diff --git a/compute/pzcesca.c b/compute/pzcesca.c
index b61c54aa664e008a053e1aaf5a9495db28d47e6f..d09505c2ff68796825105f20093e9f2d4d783a15 100644
--- a/compute/pzcesca.c
+++ b/compute/pzcesca.c
@@ -37,8 +37,8 @@ chameleon_pzcesca_internal( int center,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = A->p;
- int Q = A->q;
+ int P = chameleon_desc_datadist_get_iparam(A, 0);
+ int Q = chameleon_desc_datadist_get_iparam(A, 1);
/**
* 1) compute sums and sum-square (scl,ssq) in each tile
diff --git a/compute/pzgemm.c b/compute/pzgemm.c
index 5c6563d2d6d6280f0f08a35b9a25f266eb5e9764..a4a12eca743aa1e9ef69a15ddfec7bc1564edc92 100644
--- a/compute/pzgemm.c
+++ b/compute/pzgemm.c
@@ -195,15 +195,15 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
KT = transA == ChamNoTrans ? A->nt : A->mt;
K = transA == ChamNoTrans ? A->n : A->m;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
/*
* A: ChamNoTrans / B: ChamNoTrans
*/
for (k = 0; k < KT; k++ ) {
- lp = (k % lookahead) * C->p;
- lq = (k % lookahead) * C->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 1);
tempkk = k == KT - 1 ? K - k * A->nb : A->nb;
zbeta = k == 0 ? beta : zone;
@@ -216,16 +216,16 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempmm, tempkk,
A( m, k ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (k % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, A( m, k ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempmm, tempkk,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((k+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((k+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
else {
@@ -233,16 +233,16 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempkk, tempmm,
A( k, m ),
- WA( m, (m % C->q) + lq ) );
+ WA( m, (m % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, A( k, m ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempkk, tempmm,
- WA( m, ((m+q-1) % C->q) + lq ),
- WA( m, ((m+q) % C->q) + lq ) );
+ WA( m, ((m+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((m+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
}
@@ -256,16 +256,16 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempkk, tempnn,
B( k, n ),
- WB( (k % C->p) + lp, n ) );
+ WB( (k % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, B( k, n ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempkk, tempnn,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((k+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((k+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
else {
@@ -273,24 +273,24 @@ chameleon_pzgemm_summa( CHAM_context_t *chamctxt, cham_trans_t transA, cham_tran
options,
ChamUpperLower, tempnn, tempkk,
B( n, k ),
- WB( (n % C->p) + lp, n ) );
+ WB( (n % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, B( n, k ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempnn, tempkk,
- WB( ((n+p-1) % C->p) + lp, n ),
- WB( ((n+p) % C->p) + lp, n ) );
+ WB( ((n+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((n+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
}
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zgemm(
diff --git a/compute/pzgenm2.c b/compute/pzgenm2.c
index 97f8ae8937b6c5c5821bb2e5ae8226fcbb0ca65b..99455940aae73822a1e7dfa937873b30715dce9d 100644
--- a/compute/pzgenm2.c
+++ b/compute/pzgenm2.c
@@ -38,8 +38,8 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
RUNTIME_option_t options;
CHAM_desc_t X, SX, NRMX, NRMSX, DROW;
int m, n, k;
- int myp = A->myrank / A->q;
- int myq = A->myrank % A->q;
+ int myp = A->myrank / chameleon_desc_datadist_get_iparam(A, 1);
+ int myq = A->myrank % chameleon_desc_datadist_get_iparam(A, 1);
int tempmm, tempnn;
int cnt, maxiter;
double e0, normx, normsx, beta, scl;
@@ -67,7 +67,9 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
* This is the same issue for X and SX to be reused from one iteration to another.
*/
chameleon_desc_init( &DROW, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, A->nb, A->nb,
- A->p, A->n, 0, 0, A->p, A->n, A->p, A->q,
+ chameleon_desc_datadist_get_iparam(A, 0), A->n, 0, 0, chameleon_desc_datadist_get_iparam(A, 0), A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
/**
* NRMX must be allocated with GLOBAL to be able to access the norm value
@@ -75,7 +77,10 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
* This is the same issue for NRMSX.
*/
chameleon_desc_init( &NRMX, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- A->p * 2, A->q, 0, 0, A->p * 2, A->q, A->p, A->q,
+ chameleon_desc_datadist_get_iparam(A, 0) * 2, chameleon_desc_datadist_get_iparam(A, 1), 0, 0
+ , chameleon_desc_datadist_get_iparam(A, 0) * 2, chameleon_desc_datadist_get_iparam(A, 1),
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
/**
@@ -86,7 +91,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
* So drow[j] = sum( S_{p,j}, p=0..P-1 ) with S_{p,j} = sum( |A_{i,j}|, i=0..m-1 \ i%P = p )
*
*/
- for(n = myq; n < A->nt; n += A->q) {
+ for(n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1)) {
tempnn = n == A->nt-1 ? A->n - n * A->nb : A->nb;
/* Zeroes the local intermediate vector */
@@ -97,7 +102,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
DROW( myp, n ) );
/* Computes the sums of the local tiles into the local vector */
- for(m = myp; m < A->mt; m += A->p) {
+ for(m = myp; m < A->mt; m += chameleon_desc_datadist_get_iparam(A, 0)) {
tempmm = m == A->mt-1 ? A->m - m * A->mb : A->mb;
INSERT_TASK_dzasum(
&options,
@@ -106,7 +111,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce on first row of nodes */
- for(m = 1; m < A->p; m++) {
+ for(m = 1; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
INSERT_TASK_daxpy(
&options, tempnn, 1.,
DROW( m, n ), 1,
@@ -125,7 +130,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
1., 0.,
NRMX( myp, myq ) );
- for( n = myq; n < A->nt; n += A->q ) {
+ for( n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1) ) {
tempnn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
INSERT_TASK_dgessq(
&options, ChamEltwise, 1, tempnn,
@@ -134,7 +139,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce on first row of nodes */
- for(n = 1; n < A->q; n++) {
+ for(n = 1; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
INSERT_TASK_dplssq(
&options, ChamEltwise, 1, 1,
NRMX( myp, n ),
@@ -146,8 +151,8 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Bcast norm over processes from node (0,0) */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ for(m = 0; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
+ for(n = 0; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
if ( (m != 0) || (n != 0) ) {
INSERT_TASK_dlacpy(
&options,
@@ -171,13 +176,22 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
chameleon_desc_init( &NRMSX, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- A->p * 2, A->q, 0, 0, A->p * 2, A->q, A->p, A->q,
+ chameleon_desc_datadist_get_iparam(A, 0) * 2, chameleon_desc_datadist_get_iparam(A, 1), 0, 0,
+ chameleon_desc_datadist_get_iparam(A, 0) * 2, chameleon_desc_datadist_get_iparam(A, 1),
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &X, CHAMELEON_MAT_ALLOC_GLOBAL, ChamComplexDouble, 1, A->nb, A->nb,
- A->p, A->n, 0, 0, A->p, A->n, A->p, A->q,
+ chameleon_desc_datadist_get_iparam(A, 0), A->n, 0, 0,
+ chameleon_desc_datadist_get_iparam(A, 0), A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &SX, CHAMELEON_MAT_ALLOC_GLOBAL, ChamComplexDouble, A->mb, 1, A->mb,
- A->m, A->q, 0, 0, A->m, A->q, A->p, A->q,
+ A->m, chameleon_desc_datadist_get_iparam(A, 1), 0, 0,
+ A->m, chameleon_desc_datadist_get_iparam(A, 1),
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
cnt = 0;
@@ -192,7 +206,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
/* Initialization of X in the first loop */
if ( cnt == 0 )
{
- for (n = myq; n < A->nt; n += A->q) {
+ for (n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1)) {
tempnn = n == A->nt-1 ? A->n - n * A->nb : A->nb;
if ( myp == 0 ) {
@@ -212,7 +226,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Broadcast X */
- for (m = 1; m < A->p; m++) {
+ for (m = 1; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, 1, tempnn,
@@ -230,7 +244,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
* copy of the scaled X.
*/
scl = 1. / e0;
- for (n = myq; n < A->nt; n += A->q) {
+ for (n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1)) {
tempnn = n == A->nt-1 ? A->n - n * A->nb : A->nb;
INSERT_TASK_zlascal(
@@ -242,10 +256,10 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
/**
* Compute Sx = S * x
*/
- for(m = myp; m < A->mt; m+=A->p) {
+ for(m = myp; m < A->mt; m+=chameleon_desc_datadist_get_iparam(A, 0)) {
tempmm = m == A->mt-1 ? A->m - m * A->mb : A->mb;
- for (n = myq; n < A->nt; n += A->q ) {
+ for (n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1) ) {
tempnn = n == A->nt-1 ? A->n - n * A->nb : A->nb;
beta = n == myq ? 0. : 1.;
@@ -258,14 +272,14 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce columns */
- for (k = 1; k < chameleon_min( A->q, A->nt ); k++) {
+ for (k = 1; k < chameleon_min( chameleon_desc_datadist_get_iparam(A, 1), A->nt ); k++) {
INSERT_TASK_zaxpy(
&options, tempmm, 1.,
SX( m, k ), 1,
SX( m, 0 ), 1 );
}
/* Broadcast SX to ease the following gemv */
- for (k = 1; k < A->q; k++) {
+ for (k = 1; k < chameleon_desc_datadist_get_iparam(A, 1); k++) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempmm, 1,
@@ -277,10 +291,10 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
/**
* Compute x = S' * S * x = S' * Sx
*/
- for ( n = myq; n < A->nt; n += A->q ) {
+ for ( n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1) ) {
tempnn = n == A->nt-1 ? A->n - n * A->nb : A->nb;
- for( m = myp; m < A->mt; m += A->p ) {
+ for( m = myp; m < A->mt; m += chameleon_desc_datadist_get_iparam(A, 0) ) {
tempmm = m == A->mt-1 ? A->m - m * A->mb : A->mb;
beta = m == myp ? 0. : 1.;
@@ -293,14 +307,14 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce rows */
- for (k = 1; k < chameleon_min( A->p, A->mt ); k++) {
+ for (k = 1; k < chameleon_min( chameleon_desc_datadist_get_iparam(A, 0), A->mt ); k++) {
INSERT_TASK_zaxpy(
&options, tempnn, 1.,
X( k, n ), 1,
X( 0, n ), 1 );
}
/* Broadcast */
- for (k = 1; k < A->p; k++) {
+ for (k = 1; k < chameleon_desc_datadist_get_iparam(A, 0); k++) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, 1, tempnn,
@@ -321,7 +335,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
1., 0.,
NRMX( myp, myq ) );
- for( n = myq; n < A->nt; n += A->q ) {
+ for( n = myq; n < A->nt; n += chameleon_desc_datadist_get_iparam(A, 1) ) {
tempnn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
