int main(int argc, char **argv) {
int heap=300000, stack=300000;
int me, nprocs, i;
double start, end;
/* Initialize Message Passing library */
MPI_Init(&argc, &argv);
/* Initialize GA */
GA_Initialize();
/* Initialize Memory Allocator (MA) */
if(! MA_init(C_DBL, stack, heap) ) GA_Error("MA_init failed",stack+heap);
me = GA_Nodeid();
nprocs = GA_Nnodes();
if(me==0) {
printf("\nUsing %d processes\n\n", nprocs); fflush(stdout);
}
start = MPI_Wtime();
for(i =0; i< 1000; i++) {
TRANSPOSE1D();
}
end = MPI_Wtime();
if(me==0) printf(" Time=%2.5e secs\n\n",end-start);
if(me==0)printf("\nTerminating ..\n");
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char** argv)
{
int nprocs,myid,nprocssq;
int dims[2],chunk[2];
int i,j,k;
int stack = 100000, heap = 100000;
MPI_Init(&argc,&argv);
GA_Initialize();
MA_init(C_DBL,stack,heap);
nprocssq = GA_Nnodes();
nprocs = sqrt(nprocssq);
myid = GA_Nodeid();
dims[0] = N; dims[1] = N;
chunk[0] = N/nprocs;
chunk[1] = N/nprocs;
int g_a = NGA_Create(C_DBL,2,dims,"Array A",chunk);
int lo[2],hi[2];
NGA_Distribution(g_a,myid,lo,hi);
int ld[1] = {N/nprocs};
void *ptr;
double *local;
printf("Myid = %d, lo = [%d,%d] , hi = [%d,%d] , ld = %d \n",myid,lo[0],lo[1],hi[0],hi[1],ld[0]);
NGA_Access(g_a,lo,hi,&ptr,ld);
local = (double*) ptr;
printf("Myid = %d , local[0][0] = %f\n",*local);
GA_Sync();
GA_Destroy(g_a);
GA_Terminate();
MPI_Finalize();
return 0;
}
int main(int argc, char **argv)
{
int rank, nprocs, n=1;
int g_A, dims[DIM]={5,5};
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
printf("check %d \n", n);
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
if(GA_Ndim(g_A)!=DIM)
printf("ERROR: GA_Ndim didnt return nDimension after GA_Initialize\n");
printf("%d : %d \n", rank, GA_Ndim(g_A));
GA_Terminate();
if(rank==0)
printf(" GA: Test Completed \n");
MPI_Finalize();
return 0;
}
int main(int argc, char **argv)
{
int rank, nprocs, n=10;
MPI_Init(&argc, &argv);
GA_Initialize();
MA_init(C_INT, 1000, 1000);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
/*
addition_operator(rank, nprocs, n);
multiply_operator(rank, nprocs, n);
max_operator(rank, nprocs, n);
min_operator(rank, nprocs, n);
absmax_operator(rank, nprocs, n);
absmin_operator(rank, nprocs, n);
*/
checking_operator (rank, nprocs);
if(rank==0)
GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
//irregular_array1(rank);
// irregular_array2(rank);
// auto_number1(rank, nprocs);
auto_number2(rank, nprocs);
if(rank == 0)
printf("Test Completed \n");
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char **argv)
{
int rank, nprocs, i;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
for(i=1; i<=NDIM; i++)
{
verify_ga_dim(i);
}
GA_Sync();
if(rank == 0)
printf("Test Completed \n");
GA_Terminate();
MPI_Finalize();
return 0;
}
// -------------------------------------------------------------
// Environment:: constructors / destructor
// -------------------------------------------------------------
Environment::Environment(int& argc, char **argv,
const long int& ma_stack,
const long int& ma_heap)
: p_boostEnv(argc, argv)
{
GA_Initialize();
MA_init(C_DBL, ma_stack, ma_heap);
}
extern "C" void
gridpack_initialize_parallel(int ma_stack, int ma_heap)
{
int argc(0);
int ierr = MPI_Init(&argc, NULL);
GA_Initialize();
MA_init(C_DBL, ma_stack, ma_heap);
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int p_Geven, p_Godd, p_size, mod, p_size_mod, *list_even=NULL, *list_odd=NULL;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
p_size=nprocs/2;
mod=nprocs%2;
p_size_mod=p_size+mod;
list_even = (int*)malloc(p_size*sizeof(int));
list_odd = (int*)malloc(p_size*sizeof(int));
j=0;
for(i=0; i<nprocs; i++)
{
if(i%2==0)
list_even[j]=i;
else if(i%2==1)
