annotate src/fftw-3.3.8/mpi/rearrange.c @ 169:223a55898ab9 tip default

Add null config files
author Chris Cannam <cannam@all-day-breakfast.com>
date Mon, 02 Mar 2020 14:03:47 +0000
parents bd3cc4d1df30
children
rev   line source
cannam@167 1 /*
cannam@167 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
cannam@167 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
cannam@167 4 *
cannam@167 5 * This program is free software; you can redistribute it and/or modify
cannam@167 6 * it under the terms of the GNU General Public License as published by
cannam@167 7 * the Free Software Foundation; either version 2 of the License, or
cannam@167 8 * (at your option) any later version.
cannam@167 9 *
cannam@167 10 * This program is distributed in the hope that it will be useful,
cannam@167 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@167 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
cannam@167 13 * GNU General Public License for more details.
cannam@167 14 *
cannam@167 15 * You should have received a copy of the GNU General Public License
cannam@167 16 * along with this program; if not, write to the Free Software
cannam@167 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
cannam@167 18 *
cannam@167 19 */
cannam@167 20
cannam@167 21 #include "ifftw-mpi.h"
cannam@167 22
cannam@167 23 /* common functions for rearrangements of the data for the *-rank1-bigvec
cannam@167 24 solvers */
cannam@167 25
cannam@167 26 static int div_mult(INT b, INT a) {
cannam@167 27 return (a > b && a % b == 0);
cannam@167 28 }
cannam@167 29 static int div_mult2(INT b, INT a, INT n) {
cannam@167 30 return (div_mult(b, a) && div_mult(n, b));
cannam@167 31 }
cannam@167 32
cannam@167 33 int XM(rearrange_applicable)(rearrangement rearrange,
cannam@167 34 ddim dim0, INT vn, int n_pes)
cannam@167 35 {
cannam@167 36 /* note: it is important that cases other than CONTIG be
cannam@167 37 applicable only when the resulting transpose dimension
cannam@167 38 is divisible by n_pes; otherwise, the allocation size
cannam@167 39 returned by the API will be incorrect */
cannam@167 40 return ((rearrange != DISCONTIG || div_mult(n_pes, vn))
cannam@167 41 && (rearrange != SQUARE_BEFORE
cannam@167 42 || div_mult2(dim0.b[IB], vn, n_pes))
cannam@167 43 && (rearrange != SQUARE_AFTER
cannam@167 44 || (dim0.b[IB] != dim0.b[OB]
cannam@167 45 && div_mult2(dim0.b[OB], vn, n_pes)))
cannam@167 46 && (rearrange != SQUARE_MIDDLE
cannam@167 47 || div_mult(dim0.n * n_pes, vn)));
cannam@167 48 }
cannam@167 49
cannam@167 50 INT XM(rearrange_ny)(rearrangement rearrange, ddim dim0, INT vn, int n_pes)
cannam@167 51 {
cannam@167 52 switch (rearrange) {
cannam@167 53 case CONTIG:
cannam@167 54 return vn;
cannam@167 55 case DISCONTIG:
cannam@167 56 return n_pes;
cannam@167 57 case SQUARE_BEFORE:
cannam@167 58 return dim0.b[IB];
cannam@167 59 case SQUARE_AFTER:
cannam@167 60 return dim0.b[OB];
cannam@167 61 case SQUARE_MIDDLE:
cannam@167 62 return dim0.n * n_pes;
cannam@167 63 }
cannam@167 64 return 0;
cannam@167 65 }