INSERT_TASK_zgessq(
@@ -331,7 +345,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce columns */
- for(n = 1; n < chameleon_min( A->q, A->nt ); n++) {
+ for(n = 1; n < chameleon_min( chameleon_desc_datadist_get_iparam(A, 1), A->nt ); n++) {
INSERT_TASK_dplssq(
&options, ChamEltwise, 1, 1,
NRMX( myp, n ),
@@ -342,7 +356,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
&options, 1, NRMX( myp, 0 ) );
/* Broadcast the results to processes in the same row */
- for(n = 1; n < A->q; n++) {
+ for(n = 1; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
INSERT_TASK_dlacpy(
&options,
ChamUpperLower, 1, 1,
@@ -363,7 +377,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
1., 0.,
NRMSX( myp, myq ) );
- for( m = myp; m < A->mt; m += A->p ) {
+ for( m = myp; m < A->mt; m += chameleon_desc_datadist_get_iparam(A, 0) ) {
tempmm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
INSERT_TASK_zgessq(
&options, ChamEltwise, tempmm, 1,
@@ -372,7 +386,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
}
/* Reduce rows */
- for( m = 1; m < chameleon_min( A->p, A->mt ); m++ ) {
+ for( m = 1; m < chameleon_min( chameleon_desc_datadist_get_iparam(A, 0), A->mt ); m++ ) {
INSERT_TASK_dplssq(
&options, ChamEltwise, 1, 1,
NRMSX( m, myq ),
@@ -383,7 +397,7 @@ chameleon_pzgenm2( double tol, const CHAM_desc_t *A, double *result,
&options, 1, NRMSX( 0, myq ) );
/* Broadcast the results to processes in the same column */
- for(m = 1; m < A->p; m++) {
+ for(m = 1; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
INSERT_TASK_dlacpy(
&options,
ChamUpperLower, 1, 1,
diff --git a/compute/pzgepdf_qdwh.c b/compute/pzgepdf_qdwh.c
index 0d1b66cb267f374b5ce493c01a2ecb619603d645..a96afd5b56ff21bd8c04a200bca9d353aafd697a 100644
--- a/compute/pzgepdf_qdwh.c
+++ b/compute/pzgepdf_qdwh.c
@@ -109,23 +109,31 @@ chameleon_pzgepdf_qdwh_init( const CHAM_desc_t *U, const CHAM_desc_t *H,
chameleon_desc_init( TS1, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, nb, ib * nb,
ib * U->mt, nb * U->nt, 0, 0,
- ib * U->mt, nb * U->nt, U->p, U->q,
+ ib * U->mt, nb * U->nt,
+ chameleon_desc_datadist_get_iparam(U, 0),
+ chameleon_desc_datadist_get_iparam(U, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( TT1, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, nb, ib * nb,
ib * U->mt, nb * U->nt, 0, 0,
- ib * U->mt, nb * U->nt, H->p, H->q,
+ ib * U->mt, nb * U->nt,
+ chameleon_desc_datadist_get_iparam(H, 0),
+ chameleon_desc_datadist_get_iparam(H, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( TS2, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, nb, ib * nb,
ib * H->mt, nb * H->nt, 0, 0,
- ib * H->mt, nb * H->nt, U->p, U->q,
+ ib * H->mt, nb * H->nt,
+ chameleon_desc_datadist_get_iparam(U, 0),
+ chameleon_desc_datadist_get_iparam(U, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( TT2, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, nb, ib * nb,
ib * H->mt, nb * H->nt, 0, 0,
- ib * H->mt, nb * H->nt, H->p, H->q,
+ ib * H->mt, nb * H->nt,
+ chameleon_desc_datadist_get_iparam(H, 0),
+ chameleon_desc_datadist_get_iparam(H, 1),
NULL, NULL, NULL, NULL );
/*
@@ -135,8 +143,8 @@ chameleon_pzgepdf_qdwh_init( const CHAM_desc_t *U, const CHAM_desc_t *H,
libhqr_matrix_t mat = {
.mt = B1->mt,
.nt = B1->nt,
- .nodes = B1->p * B1-> q,
- .p = B1->p,
+ .nodes = chameleon_desc_datadist_get_iparam(B1, 0) * chameleon_desc_datadist_get_iparam(B1,1),
+ .p = chameleon_desc_datadist_get_iparam(B1, 0),
};
/* Tree for the top matrix */
@@ -144,7 +152,7 @@ chameleon_pzgepdf_qdwh_init( const CHAM_desc_t *U, const CHAM_desc_t *H,
-1, /*low level tree */
-1, /* high level tree */
-1, /* TS tree size */
- B1->p, /* High level size */
+ chameleon_desc_datadist_get_iparam(B1, 0), /* High level size */
-1, /* Domino */
0 /* TSRR (unstable) */ );
@@ -156,7 +164,7 @@ chameleon_pzgepdf_qdwh_init( const CHAM_desc_t *U, const CHAM_desc_t *H,
/* high level tree (Could be greedy, but flat should reduce the volume of comm) */
LIBHQR_FLAT_TREE,
-1, /* TS tree size */
- B2->p /* High level size */ );
+ chameleon_desc_datadist_get_iparam(B2, 0) /* High level size */ );
}
/*
@@ -170,7 +178,9 @@ chameleon_pzgepdf_qdwh_init( const CHAM_desc_t *U, const CHAM_desc_t *H,
chameleon_desc_init( Ut, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, U->mb, U->nb, U->mb * U->nb,
U->n, U->m, 0, 0,
- U->n, U->m, U->p, U->q,
+ U->n, U->m,
+ chameleon_desc_datadist_get_iparam(U, 0),
+ chameleon_desc_datadist_get_iparam(U, 1),
NULL, NULL, NULL, NULL );
/*
diff --git a/compute/pzgered.c b/compute/pzgered.c
index e7feeed9d7ba8f0a3e433295e1d5af449035d531..dd2c4a597c063ae33f468c529a6302808e3f7356 100644
--- a/compute/pzgered.c
+++ b/compute/pzgered.c
@@ -44,8 +44,8 @@ chameleon_pzgered_frb( cham_uplo_t uplo,
int NT = (uplo == ChamLower) ? minMNT : A->nt;
int M = (uplo == ChamUpper) ? minMN : A->m;
int N = (uplo == ChamLower) ? minMN : A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/* Initialize workspaces for tile norms */
for(m = 0; m < Wnorm->mt; m++) {
@@ -142,8 +142,8 @@ chameleon_pzgered_frb( cham_uplo_t uplo,
/**
* Broadcast the result
*/
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ for(m = 0; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
+ for(n = 0; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
if ( (m != 0) || (n != 0) ) {
INSERT_TASK_dlacpy(
options,
@@ -178,19 +178,23 @@ void chameleon_pzgered( cham_uplo_t uplo,
}
RUNTIME_options_init(&options, chamctxt, sequence, request);
- workmt = chameleon_max( A->mt, A->p );
- worknt = chameleon_max( A->nt, A->q );
+ workmt = chameleon_max( A->mt, chameleon_desc_datadist_get_iparam(A, 0) );
+ worknt = chameleon_max( A->nt, chameleon_desc_datadist_get_iparam(A, 1) );
RUNTIME_options_ws_alloc( &options, 1, 0 );
/* Matrix to store the norm of each element */
chameleon_desc_init( &Wcol, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- A->mt * 2, A->nt, 0, 0, A->mt * 2, A->nt, A->p, A->q,
+ A->mt * 2, A->nt, 0, 0, A->mt * 2, A->nt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
/* Matrix to compute the global frobenius norm */
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- workmt*2, worknt, 0, 0, workmt*2, worknt, A->p, A->q,
+ workmt*2, worknt, 0, 0, workmt*2, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_pzgered_frb( uplo, A, &Wcol, &Welt, &options );
@@ -201,7 +205,7 @@ void chameleon_pzgered( cham_uplo_t uplo,
RUNTIME_sequence_wait( chamctxt, sequence );
- gnorm = *((double *)Welt.get_blkaddr( &Welt, A->myrank / A->q, A->myrank % A->q ));
+ gnorm = *((double *)Welt.get_blkaddr( &Welt, A->myrank / chameleon_desc_datadist_get_iparam(A, 1), A->myrank % chameleon_desc_datadist_get_iparam(A, 1) ));
chameleon_desc_destroy( &Welt );
/**
diff --git a/compute/pzgetrf_nopiv.c b/compute/pzgetrf_nopiv.c
index 18fb33ed81656b462cdcff0f52fe82e6080b90de..424842bc06c9639442124d13509e1f2308b11820 100644
--- a/compute/pzgetrf_nopiv.c
+++ b/compute/pzgetrf_nopiv.c
@@ -57,7 +57,7 @@ void chameleon_pzgetrf_nopiv_generic( CHAM_desc_t *A,
if ( chamctxt->autominmax_enabled && (chamctxt->scheduler == RUNTIME_SCHED_STARPU) ) {
int lookahead = chamctxt->lookahead;
- int nbtasks_per_step = (A->mt * A->nt) / (A->p * A->q);
+ int nbtasks_per_step = (A->mt * A->nt) / (chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1));
int mintasks = nbtasks_per_step * lookahead;
int maxtasks = nbtasks_per_step * (lookahead+1);
@@ -149,13 +149,13 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
ib = CHAMELEON_IB;
lookahead = chamctxt->lookahead;
- myp = A->myrank / A->q;
- myq = A->myrank % A->q;
+ myp = A->myrank / chameleon_desc_datadist_get_iparam(A, 1);
+ myq = A->myrank % chameleon_desc_datadist_get_iparam(A, 1);
for (k = 0; k < chameleon_min(A->mt, A->nt); k++) {
RUNTIME_iteration_push(chamctxt, k);
- lp = (k % lookahead) * A->p;
- lq = (k % lookahead) * A->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(A, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(A, 1);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
@@ -174,27 +174,27 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
&options,
ChamUpperLower, tempkm, tempkn,
A( k, k ),
- WL( k, (k % A->q) + lq ) );
+ WL( k, (k % chameleon_desc_datadist_get_iparam(A, 1)) + lq ) );
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempkm, tempkn,
A( k, k ),
- WU( (k % A->p) + lp, k ) );
+ WU( (k % chameleon_desc_datadist_get_iparam(A, 0)) + lp, k ) );
- for ( q=1; q < A->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(A, 1); q++ ) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempkm, tempkn,
- WL( k, ((k+q-1) % A->q) + lq ),
- WL( k, ((k+q) % A->q) + lq ) );
+ WL( k, ((k+q-1) % chameleon_desc_datadist_get_iparam(A, 1)) + lq ),
+ WL( k, ((k+q) % chameleon_desc_datadist_get_iparam(A, 1)) + lq ) );
}
- for ( p=1; p < A->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(A, 0); p++ ) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempkm, tempkn,
- WU( ((k+p-1) % A->p) + lp, k ),
- WU( ((k+p) % A->p) + lp, k ) );
+ WU( ((k+p-1) % chameleon_desc_datadist_get_iparam(A, 0)) + lp, k ),
+ WU( ((k+p) % chameleon_desc_datadist_get_iparam(A, 0)) + lp, k ) );
}
}
RUNTIME_data_flush( sequence, A( k, k ) );
@@ -202,7 +202,7 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
for (m = k+1; m < A->mt; m++) {
/* Skip the row if you are not involved with */
- if ( m%A->p != myp ) {
+ if ( m%chameleon_desc_datadist_get_iparam(A, 0) != myp ) {
continue;
}
@@ -219,19 +219,19 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
/* Broadcast A(m,k) into temp buffers through a ring */
{
- assert( A->get_rankof( A, m, k ) == WL->get_rankof( WL, m, (k % A->q) + lq) );
+ assert( A->get_rankof( A, m, k ) == WL->get_rankof( WL, m, (k % chameleon_desc_datadist_get_iparam(A, 1)) + lq) );
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempmm, tempkn,
A( m, k ),
- WL( m, (k % A->q) + lq) );
+ WL( m, (k % chameleon_desc_datadist_get_iparam(A, 1)) + lq) );
- for ( q=1; q < A->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(A, 1); q++ ) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempmm, tempkn,
- WL( m, ((k+q-1) % A->q) + lq ),
- WL( m, ((k+q) % A->q) + lq ) );
+ WL( m, ((k+q-1) % chameleon_desc_datadist_get_iparam(A, 1)) + lq ),
+ WL( m, ((k+q) % chameleon_desc_datadist_get_iparam(A, 1)) + lq ) );
}
}
RUNTIME_data_flush( sequence, A( m, k ) );
@@ -240,7 +240,7 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
for (n = k+1; n < A->nt; n++) {
/* Skip the column if you are not involved with */
- if ( n%A->q != myq ) {