list_odd[j]=i;
else
break;
j++;
}
j=0;
/*
for(i=0; i<nprocs; i++)
{
if(i%2==1)
{
j++;
}
}
*/
p_Geven=GA_Pgroup_create(list_even, p_size_mod);
p_Godd=GA_Pgroup_create(list_odd, p_size);
if(rank%2==0)
printf("%d: My ID is %d :: %d -- even \n", rank, GA_Pgroup_nodeid(p_Geven), GA_Pgroup_nnodes(p_Geven));
else
printf("%d: My ID is %d :: %d --- odd\n", rank, GA_Pgroup_nodeid(p_Godd), GA_Pgroup_nnodes(p_Godd));
GA_Sync();
if(rank==0)
GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char **argv)
{
int status, me;
int max_arrays = 10;
double max_sz = 1e8, max_disk = 1e10, max_mem = 1e6;
int stack = 120000, heap = 3200000;
int numfiles, numioprocs;
int total, nproc;
MP_INIT(argc,argv);
GA_Initialize();
me = GA_Nodeid();
nproc = GA_Nnodes();
total = pow(SIZE,NDIM)*sizeof(double);
if (!GA_Uses_ma()) {
if (GA_Nodeid() == 0) {
printf("GA is not using MA\n");
}
stack = 100000;
heap = (int)(2.2*(float)(total));
}
if (MA_init(MT_F_DBL, stack, heap) ) {
if (DRA_Init(max_arrays, max_sz, max_disk, max_mem) != 0)
GA_Error("DRA_Init failed: ",0);
if (USER_CONFIG == 0) {
numfiles = -1;
numioprocs = -1;
} else if (USER_CONFIG == 1) {
numfiles = 1;
numioprocs = GA_Cluster_nnodes();
} else if (USER_CONFIG == 2) {
numfiles = GA_Cluster_nnodes();
numioprocs = GA_Cluster_nnodes();
} else {
numfiles = 1;
numioprocs = 1;
}
if (me==0) {
printf("Disk resident arrays configured as:\n");
printf(" Number of files: %d\n",numfiles);
printf(" Number of I/O processors: %d\n",numioprocs);
}
DRA_Set_default_config(numfiles,numioprocs);
if (me == 0) printf("\n");
if (me == 0) printf("TESTING PERFORMANCE OF DISK ARRAYS\n");
if (me == 0) printf("\n");
test_io_dbl();
status = DRA_Terminate();
GA_Terminate();
} else {
printf("MA_init failed\n");
}
if(me == 0) printf("all done ...\n");
MP_FINALIZE();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, g_B, g_C, local_C[DIM][DIM], dims[DIM]={5,5};
int val_A=5, val_B=3, ld=DIM, max;
int lo[DIM]={2,2}, hi[DIM]={4,4}, blo[DIM]={0,0}, bhi[DIM]={2,2}, clo[DIM]={1,1}, chi[DIM]={3,3};
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
g_B = NGA_Create(C_INT, DIM, dims, "array_B", NULL);
g_C = NGA_Create(C_INT, DIM, dims, "array_C", NULL);
GA_Fill(g_A, &val_A);
GA_Fill(g_B, &val_B);
GA_Zero(g_C);
GA_Elem_maximum_patch(g_A, lo, hi, g_B, blo, bhi, g_C, clo, chi);
GA_Print(g_C);
GA_Sync();
NGA_Get(g_C, clo, chi, local_C, &ld);
if(rank==1)
{
for(i=0; i<DIM; i++)
{
for(j=0; j<DIM; j++)printf("%d ", local_C[i][j]);
printf("\n");
}
if(val_A>val_B) max=val_A;
else max=val_B;
for(i=0; i<DIM; i++)
{
for(j=0; j<DIM; j++)
if(local_C[i][j]!=max) printf("GA Error : \n");
}
}
GA_Sync();
if(rank == 0)
printf("Test Completed \n");
GA_Terminate();
MPI_Finalize();
}
main(int argc, char **argv)
{
int rank, nprocs;
int g_A, dims[D]={SIZE,SIZE}, *local_A=NULL, *local_G=NULL, **sub_array=NULL, **s_array=NULL;
int i, j, value=5;
MPI_Init(&argc, &argv);
GA_Initialize();
MA_init(C_INT, 1000, 1000);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
s_array=(int**)malloc(N*sizeof(int*));
for(i=0; i<N; i++)
{
s_array[i]=(int*)malloc(D*sizeof(int));
for(j=0; j<D; j++) s_array[i][j]=rand()%10;
}
sub_array=(int**)malloc(N*sizeof(int*));
for(i=0; i<N; i++)
{
sub_array[i]=(int*)malloc(D*sizeof(int));
for(j=0; j<D; j++) sub_array[i][j]=rand()%10;
}
for(i=0; i<N; i++) local_A=(int*)malloc(N*sizeof(int));
for(i=0; i<N; i++) local_G=(int*)malloc(N*sizeof(int));
g_A=NGA_Create(C_INT, D, dims, "array_A", NULL);
GA_Fill(g_A, &value);
GA_Sync();
NGA_Scatter(g_A, local_A, s_array, N);
NGA_Gather(g_A, local_G, s_array, N);
GA_Sync();
GA_Print(g_A);
if(rank==0)
{
for(i=0; i<N; i++)
if(local_G[i]!=local_A[i]) printf("GA Error: \n");
}
GA_Sync();
if(rank==0)
GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, **local_A=NULL, **local_B=NULL;
int dims[DIM]={SIZE,SIZE}, dims2[DIM], lo[DIM]={SIZE-SIZE,SIZE-SIZE}, hi[DIM]={SIZE-1,SIZE-1}, ld=5, value=5;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
local_A=(int**)malloc(SIZE*sizeof(int*));
for(i=0; i<SIZE; i++)
{
local_A[i]=(int*)malloc(SIZE*sizeof(int));
for(j=0; j<SIZE; j++) local_A[i][j]=rand()%10;
}
local_B=(int**)malloc(SIZE*sizeof(int*));
for(i=0; i<SIZE; i++)
{
local_B[i]=(int*)malloc(SIZE*sizeof(int));
for(j=0; j<SIZE; j++) local_B[i][j]=rand()%10;
}
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
GA_Zero(g_A);
if(rank==0)
{
NGA_Put(g_A, lo, hi, local_A, &ld);
NGA_Get(g_A, lo, hi, local_B, &ld);
for(i=0; i<SIZE; i++)
{
for(j=0; j<SIZE; j++)
if(local_A[i][j]!=local_B[i][j]) GA_ERROR_MSG();
}
}
GA_Sync();
GA_Destroy(g_A);
if(rank == 0) GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, g_B;
int dims[MAX_DIM], val=4, ndim, re;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
for(i=1; i<=MAX_DIM; i++)
{
ndim=i;
dims[i]=SIZE;
// for(j=0; j<ndim; j++)
g_A = NGA_Create(C_INT, ndim, dims, "array_A", NULL);
g_B = NGA_Create(C_INT, ndim, dims, "array_B", NULL);
if(!g_A)
GA_Error("GA Error: no global array exists \n", ndim);
if(!g_B)
GA_Error("GA Error: no global array exists \n", ndim);
}
GA_Sync();
GA_Fill(g_A, &val);
re=GA_Solve(g_A, g_B);
if(re==0)
printf("Cholesky Fact is Successful \n");
else if (re >0)
printf("Cholesky Fact couldn't be completed \n");
else
printf("An Error occured\n");
if(rank == 0)
GA_PRINT_MSG();
GA_Destroy(g_A);
GA_Destroy(g_B);
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char **argv)
{
int proc, nprocs;
int M, N, K; /* */
int blockX_len, blockY_len;
int heap=3000000, stack=3000000;
if (argc == 6) {
M = atoi(argv[1]);
N = atoi(argv[2]);
K = atoi(argv[3]);
blockX_len = atoi(argv[4]);
blockY_len = atoi(argv[5]);
}
else {
printf("Please enter ./a.out <M> <N> <K> <BLOCK-X-LEN> <BLOCK-Y-LEN>");
exit(-1);
}
MPI_Init(&argc, &argv);
GA_Initialize();
if(! MA_init(C_DBL, stack, heap) ) GA_Error("MA_init failed",stack+heap);
proc = GA_Nodeid();
nprocs = GA_Nnodes();
if(proc == 0) {
printf("Using %d processes\n", nprocs);
fflush(stdout);
}
matrix_multiply(M, N, K, blockX_len, blockY_len);
if(proc == 0)
printf("\nTerminating ..\n");
GA_Terminate();
MPI_Finalize();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs;
int g_A, g_V, val1=5, val2=5, local_A[SIZE][SIZE], dims_V=SIZE, local_V[dims_V];
int dims[DIM]={SIZE,SIZE}, dims2[DIM], lo[DIM]={1,1}, hi[DIM]={2,2}, ld=5, i, j;
int loV=0, hiV=dims_V-1;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
g_V = NGA_Create(C_INT, 1, &dims_V, "array_A", NULL);
GA_Fill(g_A, &val1);
GA_Print(g_A);
printf("\n");
GA_Scale(g_A, &val2);
GA_Print(g_A);
GA_Get_diag(g_A, g_V);
GA_Print(g_V);
NGA_Get(g_A, lo, hi, local_A, &ld);
NGA_Get(g_V, &loV, &hiV, local_V, &ld);
if(rank==1)
{
for(i=0; i<dims_V; i++)
if(local_V[i]!=val1*val2) printf(" GA Error: \n");
}
if(rank == 0)
printf("Test Completed \n");
GA_Terminate();
MPI_Finalize();
}
main(int argc, char **argv)
{
int rank, nprocs;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
// MA_init(C_DBL, 1000, 1000);
GA_Initialize();
integer_dot(rank, nprocs);
//float_dot(rank, nprocs);
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char **args)
{
PetscErrorCode ierr;
ierr = PetscInitialize(&argc, &args, (char *) 0, ""); CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD, "Start %s\n", __FILE__); CHKERRQ(ierr);
GA_Initialize();
ierr = MA_init(MT_DBL, 1000, 1000 );
if( !ierr ) GA_Error("\n\n\nga allocaltion failed\n\n",ierr);
size_t len = 1024;
char name[1024];
PetscGetHostName(name,len);
