Chris@42: /*
Chris@42:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42:  *
Chris@42:  * This program is free software; you can redistribute it and/or modify
Chris@42:  * it under the terms of the GNU General Public License as published by
Chris@42:  * the Free Software Foundation; either version 2 of the License, or
Chris@42:  * (at your option) any later version.
Chris@42:  *
Chris@42:  * This program is distributed in the hope that it will be useful,
Chris@42:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@42:  * GNU General Public License for more details.
Chris@42:  *
Chris@42:  * You should have received a copy of the GNU General Public License
Chris@42:  * along with this program; if not, write to the Free Software
Chris@42:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@42:  *
Chris@42:  */
Chris@42: 
Chris@42: /* routines shared by the various buffered solvers */
Chris@42: 
Chris@42: #include "ifftw.h"
Chris@42: 
Chris@42: #define DEFAULT_MAXNBUF ((INT)256)
Chris@42: 
Chris@42: /* approx. 512KB of buffers for complex data */
Chris@42: #define MAXBUFSZ (256 * 1024 / (INT)(sizeof(R)))
Chris@42: 
Chris@42: INT X(nbuf)(INT n, INT vl, INT maxnbuf)
Chris@42: {
Chris@42:      INT i, nbuf, lb; 
Chris@42: 
Chris@42:      if (!maxnbuf) 
Chris@42: 	  maxnbuf = DEFAULT_MAXNBUF;
Chris@42: 
Chris@42:      nbuf = X(imin)(maxnbuf,
Chris@42: 		    X(imin)(vl, X(imax)((INT)1, MAXBUFSZ / n)));
Chris@42: 
Chris@42:      /*
Chris@42:       * Look for a buffer number (not too small) that divides the
Chris@42:       * vector length, in order that we only need one child plan:
Chris@42:       */
Chris@42:      lb = X(imax)(1, nbuf / 4);
Chris@42:      for (i = nbuf; i >= lb; --i)
Chris@42:           if (vl % i == 0)
Chris@42:                return i;
Chris@42: 
Chris@42:      /* whatever... */
Chris@42:      return nbuf;
Chris@42: }
Chris@42: 
Chris@42: #define SKEW 6 /* need to be even for SIMD */
Chris@42: #define SKEWMOD 8 
Chris@42: 
Chris@42: INT X(bufdist)(INT n, INT vl)
Chris@42: {
Chris@42:      if (vl == 1)
Chris@42: 	  return n;
Chris@42:      else 
Chris@42: 	  /* return smallest X such that X >= N and X == SKEW (mod SKEWMOD) */
Chris@42: 	  return n + X(modulo)(SKEW - n, SKEWMOD);
Chris@42: }
Chris@42: 
Chris@42: int X(toobig)(INT n)
Chris@42: {
Chris@42:      return n > MAXBUFSZ;
Chris@42: }
Chris@42: 
Chris@42: /* TRUE if there exists i < which such that maxnbuf[i] and
Chris@42:    maxnbuf[which] yield the same value, in which case we canonicalize
Chris@42:    on the minimum value */
Chris@42: int X(nbuf_redundant)(INT n, INT vl, size_t which, 
Chris@42: 		      const INT *maxnbuf, size_t nmaxnbuf)
Chris@42: {
Chris@42:      size_t i;
Chris@42:      (void)nmaxnbuf; /* UNUSED */
Chris@42:      for (i = 0; i < which; ++i)
Chris@42: 	  if (X(nbuf)(n, vl, maxnbuf[i]) == X(nbuf)(n, vl, maxnbuf[which]))
Chris@42: 	       return 1;
Chris@42:      return 0;
Chris@42: }