+ if ( n%chameleon_desc_datadist_get_iparam(A, 1) != myq ) {
continue;
}
@@ -257,19 +257,19 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
/* Broadcast A(k,n) into temp buffers through a ring */
{
- assert( A->get_rankof( A, k, n ) == WU->get_rankof( WU, (k%A->p) + lp, n) );
+ assert( A->get_rankof( A, k, n ) == WU->get_rankof( WU, (k%chameleon_desc_datadist_get_iparam(A, 0)) + lp, n) );
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempkm, tempnn,
A( k, n ),
- WU( (k % A->p) + lp, n ) );
+ WU( (k % chameleon_desc_datadist_get_iparam(A, 0)) + lp, n ) );
- for ( p=1; p < A->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(A, 0); p++ ) {
INSERT_TASK_zlacpy(
&options,
ChamUpperLower, tempkm, tempnn,
- WU( ((k+p-1) % A->p) + lp, n ),
- WU( ((k+p) % A->p) + lp, n ) );
+ WU( ((k+p-1) % chameleon_desc_datadist_get_iparam(A, 0)) + lp, n ),
+ WU( ((k+p) % chameleon_desc_datadist_get_iparam(A, 0)) + lp, n ) );
}
}
RUNTIME_data_flush( sequence, A( k, n ) );
@@ -277,7 +277,7 @@ void chameleon_pzgetrf_nopiv_ws( CHAM_desc_t *A,
for (m = k+1; m < A->mt; m++) {
/* Skip the row if you are not involved with */
- if ( m%A->p != myp ) {
+ if ( m%chameleon_desc_datadist_get_iparam(A, 0) != myp ) {
continue;
}
diff --git a/compute/pzgram.c b/compute/pzgram.c
index 0ceda1b41e389fa4063f10b01357b909ded9edcf..04f849c6ba3a0092ec846fd31a56fdc259e7bc59 100644
--- a/compute/pzgram.c
+++ b/compute/pzgram.c
@@ -33,8 +33,8 @@ chameleon_pzgram_internal( cham_uplo_t uplo,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* 1) compute (scl,ssq) over columns in each tile
diff --git a/compute/pzhemm.c b/compute/pzhemm.c
index 12269d34a3abf4de45163eb75657441cceac5e74..1d9205311c3e108a0fe4162a5d2f32ef9ed528e0 100644
--- a/compute/pzhemm.c
+++ b/compute/pzhemm.c
@@ -302,12 +302,12 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
lookahead = chamctxt->lookahead;
KT = A->nt;
K = A->n;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
for (k = 0; k < KT; k++ ) {
- lp = (k % lookahead) * C->p;
- lq = (k % lookahead) * C->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 1);
tempkk = k == KT - 1 ? K - k * A->nb : A->nb;
zbeta = k == 0 ? beta : zone;
@@ -339,16 +339,16 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempam, tempak,
A( Am, Ak ),
- WA( m, (Ak % C->q) + lq ) );
+ WA( m, (Ak % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, A( Am, Ak ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempam, tempak,
- WA( m, ((Ak+q-1) % C->q) + lq ),
- WA( m, ((Ak+q) % C->q) + lq ) );
+ WA( m, ((Ak+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((Ak+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
@@ -361,25 +361,25 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempkk, tempnn,
B( k, n ),
- WB( (k % C->p) + lp, n ) );
+ WB( (k % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, B( k, n ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempkk, tempnn,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((k+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((k+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
/* Perform the update of this iteration */
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
if ( k == m ) {
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zhemm(
@@ -400,7 +400,7 @@ chameleon_pzhemm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
transA = ChamNoTrans;
}
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zgemm(
@@ -438,12 +438,12 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
lookahead = chamctxt->lookahead;
KT = A->mt;
K = A->m;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
for (k = 0; k < KT; k++ ) {
- lp = (k % lookahead) * C->p;
- lq = (k % lookahead) * C->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 1);
tempkk = k == KT - 1 ? K - k * A->nb : A->nb;
zbeta = k == 0 ? beta : zone;
@@ -456,16 +456,16 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempmm, tempkk,
B( m, k ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (k % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, B( m, k ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempmm, tempkk,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((k+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((k+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
@@ -496,25 +496,25 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempak, tempan,
A( Ak, An ),
- WB( (Ak % C->p) + lp, n ) );
+ WB( (Ak % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, A( Ak, An ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempak, tempan,
- WB( ((Ak+p-1) % C->p) + lp, n ),
- WB( ((Ak+p) % C->p) + lp, n ) );
+ WB( ((Ak+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((Ak+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
/* Perform the update of this iteration */
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
if ( k == n ) {
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
/* A has been stored in WA or WB for the summa ring */
@@ -536,7 +536,7 @@ chameleon_pzhemm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
transA = ChamNoTrans;
}
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zgemm(
diff --git a/compute/pzhered.c b/compute/pzhered.c
index 869d748cf49c4d131ccb90292b492bb3658167da..621cb87ae8a3e3136cf95b9acc12502ee6b7543f 100644
--- a/compute/pzhered.c
+++ b/compute/pzhered.c
@@ -43,8 +43,8 @@ chameleon_pzhered_frb( cham_trans_t trans,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/* Initialize workspaces for tile norms */
for (m = 0; m < Wnorm->mt; m++)
@@ -163,9 +163,9 @@ chameleon_pzhered_frb( cham_trans_t trans,
/**
* Broadcast the result
*/
- for (m = 0; m < A->p; m++)
+ for (m = 0; m < chameleon_desc_datadist_get_iparam(A, 0); m++)
{
- for (n = 0; n < A->q; n++)
+ for (n = 0; n < chameleon_desc_datadist_get_iparam(A, 1); n++)
{
if ( ( m != 0 ) || ( n != 0 ) )
{
@@ -204,20 +204,20 @@ void chameleon_pzhered( cham_trans_t trans,
}
RUNTIME_options_init(&options, chamctxt, sequence, request);
- workmt = chameleon_max(A->mt, A->p);
- worknt = chameleon_max(A->nt, A->q);
+ workmt = chameleon_max(A->mt, chameleon_desc_datadist_get_iparam(A, 0));
+ worknt = chameleon_max(A->nt, chameleon_desc_datadist_get_iparam(A, 1));
RUNTIME_options_ws_alloc(&options, 1, 0);
/* Matrix to store the norm of each element */
- chameleon_desc_init( &Wcol, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- A->mt * 2, A->nt, 0, 0, A->mt * 2, A->nt, A->p, A->q,
- NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
+ chameleon_desc_init(&Wcol, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
+ A->mt * 2, A->nt, 0, 0, A->mt * 2, A->nt, chameleon_desc_datadist_get_iparam(A, 0), chameleon_desc_datadist_get_iparam(A, 1),
+ NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg);
/* Matrix to compute the global frobenius norm */
- chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- workmt * 2, worknt, 0, 0, workmt * 2, worknt, A->p, A->q,
- NULL, NULL, NULL, NULL );
+ chameleon_desc_init(&Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
+ workmt * 2, worknt, 0, 0, workmt * 2, worknt, chameleon_desc_datadist_get_iparam(A, 0), chameleon_desc_datadist_get_iparam(A, 1),
+ NULL, NULL, NULL, NULL);
chameleon_pzhered_frb( trans, uplo, A, &Wcol, &Welt, &options );
@@ -227,7 +227,7 @@ void chameleon_pzhered( cham_trans_t trans,
RUNTIME_sequence_wait( chamctxt, sequence );
- gnorm = *((double *)Welt.get_blkaddr(&Welt, A->myrank / A->q, A->myrank % A->q));
+ gnorm = *((double *)Welt.get_blkaddr(&Welt, A->myrank / chameleon_desc_datadist_get_iparam(A, 1), A->myrank % chameleon_desc_datadist_get_iparam(A, 1)));
chameleon_desc_destroy(&Welt);
/**
diff --git a/compute/pzhetrd_he2hb.c b/compute/pzhetrd_he2hb.c
index ca26a7f0af1a87ff0c5b43af8529ff02d13d58dc..706740891ade6bc8d3cd10ac1ff1be2338ce0c55 100644
--- a/compute/pzhetrd_he2hb.c
+++ b/compute/pzhetrd_he2hb.c
@@ -90,7 +90,9 @@ void chameleon_pzhetrd_he2hb(cham_uplo_t uplo,
RUNTIME_options_ws_alloc( &options, ws_worker, ws_host );
/* Copy of the diagonal tiles to keep the general version of the tile all along the computation */
- chameleon_zdesc_alloc_diag( &D, A->mb, A->m, A->n, A->p, A->q );
+ chameleon_zdesc_alloc_diag( &D, A->mb, A->m, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1) );
chameleon_desc_init( &AT, CHAMELEON_MAT_ALLOC_GLOBAL, ChamComplexDouble, A->mb, A->nb, (A->mb*A->nb),
chameleon_min(A->mt, A->nt) * A->mb, A->nb, 0, 0,
diff --git a/compute/pzlange.c b/compute/pzlange.c
index 0917c5f40e017a146bbe8aa901586512033dd7c0..abfb27a366655537e0fa0e6e4a48d7e3626aa2bf 100644
--- a/compute/pzlange.c
+++ b/compute/pzlange.c
@@ -43,8 +43,8 @@ chameleon_pzlange_one( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A,
int NT = (uplo == ChamLower) ? minMNT : A->nt;
int M = (uplo == ChamUpper) ? minMN : A->m;
int N = (uplo == ChamLower) ? minMN : A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -133,8 +133,8 @@ chameleon_pzlange_inf( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A,
int NT = (uplo == ChamLower) ? minMNT : A->nt;
int M = (uplo == ChamUpper) ? minMN : A->m;
int N = (uplo == ChamLower) ? minMN : A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -219,8 +219,8 @@ chameleon_pzlange_max( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A, CHAM_
int NT = (uplo == ChamLower) ? minMNT : A->nt;
int M = (uplo == ChamUpper) ? minMN : A->m;
int N = (uplo == ChamLower) ? minMN : A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -302,8 +302,8 @@ chameleon_pzlange_frb( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A, CHAM_
int NT = (uplo == ChamLower) ? minMNT : A->nt;
int M = (uplo == ChamUpper) ? minMN : A->m;
int N = (uplo == ChamLower) ? minMN : A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -395,15 +395,17 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
*result = 0.0;
- workmt = chameleon_max( A->mt, A->p );
- worknt = chameleon_max( A->nt, A->q );
+ workmt = chameleon_max( A->mt, chameleon_desc_datadist_get_iparam(A, 0) );
+ worknt = chameleon_max( A->nt, chameleon_desc_datadist_get_iparam(A, 1) );
switch ( norm ) {
case ChamOneNorm:
RUNTIME_options_ws_alloc( &options, 1, 0 );
chameleon_desc_init( &Wcol, CHAMELEON_MAT_ALLOC_TILE, ChamRealDouble, 1, A->nb, A->nb,
- workmt, worknt * A->nb, 0, 0, workmt, worknt * A->nb, A->p, A->q,
+ workmt, worknt * A->nb, 0, 0, workmt, worknt * A->nb,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
wcol_init = 1;
@@ -411,7 +413,9 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
* Use the global allocator for Welt, otherwise flush may free the data before the result is read.