PetscPrintf(PETSC_COMM_SELF, "Hello from %s\n", name);
ierr = testCreate3D(); CHKERRQ(ierr);
GA_Terminate();
ierr = PetscPrintf(PETSC_COMM_WORLD, "End %s\n", __FILE__); CHKERRQ(ierr);
ierr = PetscFinalize(); CHKERRQ(ierr);
}
int main(int argc, char **argv) {
int heap=300000, stack=300000;
int me, nprocs;
/* Step1: Initialize Message Passing library */
#ifdef MPI
MPI_Init(&argc, &argv); /* initialize MPI */
#else
PBEGIN_(argc, argv); /* initialize TCGMSG */
#endif
/* Step2: Initialize GA */
/* ### intialize the GA library */
GA_Initialize();
/* Step3: Initialize Memory Allocator (MA) */
if(! MA_init(C_DBL, stack, heap) ) GA_Error("MA_init failed",stack+heap);
/* ### assign the local processor ID to the int variable "me"
* ### and the total number of processors to the int variable
* ### "nprocs" */
me = GA_Nodeid();
nprocs = GA_Nnodes();
if(me==0) {
printf("\nUsing %d processes\n\n", nprocs); fflush(stdout);
}
TRANSPOSE1D();
if(me==0)printf("\nTerminating ..\n");
/* ### terminate the GA library */
GA_Terminate();
#ifdef MPI
MPI_Finalize();
#else
PEND_();
#endif
}
main(int argc, char **argv)
{
// Initialize MPI libraries
int ierr = MPI_Init(&argc, &argv);
// Initialize Math libraries
gridpack::math::Initialize();
GA_Initialize();
int stack = 200000, heap = 200000;
MA_init(C_DBL, stack, heap);
gridpack::resistor_grid::RGApp app;
app.execute(argc, argv);
GA_Terminate();
// Terminate Math libraries
gridpack::math::Finalize();
// Clean up MPI libraries
ierr = MPI_Finalize();
}
main(int argc, char **argv)
{
int rank, nprocs;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
fillonly(rank, nprocs);
fillandscale(rank, nprocs);
GA_Sync();
if(rank == 0)
printf("Test Completed \n");
GA_Terminate();
MPI_Finalize();
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, g_B, **local_value=NULL;
int dims[DIM]={SIZE,SIZE}, lo[DIM]={SIZE-SIZE,SIZE-SIZE}, hi[DIM]={SIZE-1,SIZE-1}, ld=SIZE;
local_value=(int**)malloc(SIZE*sizeof(int*));
for(i=0; i<SIZE; i++)
local_value[i]=(int*)malloc(SIZE*sizeof(int));
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
g_B = NGA_Create(C_INT, DIM, dims, "array_B", NULL);
for(i=0; i<SIZE; i++)
for(j=0; j<SIZE; j++)
local_value[i][j]=rand()%10;
if(rank==0) NGA_Put(g_A, lo, hi, local_value, &ld);
GA_Transpose(g_A, g_B);
if(rank==0) validate_transpose(g_A, g_B, lo, hi, ld);
GA_Sync();
if(rank == 1) GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char **argv)
{
int rank, nprocs;
int g_A;
int *local_A=NULL, *local_B=NULL, *output_A=NULL;
int dims[DIM]={SIZE,SIZE}, dims2[DIM], lo[DIM]={SIZE-SIZE,SIZE-SIZE}, hi[DIM]={SIZE-1,SIZE-1}, ld=SIZE;
int value=SIZE;
#if defined(USE_ELEMENTAL)
// initialize Elemental (which will initialize MPI)
ElInitialize( &argc, &argv );
ElMPICommRank( MPI_COMM_WORLD, &rank );
ElMPICommSize( MPI_COMM_WORLD, &nprocs );
// instantiate el::global array
ElGlobalArraysConstruct_i( &eliga );
// initialize global arrays
ElGlobalArraysInitialize_i( eliga );
#else
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
#endif
local_A=(int*)malloc(SIZE*SIZE*sizeof(int));
output_A=(int*)malloc(SIZE*SIZE*sizeof(int));
memset (output_A, 0, SIZE*SIZE*sizeof(int));
for(int j=0; j<SIZE; j++)
for(int i=0; i<SIZE; i++) local_A[i+j*ld]=(i + j);
//for(int i=0; i<SIZE; i++) local_A[i+j*ld]=(rand()%10);
local_B=(int*)malloc(SIZE*SIZE*sizeof(int));
memset (local_B, 0, SIZE*SIZE*sizeof(int));
// nb handle
#if defined(USE_ELEMENTAL)
typedef ElInt ga_nbhdl_t;
#endif
ga_nbhdl_t nbnb;
#if defined(USE_ELEMENTAL)
ElGlobalArraysCreate_i( eliga, DIM, dims, "array_A", NULL, &g_A );
ElGlobalArraysFill_i( eliga, g_A, &value );
#else
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
GA_Fill(g_A, &value);
#endif
