int test_pluq_solve_left(rci_t m, rci_t n, int offsetA, int offsetB) {
mzd_t* Abase = mzd_init(2048, 2048);
mzd_t* Bbase = mzd_init(2048, 2048);
mzd_randomize(Abase);
mzd_randomize(Bbase);
mzd_t* A = mzd_init_window(Abase, 0, offsetA, m, m + offsetA);
mzd_t* B = mzd_init_window(Bbase, 0, offsetB, m, n + offsetB);
// copy B
mzd_t* Bcopy = mzd_init(B->nrows, B->ncols);
for (rci_t i = 0; i < B->nrows; ++i)
for (rci_t j = 0; j < B->ncols; ++j)
mzd_write_bit(Bcopy,i,j, mzd_read_bit (B,i,j));
for (rci_t i = 0; i < m; ++i) {
mzd_write_bit(A,i,i, 1);
}
mzd_t *Acopy = mzd_copy(NULL, A);
rci_t r = mzd_echelonize(Acopy,1);
printf("solve_left m: %4d, n: %4d, r: %4d da: %4d db: %4d ", m, n, r, offsetA, offsetB);
mzd_free(Acopy);
Acopy = mzd_copy(NULL, A);
int consistency = mzd_solve_left(A, B, 0, 1);
//copy B
mzd_t *X = mzd_init(B->nrows,B->ncols);
for (rci_t i = 0; i < B->nrows; ++i)
for (rci_t j = 0; j < B->ncols; ++j)
mzd_write_bit(X,i,j, mzd_read_bit (B,i,j));
mzd_t *B1 = mzd_mul(NULL, Acopy, X, 0);
mzd_t *Z = mzd_add(NULL, Bcopy, B1);
int status = 0;
if(consistency == 0) {
status = 1 - mzd_is_zero(Z);
if (status == 0) {
printf("passed\n");
} else {
printf("FAILED\n");
}
} else {
printf("skipped (no solution)\n");
}
mzd_free(Bcopy);
mzd_free(B1);
mzd_free(Z);
mzd_free_window(A);
mzd_free_window(B);
mzd_free(Acopy);
mzd_free(Abase);
mzd_free(Bbase);
mzd_free(X);
return status;
}
int test_lqup_random(size_t m, size_t n) {
printf("pluq: testing random m: %5zd, n: %5zd",m,n);
size_t i,j;
mzd_t* U = mzd_init(m, n);
mzd_t* L = mzd_init(m, m);
mzd_t* A = mzd_init(m, n);
mzd_randomize(A);
mzd_t* Acopy = mzd_copy (NULL,A);
mzp_t* P = mzp_init(m);
mzp_t* Q = mzp_init(n);
int r;
r=mzd_pluq(A, P, Q, 0);
printf(", rank: %5d ",r);
for (i=0; i<r; ++i){
for (j=0; j<i;++j)
mzd_write_bit(L, i, j, mzd_read_bit(A,i,j));
for (j=i+1; j<n;++j)
mzd_write_bit(U, i, j, mzd_read_bit(A,i,j));
}
for (i=r; i<m; i++)
for (j=0; j<r;++j)
mzd_write_bit(L, i, j, mzd_read_bit(A,i,j));
for (i=0; i<r; ++i){
mzd_write_bit(L,i,i, 1);
mzd_write_bit(U,i,i, 1);
}
mzd_apply_p_left(Acopy, P);
mzd_apply_p_right_trans(Acopy, Q);
mzd_addmul(Acopy, L, U, 0);
int status = 0;
for ( i=0; i<m; ++i)
for ( j=0; j<n; ++j){
if (mzd_read_bit (Acopy,i,j)){
status = 1;
break;
}
}
if (status) {
printf(" ... FAILED\n");
} else
printf (" ... passed\n");
mzd_free(U);
mzd_free(L);
mzd_free(A);
mzd_free(Acopy);
mzp_free(P);
mzp_free(Q);
return status;
}
int test_trsm_upper_right (int m, int n, int offset, const char* description){
printf("upper_right: %s m: %4d n: %4d offset: %4d ... ",description, m, n, offset);
mzd_t* Ubase = mzd_init (2048,2048);