*/
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, 1, 1,
- A->p, worknt, 0, 0, A->p, worknt, A->p, A->q,
+ chameleon_desc_datadist_get_iparam(A, 0), worknt, 0, 0, chameleon_desc_datadist_get_iparam(A, 0), worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
break;
@@ -423,12 +427,16 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
RUNTIME_options_ws_alloc( &options, A->mb, 0 );
chameleon_desc_init( &Wcol, CHAMELEON_MAT_ALLOC_TILE, ChamRealDouble, A->mb, 1, A->mb,
- workmt * A->mb, worknt, 0, 0, workmt * A->mb, worknt, A->p, A->q,
+ workmt * A->mb, worknt, 0, 0, workmt * A->mb, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
wcol_init = 1;
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, 1, 1,
- workmt, A->q, 0, 0, workmt, A->q, A->p, A->q,
+ workmt, chameleon_desc_datadist_get_iparam(A, 1), 0, 0, workmt, chameleon_desc_datadist_get_iparam(A, 1),
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
break;
@@ -440,7 +448,9 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
alpha = 1.;
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- workmt*2, worknt, 0, 0, workmt*2, worknt, A->p, A->q,
+ workmt*2, worknt, 0, 0, workmt*2, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
break;
@@ -452,7 +462,9 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
RUNTIME_options_ws_alloc( &options, 1, 0 );
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, 1, 1,
- workmt, worknt, 0, 0, workmt, worknt, A->p, A->q,
+ workmt, worknt, 0, 0, workmt, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
}
@@ -504,8 +516,8 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
/**
* Broadcast the result
*/
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ for(m = 0; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
+ for(n = 0; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
if ( (m != 0) || (n != 0) ) {
INSERT_TASK_dlacpy(
&options,
@@ -522,7 +534,7 @@ void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_dia
CHAMELEON_Desc_Flush( A, sequence );
RUNTIME_sequence_wait( chamctxt, sequence );
- *result = *((double *)Welt.get_blkaddr( &Welt, A->myrank / A->q, A->myrank % A->q ));
+ *result = *((double *)Welt.get_blkaddr( &Welt, A->myrank / chameleon_desc_datadist_get_iparam(A, 1), A->myrank % chameleon_desc_datadist_get_iparam(A, 1) ));
if ( wcol_init ) {
chameleon_desc_destroy( &Wcol );
diff --git a/compute/pzlansy.c b/compute/pzlansy.c
index fb4b0f4e97a8db5d73a35ef4716a2470eb3d1f54..ab709ac9e396a578ad3e5db31c7577661505848b 100644
--- a/compute/pzlansy.c
+++ b/compute/pzlansy.c
@@ -40,8 +40,8 @@ chameleon_pzlansy_inf( cham_uplo_t uplo, CHAM_desc_t *A,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -133,8 +133,8 @@ chameleon_pzlansy_max( cham_trans_t trans, cham_uplo_t uplo, CHAM_desc_t *A,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -219,8 +219,8 @@ chameleon_pzlansy_frb( cham_trans_t trans, cham_uplo_t uplo,
int NT = A->nt;
int M = A->m;
int N = A->n;
- int P = Welt->p;
- int Q = Welt->q;
+ int P = chameleon_desc_datadist_get_iparam(Welt, 0);
+ int Q = chameleon_desc_datadist_get_iparam(Welt, 1);
/**
* Step 1:
@@ -321,8 +321,8 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
*result = 0.0;
- workmt = chameleon_max( A->mt, A->p );
- worknt = chameleon_max( A->nt, A->q );
+ workmt = chameleon_max( A->mt, chameleon_desc_datadist_get_iparam(A, 0) );
+ worknt = chameleon_max( A->nt, chameleon_desc_datadist_get_iparam(A, 1) );
switch ( norm ) {
case ChamOneNorm:
@@ -330,7 +330,9 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
RUNTIME_options_ws_alloc( &options, 1, 0 );
chameleon_desc_init( &Wcol, CHAMELEON_MAT_ALLOC_TILE, ChamRealDouble, A->mb, 1, A->mb,
- workmt * A->mb, worknt, 0, 0, workmt * A->mb, worknt, A->p, A->q,
+ workmt * A->mb, worknt, 0, 0, workmt * A->mb, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
wcol_init = 1;
@@ -338,7 +340,9 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
* Use the global allocator for Welt, otherwise flush may free the data before the result is read.
*/
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, 1, 1,
- workmt, A->q, 0, 0, workmt, A->q, A->p, A->q,
+ workmt, chameleon_desc_datadist_get_iparam(A, 1), 0, 0, workmt, chameleon_desc_datadist_get_iparam(A, 1),
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
break;
@@ -350,7 +354,9 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
alpha = 1.;
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 2, 1, 2,
- workmt*2, worknt, 0, 0, workmt*2, worknt, A->p, A->q,
+ workmt*2, worknt, 0, 0, workmt*2, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
break;
@@ -362,7 +368,9 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
RUNTIME_options_ws_alloc( &options, 1, 0 );
chameleon_desc_init( &Welt, CHAMELEON_MAT_ALLOC_GLOBAL, ChamRealDouble, 1, 1, 1,
- workmt, worknt, 0, 0, workmt, worknt, A->p, A->q,
+ workmt, worknt, 0, 0, workmt, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
}
@@ -410,8 +418,8 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
/**
* Broadcast the result
*/
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ for(m = 0; m < chameleon_desc_datadist_get_iparam(A, 0); m++) {
+ for(n = 0; n < chameleon_desc_datadist_get_iparam(A, 1); n++) {
if ( (m != 0) || (n != 0) ) {
INSERT_TASK_dlacpy(
&options,
@@ -428,7 +436,7 @@ void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_tra
CHAMELEON_Desc_Flush( A, sequence );
RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)Welt.get_blkaddr( &Welt, A->myrank / A->q, A->myrank % A->q );
+ *result = *(double *)Welt.get_blkaddr( &Welt, A->myrank / chameleon_desc_datadist_get_iparam(A, 1), A->myrank % chameleon_desc_datadist_get_iparam(A, 1) );
if ( wcol_init ) {
chameleon_desc_destroy( &Wcol );
diff --git a/compute/pzlatms.c b/compute/pzlatms.c
index f974d04175a619bf27cb89898643a302a4817be7..483ce2a1f7afa2b79623bba80bb22b86afc51fec 100644
--- a/compute/pzlatms.c
+++ b/compute/pzlatms.c
@@ -198,8 +198,8 @@ void chameleon_pzlatms( cham_dist_t idist, unsigned long long int seed, cham_sym
/* Apply a QR factorization */
mat.mt = descU.mt;
mat.nt = descU.nt;
- mat.nodes = descU.p * descU.q;
- mat.p = descU.p;
+ mat.nodes = chameleon_desc_datadist_get_iparam(&descU, 0) * chameleon_desc_datadist_get_iparam(&descU, 1);
+ mat.p = chameleon_desc_datadist_get_iparam(&descU, 0);
libhqr_init_hqr( &qrtree, LIBHQR_QR, &mat,
-1, /*low level tree */
@@ -217,12 +217,16 @@ void chameleon_pzlatms( cham_dist_t idist, unsigned long long int seed, cham_sym
chameleon_desc_init( &descTS, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, descU.nb, ib * descU.nb,
ib * descU.mt, descU.nb * descU.nt, 0, 0,
- ib * descU.mt, descU.nb * descU.nt, descU.p, descU.q,
+ ib * descU.mt, descU.nb * descU.nt,
+ chameleon_desc_datadist_get_iparam(&descU, 0),
+ chameleon_desc_datadist_get_iparam(&descU, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &descTT, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, descU.nb, ib * descU.nb,
ib * descU.mt, descU.nb * descU.nt, 0, 0,
- ib * descU.mt, descU.nb * descU.nt, descU.p, descU.q,
+ ib * descU.mt, descU.nb * descU.nt,
+ chameleon_desc_datadist_get_iparam(&descU, 0),
+ chameleon_desc_datadist_get_iparam(&descU, 1),
NULL, NULL, NULL, NULL );
/* U <= qr(U) */
@@ -269,8 +273,8 @@ void chameleon_pzlatms( cham_dist_t idist, unsigned long long int seed, cham_sym
/* Apply a QR factorization */
mat.mt = descV.mt;
mat.nt = descV.nt;
- mat.nodes = descV.p * descV.q;
- mat.p = descV.q;
+ mat.nodes = chameleon_desc_datadist_get_iparam(&descV, 0) * chameleon_desc_datadist_get_iparam(&descV, 1);
+ mat.p = chameleon_desc_datadist_get_iparam(&descV, 1);
libhqr_init_hqr( &qrtree, LIBHQR_LQ, &mat,
-1, /*low level tree */
@@ -288,12 +292,16 @@ void chameleon_pzlatms( cham_dist_t idist, unsigned long long int seed, cham_sym
chameleon_desc_init( &descTS, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, descV.nb, ib * descV.nb,
ib * descV.mt, descV.nb * descV.nt, 0, 0,
- ib * descV.mt, descV.nb * descV.nt, descV.p, descV.q,
+ ib * descV.mt, descV.nb * descV.nt,
+ chameleon_desc_datadist_get_iparam(&descV, 0),
+ chameleon_desc_datadist_get_iparam(&descV, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &descTT, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ib, descV.nb, ib * descV.nb,
ib * descV.mt, descV.nb * descV.nt, 0, 0,
- ib * descV.mt, descV.nb * descV.nt, descV.p, descV.q,
+ ib * descV.mt, descV.nb * descV.nt,
+ chameleon_desc_datadist_get_iparam(&descV, 0),
+ chameleon_desc_datadist_get_iparam(&descV, 1),
NULL, NULL, NULL, NULL );
/* V <= qr(V) */
diff --git a/compute/pzplrnk.c b/compute/pzplrnk.c
index 2705dd029afcc05805ecda7ddbfd2f548059a69d..86f6c3364e07e81d2a12d2acd741cbd68d1be5c0 100644
--- a/compute/pzplrnk.c
+++ b/compute/pzplrnk.c
@@ -47,24 +47,28 @@ void chameleon_pzplrnk( int K, CHAM_desc_t *C,
chameleon_desc_init( &WA, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mt * C->mb, C->nb * C->q, 0, 0,
- C->mt * C->mb, C->nb * C->q, C->p, C->q,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1), 0, 0,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1),
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &WB, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mb * C->p, C->nt * C->nb, 0, 0,
- C->mb * C->p, C->nt * C->nb, C->p, C->q,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0), C->nt * C->nb, 0, 0,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0), C->nt * C->nb,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
KT = (K + C->mb - 1) / C->mb;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
for (k = 0; k < KT; k++) {
tempkk = k == KT-1 ? K - k * WA.nb : WA.nb;
zbeta = k == 0 ? 0. : 1.;
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zplrnt(
@@ -73,7 +77,7 @@ void chameleon_pzplrnk( int K, CHAM_desc_t *C,
WB.m, k * WB.mb, n * WB.nb, seedB );
}
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zplrnt(
@@ -81,7 +85,7 @@ void chameleon_pzplrnk( int K, CHAM_desc_t *C,
tempmm, tempkk, WA(m, myq),
WA.m, m * WA.mb, k * WA.nb, seedA );
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zgemm(
@@ -94,7 +98,7 @@ void chameleon_pzplrnk( int K, CHAM_desc_t *C,
}
RUNTIME_data_flush( sequence, WA(m, 0) );
}
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
RUNTIME_data_flush( sequence, WB(0, n) );
}
}
diff --git a/compute/pzsymm.c b/compute/pzsymm.c
index 944ac75ea2daef41f206207d420e9d0543fd0ae1..8f16daf6fa1a1ed601ea31e6425f14c949891e3c 100644
--- a/compute/pzsymm.c
+++ b/compute/pzsymm.c
@@ -303,12 +303,12 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
lookahead = chamctxt->lookahead;
KT = A->nt;
K = A->n;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
for (k = 0; k < KT; k++ ) {
- lp = (k % lookahead) * C->p;
- lq = (k % lookahead) * C->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 1);
tempkk = k == KT - 1 ? K - k * A->nb : A->nb;
zbeta = k == 0 ? beta : zone;
@@ -340,16 +340,16 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempam, tempak,
A( Am, Ak ),
- WA( m, (Ak % C->q) + lq ) );
+ WA( m, (Ak % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, A( Am, Ak ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempam, tempak,
- WA( m, ((Ak+q-1) % C->q) + lq ),
- WA( m, ((Ak+q) % C->q) + lq ) );
+ WA( m, ((Ak+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((Ak+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
@@ -362,25 +362,25 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempkk, tempnn,
B( k, n ),
- WB( (k % C->p) + lp, n ) );
+ WB( (k % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, B( k, n ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempkk, tempnn,
- WB( ((k+p-1) % C->p) + lp, n ),
- WB( ((k+p) % C->p) + lp, n ) );