if (rank == 0) printf ("Initial global array:\n");
#if defined(USE_ELEMENTAL)
ElGlobalArraysPrint_i( eliga, g_A );
#else
GA_Print(g_A);
#endif
for (int i = 0; i < NITERS; i++)
{
// acc data
#if defined(USE_ELEMENTAL)
ElGlobalArraysNBAccumulate_i( eliga, g_A, lo, hi, local_A, &ld, &value, &nbnb );
#else
NGA_NbAcc(g_A, lo, hi, local_A, &ld, &value, &nbnb);
#endif
// updated output
MPI_Reduce (local_A, output_A, SIZE*SIZE, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);
#if defined(USE_ELEMENTAL)
ElGlobalArraysNBWait_i( eliga, &nbnb );
#else
NGA_NbWait (&nbnb);
#endif
// get
if (rank == 0) printf ("Get in iter #%d\n", i);
#if defined(USE_ELEMENTAL)
ElGlobalArraysSync_i( eliga );
ElGlobalArraysGet_i( eliga, g_A, lo, hi, local_B, &ld );
ElGlobalArraysPrint_i( eliga, g_A );
#else
GA_Sync();
NGA_Get(g_A, lo, hi, local_B, &ld);
GA_Print(g_A);
#endif
} // end of iters
if(rank==0)
{
printf(" Alpha (multiplier): %d\n", value);
printf(" Original local buffer (before accumulation): \n");
for(int i=0; i<SIZE; i++)
{
for(int j=0; j<SIZE; j++)
printf("%d ", local_A[i*ld+j]);
printf("\n");
}
printf("\n");
printf(" Get returns: \n");
for(int i=0; i<SIZE; i++)
{
for(int j=0; j<SIZE; j++)
printf("%d ", local_B[i*ld + j]);
printf("\n");
}
printf("\n");
for(int i=0; i<SIZE; i++)
{
for(int j=0; j<SIZE; j++)
{
if(local_B[i*ld+j]!=(value + (NITERS * value * (output_A[i*ld+j]))))
GA_Error("ERROR", -99);
}
}
}
#if defined(USE_ELEMENTAL)
ElGlobalArraysDestroy_i( eliga, g_A );
#else
GA_Destroy(g_A);
#endif
if(rank == 0)
printf ("OK. Test passed\n");
free (local_A);
free (local_B);
free (output_A);
#if defined(USE_ELEMENTAL)
ElGlobalArraysTerminate_i( eliga );
// call el::global arrays destructor
ElGlobalArraysDestruct_i( eliga );
ElFinalize();
#else
GA_Terminate();
MPI_Finalize();
#endif
}
int main(int argc, char **argv) {
int me;
int g_a;
int status;
int i,j;
int dims[] = {n,n};
int proc_group[PROC_LIST_SIZE],proclist[PROC_LIST_SIZE],inode;
int sbuf[1],rbuf[1];
MPI_Comm comm;
MP_INIT(argc,argv);
GA_Initialize();
me = GA_Nodeid();
status = MA_init(MT_DBL, 100000, 100000);
if (!status) GA_Error("ma_init failed",-1);
status = MA_set_auto_verify(1);
status = MA_set_hard_fail(1);
status = MA_set_error_print(1);
inode = GA_Cluster_nodeid();
if (me == 0) {
printf("there are %d nodes, node 0 has %d procs\n",
GA_Cluster_nnodes(), GA_Cluster_nprocs(0));
fflush(stdout);
}
GA_Sync();
for (i=0; i<GA_Cluster_nnodes(); ++i) {
for (j=0; j<GA_Cluster_nprocs(i); ++j) {
proclist[j]=GA_Cluster_procid(i,j);
}
proc_group[i]=GA_Pgroup_create(proclist,GA_Cluster_nprocs(i));
}
GA_Sync();
for (i=0; i<GA_Cluster_nnodes(); ++i) {
if (i == inode) {
printf("%d joining group %d\n", me, proc_group[inode]);
GA_Pgroup_set_default(proc_group[inode]);
g_a = NGA_Create(C_DBL, 2, dims, "a", NULL);
if (!g_a) GA_Error("NGA_Create failed",-1);
printf("%d Created array of group %d as proc no. %d\n",
me, proc_group[inode], GA_Nodeid());
GA_Print_distribution(g_a);
comm = GA_MPI_Comm_pgroup_default();
if (comm != MPI_COMM_NULL) {
sbuf[0] = GA_Nodeid();
status = MPI_Allreduce(sbuf, rbuf, 1, MPI_INT, MPI_MAX, comm);
printf("%d max nodeid is %d\n", me, rbuf[0]);
if ((rbuf[0]+1) != GA_Cluster_nprocs(i)) {
GA_Error("MPI_Allreduce failed",1);
}
}
else {
printf("MPI_Comm was null!\n");
}
GA_Pgroup_set_default(GA_Pgroup_get_world());
}
GA_Sync();
}
GA_Terminate();
MP_FINALIZE();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, g_B, g_C, local_C[DIM][DIM], dims[DIM]={5,5}, val1=5, val2=4, alpha=3, beta=2, ld=5;
int alo[DIM]={2,2}, ahi[DIM]={3,3}, blo[DIM]={2,2}, bhi[DIM]={3,3}, clo[DIM]={1,1}, chi[DIM]={2,2};
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
g_B = GA_Duplicate(g_A, "array_B");