mzd_t* Bbase = mzd_init (2048,2048);
mzd_randomize (Ubase);
mzd_randomize (Bbase);
mzd_t* Bbasecopy = mzd_copy (NULL, Bbase);
mzd_t* U = mzd_init_window (Ubase, 0, offset, n, offset + n);
mzd_t* B = mzd_init_window (Bbase, 0, offset, m, offset + n);
mzd_t* W = mzd_copy (NULL, B);
size_t i,j;
for (i=0; i<n; ++i){
for (j=0; j<i;++j)
mzd_write_bit(U,i,j, 0);
mzd_write_bit(U,i,i, 1);
}
mzd_trsm_upper_right (U, B, 2048);
mzd_addmul(W, B, U, 2048);
int status = 0;
for ( i=0; i<m; ++i)
for ( j=0; j<n; ++j){
if (mzd_read_bit (W,i,j)){
status = 1;
}
}
// Verifiying that nothing has been changed around the submatrices
mzd_addmul(W, B, U, 2048);
mzd_copy (B, W);
for ( i=0; i<2048; ++i)
for ( j=0; j<2048/RADIX; ++j){
if (Bbase->rows[i][j] != Bbasecopy->rows[i][j]){
status = 1;
}
}
mzd_free_window (U);
mzd_free_window (B);
mzd_free (W);
mzd_free(Ubase);
mzd_free(Bbase);
mzd_free(Bbasecopy);
if (!status)
printf("passed\n");
else
printf("FAILED\n");
return status;
}
int test_trsm_lower_left (int m, int n, int offsetL, int offsetB){
mzd_t* Lbase = mzd_init (2048,2048);
mzd_t* Bbase = mzd_init (2048,2048);
mzd_randomize (Lbase);
mzd_randomize (Bbase);
mzd_t* Bbasecopy = mzd_copy (NULL, Bbase);
mzd_t* L = mzd_init_window (Lbase, 0, offsetL, m, offsetL + m);
mzd_t* B = mzd_init_window (Bbase, 0, offsetB, m, offsetB + n);
mzd_t* W = mzd_copy (NULL, B);
size_t i,j;
for (i=0; i<m; ++i){
for (j=i+1; j<m;++j)
mzd_write_bit(L,i,j, 0);
mzd_write_bit(L,i,i, 1);
}
mzd_trsm_lower_left(L, B, 2048);
mzd_addmul(W, L, B, 2048);
int status = 0;
for ( i=0; i<m; ++i)
for ( j=0; j<n; ++j){
if (mzd_read_bit (W,i,j)){
status = 1;
}
}
// Verifiying that nothing has been changed around the submatrices
mzd_addmul(W, L, B, 2048);
mzd_copy (B, W);
for ( i=0; i<2048; ++i)
for ( j=0; j<2048/RADIX; ++j){
if (Bbase->rows[i][j] != Bbasecopy->rows[i][j]){
status = 1;
}
}
mzd_free_window (L);
mzd_free_window (B);
mzd_free_window (W);
mzd_free(Lbase);
mzd_free(Bbase);
mzd_free(Bbasecopy);
if (!status)
printf(" ... passed\n");
else
printf(" ... FAILED\n");
return status;
}
mzd_t *mzd_kernel_left_pluq(mzd_t *A, int const cutoff) {
mzp_t *P = mzp_init(A->nrows);
mzp_t *Q = mzp_init(A->ncols);
rci_t r = mzd_pluq(A, P, Q, cutoff);
if (r == A->ncols) {
mzp_free(P);
mzp_free(Q);
__M4RI_DD_MZD(A);
return NULL;
}
mzd_t *U = mzd_init_window(A, 0, 0, r, r);
mzd_t *B = mzd_init_window(A, 0, r, r, A->ncols);
mzd_trsm_upper_left(U, B, cutoff);
mzd_t *R = mzd_init(A->ncols, A->ncols - r);
mzd_t *RU = mzd_init_window(R, 0, 0, r, R->ncols);
mzd_copy(RU, B);
for(rci_t i = 0; i < R->ncols; ++i) {
mzd_write_bit(R, r + i, i, 1);
}
mzd_apply_p_left_trans(R, Q);