+ WB( ((k+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((k+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
/* Perform the update of this iteration */
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
if ( k == m ) {
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zsymm(
@@ -401,7 +401,7 @@ chameleon_pzsymm_summa_left( CHAM_context_t *chamctxt, cham_uplo_t uplo,
transA = ChamNoTrans;
}
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zgemm(
@@ -439,12 +439,12 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
lookahead = chamctxt->lookahead;
KT = A->mt;
K = A->m;
- myp = C->myrank / C->q;
- myq = C->myrank % C->q;
+ myp = C->myrank / chameleon_desc_datadist_get_iparam(C, 1);
+ myq = C->myrank % chameleon_desc_datadist_get_iparam(C, 1);
for (k = 0; k < KT; k++ ) {
- lp = (k % lookahead) * C->p;
- lq = (k % lookahead) * C->q;
+ lp = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 0);
+ lq = (k % lookahead) * chameleon_desc_datadist_get_iparam(C, 1);
tempkk = k == KT - 1 ? K - k * A->nb : A->nb;
zbeta = k == 0 ? beta : zone;
@@ -457,16 +457,16 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempmm, tempkk,
B( m, k ),
- WA( m, (k % C->q) + lq ) );
+ WA( m, (k % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
RUNTIME_data_flush( sequence, B( m, k ) );
- for ( q=1; q < C->q; q++ ) {
+ for ( q=1; q < chameleon_desc_datadist_get_iparam(C, 1); q++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempmm, tempkk,
- WA( m, ((k+q-1) % C->q) + lq ),
- WA( m, ((k+q) % C->q) + lq ) );
+ WA( m, ((k+q-1) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ),
+ WA( m, ((k+q) % chameleon_desc_datadist_get_iparam(C, 1)) + lq ) );
}
}
@@ -497,25 +497,25 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
options,
ChamUpperLower, tempak, tempan,
A( Ak, An ),
- WB( (Ak % C->p) + lp, n ) );
+ WB( (Ak % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
RUNTIME_data_flush( sequence, A( Ak, An ) );
- for ( p=1; p < C->p; p++ ) {
+ for ( p=1; p < chameleon_desc_datadist_get_iparam(C, 0); p++ ) {
INSERT_TASK_zlacpy(
options,
ChamUpperLower, tempak, tempan,
- WB( ((Ak+p-1) % C->p) + lp, n ),
- WB( ((Ak+p) % C->p) + lp, n ) );
+ WB( ((Ak+p-1) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ),
+ WB( ((Ak+p) % chameleon_desc_datadist_get_iparam(C, 0)) + lp, n ) );
}
}
/* Perform the update of this iteration */
- for (n = myq; n < C->nt; n+=C->q) {
+ for (n = myq; n < C->nt; n+=chameleon_desc_datadist_get_iparam(C, 1)) {
tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
if ( k == n ) {
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
/* A has been stored in WA or WB for the summa ring */
@@ -537,7 +537,7 @@ chameleon_pzsymm_summa_right( CHAM_context_t *chamctxt, cham_uplo_t uplo,
transA = ChamNoTrans;
}
- for (m = myp; m < C->mt; m+=C->p) {
+ for (m = myp; m < C->mt; m+=chameleon_desc_datadist_get_iparam(C, 0)) {
tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zgemm(
diff --git a/compute/zcesca.c b/compute/zcesca.c
index 92de039f37269152768055de00a5a123e94bf0fe..a0c8694dabdca07c8e8b9d21b516d75c5afbdd48 100644
--- a/compute/zcesca.c
+++ b/compute/zcesca.c
@@ -55,37 +55,47 @@ void *CHAMELEON_zcesca_WS_Alloc( const CHAM_desc_t *A )
options = calloc( 1, sizeof(struct chameleon_pzcesca_s) );
- workmt = chameleon_max( A->mt, A->p );
- worknt = chameleon_max( A->nt, A->q );
+ workmt = chameleon_max( A->mt, chameleon_desc_datadist_get_iparam(A, 0) );
+ worknt = chameleon_max( A->nt, chameleon_desc_datadist_get_iparam(A, 1) );
chameleon_desc_init( &(options->Wgcol), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, 1, A->nb, A->nb,
workmt, A->n, 0, 0,
- workmt, A->n, A->p, A->q,
+ workmt, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->Wgrow), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, A->mb, 1, A->mb,
A->m, worknt, 0, 0,
- A->m, worknt, A->p, A->q,
+ A->m, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->Wgelt), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, 1, 1, 1,
1, worknt, 0, 0,
- 1, worknt, A->p, A->q,
+ 1, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->Wdcol), CHAMELEON_MAT_ALLOC_TILE,
ChamRealDouble, 2, A->nb, 2*A->nb,
2*workmt, A->n, 0, 0,
- 2*workmt, A->n, A->p, A->q,
+ 2*workmt, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->Wdrow), CHAMELEON_MAT_ALLOC_TILE,
ChamRealDouble, A->mb, 2, 2*A->mb,
A->m, 2*worknt, 0, 0,
- A->m, 2*worknt, A->p, A->q,
+ A->m, 2*worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
return (void*)options;
diff --git a/compute/zgemm.c b/compute/zgemm.c
index fb5ebff37b416841e891f59bb3c191b537a36bd4..98519384e2ced2c0efc92d965b23247f3bee6f9c 100644
--- a/compute/zgemm.c
+++ b/compute/zgemm.c
@@ -110,7 +110,7 @@ void *CHAMELEON_zgemm_WS_Alloc( cham_trans_t transA __attribute__((unused)
* If only one process, or if generic has been globally enforced, we switch
* to generic immediately.
*/
- if ( ((C->p == 1) && (C->q == 1)) ||
+ if ( ((chameleon_desc_datadist_get_iparam(C, 0) == 1) && (chameleon_desc_datadist_get_iparam(C, 1) == 1)) ||
(chamctxt->generic_enabled == CHAMELEON_TRUE) )
{
options->alg = ChamGemmAlgGeneric;
@@ -151,9 +151,9 @@ void *CHAMELEON_zgemm_WS_Alloc( cham_trans_t transA __attribute__((unused)
double ratio = 1.5; /* Arbitrary ratio to give more weight to writes wrt reads. */
/* Compute the average array per node for each matrix */
- sizeA = ((double)A->m * (double)A->n) / (double)(A->p * A->q);
- sizeB = ((double)B->m * (double)B->n) / (double)(B->p * B->q);
- sizeC = ((double)C->m * (double)C->n) / (double)(C->p * C->q) * ratio;
+ sizeA = ((double)A->m * (double)A->n) / (double)(chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1));
+ sizeB = ((double)B->m * (double)B->n) / (double)(chameleon_desc_datadist_get_iparam(B, 0) * chameleon_desc_datadist_get_iparam(B, 1));
+ sizeC = ((double)C->m * (double)C->n) / (double)(chameleon_desc_datadist_get_iparam(C, 0) * chameleon_desc_datadist_get_iparam(C, 1)) * ratio;
options->alg = ChamGemmAlgGeneric;
if ( (sizeC > sizeA) && (sizeC > sizeB) )
@@ -192,13 +192,17 @@ void *CHAMELEON_zgemm_WS_Alloc( cham_trans_t transA __attribute__((unused)
chameleon_desc_init( &(options->WA), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mt * C->mb, C->nb * C->q * lookahead, 0, 0,
- C->mt * C->mb, C->nb * C->q * lookahead, C->p, C->q,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead, 0, 0,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->WB), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mb * C->p * lookahead, C->nt * C->nb, 0, 0,
- C->mb * C->p * lookahead, C->nt * C->nb, C->p, C->q,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb, 0, 0,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
}
diff --git a/compute/zgetrf.c b/compute/zgetrf.c
index b7e8f87b622c35d68f557f9c59393eabc017c679..9dabf735d3280d77f92ea30388a291a2db8cb861 100644
--- a/compute/zgetrf.c
+++ b/compute/zgetrf.c
@@ -68,7 +68,7 @@ CHAMELEON_zgetrf_WS_Alloc( const CHAM_desc_t *A )
ws->ib = CHAMELEON_IB;
#if defined (CHAMELEON_USE_MPI)
- ws->proc_involved = malloc( sizeof( int ) * A->p );
+ ws->proc_involved = malloc( sizeof( int ) * chameleon_desc_datadist_get_iparam(A, 0) );
ws->involved = 0;
ws->np_involved = 0;
#endif
@@ -107,7 +107,9 @@ CHAMELEON_zgetrf_WS_Alloc( const CHAM_desc_t *A )
chameleon_desc_init( &(ws->U), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, 1, A->nb, A->nb,
A->mt, A->nt * A->nb, 0, 0,
- A->mt, A->nt * A->nb, A->p, A->q,
+ A->mt, A->nt * A->nb,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
}
else if ( ( ws->alg == ChamGetrfPPiv ) ||
@@ -116,12 +118,14 @@ CHAMELEON_zgetrf_WS_Alloc( const CHAM_desc_t *A )
chameleon_desc_init( &(ws->U), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, A->mb, A->nb, A->mb*A->nb,
A->m, A->n, 0, 0,
- A->m, A->n, A->p, A->q,
+ A->m, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
chameleon_desc_init( &(ws->Wu), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, A->mb, A->nb, A->mb*A->nb,
- A->mb * A->p * A->q, A->n, 0, 0,
- A->mb * A->p * A->q, A->n, A->p * A->q, 1,
+ A->mb * chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1), A->n, 0, 0,
+ A->mb * chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1), A->n, chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1), 1,
NULL, NULL, NULL, A->get_rankof_init_arg );
}
@@ -140,7 +144,9 @@ CHAMELEON_zgetrf_WS_Alloc( const CHAM_desc_t *A )
chameleon_desc_init( &(ws->Up), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, ws->ib, A->nb, ws->ib * A->nb,
A->mt * ws->ib, A->nt * A->nb, 0, 0,
- A->mt * ws->ib, A->nt * A->nb, A->p, A->q,
+ A->mt * ws->ib, A->nt * A->nb,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
}
diff --git a/compute/zgetrf_nopiv.c b/compute/zgetrf_nopiv.c
index 3409e97f7a80cfb0844b4eefa66d6646be6239f1..63e40d5367bb630095c52836dc0b4bc685eaae13 100644
--- a/compute/zgetrf_nopiv.c
+++ b/compute/zgetrf_nopiv.c
@@ -61,7 +61,7 @@ void *CHAMELEON_zgetrf_nopiv_WS_Alloc( const CHAM_desc_t *A )
options = calloc( 1, sizeof(struct chameleon_pzgetrf_nopiv_s) );
options->use_workspace = 0;
- if ( ( ( A->p > 1 ) || ( A->q > 1 ) ) &&
+ if ( ( ( chameleon_desc_datadist_get_iparam(A, 0) > 1 ) || ( chameleon_desc_datadist_get_iparam(A, 1) > 1 ) ) &&
( A->get_rankof_init == chameleon_getrankof_2d ) &&
( chamctxt->generic_enabled != CHAMELEON_TRUE ) )
{
@@ -70,15 +70,19 @@ void *CHAMELEON_zgetrf_nopiv_WS_Alloc( const CHAM_desc_t *A )
chameleon_desc_init( &(options->WL), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, A->mb, A->nb, (A->mb * A->nb),
- A->mt * A->mb, A->nb * A->q * lookahead, 0, 0,
- A->mt * A->mb, A->nb * A->q * lookahead, A->p, A->q,
+ A->mt * A->mb, A->nb * chameleon_desc_datadist_get_iparam(A, 1) * lookahead, 0, 0,
+ A->mt * A->mb, A->nb * chameleon_desc_datadist_get_iparam(A, 1) * lookahead,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
chameleon_desc_init( &(options->WU), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble,
A->mb, A->nb, (A->mb * A->nb),
- A->mb * A->p * lookahead, A->nt * A->nb, 0, 0,
- A->mb * A->p * lookahead, A->nt * A->nb, A->p, A->q,
+ A->mb * chameleon_desc_datadist_get_iparam(A, 0) * lookahead, A->nt * A->nb, 0, 0,
+ A->mb * chameleon_desc_datadist_get_iparam(A, 0) * lookahead, A->nt * A->nb,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, A->get_rankof_init, A->get_rankof_init_arg );
}
diff --git a/compute/zgram.c b/compute/zgram.c
index c40c696a0c04abb5b7bb63bdf3749253da497837..07c9d7d066827578b2bcbc01beeafed80a98b10e 100644
--- a/compute/zgram.c
+++ b/compute/zgram.c
@@ -58,19 +58,23 @@ void *CHAMELEON_zgram_WS_Alloc( const CHAM_desc_t *A )
options = calloc( 1, sizeof(struct chameleon_pzgram_s) );
- workmt = chameleon_max( A->mt, A->p );
- worknt = chameleon_max( A->nt, A->q );
+ workmt = chameleon_max( A->mt, chameleon_desc_datadist_get_iparam(A, 0) );
+ worknt = chameleon_max( A->nt, chameleon_desc_datadist_get_iparam(A, 1) );
chameleon_desc_init( &(options->Wcol), CHAMELEON_MAT_ALLOC_TILE,
ChamRealDouble, 2, A->nb, 2*A->nb,
2*workmt, A->n, 0, 0,
- 2*workmt, A->n, A->p, A->q,
+ 2*workmt, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->Welt), CHAMELEON_MAT_ALLOC_TILE,
ChamRealDouble, 2, 1, 2,
2, worknt, 0, 0,
- 2, worknt, A->p, A->q,
+ 2, worknt,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1),
NULL, NULL, NULL, NULL );
return (void*)options;
diff --git a/compute/zhemm.c b/compute/zhemm.c
index a174e2cfb14e9846d2496649deb4a22c8c5862f3..c8fc553b958132224aa983fbce18605f0ce3ed72 100644
--- a/compute/zhemm.c
+++ b/compute/zhemm.c
@@ -89,7 +89,7 @@ void *CHAMELEON_zhemm_WS_Alloc( cham_side_t side __attribute__((unused)),
* If only one process, or if generic has been globally enforced, we switch
* to generic immediately.
