void divergence(int *n, int nthreads) {
int nprocs, procid;
MPI_Comm_rank(MPI_COMM_WORLD, &procid);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
/* Create Cartesian Communicator */
int c_dims[2]={0};
MPI_Comm c_comm;
accfft_create_comm(MPI_COMM_WORLD,c_dims,&c_comm);
float *data;
Complexf *data_hat;
double f_time=0*MPI_Wtime(),i_time=0, setup_time=0;
int alloc_max=0;
int isize[3],osize[3],istart[3],ostart[3];
/* Get the local pencil size and the allocation size */
alloc_max=accfft_local_size_dft_r2cf(n,isize,istart,osize,ostart,c_comm);
//data=(float*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(float));
data=(float*)accfft_alloc(alloc_max);
data_hat=(Complexf*)accfft_alloc(alloc_max);
accfft_init(nthreads);
/* Create FFT plan */
setup_time=-MPI_Wtime();
accfft_planf * plan=accfft_plan_dft_3d_r2cf(n,data,(float*)data_hat,c_comm,ACCFFT_MEASURE);
setup_time+=MPI_Wtime();
/* Initialize data */
initialize(data,n,c_comm);
MPI_Barrier(c_comm);
float * gradx=(float*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(float));
float * grady=(float*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(float));
float * gradz=(float*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(float));
float * divergence=(float*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(float));
double timings[5]={0};
std::bitset<3> XYZ=0;
XYZ[0]=1;
XYZ[1]=1;
XYZ[2]=1;
double exec_time=-MPI_Wtime();
accfft_gradf(gradx,grady,gradz,data,plan,&XYZ,timings);
accfft_divergencef(divergence,gradx,grady,gradz,plan,timings);
exec_time+=MPI_Wtime();
PCOUT<<">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>"<<std::endl;
PCOUT<<">>>>>Checking Divergence>>>>>>"<<std::endl;
check_err_laplace(divergence,n,c_comm);
PCOUT<<"<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<"<<std::endl;
PCOUT<<"<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n"<<std::endl;
/* Compute some timings statistics */
double g_setup_time,g_timings[5],g_exec_time;
MPI_Reduce(timings,g_timings,5, MPI_DOUBLE, MPI_MAX,0, c_comm);
MPI_Reduce(&setup_time,&g_setup_time,1, MPI_DOUBLE, MPI_MAX,0, c_comm);
MPI_Reduce(&exec_time,&g_exec_time,1, MPI_DOUBLE, MPI_MAX,0, c_comm);
PCOUT<<"Timing for Divergence Computation for size "<<n[0]<<"*"<<n[1]<<"*"<<n[2]<<std::endl;
PCOUT<<"Setup \t\t"<<g_setup_time<<std::endl;
PCOUT<<"Evaluation \t"<<g_exec_time<<std::endl;
accfft_free(data);
accfft_free(data_hat);
MPI_Barrier(c_comm);
accfft_free(gradx);
accfft_free(grady);
accfft_free(gradz);
accfft_destroy_plan(plan);
accfft_cleanup();
MPI_Comm_free(&c_comm);
PCOUT<<"-------------------------------------------------------"<<std::endl;
PCOUT<<"-------------------------------------------------------"<<std::endl;
PCOUT<<"-------------------------------------------------------\n"<<std::endl;
return ;
} // end divergence
void divergence(int *n) {
int nprocs, procid;
MPI_Comm_rank(MPI_COMM_WORLD, &procid);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
/* Create Cartesian Communicator */
int c_dims[2]={0};
MPI_Comm c_comm;
accfft_create_comm(MPI_COMM_WORLD,c_dims,&c_comm);
double f_time=0*MPI_Wtime(),i_time=0, setup_time=0;
int alloc_max=0;
int isize[3],osize[3],istart[3],ostart[3];
/* Get the local pencil size and the allocation size */
alloc_max=accfft_local_size_dft_r2c_gpu(n,isize,istart,osize,ostart,c_comm);
//data=(double*)accfft_alloc(isize[0]*isize[1]*isize[2]*sizeof(double));
double * data_cpu=(double*)accfft_alloc(alloc_max);
double* data;
Complex* data_hat;
cudaMalloc((void**) &data , alloc_max);
cudaMalloc((void**) &data_hat, alloc_max);
accfft_init();
/* Create FFT plan */
setup_time=-MPI_Wtime();
accfft_plan_gpu * plan=accfft_plan_dft_3d_r2c_gpu(n,data,(double*)data_hat,c_comm,ACCFFT_MEASURE);
setup_time+=MPI_Wtime();
/* Initialize data */
initialize(data_cpu,n,c_comm);
cudaMemcpy(data, data_cpu,alloc_max, cudaMemcpyHostToDevice);
MPI_Barrier(c_comm);
double * gradx,*grady, *gradz, *divergence;
cudaMalloc((void**) &gradx , alloc_max);
cudaMalloc((void**) &grady , alloc_max);
cudaMalloc((void**) &gradz , alloc_max);
cudaMalloc((void**) &divergence, alloc_max);
double timings[5]={0};
std::bitset<3> XYZ=0;
XYZ[0]=1;
XYZ[1]=1;
XYZ[2]=1;
double exec_time=-MPI_Wtime();
accfft_grad_gpu(gradx,grady,gradz,data,plan,XYZ,timings);
accfft_divergence_gpu(divergence,gradx,grady,gradz,plan,timings);
exec_time+=MPI_Wtime();
PCOUT<<">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>"<<std::endl;
PCOUT<<">>>>>Checking Divergence>>>>>>"<<std::endl;
cudaMemcpy(data_cpu, divergence, alloc_max, cudaMemcpyDeviceToHost);
check_err_laplace(data_cpu,n,c_comm);
PCOUT<<"<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<"<<std::endl;
PCOUT<<"<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<\n"<<std::endl;
/* Compute some timings statistics */
double g_setup_time,g_timings[5],g_exec_time;
MPI_Reduce(timings,g_timings,5, MPI_DOUBLE, MPI_MAX,0, c_comm);
MPI_Reduce(&setup_time,&g_setup_time,1, MPI_DOUBLE, MPI_MAX,0, c_comm);
MPI_Reduce(&exec_time,&g_exec_time,1, MPI_DOUBLE, MPI_MAX,0, c_comm);
PCOUT<<"Timing for Grad Computation for size "<<n[0]<<"*"<<n[1]<<"*"<<n[2]<<std::endl;
PCOUT<<"Setup \t\t"<<g_setup_time<<std::endl;
PCOUT<<"Evaluation \t"<<g_exec_time<<std::endl;
accfft_free(data_cpu);
cudaFree(data);
cudaFree(data_hat);
MPI_Barrier(c_comm);
cudaFree(gradx);
cudaFree(grady);
cudaFree(gradz);
accfft_destroy_plan(plan);
accfft_cleanup_gpu();
MPI_Comm_free(&c_comm);
PCOUT<<"-------------------------------------------------------"<<std::endl;
PCOUT<<"-------------------------------------------------------"<<std::endl;
PCOUT<<"-------------------------------------------------------\n"<<std::endl;
return ;
} // end divergence