mzp_free(P);
mzp_free(Q);
mzd_free_window(RU);
mzd_free_window(U);
mzd_free_window(B);
__M4RI_DD_MZD(A);
__M4RI_DD_MZD(R);
return R;
}
/*
* File: m4ri_matrix_impl
* Implementation of remaining matrix operations based on
* M4RI library
*/
int copy_matrix_to_offset(gf2matrix *dest, const gf2matrix *src,
int dest_offset_row, int dest_offset_col)
{
size_t i, j;
if (!src)
return -1;
for (i = 0; i < src->nrows; ++i) {
for (j = 0; j < src->ncols; ++j) {
BIT bit = mzd_read_bit(src, i, j);
mzd_write_bit(dest, dest_offset_row + i, dest_offset_col + j, bit);
}
}
return 0;
}
int run(void *_p, unsigned long long *data, int *data_len) {
struct trsm_params *p = (struct trsm_params *)_p;
*data_len = 2;
mzd_t *B = mzd_init(p->m, p->n);
mzd_t *L = mzd_init(p->n, p->n);
mzd_randomize(B);
mzd_randomize(L);
for (rci_t i = 0; i < p->n; ++i){
for (rci_t j = i + 1; j < p->n; ++j)
mzd_write_bit(L,i,j, 0);
mzd_write_bit(L,i,i, 1);
}
data[0] = walltime(0);
data[1] = cpucycles();
mzd_trsm_lower_right(L, B, 2048);
data[0] = walltime(data[0]);
data[1] = cpucycles() - data[1];
mzd_free(B);
mzd_free(L);
return 0;
}
/*
* Class: m4rjni_Mzd
* Method: mzd_write_bit
* Signature: (JIII)V
*/
JNIEXPORT void JNICALL Java_m4rjni_Mzd_mzd_1write_1bit(JNIEnv *env, jobject obj, jlong ptr, jint row, jint col, jint val) {
mzd_t *M = (mzd_t*)ptr;
if (M==NULL) return;
rci_t rrow = row;
rci_t rcol = col;
rci_t rval = val;
if (rrow<0 || rrow>=M->nrows) {
printf("M4RJNI ERROR: invalid row index\n");
return;
}
if (rcol<0 || rcol>=M->ncols) {
printf("M4RJNI ERROR: invalid column index\n");
return;
}
//printf("Java_m4rjni_Mzd_mzd_1write_1bit(%016lX, %d, %d, %d)\n",ptr,rrow,rcol,rval);
mzd_write_bit(M, rrow, rcol, rval);
}
void MatrixF2::set(uint r,uint c,uint64_t value)
{
inRowEchelonForm = false;
mzd_write_bit(matrix,r,c,value);
}
int run(void *_p, unsigned long long *data, int *data_len) {
struct elim_params *p = (struct elim_params *)_p;
#ifndef HAVE_LIBPAPI
*data_len = 2;
#else
*data_len = MIN(papi_array_len + 1, *data_len);
#endif
int papi_res;
mzd_t *A = mzd_init(p->m, p->n);
if(p->r != 0) {
mzd_t *L, *U;
L = mzd_init(p->m, p->m);
U = mzd_init(p->m, p->n);
mzd_randomize(U);
mzd_randomize(L);
for (rci_t i = 0; i < p->m; ++i) {
for (rci_t j = i + 1; j < p->m; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, p->m - j);
mzd_clear_bits(L, i, j, length);
}
mzd_write_bit(L,i,i, 1);
for (rci_t j = 0; j < i && j < p->n; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, i - j);
mzd_clear_bits(U, i, j, length);
}
if(i < p->r) {
mzd_write_bit(U, i, i, 1);
} else {
for (rci_t j = i; j < p->n; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, p->n - i);