*/
- if ( ((C->p == 1) && (C->q == 1)) ||
+ if ( ((chameleon_desc_datadist_get_iparam(C, 0) == 1) && (chameleon_desc_datadist_get_iparam(C, 1) == 1)) ||
(chamctxt->generic_enabled == CHAMELEON_TRUE) )
{
options->alg = ChamGemmAlgGeneric;
@@ -130,9 +130,9 @@ void *CHAMELEON_zhemm_WS_Alloc( cham_side_t side __attribute__((unused)),
double ratio = 1.5; /* Arbitrary ratio to give more weight to writes wrt reads. */
/* Compute the average array per node for each matrix */
- sizeA = ((double)A->m * (double)A->n) / (double)(A->p * A->q);
- sizeB = ((double)B->m * (double)B->n) / (double)(B->p * B->q);
- sizeC = ((double)C->m * (double)C->n) / (double)(C->p * C->q) * ratio;
+ sizeA = ((double)A->m * (double)A->n) / (double)(chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1));
+ sizeB = ((double)B->m * (double)B->n) / (double)(chameleon_desc_datadist_get_iparam(B, 0) * chameleon_desc_datadist_get_iparam(B, 1));
+ sizeC = ((double)C->m * (double)C->n) / (double)(chameleon_desc_datadist_get_iparam(C, 0) * chameleon_desc_datadist_get_iparam(C, 1)) * ratio;
if ( (sizeC > sizeA) && (sizeC > sizeB) ) {
options->alg = ChamGemmAlgSummaC;
@@ -168,13 +168,17 @@ void *CHAMELEON_zhemm_WS_Alloc( cham_side_t side __attribute__((unused)),
chameleon_desc_init( &(options->WA), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mt * C->mb, C->nb * C->q * lookahead, 0, 0,
- C->mt * C->mb, C->nb * C->q * lookahead, C->p, C->q,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead, 0, 0,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->WB), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mb * C->p * lookahead, C->nt * C->nb, 0, 0,
- C->mb * C->p * lookahead, C->nt * C->nb, C->p, C->q,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb, 0, 0,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
}
diff --git a/compute/zhetrd.c b/compute/zhetrd.c
index cc9826b2f383670f405f7ae4b35e66c4cebc0ce4..b846cc90fd3f56a09f7dab0adde5e413720f227e 100644
--- a/compute/zhetrd.c
+++ b/compute/zhetrd.c
@@ -391,7 +391,9 @@ int CHAMELEON_zhetrd_Tile_Async( cham_job_t jobz,
NB = descA.mb;
#if defined(CHAMELEON_COPY_DIAG)
{
- chameleon_zdesc_alloc_diag( &D, A->mb, A->m, A->n, A->p, A->q );
+ chameleon_zdesc_alloc_diag( &D, A->mb, A->m, A->n,
+ chameleon_desc_datadist_get_iparam(A, 0),
+ chameleon_desc_datadist_get_iparam(A, 1) );
Dptr = &D;
}
#endif
diff --git a/compute/zsymm.c b/compute/zsymm.c
index dbb1ae72212f5fe52de2a7f8d9f4f2f567e79b83..ab1f68eb4cea8632b5734b081be73f932f0b0cb8 100644
--- a/compute/zsymm.c
+++ b/compute/zsymm.c
@@ -89,7 +89,7 @@ void *CHAMELEON_zsymm_WS_Alloc( cham_side_t side __attribute__((unused)),
* If only one process, or if generic has been globally enforced, we switch
* to generic immediately.
*/
- if ( ((C->p == 1) && (C->q == 1)) ||
+ if ( ((chameleon_desc_datadist_get_iparam(C, 0) == 1) && (chameleon_desc_datadist_get_iparam(C, 1) == 1)) ||
(chamctxt->generic_enabled == CHAMELEON_TRUE) )
{
options->alg = ChamGemmAlgGeneric;
@@ -130,9 +130,9 @@ void *CHAMELEON_zsymm_WS_Alloc( cham_side_t side __attribute__((unused)),
double ratio = 1.5; /* Arbitrary ratio to give more weight to writes wrt reads. */
/* Compute the average array per node for each matrix */
- sizeA = ((double)A->m * (double)A->n) / (double)(A->p * A->q);
- sizeB = ((double)B->m * (double)B->n) / (double)(B->p * B->q);
- sizeC = ((double)C->m * (double)C->n) / (double)(C->p * C->q) * ratio;
+ sizeA = ((double)A->m * (double)A->n) / (double)(chameleon_desc_datadist_get_iparam(A, 0) * chameleon_desc_datadist_get_iparam(A, 1));
+ sizeB = ((double)B->m * (double)B->n) / (double)(chameleon_desc_datadist_get_iparam(B, 0) * chameleon_desc_datadist_get_iparam(B, 1));
+ sizeC = ((double)C->m * (double)C->n) / (double)(chameleon_desc_datadist_get_iparam(C, 0) * chameleon_desc_datadist_get_iparam(C, 1)) * ratio;
if ( (sizeC > sizeA) && (sizeC > sizeB) ) {
options->alg = ChamGemmAlgSummaC;
@@ -168,13 +168,17 @@ void *CHAMELEON_zsymm_WS_Alloc( cham_side_t side __attribute__((unused)),
chameleon_desc_init( &(options->WA), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mt * C->mb, C->nb * C->q * lookahead, 0, 0,
- C->mt * C->mb, C->nb * C->q * lookahead, C->p, C->q,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead, 0, 0,
+ C->mt * C->mb, C->nb * chameleon_desc_datadist_get_iparam(C, 1) * lookahead,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
chameleon_desc_init( &(options->WB), CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, C->mb, C->nb, (C->mb * C->nb),
- C->mb * C->p * lookahead, C->nt * C->nb, 0, 0,
- C->mb * C->p * lookahead, C->nt * C->nb, C->p, C->q,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb, 0, 0,
+ C->mb * chameleon_desc_datadist_get_iparam(C, 0) * lookahead, C->nt * C->nb,
+ chameleon_desc_datadist_get_iparam(C, 0),
+ chameleon_desc_datadist_get_iparam(C, 1),
NULL, NULL, NULL, NULL );
}
diff --git a/control/compute_z.h b/control/compute_z.h
index acb9599f295c77774acccef98c88d7db22c59362..df4b18d79010af05a0d5f482b7e9d9390442c29a 100644
--- a/control/compute_z.h
+++ b/control/compute_z.h
@@ -310,7 +310,7 @@ chameleon_zdesc_copy_and_restrict( const CHAM_desc_t *descIn,
int rc;
rc = chameleon_desc_init( descOut, CHAMELEON_MAT_ALLOC_TILE,
ChamComplexDouble, descIn->mb, descIn->nb, descIn->mb * descIn->nb,
- m, n, 0, 0, m, n, descIn->p, descIn->q,
+ m, n, 0, 0, m, n, chameleon_desc_datadist_get_iparam(descIn, 0), chameleon_desc_datadist_get_iparam(descIn, 1),
descIn->get_blkaddr,
descIn->get_blkldd,
descIn->get_rankof_init, descIn->get_rankof_init_arg );
diff --git a/control/descriptor.c b/control/descriptor.c
index 556dc8ba7671e1fd32b57f8690bc80bd14d87804..22ddfd71e15548d49d32b4af912203abade845fe 100644
--- a/control/descriptor.c
+++ b/control/descriptor.c
@@ -86,6 +86,7 @@ int chameleon_desc_mat_free( CHAM_desc_t *desc )
#endif
free( desc->tiles );
}
+ free( desc->data_dist );
return CHAMELEON_SUCCESS;
}
@@ -116,6 +117,26 @@ void chameleon_desc_init_tiles( CHAM_desc_t *desc, blkrankof_fct_t rankof )
}
}
+/* Get access to data dist */
+int chameleon_desc_datadist_get_iparam( const CHAM_desc_t *desc, int i )
+{
+ return desc->data_dist->distrib[desc->data_dist->get_distrib(desc, i)];
+}
+
+int chameleon_get_2d_block_cyclic( const CHAM_desc_t *desc, int i ) { return i; }
+
+void chameleon_desc_set_datadist( CHAM_desc_t *to, cham_data_dist_t *from )
