **argv)
|
||||
fclose(CommFile);
|
||||
*/
|
||||
if (rank==0) printf("Setting the distributions, size = : %i\n", Np);
|
||||
cout << flush;
|
||||
|
||||
//...........................................................................
|
||||
GlobalFlipScaLBL_D3Q19_Init(fq_host, Map, Np, Nx-2, Ny-2, Nz-2, iproc,jproc,kproc,nprocx,nprocy,nprocz);
|
||||
ScaLBL_CopyToDevice(fq, fq_host, 19*dist_mem_size);
|
||||
ScaLBL_DeviceBarrier();
|
||||
comm.barrier();
|
||||
//comm.barrier();
|
||||
//*************************************************************************
|
||||
// First timestep
|
||||
ScaLBL_Comm.SendD3Q19AA(fq); //READ FROM NORMAL
|
||||
@ -376,6 +384,7 @@ int main(int argc, char **argv)
|
||||
int timestep = 0;
|
||||
if (rank==0) printf("********************************************************\n");
|
||||
if (rank==0) printf("No. of timesteps for timing: %i \n", 100);
|
||||
cout << flush;
|
||||
|
||||
//.......create and start timer............
|
||||
double starttime,stoptime,cputime;
|
||||
@ -426,8 +435,11 @@ int main(int argc, char **argv)
|
||||
// communication bandwidth includes both send and recieve
|
||||
if (rank==0) printf("Communication bandwidth (per process)= %f Gbit/sec \n",ScaLBL_Comm.CommunicationCount*64*timestep/1e9);
|
||||
if (rank==0) printf("Aggregated communication bandwidth = %f Gbit/sec \n",nprocs*ScaLBL_Comm.CommunicationCount*64*timestep/1e9);
|
||||
cout << flush;
|
||||
|
||||
}
|
||||
// ****************************************************
|
||||
cout << fflush;
|
||||
comm.barrier();
|
||||
Utilities::shutdown();
|
||||
// ****************************************************
|
||||
|
@ -23,7 +23,7 @@ int main( int argc, char **argv )
|
||||
{
|
||||
|
||||
// Initialize
|
||||
Utilities::startup( argc, argv );
|
||||
Utilities::startup( argc, argv, false );
|
||||
|
||||
{ // Limit scope so variables that contain communicators will free before MPI_Finialize
|
||||
|
||||
@ -198,7 +198,7 @@ int main( int argc, char **argv )
|
||||
|
||||
|
||||
} // Limit scope so variables that contain communicators will free before MPI_Finialize
|
||||
|
||||
cout << flush;
|
||||
Utilities::shutdown();
|
||||
return 0;
|
||||
}
|
||||
|
@ -24,7 +24,7 @@ using namespace std;
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
// Initialize MPI
|
||||
Utilities::startup( argc, argv );
|
||||
Utilities::startup( argc, argv, false );
|
||||
Utilities::MPI comm( MPI_COMM_WORLD );
|
||||
int rank = comm.getRank();
|
||||
int nprocs = comm.getSize();
|
||||
@ -49,6 +49,7 @@ int main(int argc, char **argv)
|
||||
MRT.Initialize(); // initializing the model will set initial conditions for variables
|
||||
MRT.Run();
|
||||
MRT.VelocityField();
|
||||
cout << flush;
|
||||
}
|
||||
Utilities::shutdown();
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user