cannam@95: /*
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  */
cannam@95: 
cannam@95: #include "ifftw-mpi.h"
cannam@95: 
cannam@95: /* tproblem.c: */
cannam@95: typedef struct {
cannam@95:      problem super;
cannam@95:      INT vn; /* vector length (vector stride 1) */
cannam@95:      INT nx, ny; /* nx x ny transposed to ny x nx */
cannam@95:      R *I, *O; /* contiguous real arrays (both same size!) */
cannam@95: 
cannam@95:      unsigned flags; /* TRANSPOSED_IN: input is *locally* transposed
cannam@95: 			TRANSPOSED_OUT: output is *locally* transposed */
cannam@95: 
cannam@95:      INT block, tblock; /* block size, slab decomposition;
cannam@95: 			   tblock is for transposed blocks on output */
cannam@95: 
cannam@95:      MPI_Comm comm;
cannam@95: } problem_mpi_transpose;
cannam@95: 
cannam@95: problem *XM(mkproblem_transpose)(INT nx, INT ny, INT vn,
cannam@95: 				 R *I, R *O,
cannam@95: 				 INT block, INT tblock,
cannam@95: 				 MPI_Comm comm,
cannam@95: 				 unsigned flags);
cannam@95: 
cannam@95: /* tsolve.c: */
cannam@95: void XM(transpose_solve)(const plan *ego_, const problem *p_);
cannam@95: 
cannam@95: /* plans have same operands as rdft plans, so just re-use */
cannam@95: typedef plan_rdft plan_mpi_transpose;
cannam@95: #define MKPLAN_MPI_TRANSPOSE(type, adt, apply) \
cannam@95:   (type *)X(mkplan_rdft)(sizeof(type), adt, apply)
cannam@95: 
cannam@95: /* transpose-pairwise.c: */
cannam@95: int XM(mkplans_posttranspose)(const problem_mpi_transpose *p, planner *plnr,
cannam@95: 			      R *I, R *O, int my_pe,
cannam@95: 			      plan **cld2, plan **cld2rest, plan **cld3,
cannam@95: 			      INT *rest_Ioff, INT *rest_Ooff);
cannam@95: /* various solvers */
cannam@95: void XM(transpose_pairwise_register)(planner *p);
cannam@95: void XM(transpose_alltoall_register)(planner *p);
cannam@95: void XM(transpose_recurse_register)(planner *p);