+{
+ int i;
+ to->data_dist = malloc(sizeof(cham_data_dist_t));
+ to->data_dist->get_distrib = from->get_distrib;
+ to->data_dist->distrib_array_size = from->distrib_array_size;
+
+ for (i = 0; i < to->data_dist->distrib_array_size; i++) {
+ to->data_dist->distrib[i] = from->distrib[i];
+ }
+}
+
/**
******************************************************************************
*
@@ -238,8 +259,11 @@ int chameleon_desc_init_internal( CHAM_desc_t *desc, const char *name, void *mat
desc->myrank = RUNTIME_comm_rank( chamctxt );
/* Grid size */
- desc->p = p;
- desc->q = q;
+ cham_data_dist_t dist = {
+ .get_distrib = (datadist_access_fct_t)chameleon_get_2d_block_cyclic,
+ .distrib_array_size = 2,
+ .distrib = {p, q} };
+ chameleon_desc_set_datadist( desc, &dist );
/* Local dimensions in tiles */
if ( desc->myrank < (p*q) ) {
@@ -808,7 +832,8 @@ CHAM_desc_t *CHAMELEON_Desc_Copy( const CHAM_desc_t *descin, void *mat )
CHAM_desc_t *descout = NULL;
CHAMELEON_Desc_Create_User( &descout, mat,
descin->dtyp, descin->mb, descin->nb, descin->bsiz,
- descin->lm, descin->ln, descin->i, descin->j, descin->m, descin->n, descin->p, descin->q,
+ descin->lm, descin->ln, descin->i, descin->j, descin->m, descin->n,
+ chameleon_desc_datadist_get_iparam(descin, 0), chameleon_desc_datadist_get_iparam(descin, 1),
NULL, NULL, descin->get_rankof_init, descin->get_rankof_init_arg );
return descout;
}
diff --git a/control/descriptor.h b/control/descriptor.h
index 1e0315fae2c70cdec40052e49a58b47c32a46ec9..739d752830cc0917a3d366ca5cdd9bfec3ab50e4 100644
--- a/control/descriptor.h
+++ b/control/descriptor.h
@@ -106,8 +106,8 @@ inline static void* chameleon_geteltaddr(const CHAM_desc_t *A, int m, int n, int
#if defined(CHAMELEON_USE_MPI)
assert( A->myrank == A->get_rankof( A, mm, nn) );
- mm = mm / A->p;
- nn = nn / A->q;
+ mm = mm / chameleon_desc_datadist_get_iparam(A, 0);
+ nn = nn / chameleon_desc_datadist_get_iparam(A, 1);
#endif
if (mm < (size_t)(A->llm1)) {
diff --git a/control/descriptor_helpers.c b/control/descriptor_helpers.c
index b49cb69e9b751e4494ae01de14571010c64e980c..d5e1430638e8b507e4ac6163869d87b3021d97f4 100644
--- a/control/descriptor_helpers.c
+++ b/control/descriptor_helpers.c
@@ -56,7 +56,7 @@ int chameleon_getrankof_2d( const CHAM_desc_t *A, int m, int n )
{
int mm = m + A->i / A->mb;
int nn = n + A->j / A->nb;
- return (mm % A->p) * A->q + (nn % A->q);
+ return (mm % chameleon_desc_datadist_get_iparam(A,0)) * chameleon_desc_datadist_get_iparam(A,1) + (nn % chameleon_desc_datadist_get_iparam(A,1));
}
/**
@@ -79,7 +79,7 @@ int chameleon_getrankof_2d_diag( const CHAM_desc_t *A, int m, int n )
{
int mm = m + A->i / A->mb;
(void)n;
- return (mm % A->p) * A->q + (mm % A->q);
+ return (mm % chameleon_desc_datadist_get_iparam(A,0)) * chameleon_desc_datadist_get_iparam(A,1) + (mm % chameleon_desc_datadist_get_iparam(A,1));
}
/**
@@ -97,7 +97,7 @@ int chameleon_getrankof_2d_diag( const CHAM_desc_t *A, int m, int n )
*/
int chameleon_involved_in_panelk_2dbc( const CHAM_desc_t *A, int k ) {
int myrank = A->myrank;
- return ( myrank % A->q == k % A->q );
+ return ( myrank % chameleon_desc_datadist_get_iparam(A,1) == k % chameleon_desc_datadist_get_iparam(A,1) );
}
/**
@@ -128,7 +128,7 @@ void chameleon_get_proc_involved_in_panelk_2dbc( const CHAM_desc_t *A,
np = 0;
ws->involved = 0;
- for ( b = k; (b < A->mt) && ((b-k) < A->p); b ++ ) {
+ for ( b = k; (b < A->mt) && ((b-k) < chameleon_desc_datadist_get_iparam(A, 0)); b ++ ) {
rank = chameleon_getrankof_2d( A, b, n );
proc_involved[ b-k ] = rank;
np ++;
@@ -325,8 +325,8 @@ void* chameleon_getaddr_ccrb( const CHAM_desc_t *A, int m, int n )
#if defined(CHAMELEON_USE_MPI)
assert( A->myrank == A->get_rankof( A, mm, nn ) );
- mm = mm / A->p;
- nn = nn / A->q;
+ mm = mm / chameleon_desc_datadist_get_iparam(A, 0);
+ nn = nn / chameleon_desc_datadist_get_iparam(A, 1);
#endif
if (mm < (size_t)(A->llm1)) {
@@ -372,8 +372,8 @@ void *chameleon_getaddr_cm( const CHAM_desc_t *A, int m, int n )
#if defined(CHAMELEON_USE_MPI)
assert( A->myrank == A->get_rankof( A, mm, nn ) );
- mm = mm / A->p;
- nn = nn / A->q;
+ mm = mm / chameleon_desc_datadist_get_iparam(A, 0);
+ nn = nn / chameleon_desc_datadist_get_iparam(A, 1);
#endif
offset = (size_t)(A->llm * A->nb) * nn + (size_t)(A->mb) * mm;
diff --git a/example/lapack_to_chameleon/step3.h b/example/lapack_to_chameleon/step3.h
index cbeeccaad4bff639ef8c266789fc28dec17dcfb4..1fe156e99a8a8b88f556001659ef9240bd0780b2 100644
--- a/example/lapack_to_chameleon/step3.h
+++ b/example/lapack_to_chameleon/step3.h
@@ -168,8 +168,8 @@ inline static void* user_getaddr_arrayofpointers(const CHAM_desc_t *A, int m, in
#if defined(CHAMELEON_USE_MPI)
assert( A->myrank == A->get_rankof( A, mm, nn) );
- mm = mm / A->p;
- nn = nn / A->q;
+ mm = mm / chameleon_desc_datadist_get_iparam(A, 0);
+ nn = nn / chameleon_desc_datadist_get_iparam(A, 1);
#endif
offset = A->mt*nn + mm;
diff --git a/include/chameleon/struct.h b/include/chameleon/struct.h
index 19b6137f0f48951bed799c9563af05fb84e32e89..e61f95b18ba8a9e2f5c204bc1aff1428e24e6620 100644
--- a/include/chameleon/struct.h
+++ b/include/chameleon/struct.h
@@ -35,6 +35,8 @@ BEGIN_C_DECLS
#define CHAMELEON_TILE_DESC (1 << 1)
#define CHAMELEON_TILE_HMAT (1 << 2)
+#define CHAMELEON_MAX_DIMENSION 10
+
/**
* @brief CHAMELEON structure to hold pivot informations for the LU factorization with partial pivoting
*/
@@ -76,10 +78,41 @@ typedef struct chameleon_tile_s {
struct chameleon_desc_s;
typedef struct chameleon_desc_s CHAM_desc_t;
-typedef void* (*blkaddr_fct_t) ( const CHAM_desc_t*, int, int );
-typedef int (*blkldd_fct_t) ( const CHAM_desc_t*, int );
-typedef int (*blkrankof_fct_t)( const CHAM_desc_t*, int, int );
-typedef CHAM_tile_t* (*blktile_fct_t) ( const CHAM_desc_t*, int, int );
+typedef void* (*blkaddr_fct_t) ( const CHAM_desc_t*, int, int );
+typedef int (*blkldd_fct_t) ( const CHAM_desc_t*, int );
+typedef int (*blkrankof_fct_t) ( const CHAM_desc_t*, int, int );
+typedef int (*datadist_access_fct_t)( const CHAM_desc_t*, int, ... );
+typedef CHAM_tile_t* (*blktile_fct_t) ( const CHAM_desc_t*, int, int );
+
+/**
+ * Data distribution type and acces functions
+ */
+/**
+ * @brief Function discribing the indexed access to a 2D block cyclic data
+ * distribution holding the the grid size in the data distribution
+ * array [p,q] :
+ * 0 returns p, 1 returns q.
