void *split( void *hnd, c4snet_data_t *A, int A_width, int A_height, int nodes)
{
int i, first, last, block;
int *new_array;
c4snet_data_t *new_data;
int *array = C4SNetGetData(A);
block = A_height / nodes;
first = 0;
last = first + block - 1;
for (i = 0; i < nodes - 1; i++) {
new_data = C4SNetAlloc(CTYPE_int, A_width * (last - first + 1), (void **) &new_array);
memcpy(new_array, (array + A_width * first), sizeof(int) * A_width * (last - first + 1));
C4SNetOut( hnd, 1, new_data, A_width, A_height, i, first, last);
first = last + 1;
last += block;
}
last = A_height - 1;
new_data = C4SNetAlloc(CTYPE_int, A_width * (last - first + 1), (void **) &new_array);
memcpy(new_array, (array + A_width * first), sizeof(int) * A_width * (last - first + 1));
C4SNetOut(hnd, 1, new_data, A_width, A_height, i, first, last);
C4SNetFree(A);
return hnd;
}
static c4snet_data_t *MPIUnpackFun(void *buf)
{
void *tmp;
c4snet_data_t *result;
int vtype, type, count;
MPIUnpack(buf, &vtype, MPI_INT, 1);
MPIUnpack(buf, &count, MPI_INT, 1);
MPIUnpack(buf, &type, MPI_INT, 1);
if (vtype == VTYPE_array) {
result = C4SNetAlloc(type, count, &tmp);
MPIUnpack(buf, tmp, TypeToMPIType(type), count);
} else {
result = C4SNetAlloc(type, 1, &tmp);
MPIUnpack(buf, tmp, TypeToMPIType(type), 1);
}
return result;
}
void *init( void *hnd)
{
int *int_x;
c4snet_data_t *result;
result = C4SNetAlloc(CTYPE_int, X*Y, &int_x);
for (int i = 0; i < X*Y; i++) int_x[i] = i;
C4SNetOut(hnd, 1, result);
return hnd;
}
void *init( void *hnd)
{
int *int_x;
c4snet_data_t *result;
result = C4SNetAlloc(CTYPE_int, X*Y, &int_x);
for (int i = 0; i < X*Y; i++) int_x[i] = i;
printf("\n\n\narray is created\n\n\n");
C4SNetOut(hnd, 1, result);
return hnd;
}
/* Do the InitialFactorization from input tile Fac to output tile Tri. */
void *facto(void *hnd, c4snet_data_t *Fac, int N, int B, int K)
{
int j;
const int P = N / B;
c4snet_data_t *Tri;
if (verbose) {
printf("%s: %p, N=%d, B=%d, (%d, , )\n",
__func__, Fac, N, B, K);
}
if (compute) {
tile_t A = (tile_t) C4SNetGetData(Fac);
void *tile_ptr;
c4snet_data_t *tile = C4SNetAlloc(CTYPE_double, B * B, &tile_ptr);
tile_t L = (tile_t) tile_ptr;
int k_b, j_b, j_bb, i_b;
memset(L, 0, B * B * sizeof(double));
for (k_b = 0; k_b < B; ++k_b) {
if (A[k_b + k_b * B] <= 0) {
fprintf(stderr, "Not a symmetric positive definite (SPD) matrix\n");
fprintf(stderr, "K = %d, k_b = %d, B = %d, A[k_b + k_b * B] = %f\n",
K, k_b, B, A[k_b + k_b * B]);
exit(1);
}
L[k_b + k_b * B] = sqrt(A[k_b + k_b * B]);
for (j_b = k_b + 1; j_b < B; ++j_b) {
L[j_b + k_b * B] = A[j_b + k_b * B] / L[k_b + k_b * B];
}
for (j_bb = k_b + 1; j_bb < B; ++j_bb) {
for (i_b = j_bb; i_b < B; ++i_b) {
A[i_b + j_bb * B] -= L[i_b + k_b * B] * L[j_bb + k_b * B];
}
}
}
Tri = tile;
C4SNetFree(Fac);
} else {