mzd_clear_bits(U, i, j, length);
}
}
}
mzd_mul(A,L,U,0);
mzd_free(L);
mzd_free(U);
} else {
mzd_randomize(A);
}
mzp_t *P = mzp_init(A->nrows);
mzp_t *Q = mzp_init(A->ncols);
#ifndef HAVE_LIBPAPI
data[0] = walltime(0);
data[1] = cpucycles();
#else
int array_len = *data_len - 1;
unsigned long long t0 = PAPI_get_virt_usec();
papi_res = PAPI_start_counters((int*)papi_events, array_len);
if (papi_res)
m4ri_die("");
#endif
if(strcmp(p->algorithm, "m4ri") == 0)
p->r = mzd_echelonize_m4ri(A, 0, 0);
else if(strcmp(p->algorithm, "ple") == 0)
p->r = mzd_ple(A, P, Q, 0);
else if(strcmp(p->algorithm, "mmpf") == 0)
p->r = _mzd_ple_russian(A, P, Q, 0);
else
m4ri_die("unknown algorithm %s",p->algorithm);
#ifndef HAVE_LIBPAPI
data[1] = cpucycles() - data[1];
data[0] = walltime(data[0]);
#else
mzp_free(P);
mzp_free(Q);
PAPI_stop_counters((long long*)&data[1], array_len);
t0 = PAPI_get_virt_usec() - t0;
data[0] = t0;
for (int nv = 0; nv <= array_len; ++nv) {
data[nv] -= loop_calibration[nv];
}
#endif
mzd_free(A);
return 0;
}
int run_nothing(void *_p, unsigned long long *data, int *data_len) {
struct elim_params *p = (struct elim_params *)_p;
mzd_t *A = mzd_init(p->m, p->n);
if(p->r != 0) {
mzd_t *L, *U;
L = mzd_init(p->m, p->m);
U = mzd_init(p->m, p->n);
mzd_randomize(U);
mzd_randomize(L);
for (rci_t i = 0; i < p->m; ++i) {
for (rci_t j = i + 1; j < p->m; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, p->m - j);
mzd_clear_bits(L, i, j, length);
}
mzd_write_bit(L,i,i, 1);
for (rci_t j = 0; j < i && j <p->n; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, i - j);
mzd_clear_bits(U, i, j, length);
}
if(i < p->r) {
mzd_write_bit(U, i, i, 1);
} else {
for (rci_t j = i; j < p->n; j+=m4ri_radix) {
int const length = MIN(m4ri_radix, p->n - j);
mzd_clear_bits(U, i, j, length);
}
}
}
mzd_mul(A,L,U,0);
mzd_free(L);
mzd_free(U);
} else {
mzd_randomize(A);
}
#ifndef HAVE_LIBPAPI
*data_len = 2;
#else
*data_len = MIN(papi_array_len + 1, *data_len);
#endif
int papi_res;
#ifndef HAVE_LIBPAPI
data[0] = walltime(0);
data[1] = cpucycles();
#else
int array_len = *data_len - 1;
unsigned long long t0 = PAPI_get_virt_usec();
papi_res = PAPI_start_counters((int*)papi_events, array_len);
if(papi_res)
m4ri_die("");
#endif
#ifndef HAVE_LIBPAPI
data[1] = cpucycles() - data[1];
data[0] = walltime(data[0]);
#else
PAPI_stop_counters((long long*)&data[1], array_len);
t0 = PAPI_get_virt_usec() - t0;
data[0] = t0;
for (int nv = 0; nv <= array_len; ++nv) {
if (data[nv] < loop_calibration[nv])
loop_calibration[nv] = data[nv];
}
#endif
mzd_free(A);
return (0);
}
void set_bit_at(gf2matrix *m, int bit, int row, int col)
