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;
}
int test_pluq_structured(rci_t m, rci_t n) {
printf("pluq: testing structured m: %5d, n: %5d", m, n);
mzd_t* A = mzd_init(m, n);
mzd_t* L = mzd_init(m, m);
mzd_t* U = mzd_init(m, n);
for(rci_t i = 0; i < m; i += 2)
for (rci_t j = i; j < n; ++j)
mzd_write_bit(A, i, j, 1);
mzd_t* Acopy = mzd_copy (NULL,A);
mzp_t* P = mzp_init(m);
mzp_t* Q = mzp_init(n);
rci_t r = mzd_pluq(A, P, Q, 0);
printf(", rank: %5d ",r);
for (rci_t i = 0; i < r; ++i){
for (rci_t j = 0; j < i; ++j)
mzd_write_bit(L, i, j, mzd_read_bit(A,i,j));
for (rci_t j = i + 1; j < n; ++j)
mzd_write_bit(U, i, j, mzd_read_bit(A,i,j));
}
for (rci_t i = r; i < m; ++i)
for (rci_t j = 0; j < r; ++j)
mzd_write_bit(L, i, j, mzd_read_bit(A,i,j));
for (rci_t 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 (rci_t i = 0; i < m; ++i)
for (rci_t j = 0; j < n; ++j){
if (mzd_read_bit (Acopy,i,j)){
status = 1;
break;
}
}
if (status) {
printf("\n");
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 addmul_test_equality(rci_t m, rci_t l, rci_t n, int k, int cutoff) {
int ret = 0;
mzd_t *A, *B, *C, *D, *E, *F;
printf("addmul: m: %4d, l: %4d, n: %4d, k: %2d, cutoff: %4d", m, l, n, k, cutoff);
/* we create two random matrices */
A = mzd_init(m, l);
B = mzd_init(l, n);
C = mzd_init(m, n);
mzd_randomize(A);
mzd_randomize(B);
mzd_randomize(C);
/* D = C + A*B via M4RM, temporary buffers are managed internally */
D = mzd_copy(NULL, C);
D = mzd_addmul_m4rm(D, A, B, k);
/* E = C + A*B via naiv cubic multiplication */
E = mzd_mul_m4rm(NULL, A, B, k);
mzd_add(E, E, C);
/* F = C + A*B via naiv cubic multiplication */
F = mzd_copy(NULL, C);
F = mzd_addmul(F, A, B, cutoff);
mzd_free(A);
mzd_free(B);
mzd_free(C);
if (mzd_equal(D, E) != TRUE) {
printf(" M4RM != add,mul");
ret -=1;
}
if (mzd_equal(E, F) != TRUE) {
printf(" add,mul = addmul");
ret -=1;
}
if (mzd_equal(F, D) != TRUE) {
printf(" M4RM != addmul");
ret -=1;
}
if (ret==0)
printf(" ... passed\n");
else
printf(" ... FAILED\n");
mzd_free(D);
mzd_free(E);
mzd_free(F);
return ret;
}
int _mzd_pluq_solve_left(mzd_t const *A, rci_t rank,
mzp_t const *P, mzp_t const *Q,
mzd_t *B, int const cutoff, int const inconsistency_check) {
/** A is supposed to store L lower triangular and U upper triangular
* B is modified in place
* (Bi's in the comments are just modified versions of B)
* PLUQ = A
* 1) P B2 = B1
* 2) L B3 = B2
* 3) U B4 = B3
* 4) Q B5 = B4
*/
int retval = 0;
/* P B2 = B1 or B2 = P^T B1 */
mzd_apply_p_left(B, P);
/* L B3 = B2 */
/* view on the upper part of L */
mzd_t const *LU = mzd_init_window_const(A, 0, 0, rank, rank);
mzd_t *Y1 = mzd_init_window(B, 0, 0, rank, B->ncols);
mzd_trsm_lower_left(LU, Y1, cutoff);
if (inconsistency_check) { /* Check for inconsistency */
/** FASTER without this check; update with the lower part of L
*/
mzd_t const *H = mzd_init_window_const(A, rank, 0, A->nrows, rank);
mzd_t *Y2 = mzd_init_window(B, rank, 0, A->nrows, B->ncols);
if(A->nrows < B->nrows) {
mzd_t *Y3 = mzd_init_window(B, A->nrows, 0, B->nrows, B->ncols);
mzd_set_ui(Y3, 0);
mzd_free_window(Y3);
}
mzd_addmul(Y2, H, Y1, cutoff);
/*
* test whether Y2 is the zero matrix
*/