g_C = GA_Duplicate(g_A, "array_C");
GA_Fill(g_A, &val1);
GA_Fill(g_B, &val2);
GA_Zero(g_C);
NGA_Add_patch(&alpha, g_A, clo, chi, &beta, g_B, blo, bhi, g_C, clo, chi);
GA_Sync();
GA_Print(g_A);
GA_Print(g_B);
GA_Print(g_C);
NGA_Get(g_C, clo, chi, local_C, &ld);
//printf("check 1 \n");
for(i=0; i<DIM; i++)
{
for(j=0; j<DIM; j++)printf("%d ", local_C[i][j]);
printf("\n");
}
if(rank == 0)
{
printf("check 2\n");
for(i=0; i<DIM; i++)
{
for(j=0; j<DIM; j++)
if(local_C[i][j]!=(alpha*val1)+(beta*val2)) printf("GA Error : \n");
}
}
if(rank==0)
GA_PRINT_MSG();
GA_Sync();
/*
GA_Destroy(g_A);
GA_Destroy(g_B);
GA_Destroy(g_C);
*/
//*******************************************************************
/* what would be the possible reason for GA_destroy to get failed ..,
* solve this before consolidate the whole
*/
GA_Terminate();
MPI_Finalize();
}
int main(int argc, char**argv)
{
int nprocs, me;
int i,j;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&me);
GA_Initialize();
const int heap=3000000, stack=300000;
if(! MA_init(C_INT,stack,heap) ) GA_Error((char *) "MA_init failed",stack+heap /*error code*/);
int Nx=97; int Ny=97; int Nz = 97;
Nx+=3; Ny+=3; Nz+=3;
int data[Nx*Ny*Nz];
int num_splines = 32;
int g_a,dims[4]={Nx,Ny,Nz,num_splines},chunk[4]={-1,-1,-1,num_splines};
int width[4] = {3, 3, 3, 0};
int type=C_INT;
//g_a=NGA_Create(type,4,dims,"Coefs",chunk);
g_a=NGA_Create_ghosts(type, 4, dims, width, "Coefs", chunk);
int lo[4],hi[4],ld[3];
//double value=9.0; GA_Fill(g_a,&value);
GA_Print_distribution(g_a);
fflush(stdout);
if(me==0)
{
for (i=0; i<num_splines; i++)
{
int x, y, z;
for (x=0; x<Nx; x++)
for (y=0; y<Ny; y++)
for (z=0; z<Nz; z++)
{ j=x*(Ny*Nz)+y*Nz+z;
data[j] = (x*100*100+y*100+z)*100+i;}
lo[0]=lo[1]=lo[2]=0;
hi[0]=Nx-1;hi[1]=Ny-1;hi[2]=Nz-1;
lo[3]=hi[3]=i%num_splines;
ld[0]=Ny;ld[1]=Nz;ld[2]=1;
NGA_Put(g_a,lo,hi,data,ld);
}
}
GA_Update_ghosts(g_a);
GA_Sync();
printf("done\n"),fflush(stdout);
ga_coefs_t *ga_coefs = malloc(sizeof(ga_coefs_t));
ga_coefs->Mx = Nx;
ga_coefs->My = Ny;
ga_coefs->Mz = Nz;
ga_coefs->nsplines = num_splines;
ga_coefs->g_a=g_a;
int *coefs1 = (int*)malloc((size_t)1*sizeof(int)*4*4*4*num_splines);
int ix,iy,iz;
Nx-=3; Ny-=3; Nz-=3;
ga_coefs->sumt=ga_coefs->amount=0;
NGA_Distribution(g_a,me,lo,hi);
GA_Print_distribution(g_a);
int low[16][4],high[16][4];
for(i=0;i<nprocs;i++)
NGA_Distribution(g_a,i,low[i],high[i]);
srand ( time(NULL) );
int k=GA_Nodeid();
printf("%d: low[k]=%d high[k]=%d\n", GA_Nodeid(), low[k][2], high[k][2]);
int unequal=0;
for(i=0;i<1000;i++) {
ix=rand_index(low[k][0],high[k][0]);
if(ix+3>=dims[0]) ix=low[k][0];
iy=rand_index(low[k][1],high[k][1]);
if(iy+3>=dims[1]) iy=low[k][1];
iz=rand_index(low[k][2],high[k][2]);
if(iz+3>=dims[2]) iz=low[k][2];
coefs_ga_get_3d(ga_coefs,coefs1,ix,iy,iz);
long get_sum=mini_cube_sum(coefs1, ga_coefs->nsplines);
long ghost_sum=coefs_ghost_access_3d(ga_coefs->g_a, ix, iy, iz, ga_coefs->nsplines);
if(get_sum!=ghost_sum) {
printf("ixyz=\t%d\t%d\t%d\t", ix, iy, iz);
printf("get_sum=%ld ghost_sum=%ld\n", get_sum, ghost_sum);
unequal++;
}
}
printf("unequal count=%d\n", unequal);
free(coefs1);
GA_Terminate();
MPI_Finalize();
return 0;
}
int main (int argc, char **argv)
{
double startTime;
int info; /* used to check for functions returning nonzeros */
GAVec ga_x; /* solution vector */
TAO_SOLVER tao; /* TAO_SOLVER solver context */
TAO_GA_APPLICATION taoapp; /* TAO application context */
TaoTerminateReason reason;
AppCtx user; /* user-defined application context */
/*initialize GA and MPI */
int heap = 400000, stack = 400000;
MPI_Init (&argc, &argv); /* initialize MPI */
GA_Initialize (); /* initialize GA */
user.me = GA_Nodeid ();
user.nproc = GA_Nnodes ();
startTime = MPI_Wtime();