{
if (!m)
return;
mzd_write_bit(m, row, col, bit);
}
int test_lqup_half_rank(size_t m, size_t n) {
printf("pluq: testing half rank m: %5zd, n: %5zd",m,n);
mzd_t* U = mzd_init(m, n);
mzd_t* L = mzd_init(m, m);
mzd_t* U2 = mzd_init(m, n);
mzd_t* L2 = mzd_init(m, m);
mzd_t* A = mzd_init(m, n);
mzd_randomize (U);
mzd_randomize (L);
size_t i,j;
for (i=0; i<m && i<n; ++i){
mzd_write_bit(U,i,i, 1);
for (j=0; j<i;++j)
mzd_write_bit(U,i,j, 0);
if (i%2)
for (j=i; j<n;++j)
mzd_write_bit(U,i,j, 0);
for (j=i+1; j<m;++j)
mzd_write_bit(L,i,j, 0);
mzd_write_bit(L,i,i, 1);
}
mzd_mul(A, L, U, 0);
mzd_t* Acopy = mzd_copy (NULL,A);
mzp_t* Pt = mzp_init(m);
mzp_t* Q = mzp_init(n);
int r = mzd_pluq(A, Pt, Q, 0);
for (i=0; i<r; ++i){
for (j=0; j<i;++j)
mzd_write_bit (L2, i, j, mzd_read_bit(A,i,j));
for (j=i+1; j<n;++j)
mzd_write_bit (U2, i, j, mzd_read_bit(A,i,j));
}
for (i=r; i<m; i++)
for (j=0; j<r;++j)
mzd_write_bit (L2, i, j, mzd_read_bit(A,i,j));
for (i=0; i<r; ++i){
mzd_write_bit(L2,i,i, 1);
mzd_write_bit(U2,i,i, 1);
}
mzd_apply_p_left(Acopy, Pt);
mzd_apply_p_right_trans(Acopy, Q);
mzd_addmul(Acopy,L2,U2,0);
int status = 0;
for ( i=0; i<m; ++i) {
for ( j=0; j<n; ++j){
if (mzd_read_bit(Acopy,i,j)){
status = 1;
}
}
if(status)
break;
}
if (status)
printf(" ... FAILED\n");
else
printf (" ... passed\n");
mzd_free(U);
mzd_free(L);
mzd_free(U2);
mzd_free(L2);
mzd_free(A);
mzd_free(Acopy);
mzp_free(Pt);
mzp_free(Q);
return status;
}
int test_lqup_full_rank (size_t m, size_t n){
printf("pluq: testing full rank m: %5zu, n: %5zu",m,n);
mzd_t* U = mzd_init (m,n);
mzd_t* L = mzd_init (m,m);
mzd_t* U2 = mzd_init (m,n);
mzd_t* L2 = mzd_init (m,m);
mzd_t* A = mzd_init (m,n);
mzd_randomize (U);
mzd_randomize (L);
size_t i,j;
for (i=0; i<m; ++i){
for (j=0; j<i && j<n;++j)
mzd_write_bit(U,i,j, 0);
for (j=i+1; j<m;++j)
mzd_write_bit(L,i,j, 0);
if(i<n)
mzd_write_bit(U,i,i, 1);
mzd_write_bit(L,i,i, 1);
}
mzd_mul(A, L, U, 2048);
mzd_t* Acopy = mzd_copy (NULL,A);
mzp_t* P = mzp_init(m);
mzp_t* Q = mzp_init(n);
mzd_pluq(A, P, Q, 2048);
for (i=0; i<m; ++i){
for (j=0; j<i && j <n;++j)
mzd_write_bit (L2, i, j, mzd_read_bit(A,i,j));
for (j=i+1; j<n;++j)
mzd_write_bit (U2, i, j, mzd_read_bit(A,i,j));
}
for (i=0; i<n && i<m; ++i){
mzd_write_bit(L2,i,i, 1);
mzd_write_bit(U2,i,i, 1);
}
mzd_addmul(Acopy,L2,U2,0);
int status = 0;
for ( i=0; i<m; ++i)
for ( j=0; j<n; ++j){
if (mzd_read_bit (Acopy,i,j)){
status = 1;
}
}
if (status){
printf(" ... FAILED\n");
} else
printf (" ... passed\n");
mzd_free(U);
mzd_free(L);
mzd_free(U2);
mzd_free(L2);
mzd_free(A);
mzd_free(Acopy);
mzp_free(P);
mzp_free(Q);
return status;
}
