Add ztpgqrt function to generate Q from ztpqrt

compute/CMakeLists.txt

@@ -130,6 +130,7 @@ set(ZSRC
     pzunmlqrh.c
     pzunmqr.c
     pzunmqrrh.c
+    pztpgqrt.c
     pztpqrt.c
     ###
     zgels.c
@@ -168,6 +169,7 @@ set(ZSRC
     zungqr.c
     zunmlq.c
     zunmqr.c
+    ztpgqrt.c
     ztpqrt.c
     ##################
     # MIXED PRECISION

compute/pztpgqrt.c

+/**
+ *
+ * @copyright (c) 2009-2016 The University of Tennessee and The University
+ *                          of Tennessee Research Foundation.
+ *                          All rights reserved.
+ * @copyright (c) 2012-2016 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ *                          Univ. Bordeaux. All rights reserved.
+ *
+ **/
+
+/**
+ *
+ * @file pztpgqrt.c
+ *
+ *  MORSE computational routines
+ *  MORSE is a software package provided by Univ. of Tennessee,
+ *  Univ. of California Berkeley and Univ. of Colorado Denver
+ *
+ * @version 0.9.0
+ * @author Mathieu Faverge
+ * @date 2016-12-15
+ * @precisions normal z -> s d c
+ *
+ **/
+#include "control/common.h"
+
+#define V(m,n) V,  m,  n
+#define T(m,n) T,  m,  n
+#define A(m,n) A,  m,  n
+#define B(m,n) B,  m,  n
+
+/***************************************************************************//**
+ *  Parallel tile QR factorization - dynamic scheduling
+ **/
+void morse_pztpgqrt( int L, MORSE_desc_t *V, MORSE_desc_t *T, MORSE_desc_t *A, MORSE_desc_t *B,
+                     MORSE_sequence_t *sequence, MORSE_request_t *request )
+{
+    MORSE_context_t *morse;
+    MORSE_option_t options;
+    size_t ws_worker = 0;
+    size_t ws_host = 0;
+
+    int k, m, n;
+    int ldak, ldvm, ldbm;
+    int tempkn, tempnn, tempmm, templm;
+    int ib;
+
+    /* Dimension of the first column */
+    int maxm  = B->m - L;
+    int maxmt = (maxm % B->mb == 0) ? (maxm / B->mb) : (maxm / B->mb + 1);
+    int maxmtk;
+
+    morse = morse_context_self();
+    if (sequence->status != MORSE_SUCCESS)
+        return;
+    RUNTIME_options_init(&options, morse, sequence, request);
+
+    ib = MORSE_IB;
+
+    /*
+     * ztsmqr = A->nb * ib
+     */
+    ws_worker = A->nb * ib;
+
+    /* Allocation of temporary (scratch) working space */
+#if defined(CHAMELEON_USE_CUDA)
+    /* Worker space
+     *
+     * ztsmqr = 2 * A->nb * ib
+     */
+    ws_worker = max( ws_worker, ib * A->nb * 2 );
+#endif
+
+    ws_worker *= sizeof(MORSE_Complex64_t);
+    ws_host   *= sizeof(MORSE_Complex64_t);
+
+    RUNTIME_options_ws_alloc( &options, ws_worker, ws_host );
+
+    for (k = V->nt-1; k >= 0; k--) {
+        tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
+        ldak = BLKLDD(A, k);
+
+        maxmtk = min( B->mt, maxmt+k ) - 1;
+        for (m = maxmtk; m > -1; m--) {
+            tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+            templm = m == maxmtk  ? tempmm       : 0;
+            ldvm = BLKLDD(V, m);
+            ldbm = BLKLDD(B, m);
+
+            for (n = k; n < B->nt; n++) {
+                tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+                MORSE_TASK_ztpmqrt(
+                    &options,
+                    MorseLeft, MorseConjTrans,
+                    tempmm, tempnn, tempkn, templm, ib, T->nb,
+                    V(m, k), ldvm,
+                    T(m, k), T->mb,
+                    A(k, n), ldak,
+                    B(m, n), ldbm );
+            }
+        }
+    }
+    RUNTIME_options_ws_free(&options);
+    RUNTIME_options_finalize(&options, morse);
+    MORSE_TASK_dataflush_all();
+}