+ */
+int chameleon_get_2d_block_cyclic(const CHAM_desc_t *desc, int i );
+
+/**
+ * @brief Function call which forward the index i to the access function
+ */
+int chameleon_desc_datadist_get_iparam( const CHAM_desc_t *desc, int i );
+
+/**
+ * @brief Data distribution type
+ */
+typedef struct cham_data_dist_s {
+ datadist_access_fct_t get_distrib; /**> function describing how to index the distribution array */
+ int distrib_array_size; /**> number of parameters stored in the distribution array */
+ int distrib[CHAMELEON_MAX_DIMENSION]; /**> array holding the parameters */
+} cham_data_dist_t;
+
+/**
+ * @brief Function for initialising the data distribution
+ */
+void chameleon_desc_set_datadist( CHAM_desc_t *to, cham_data_dist_t *from );
struct chameleon_desc_s {
const char *name;
@@ -116,14 +149,13 @@ struct chameleon_desc_s {
int lnt; /**> number of tile columns of the entire matrix - derived parameter */
/* Distributed case */
- int p; /**> number of rows of the 2D distribution grid */
- int q; /**> number of columns of the 2D distribution grid */
- int llm; /**> local number of rows of the full matrix - derived parameter */
- int lln; /**> local number of columns of the full matrix - derived parameter */
- int llm1; /**> local number of tile rows of the A11 matrix - derived parameter */
- int lln1; /**> local number of tile columns of the A11 matrix - derived parameter */
- int llmt; /**> local number of tile rows of the full matrix - derived parameter */
- int llnt; /**> local number of tile columns of the full matrix - derived parameter */
+ cham_data_dist_t *data_dist; /**> data distribution type used to retrieve the distributed layout */
+ int llm; /**> local number of rows of the full matrix - derived parameter */
+ int lln; /**> local number of columns of the full matrix - derived parameter */
+ int llm1; /**> local number of tile rows of the A11 matrix - derived parameter */
+ int lln1; /**> local number of tile columns of the A11 matrix - derived parameter */
+ int llmt; /**> local number of tile rows of the full matrix - derived parameter */
+ int llnt; /**> local number of tile columns of the full matrix - derived parameter */
int id; /**> identification number of the descriptor */
int occurences; /**> identify main matrix desc (occurances=1) or */
diff --git a/runtime/starpu/codelets/codelet_zipiv_allreduce.c b/runtime/starpu/codelets/codelet_zipiv_allreduce.c
index f296bb3df7433bf6eb8d8c47e468b262ee741f16..5e7254a28548bfac4778c49850daa543197f69fe 100644
--- a/runtime/starpu/codelets/codelet_zipiv_allreduce.c
+++ b/runtime/starpu/codelets/codelet_zipiv_allreduce.c
@@ -114,7 +114,7 @@ void INSERT_TASK_zipiv_allreduce( CHAM_desc_t *A,
int h,
int n )
{
- int np_involved = chameleon_min( A->p, A->mt - k);
+ int np_involved = chameleon_min( chameleon_desc_datadist_get_iparam(A, 0), A->mt - k);
int np_iter = np_involved;
int p_recv, p_send, me;
int shift = 1;
diff --git a/runtime/starpu/codelets/codelet_zperm_allreduce.c b/runtime/starpu/codelets/codelet_zperm_allreduce.c
index ab9cf702294f7a54348b8d7995f45aca5afc32e3..87c4cc61af30b117d86b54f4a5ae517c6dbd5f0e 100644
--- a/runtime/starpu/codelets/codelet_zperm_allreduce.c
+++ b/runtime/starpu/codelets/codelet_zperm_allreduce.c
@@ -138,7 +138,7 @@ INSERT_TASK_zperm_allreduce( const RUNTIME_option_t *options,
{
struct chameleon_pzgetrf_s *tmp = (struct chameleon_pzgetrf_s *)ws;
int *proc_involved = tmp->proc_involved;
- int np_involved = chameleon_min( A->p, A->mt - k);
+ int np_involved = chameleon_min( chameleon_desc_datadist_get_iparam(A, 0), A->mt - k);
int np_iter = np_involved;
int p_recv, p_send, me, p_first;
int shift = 1;
@@ -161,7 +161,7 @@ INSERT_TASK_zperm_allreduce( const RUNTIME_option_t *options,
INSERT_TASK_zperm_allreduce_send( options, U, A->myrank, p_send, n );
INSERT_TASK_zperm_allreduce_recv( options, U, ipiv, ipivk, A->myrank, p_recv,
n, k == (A->mt-1) ? A->m - k * A->mb : A->mb,
- A->p, A->q, shift, np_involved, p_first );
+ chameleon_desc_datadist_get_iparam(A, 0), chameleon_desc_datadist_get_iparam(A, 1), shift, np_involved, p_first );
shift = shift << 1;
np_iter = chameleon_ceil( np_iter, 2 );
@@ -220,7 +220,7 @@ INSERT_TASK_zperm_allreduce_send_invp( const RUNTIME_option_t *options,
{
int b, rank;
- for ( b = k+1; (b < A->mt) && ((b-(k+1)) < A->p); b ++ ) {
+ for ( b = k+1; (b < A->mt) && ((b-(k+1)) < chameleon_desc_datadist_get_iparam(A, 0)); b ++ ) {
rank = A->get_rankof( A, b, n );
if ( rank == A->myrank ) {
continue;
diff --git a/runtime/starpu/control/runtime_descriptor_ipiv.c b/runtime/starpu/control/runtime_descriptor_ipiv.c
index e8cc1a1b3500ab1a742563bc9d3ad86a00929e12..977c8676e3acad779f2c1f9e27ec59c92026e3d1 100644
--- a/runtime/starpu/control/runtime_descriptor_ipiv.c
+++ b/runtime/starpu/control/runtime_descriptor_ipiv.c
@@ -26,14 +26,14 @@ void RUNTIME_ipiv_create( CHAM_ipiv_t *ipiv,
const CHAM_desc_t *desc )
{
assert( ipiv );
- size_t nbhandles = 3 * ipiv->mt + 2 * desc->p;
+ size_t nbhandles = 3 * ipiv->mt + 2 * chameleon_desc_datadist_get_iparam(desc, 0);
starpu_data_handle_t *handles = calloc( nbhandles, sizeof(starpu_data_handle_t) );
ipiv->ipiv = handles;
handles += ipiv->mt;
ipiv->nextpiv = handles;
- handles += desc->p;
+ handles += chameleon_desc_datadist_get_iparam(desc, 0);
ipiv->prevpiv = handles;
- handles += desc->p;
+ handles += chameleon_desc_datadist_get_iparam(desc, 0);
ipiv->perm = handles;
handles += ipiv->mt;
ipiv->invp = handles;
@@ -50,8 +50,8 @@ void RUNTIME_ipiv_create( CHAM_ipiv_t *ipiv,
return;
}
ipiv->mpitag_nextpiv = ipiv->mpitag_ipiv + ipiv->mt;
- ipiv->mpitag_prevpiv = ipiv->mpitag_nextpiv + desc->p;
- ipiv->mpitag_perm = ipiv->mpitag_prevpiv + desc->p;
+ ipiv->mpitag_prevpiv = ipiv->mpitag_nextpiv + chameleon_desc_datadist_get_iparam(desc, 0);
+ ipiv->mpitag_perm = ipiv->mpitag_prevpiv + chameleon_desc_datadist_get_iparam(desc, 0);
ipiv->mpitag_invp = ipiv->mpitag_perm + ipiv->mt;
}
#endif
@@ -65,7 +65,7 @@ void RUNTIME_ipiv_destroy( CHAM_ipiv_t *ipiv,
{
int i;
starpu_data_handle_t *handle = (starpu_data_handle_t*)(ipiv->ipiv);
- size_t nbhandles = 3 * ipiv->mt + 2 * desc->p;
+ size_t nbhandles = 3 * ipiv->mt + 2 * chameleon_desc_datadist_get_iparam(desc, 0);
for(i=0; i<nbhandles; i++) {
if ( *handle != NULL ) {
@@ -118,7 +118,7 @@ void *RUNTIME_nextpiv_getaddr( const CHAM_ipiv_t *ipiv, int rank, int k, int h )
starpu_data_handle_t *nextpiv = (starpu_data_handle_t*)(ipiv->nextpiv);
const CHAM_desc_t *A = ipiv->desc;
- nextpiv += rank/A->q;
+ nextpiv += rank/chameleon_desc_datadist_get_iparam(A, 1);
assert( nextpiv );
if ( *nextpiv != NULL ) {
@@ -128,7 +128,7 @@ void *RUNTIME_nextpiv_getaddr( const CHAM_ipiv_t *ipiv, int rank, int k, int h )
int64_t kk = k + (ipiv->i / ipiv->mb);
int owner = rank;
int ncols = (kk == (A->nt-1)) ? A->n - kk * A->nb : A->nb;
- int64_t tag = ipiv->mpitag_nextpiv + owner/A->q;
+ int64_t tag = ipiv->mpitag_nextpiv + owner/chameleon_desc_datadist_get_iparam(A, 1);
cppi_register( nextpiv, A->dtyp, ncols, tag, owner );
@@ -142,7 +142,7 @@ void *RUNTIME_prevpiv_getaddr( const CHAM_ipiv_t *ipiv, int rank, int k, int h )
starpu_data_handle_t *prevpiv = (starpu_data_handle_t*)(ipiv->prevpiv);
const CHAM_desc_t *A = ipiv->desc;
- prevpiv += rank/A->q;
+ prevpiv += rank/chameleon_desc_datadist_get_iparam(A, 1);
assert( prevpiv );
if ( *prevpiv != NULL ) {
@@ -152,7 +152,7 @@ void *RUNTIME_prevpiv_getaddr( const CHAM_ipiv_t *ipiv, int rank, int k, int h )
int64_t kk = k + (ipiv->i / ipiv->mb);
int owner = rank;
int ncols = (kk == (A->nt-1)) ? A->n - kk * A->nb : A->nb;
- int64_t tag = ipiv->mpitag_prevpiv + owner/A->q;
+ int64_t tag = ipiv->mpitag_prevpiv + owner/chameleon_desc_datadist_get_iparam(A, 1);
cppi_register( prevpiv, A->dtyp, ncols, tag, owner );
@@ -226,7 +226,7 @@ void RUNTIME_ipiv_flushk( const RUNTIME_sequence_t *sequence,
const CHAM_desc_t *A = ipiv->desc;
handle = (starpu_data_handle_t*)(ipiv->nextpiv);
- handle += rank/A->q;
+ handle += rank/chameleon_desc_datadist_get_iparam(A, 1);
if ( *handle != NULL ) {
#if defined(CHAMELEON_USE_MPI)
@@ -239,7 +239,7 @@ void RUNTIME_ipiv_flushk( const RUNTIME_sequence_t *sequence,
}
handle = (starpu_data_handle_t*)(ipiv->prevpiv);
- handle += rank/A->q;
+ handle += rank/chameleon_desc_datadist_get_iparam(A, 1);
if ( *handle != NULL ) {
#if defined(CHAMELEON_USE_MPI)
diff --git a/testing/testing_zcheck_qr_lq.c b/testing/testing_zcheck_qr_lq.c
index 48162e3d79b298b98f5f7d1bab21551b6a92c23e..502b19fa035a5bea8e3a6619baacfc191fea2f0d 100644
--- a/testing/testing_zcheck_qr_lq.c
+++ b/testing/testing_zcheck_qr_lq.c
@@ -683,7 +683,7 @@ int check_zgeqrs( run_arg_list_t *args, cham_trans_t trans, CHAM_desc_t *descA,
*
*/
CHAMELEON_Desc_Create( &descRR, NULL, ChamComplexDouble, nb, nb, nb*nb,
- NRHS, N, 0, 0, NRHS, N, descA->p, descA->q );
+ NRHS, N, 0, 0, NRHS, N, chameleon_desc_datadist_get_iparam(descA, 0), chameleon_desc_datadist_get_iparam(descA, 1) );
CHAMELEON_zgemm_Tile( ChamConjTrans, trans, 1., descR, descA, 0., descRR );
@@ -776,7 +776,8 @@ int check_zgelqs( run_arg_list_t *args, cham_trans_t trans, CHAM_desc_t *descA,
* where R = op(A)*X - B, op(A) is A or A', and alpha = ||B||
*
*/
- CHAMELEON_Desc_Create( &descRR, NULL, ChamComplexDouble, nb, nb, nb*nb, NRHS, M, 0, 0, NRHS, M, descA->p, descA->q );
+ CHAMELEON_Desc_Create( &descRR, NULL, ChamComplexDouble, nb, nb, nb*nb, NRHS, M, 0, 0, NRHS, M,
+ chameleon_desc_datadist_get_iparam(descA, 0), chameleon_desc_datadist_get_iparam(descA, 1) );
CHAMELEON_zgemm_Tile( ChamConjTrans, trans, 1., descR, descA, 0., descRR );
diff --git a/testing/testing_zgepdf_qr.c b/testing/testing_zgepdf_qr.c
index 62553f45bb0c15067d16dd1187722ea67b0fa19b..24b4fc823b9db2acd34ae7df388f508202eba200 100644
--- a/testing/testing_zgepdf_qr.c
+++ b/testing/testing_zgepdf_qr.c
@@ -108,8 +108,8 @@ testing_zgepdf_qr_desc( run_arg_list_t *args, int check )
libhqr_matrix_t mat = {
.mt = descA1->mt,
.nt = descA1->nt,
- .nodes = descA1->p * descA1->q,
- .p = descA1->p,
+ .nodes = chameleon_desc_datadist_get_iparam(descA1, 0) * chameleon_desc_datadist_get_iparam(descA1, 1),
+ .p = chameleon_desc_datadist_get_iparam(descA1, 0),
};
/* Tree for the top matrix */
@@ -117,7 +117,7 @@ testing_zgepdf_qr_desc( run_arg_list_t *args, int check )
-1, /*low level tree */
-1, /* high level tree */
-1, /* TS tree size */
- descA1->p, /* High level size */
+ chameleon_desc_datadist_get_iparam(descA1, 0), /* High level size */
-1, /* Domino */
0 /* TSRR (unstable) */ );
@@ -129,7 +129,7 @@ testing_zgepdf_qr_desc( run_arg_list_t *args, int check )
/* high level tree (Could be greedy, but flat should reduce the volume of comm) */
LIBHQR_FLAT_TREE,
-1, /* TS tree size */
- descA2->p /* High level size */ );
+ chameleon_desc_datadist_get_iparam(descA2, 0) /* High level size */ );
}
/* Calculates the solution */