if (user.me == 0) {
if (GA_Uses_fapi ())
GA_Error ("Program runs with C array API only", 0);
printf ("Using %ld processes\n", (long) user.nproc);
fflush (stdout);
}
heap /= user.nproc;
stack /= user.nproc;
if (!MA_init (MT_F_DBL, stack, heap))
GA_Error ("MA_init failed", stack + heap); /* initialize memory allocator */
/* Initialize TAO */
TaoInitialize (&argc, &argv, (char *) 0, help);
/* Initialize problem parameters */
user.ndim = NDIM;
user.natoms = NATOMS;
user.BlockSize = BLOCKSIZE;
/* Allocate vectors for the solution and gradient */
int dims[2];
dims[0] = user.ndim*user.natoms;
ga_x = NGA_Create (C_DBL, 1, dims, "GA_X", NULL);
if (!ga_x) GA_Error ("lennard-jones.main::NGA_Create ga_x", ga_x);
/* Set up structures for data distribution */
info = SetupBlocks(&user); CHKERRQ(info);
/* The TAO code begins here */
/* Create TAO solver with desired solution method */
info = TaoCreate (MPI_COMM_WORLD, "tao_lmvm", &tao); CHKERRQ(info);
info = TaoGAApplicationCreate (MPI_COMM_WORLD, &taoapp); CHKERRQ(info);
/* Set the initial solution */
info = InitializeVariables(ga_x, &user); CHKERRQ(info);
info = TaoGAAppSetInitialSolutionVec(taoapp, ga_x); CHKERRQ(info);
/* Set routines for function, gradient */
info = TaoGAAppSetObjectiveAndGradientRoutine (taoapp, FormFunctionGradient,
(void *) &user); CHKERRQ(info);
/* Check for TAO command line options */
info = TaoSetFromOptions (tao); CHKERRQ(info);
/* SOLVE THE APPLICATION */
info = TaoSolveGAApplication (taoapp, tao); CHKERRQ(info);
/* To View TAO solver information use */
info = TaoView(tao); CHKERRQ(info);
/* Get termination information */
info = TaoGetTerminationReason (tao, &reason);
if(info) GA_Error("lennard-jones.main.TaoGetTerminationReason",info);
if (user.me == 0) {
if (reason <= 0)
printf("Try a different TAO method, adjust some parameters, or check the function evaluation routines\n");
printf("WALL TIME TAKEN = %lf\n", MPI_Wtime()-startTime);
/*output the solutions */
printf ("The solution is :\n");
}
GA_Print (ga_x);
/* Free TAO data structures */
info = TaoDestroy (tao); CHKERRQ(info);
info = TaoGAAppDestroy (taoapp); CHKERRQ(info);
/* Free GA data structures */
GA_Destroy (ga_x);
if (!MA_pop_stack(user.memHandle))
ga_error("Main::MA_pop_stack for memHandle failed",0);
/* Finalize TAO, GA, and MPI */
TaoFinalize ();
GA_Terminate ();
MPI_Finalize ();
return 0;
}
int main(int argc, char **argv) {
int i, matrix_size;
int heap=20000000, stack=200000000;
double *A=NULL;
if(argc != 2)
{
printf("Usage Error\n\t Usage: <program> <matrix_size>\n");
exit(0);
}
matrix_size = atoi(argv[1]);
if(matrix_size <= 0)
{
printf("Error: matrix size (%d) should be > 0\n",
matrix_size);
GA_Error("matrix size should be >0", 1);
}
/* *****************************************************************
* Initialize MPI/TCGMSG-MPI, GA and MA
* *****************************************************************/
#ifdef MPI
#ifdef DCMF
int desired = MPI_THREAD_MULTIPLE;
int provided;
MPI_Init_thread(&argc, &argv, desired, &provided);
if ( provided != MPI_THREAD_MULTIPLE ) printf("provided != MPI_THREAD_MULTIPLE\n");
#else
MPI_Init (&argc, &argv); /* initialize MPI */
#endif
#else
PBEGIN_(argc, argv); /* initialize TCGMSG-MPI */
#endif
GA_Initialize(); /* initialize GA */
me = GA_Nodeid();
nprocs = GA_Nnodes();
heap /= nprocs;
stack /= nprocs;
if(! MA_init(MT_F_DBL, stack, heap)) /* initialize MA */
{
GA_Error("MA_init failed",stack+heap);
}
/* create/initialize the matrix */
if((A = (double*)malloc(matrix_size*matrix_size*sizeof(double))) == NULL)
{
GA_Error("malloc failed", matrix_size*matrix_size*sizeof(double));
}
for(i=0; i<2; i++) /* 5 runs */
{
init_array(A, matrix_size);
#if DEBUG