Tri = Fac;
}
for (j = K; j < P; ++j) {
if (verbose) {
printf("%s: -> Tri, N=%d, B=%d, (%d, %d, )\n", __func__, N, B, K, j);
}
C4SNetOut(hnd, 1, C4SNetShallowCopy(Tri), N, B, K, j);
}
/* printf("%s: -> Out, N=%d, B=%d, (%d, %d, %d)\n",
__func__, N, B, K, K - 1, K - 1); */
C4SNetOut(hnd, 2, Tri, N, B, K, K - 1, K - 1);
return hnd;
}
void *split(void *hnd, c4snet_data_t *entries, int num_entries)
{
c4snet_data_t *hash_cdata;
c4snet_data_t *salt_cdata;
char *data, *hash, *salt, *first, *middle, *last;
first = data = C4SNetGetData(entries)
for (int i = 0; i < num_entries; i++) {
last = strchr(first, '\n');
if (last == NULL) last = &data[strlen(data)];
for (int j = 0, middle = first; j < 3; j++) {
middle = strchr(middle, '$') + 1;
}
hash_cdata = C4SNetAlloc(CTYPE_char, middle - first + 1, &hash);
salt_cdata = C4SNetAlloc(CTYPE_char, last - first + 1, &salt);
strncpy(salt, first, middle - first);
salt[middle - first] = '\0';
strncpy(hash, first, last - first);
hash[last - first] = '\0';
C4SNetOut(hnd, 1, hash_cdata, salt_cdata, i);
first = last + 1;
}
C4SNetFree(entries);
return hnd;
}
static c4snet_data_t *MPIUnpackFun(void *buf)
{
void *tmp;
c4snet_data_t *result;
int vtype, type, count;
SNetDistribUnpack(&vtype, buf, MPI_INT, 1);
SNetDistribUnpack(&count, buf, MPI_INT, 1);
SNetDistribUnpack(&type, buf, MPI_INT, 1);
if (vtype == VTYPE_array) {
result = C4SNetAlloc(type, count, &tmp);
SNetDistribUnpack(tmp, buf, TypeToMPIType(type), count);
} else {
result = C4SNetCreate(type, 1, &tmp);
SNetDistribUnpack(tmp, buf, TypeToMPIType(type), 1);
}
return result;
}
/* Do the TriangularSolve from input tiles Tri and Tri2. */
void *trisol(void *hnd, c4snet_data_t *Tri, c4snet_data_t *Tri2,
int N, int B, int K, int J)
{
c4snet_data_t *Out;
c4snet_data_t *Sym;
c4snet_data_t *Sym2;
const int P = N / B;
int i, j;
if (verbose) {
printf("%s: %p, %p, N=%d, B=%d, (%d, %d, )\n",
__func__, Tri, Tri2, N, B, K, J);
}
if (compute) {
tile_t A = C4SNetGetData(Tri2);
tile_t Li = C4SNetGetData(Tri);
void *tile_ptr;
c4snet_data_t *tile = C4SNetAlloc(CTYPE_double, B * B, &tile_ptr);
tile_t Lo = tile_ptr;
int k_b, i_b, j_b;
for (k_b = 0; k_b < B; ++k_b) {
for (i_b = 0; i_b < B; ++i_b) {
Lo[i_b + k_b * B] = A[i_b + k_b * B] / Li[k_b + k_b * B];
}
for (j_b = k_b + 1; j_b < B; ++j_b) {
for (i_b = 0; i_b < B; ++i_b) {
A[i_b + j_b * B] -= Li[j_b + k_b * B] * Lo[i_b + k_b * B];
}
}
}
Out = Sym = Sym2 = tile;
C4SNetFree(Tri);
C4SNetFree(Tri2);
} else {
Out = Sym = Sym2 = Tri;
C4SNetFree(Tri2);
}
for (i = K + 1; i <= J; ++i) {
if (verbose) {
printf("%s: -> Sym(%d,%d)a, N=%d, B=%d, (%d, %d, %d)\n",
__func__, K, J, N, B, K, J, i);
}
C4SNetOut(hnd, 1, C4SNetShallowCopy(Sym), N, B, K, J, i);
if (i == J && i != K) {
if (verbose) {
printf("%s: -> Sym2(%d,%d)a, (%d, %d, %d)\n",