if(!mzd_is_zero(Y2)) {
retval = -1;
}
mzd_free_window((mzd_t*)H);
mzd_free_window(Y2);
}
/* U B4 = B3 */
mzd_trsm_upper_left(LU, Y1, cutoff);
mzd_free_window((mzd_t*)LU);
mzd_free_window(Y1);
if (!inconsistency_check) {
/** Default is to set the undefined bits to zero if inconsistency
* has been checked then Y2 bits are already all zeroes thus this
* clearing is not needed
*/
for(rci_t i = rank; i < B->nrows; ++i) {
for(rci_t j = 0; j < B->ncols; j += m4ri_radix) {
mzd_clear_bits(B, i, j, MIN(m4ri_radix, B->ncols - j));
}
}
}
/* Q B5 = B4 or B5 = Q^T B4 */
mzd_apply_p_left_trans(B, Q);
/* P L U Q B5 = B1 */
__M4RI_DD_MZD(B);
__M4RI_DD_INT(retval);
return retval;
}
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;
}
int smallops_test_add(rci_t M, rci_t N, rci_t m, rci_t n, rci_t offset, word pattern) {
int ret = 0;
printf(" mzd_add: M: %4d, N: %4d, m: %4d, n: %4d, offset: %4d, pattern: 0x%" PRIx64 " ", M, N, m, n, offset, pattern);
mzd_t *AA;
mzd_t *A = mzd_init_test_matrix_random(M, N, m, n, offset, pattern, &AA);
mzd_t *BB;
mzd_t *B = mzd_init_test_matrix_random(M, N, m, n, offset, pattern, &BB);
mzd_t *CC;
mzd_t *C = mzd_init_test_matrix_random(M, N, m, n, offset, pattern, &CC);
mzd_t *DD;
mzd_t *D = mzd_init_test_matrix_random(M, N, m, n, offset, pattern, &DD);
/* Creation went okay? */
ret += mzd_check_pattern(AA, m, n, offset, pattern);
ret += mzd_check_pattern(BB, m, n, offset, pattern);
ret += mzd_check_pattern(CC, m, n, offset, pattern);
ret += mzd_check_pattern(DD, m, n, offset, pattern);
/* Testing equality A+A == 0 */
mzd_add(C, A, A);
if(!mzd_is_zero(C)) {
ret +=1;
}
ret += mzd_check_pattern(AA, m, n, offset, pattern);
ret += mzd_check_pattern(BB, m, n, offset, pattern);
ret += mzd_check_pattern(CC, m, n, offset, pattern);
/* Testing equality A+A == 0 but this time C is already zero */
mzd_add(C, B, B);
if(!mzd_is_zero(C)) {
ret +=1;
}
ret += mzd_check_pattern(AA, m, n, offset, pattern);
ret += mzd_check_pattern(BB, m, n, offset, pattern);
ret += mzd_check_pattern(CC, m, n, offset, pattern);
/* Testing in place add. C is zero, so afterwards C == A */
mzd_add(C, C, A);
if(!mzd_equal(C,A)) {
ret +=1;
}
ret += mzd_check_pattern(AA, m, n, offset, pattern);
ret += mzd_check_pattern(BB, m, n, offset, pattern);
ret += mzd_check_pattern(CC, m, n, offset, pattern);
/* Testing equality C (== A) + A == 0 */
mzd_add(B, C, A);
if(!mzd_is_zero(B)) {
ret +=1;
}
if(m == n) {
/* Testing equality (A + B)^2 == A^2 + BA + AB + B^2 */
mzd_randomize(A);
mzd_randomize(B);
mzd_add(C,A,B);
mzd_mul(D,C,C, 0); // (A+B)^2
mzd_mul(C,A,A, 0);
mzd_addmul(C, B, A, 0);
mzd_addmul(C, A, B, 0);
mzd_addmul(C, B, B, 0);
if(!mzd_equal(C,D)) {
ret += 1;
}
ret += mzd_check_pattern(AA, m, n, offset, pattern);
ret += mzd_check_pattern(BB, m, n, offset, pattern);
ret += mzd_check_pattern(CC, m, n, offset, pattern);
ret += mzd_check_pattern(DD, m, n, offset, pattern);
}
mzd_free_test_matrix_random(AA, A);
mzd_free_test_matrix_random(BB, B);
mzd_free_test_matrix_random(CC, C);
mzd_free_test_matrix_random(DD, D);
if(ret == 0) {
printf(" ... passed\n");
} else {
printf(" ... FAILED\n");
}
#ifdef ABORT_ON_FAIL
if (ret) abort();
#endif
return ret;
}