if(me==0) print_array(A, matrix_size);
#endif
/* *****************************************************************
* Perform LU Factorization
* *****************************************************************/
ga_lu(A, matrix_size);
}
free(A);
/* *****************************************************************
* Terminate MPI/TCGMSG-MPI, GA and MA
* *****************************************************************/
if(me==0)printf("Terminating ..\n");
GA_Terminate();
#ifdef MPI
MPI_Finalize();
#else
PEND_();
#endif
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs;
int g_A, dims[D]={5,10}, local_A[N], local_G[N], **sub_array=NULL, **s_array=NULL;
int i, j, value=5;
MPI_Init(&argc, &argv);
GA_Initialize();
MA_init(C_INT, 1000, 1000);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
s_array=(int**)malloc(N*sizeof(int*));
for(i=0; i<N; i++)
{
s_array[i]=(int*)malloc(D*sizeof(int));
for(j=0; j<D; j++) s_array[i][j]=rand()%5;
}
sub_array=(int**)malloc(N*sizeof(int*));
for(i=0; i<N; i++)
{
sub_array[i]=(int*)malloc(D*sizeof(int));
for(j=0; j<D; j++) sub_array[i][j]=rand()%5;
}
for(i=0; i<N; i++)
//local_A=(int*)malloc(N*sizeof(int));
/*
* depends on the value of array ..we can generate the location values in randon
* we can also use the if-condition
*/
// PRINTing all the genrated array for reference
for(i=0; i<N; i++)
{
for(j=0; j<D; j++)printf("%d ",s_array[i][j]);
printf("\n");
}
printf("\n");
for(i=0; i<N; i++)
{
for(j=0; j<D; j++)printf("%d ",sub_array[i][j]);
printf("\n");
}
printf("\n");
for(i=0; i<N; i++)printf("%d \n",local_A[i]=rand()%5+1);
// PRINT done - now creating array
g_A=NGA_Create(C_INT, D, dims, "array_A", NULL);
GA_Fill(g_A, &value);
GA_Sync();
NGA_Scatter(g_A, local_A, s_array, N);
NGA_Gather(g_A, local_G, s_array, N);
GA_Sync();
GA_Print(g_A);
for(i=0; i<N; i++)printf("%d \n",local_G[i]);
printf("\n");
if(rank==0)
{
for(i=0; i<N; i++)
if(local_G[i]!=local_A[i]) printf("GA Error: \n");
}
GA_Sync();
if(rank==0)
GA_PRINT_MSG();
GA_Terminate();
MPI_Finalize();
return 0;
}
main(int argc, char **argv)
{
int rank, nprocs, i, j;
int g_A, g_B, g_C, **local_C=NULL, dims[DIM]={SIZE,SIZE}, val1=5, val2=4, alpha=3, beta=2;
int clo[DIM]={SIZE-SIZE,SIZE-SIZE}, chi[DIM]={SIZE-1,SIZE-1}, ld=SIZE;
int **local_tm=NULL;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MA_init(C_INT, 1000, 1000);
GA_Initialize();
local_C=(int**)malloc(SIZE*sizeof(int*));
for(i=0; i<SIZE; i++)
local_C[i]=(int*)malloc(SIZE*sizeof(int));
local_tm=(int**)malloc(SIZE*sizeof(int*));
for(i=0; i<SIZE; i++)
local_tm[i]=(int*)malloc(SIZE*sizeof(int));
g_A = NGA_Create(C_INT, DIM, dims, "array_A", NULL);
g_B = GA_Duplicate(g_A, "array_B");
g_C = GA_Duplicate(g_A, "array_C");
GA_Fill(g_A, &val1);
GA_Fill(g_B, &val2);
GA_Add(&alpha, g_A, &beta, g_B, g_C);
GA_Sync();
GA_Print(g_A);
GA_Print(g_B);
GA_Print(g_C);
//printf("check 1\n");
NGA_Get(g_C, clo, chi, local_tm, &ld);
//printf("check 2\n");
// GA_Sync();
if(rank==0)
{
for(i=0; i<SIZE; i++)
{
for(j=0; j<SIZE; j++)printf("%d ", local_tm[i][j]);
printf("\n");
}
}
/*
if(rank==0)
{
NGA_Get(g_C, clo, chi, local_C, &ld);
printf("check 1 \n");
for(i=0; i<SIZE; i++)
{
for(j=0; j<SIZE; j++)printf("%d ", local_C[i][j]);
printf("\n");
}
printf("check 2\n");
for(i=0; i<SIZE; i++)
{
for(j=0; j<SIZE; j++)
if(local_C[i][j]!=(alpha*val1)+(beta*val2)) printf("GA Error : \n");
}
}
*/
//GA_Sync();
if(rank==0)
printf("Test Completed \n");
// GA_Sync();
/*
GA_Destroy(g_A);
GA_Destroy(g_B);
GA_Destroy(g_C);
*/
//*******************************************************************
/* what would be the possible reason for GA_destroy to get failed ..,
* solve this before consolidate the whole
*/
GA_Terminate();
MPI_Finalize();
}
