Commit 52710748 authored by Luca Tornatore's avatar Luca Tornatore
Browse files

step in polishing

parent 90e669b8

Makefile

+13 −13
Original line number Diff line number Diff line
@@ -71,18 +71,18 @@ OPT += -DPHASE_ON 
# ========================================================





DEPS = w-stacking.h  main.c phase_correction.cu allvars.h

COBJ = allvars.o main.o init.o gridding.o gridding_std.o fourier_transform.o result.o numa.o reduce.o w-stacking.o phase_correction.o

DEPS = w-stacking.h  main.c allvars.h

OBJ = allvars.o main.o init.o gridding.o gridding_cpu.o fourier_transform.o result.o numa.o reduce.o w-stacking.o phase_correction.o



DEPS_ACC_CUDA = w-stacking.h w-stacking.cu

COBJ_ACC_CUDA = phase_correction.o w-stacking.o

OBJ_ACC_CUDA = phase_correction.o w-stacking.o



DEPS_ACC_OMP = w-stacking_omp.h 

COBJ_ACC_OMP = phase_correction.o w-stacking_omp.o

OBJ_ACC_OMP = phase_correction.o w-stacking_omp.o



COBJ_NCCL_REDUCE = gridding_nccl.o

OBJ_NCCL_REDUCE = gridding_nccl.o



COBJ_RCCL_REDUCE = gridding_rccl.o

OBJ_RCCL_REDUCE = gridding_rccl.o



ifeq (USE_FFTW,$(findstring USE_FFTW,$(OPT)))

CFLAGS += $(FFTW_MPI_INC)
@@ -116,7 +116,7 @@ EXEC_EXT := $(EXEC_EXT)_acc-cuda 
LINKER=$(NVCC)

FLAGS=$(NVFLAGS) $(CFLAGS)

LIBS=$(NVLIB)

compile_cuda: $(COBJ_ACC_CUDA)

compile_cuda: $(OBJ_ACC_CUDA)

	$(NVCC) $(OPT) $(NVFLAGS) -c *.cu $(NVLIB)

endif
@@ -125,7 +125,7 @@ EXEC_EXT := $(EXEC_EXT)_acc-omp 
LINKER=$(NVC)

FLAGS=$(NVFLAGS) $(CFLAGS)

LIBS=$(NVLIB)

compile_accomp: $(COBJ_ACC_OMP)

compile_accomp: $(OBJ_ACC_OMP)

	$(NVC) $(NVFLAGS) $(OPT) -c $^ $(CFLAGS) $(NVLIB)

endif
@@ -134,7 +134,7 @@ EXEC_EXT := $(EXEC_EXT)_acc-reduce 
LINKER=$(NVC++)

FLAGS=$(NVFLAGS) $(CFLAGS)

LIBS=$(NVLIB) $(NVLIB_3)

compile_accreduce: $(COBJ_NCCL_REDUCE)

compile_accreduce: $(OBJ_NCCL_REDUCE)

	$(NVC++) $(NVFLAGS) $(OPT) -c $^ $(CFLAGS) $(NVLIB_3)

endif
@@ -143,17 +143,17 @@ EXEC_EXT := $(EXEC_EXT)_acc-reduce 
LINKER=$(NVC++)

FLAGS=$(NVFLAGS) $(CFLAGS)

LIBS=$(NVLIB) $(NVLIB_3)

compile_accreduce: $(COBJ_RCCL_REDCUE)

compile_accreduce: $(OBJ_RCCL_REDCUE)

	$(NVC++) $(NVFLAGS) $(OPT) -c $^ $(CFLAGS) $(NVLIB_3)

endif





###################################################################################



w-stacking: $(COBJ) $(DEPS) Makefile

	@$(LINKER) $(FLAGS) $(OPT) $(FFTWLIBS) $(LIBS) -lmpi -o $(EXEC)$(EXEC_EXT)

w-stacking: $(OBJ) $(DEPS) Makefile

	$(LINKER) $(FLAGS) $(OPTIMIZE) $(OPT) $(FFTWLIBS) $(LIBS) $(OBJ) -lmpi -lm -o $(EXEC)$(EXEC_EXT)



$(COBJ): $(DEPS) Makefile

$(OBJ): $(DEPS) Makefile



%.o: %.c $(DEPS)

	$(MPICC) $(OPTIMIZE) $(OPT) -c -o $@ $< $(CFLAGS)

gridding_std.c→gridding_cpu.c

+0 −0

File moved.

View file

init.c

+1 −3
Original line number Diff line number Diff line
@@ -5,9 +5,7 @@ 
void init(int index)

{



  double start_tot;

  start_tot = CPU_TIME_wt;

  double begin = CPU_TIME_pr;

  double begin = CPU_TIME_wt;

  

  // DAV: the corresponding KernelLen is calculated within the wstack function. It can be anyway hardcoded for optimization

  dx = 1.0/(double)param.grid_size_x;

main.c

+9 −8
Original line number Diff line number Diff line
@@ -44,6 +44,8 @@ int main(int argc, char * argv[]) 
 

  /* Initializing MPI Environment */



  double time_tot = CPU_TIME_wt;

  

 #ifdef _OPENMP

  {

    int thread_level;
@@ -104,6 +106,7 @@ int main(int argc, char * argv[]) 
    }





  

  for(int ifiles=0; ifiles<param.ndatasets; ifiles++)

    {

      if(rank == 0)
@@ -124,16 +127,14 @@ int main(int argc, char * argv[]) 
      /* WRITE_FFTW_DATA function */

      write_fftw_data();

                



      /* WRITE_RESULT function */

      write_result();



      

      if(rank == 0)

	printf("*************************************************************\n"); 



    }



  /* WRITE_RESULT function */

  timing_wt.total = CPU_TIME_wt - time_tot;

  write_result();

  

  shutdown(0, NULL, 0, 0);

numa.c

+11 −1
Original line number Diff line number Diff line
@@ -24,6 +24,16 @@ void *win_hostmaster_ptr; 


win_t      *win_ctrl;



MPI_Request *requests;

int          thid;

int          Ntasks_local;

blocks_t     blocks;

double     **swins;

int        **cwins;

int          max_level;

double      *end_4, *end_reduce;

int          dsize_4, iter; 



int numa_build_mapping( int, int, MPI_Comm *, map_t *);

int numa_map_hostnames( MPI_Comm *, int, int, map_t *);

int get_cpu_id( void );