control/compute_z.h

@@ -134,6 +134,7 @@ void morse_pzsyrk(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MO
 void morse_pzsyr2k(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_Complex64_t beta, MORSE_desc_t *C, MORSE_sequence_t *sequence, MORSE_request_t *request);
 void morse_pzsytrf(MORSE_enum uplo, MORSE_desc_t *A, MORSE_sequence_t *sequence, MORSE_request_t *request);
 void morse_pztile2band(MORSE_enum uplo, MORSE_desc_t *A, MORSE_desc_t *descAB, MORSE_sequence_t *sequence, MORSE_request_t *request);
+void morse_pztpgqrt( int L, MORSE_desc_t *V, MORSE_desc_t *T, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request );
 void morse_pztpqrt( int L, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_desc_t *T, MORSE_sequence_t *sequence, MORSE_request_t *request );
 void morse_pztradd(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_Complex64_t beta, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request);
 void morse_pztrmm(MORSE_enum side, MORSE_enum uplo, MORSE_enum transA, MORSE_enum diag, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request);

include/morse_z.h

@@ -102,6 +102,7 @@ int MORSE_zsyrk(MORSE_enum uplo, MORSE_enum trans, int N, int K, MORSE_Complex64
 int MORSE_zsyr2k(MORSE_enum uplo, MORSE_enum trans, int N, int K, MORSE_Complex64_t alpha, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB, MORSE_Complex64_t beta, MORSE_Complex64_t *C, int LDC);
 int MORSE_zsysv(MORSE_enum uplo, int N, int NRHS, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB);
 int MORSE_zsytrs(MORSE_enum uplo, int N, int NRHS, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB);
+int MORSE_ztpgqrt( int M, int N, int K, int L, MORSE_Complex64_t *V, int LDV, MORSE_desc_t *descT, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB );
 int MORSE_ztpqrt( int M, int N, int L, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB, MORSE_desc_t *descT );
 int MORSE_ztradd(MORSE_enum uplo, MORSE_enum trans, int M, int N, MORSE_Complex64_t alpha, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t beta, MORSE_Complex64_t *B, int LDB);
 int MORSE_ztrmm(MORSE_enum side, MORSE_enum uplo, MORSE_enum transA, MORSE_enum diag, int N, int NRHS, MORSE_Complex64_t alpha, MORSE_Complex64_t *A, int LDA, MORSE_Complex64_t *B, int LDB);
@@ -180,6 +181,7 @@ int MORSE_zsyrk_Tile(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha,
 int MORSE_zsyr2k_Tile(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_Complex64_t beta, MORSE_desc_t *C);
 int MORSE_zsysv_Tile(MORSE_enum uplo, MORSE_desc_t *A, MORSE_desc_t *B);
 int MORSE_zsytrs_Tile(MORSE_enum uplo, MORSE_desc_t *A, MORSE_desc_t *B);
+int MORSE_ztpgqrt_Tile( int L, MORSE_desc_t *V, MORSE_desc_t *T, MORSE_desc_t *A, MORSE_desc_t *B );
 int MORSE_ztpqrt_Tile( int L, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_desc_t *T );
 int MORSE_ztradd_Tile(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_Complex64_t beta, MORSE_desc_t *B);
 int MORSE_ztrmm_Tile(MORSE_enum side, MORSE_enum uplo, MORSE_enum transA, MORSE_enum diag, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B);
@@ -255,6 +257,7 @@ int MORSE_zsytrs_Tile_Async(MORSE_enum uplo, MORSE_desc_t *A, MORSE_desc_t *B, M
 int MORSE_zsymm_Tile_Async(MORSE_enum side, MORSE_enum uplo, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_Complex64_t beta, MORSE_desc_t *C, MORSE_sequence_t *sequence, MORSE_request_t *request);
 int MORSE_zsyrk_Tile_Async(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_Complex64_t beta, MORSE_desc_t *C, MORSE_sequence_t *sequence, MORSE_request_t *request);
 int MORSE_zsyr2k_Tile_Async(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_Complex64_t beta, MORSE_desc_t *C, MORSE_sequence_t *sequence, MORSE_request_t *request);
+int MORSE_ztpgqrt_Tile_Async( int L, MORSE_desc_t *V, MORSE_desc_t *T, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request );
 int MORSE_ztpqrt_Tile_Async( int L, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_desc_t *T, MORSE_sequence_t *sequence, MORSE_request_t *request );
 int MORSE_ztradd_Tile_Async(MORSE_enum uplo, MORSE_enum trans, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_Complex64_t beta, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request);
 int MORSE_ztrmm_Tile_Async(MORSE_enum side, MORSE_enum uplo, MORSE_enum transA, MORSE_enum diag, MORSE_Complex64_t alpha, MORSE_desc_t *A, MORSE_desc_t *B, MORSE_sequence_t *sequence, MORSE_request_t *request);