__func__, K, J, K, J, i);
}
C4SNetOut(hnd, 2, C4SNetShallowCopy(Sym2), K, J, i);
}
}
for (j = J; j < P; ++j) {
if (j == J && j != K) {
if (verbose) {
printf("%s: -> Sym(%d,%d)b, N=%d, B=%d, (%d, %d, %d)\n",
__func__, K, J, N, B, K, j, K);
}
C4SNetOut(hnd, 1, C4SNetShallowCopy(Sym2), N, B, K, j, K);
} else {
if (verbose) {
printf("%s: -> Sym2(%d,%d)c, (%d, %d, %d)\n",
__func__, K, J, K, j, J);
}
C4SNetOut(hnd, 2, C4SNetShallowCopy(Sym2), K, j, J);
}
if (j == K) {
if (verbose) {
printf("%s: -> Sym(%d,%d)d, N=%d, B=%d, (%d, %d, %d)\n",
__func__, K, J, N, B, K, j, K);
}
C4SNetOut(hnd, 1, C4SNetShallowCopy(Sym), N, B, K, j, K);
}
}
C4SNetOut(hnd, 3, Out, N, B, K, J, K - 1);
return hnd;
}
/* Setup the S-Net computation: read matrix data from file and distribute. */
void *start(void *hnd, c4snet_data_t *InFile, c4snet_data_t *OutFile, int N, int B)
{
int P, X;
int j, i;
void *result_data;
char *infile;
c4snet_data_t *result;
double *array_data;
get_options();
/* Matrix size N must be a multiple of the block size B. */
if (B < 1 || N < 1 || N % B) {
fprintf(stderr, "%s: matrix size %d is not a multiple of the block size %d\n",
__func__, N, B);
exit(1);
}
/* P gives the number of tiles in either dimension. */
P = N / B;
/* X gives the number of tiles needed to construct the output. */
X = (P * P + P) / 2;
/* Get the input filename as a string. */
infile = chars_to_string(InFile);
C4SNetFree(InFile);
if (verbose) {
printf("%s: N=%d, B=%d, P=%d, X=%d, I=%s\n", __func__, N, B, P, X, infile);
}
result = C4SNetAlloc(CTYPE_long, P * P, &result_data);
memset(result_data, 0, sizeof(long) * P * P);
array_data = SNetMemAlloc(N * N * sizeof(double));
if (compute) {
read_matrix(N, array_data, infile);
}
free(infile);
if (clock_gettime(CLOCK_REALTIME, &begin)) {
pexit("clock_gettime");
}
/* Loop over the rows as j. */
for (j = 0; j < P; ++j) {
/* Loop over the columns up to the diagonal as i. */
for (i = 0; i <= j; ++i) {
void *tile_ptr;
c4snet_data_t *tile = C4SNetAlloc(CTYPE_double, B * B, &tile_ptr);
if (compute) {
tile_t tile_data = (tile_t) tile_ptr;
int A_y, T_y, A_x, T_x;
for (A_y = j * B, T_y = 0; T_y < B; ++A_y, ++T_y) {
for (A_x = i * B, T_x = 0; T_x < B; ++A_x, ++T_x) {
tile_data[T_y + T_x * B] = array_data[A_y * N + A_x];
}
}
} else {
memset(tile_ptr, 0, B * B * sizeof(double));
}
if (j == 0 && i == 0) {
if (verbose) {
printf("%s: -> Fac, N=%d, B=%d, (1, , )\n", __func__, N, B);
}
C4SNetOut(hnd, 2, tile, N, B, 1);
}
else if (i == 0) {
if (verbose) {
printf("%s: -> Tri2, (1, %d, )\n", __func__, j);
}
C4SNetOut(hnd, 3, tile, 1, j);
}
else {
if (verbose) {
printf("%s: -> Sym3, (1, %d, %d)\n", __func__, j, i);
}
C4SNetOut(hnd, 4, tile, 1, j, i);
}
}
}
C4SNetOut(hnd, 1, OutFile, result, X);
free(array_data);
return hnd;
}
