Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: 
Chris@10: /* Twiddle manipulation */
Chris@10: 
Chris@10: #include "ifftw.h"
Chris@10: #include <math.h>
Chris@10: 
Chris@10: #define HASHSZ 109
Chris@10: 
Chris@10: /* hash table of known twiddle factors */
Chris@10: static twid *twlist[HASHSZ];
Chris@10: 
Chris@10: static INT hash(INT n, INT r)
Chris@10: {
Chris@10:      INT h = n * 17 + r;
Chris@10: 
Chris@10:      if (h < 0) h = -h;
Chris@10: 
Chris@10:      return (h % HASHSZ);
Chris@10: }
Chris@10: 
Chris@10: static int equal_instr(const tw_instr *p, const tw_instr *q)
Chris@10: {
Chris@10:      if (p == q)
Chris@10:           return 1;
Chris@10: 
Chris@10:      for (;; ++p, ++q) {
Chris@10:           if (p->op != q->op)
Chris@10: 	       return 0;
Chris@10: 
Chris@10: 	  switch (p->op) {
Chris@10: 	      case TW_NEXT:
Chris@10: 		   return (p->v == q->v); /* p->i is ignored */
Chris@10: 
Chris@10: 	      case TW_FULL:
Chris@10: 	      case TW_HALF:
Chris@10: 		   if (p->v != q->v) return 0; /* p->i is ignored */
Chris@10: 		   break;
Chris@10: 
Chris@10: 	      default:
Chris@10: 		   if (p->v != q->v || p->i != q->i) return 0;
Chris@10: 		   break;
Chris@10: 	  }
Chris@10:      }
Chris@10:      A(0 /* can't happen */);
Chris@10: }
Chris@10: 
Chris@10: static int ok_twid(const twid *t, 
Chris@10: 		   enum wakefulness wakefulness,
Chris@10: 		   const tw_instr *q, INT n, INT r, INT m)
Chris@10: {
Chris@10:      return (wakefulness == t->wakefulness &&
Chris@10: 	     n == t->n &&
Chris@10: 	     r == t->r && 
Chris@10: 	     m <= t->m && 
Chris@10: 	     equal_instr(t->instr, q));
Chris@10: }
Chris@10: 
Chris@10: static twid *lookup(enum wakefulness wakefulness,
Chris@10: 		    const tw_instr *q, INT n, INT r, INT m)
Chris@10: {
Chris@10:      twid *p;
Chris@10: 
Chris@10:      for (p = twlist[hash(n,r)]; 
Chris@10: 	  p && !ok_twid(p, wakefulness, q, n, r, m); 
Chris@10: 	  p = p->cdr)
Chris@10:           ;
Chris@10:      return p;
Chris@10: }
Chris@10: 
Chris@10: static INT twlen0(INT r, const tw_instr *p, INT *vl)
Chris@10: {
Chris@10:      INT ntwiddle = 0;
Chris@10: 
Chris@10:      /* compute length of bytecode program */
Chris@10:      A(r > 0);
Chris@10:      for ( ; p->op != TW_NEXT; ++p) {
Chris@10: 	  switch (p->op) {
Chris@10: 	      case TW_FULL:
Chris@10: 		   ntwiddle += (r - 1) * 2;
Chris@10: 		   break;
Chris@10: 	      case TW_HALF:
Chris@10: 		   ntwiddle += (r - 1);
Chris@10: 		   break;
Chris@10: 	      case TW_CEXP:
Chris@10: 		   ntwiddle += 2;
Chris@10: 		   break;
Chris@10: 	      case TW_COS:
Chris@10: 	      case TW_SIN:
Chris@10: 		   ntwiddle += 1;
Chris@10: 		   break;
Chris@10: 	  }
Chris@10:      }
Chris@10: 
Chris@10:      *vl = (INT)p->v;
Chris@10:      return ntwiddle;
Chris@10: }
Chris@10: 
Chris@10: INT X(twiddle_length)(INT r, const tw_instr *p)
Chris@10: {
Chris@10:      INT vl;
Chris@10:      return twlen0(r, p, &vl);
Chris@10: }
Chris@10: 
Chris@10: static R *compute(enum wakefulness wakefulness,
Chris@10: 		  const tw_instr *instr, INT n, INT r, INT m)
Chris@10: {
Chris@10:      INT ntwiddle, j, vl;
Chris@10:      R *W, *W0;
Chris@10:      const tw_instr *p;
Chris@10:      triggen *t = X(mktriggen)(wakefulness, n);
Chris@10: 
Chris@10:      p = instr;
Chris@10:      ntwiddle = twlen0(r, p, &vl);
Chris@10: 
Chris@10:      A(m % vl == 0);
Chris@10: 
Chris@10:      W0 = W = (R *)MALLOC((ntwiddle * (m / vl)) * sizeof(R), TWIDDLES);
Chris@10: 
Chris@10:      for (j = 0; j < m; j += vl) {
Chris@10:           for (p = instr; p->op != TW_NEXT; ++p) {
Chris@10: 	       switch (p->op) {
Chris@10: 		   case TW_FULL: {
Chris@10: 			INT i;
Chris@10: 			for (i = 1; i < r; ++i) {
Chris@10: 			     A((j + (INT)p->v) * i < n);
Chris@10: 			     A((j + (INT)p->v) * i > -n);
Chris@10: 			     t->cexp(t, (j + (INT)p->v) * i, W);
Chris@10: 			     W += 2;
Chris@10: 			}
Chris@10: 			break;
Chris@10: 		   }
Chris@10: 
Chris@10: 		   case TW_HALF: {
Chris@10: 			INT i;
Chris@10: 			A((r % 2) == 1);
Chris@10: 			for (i = 1; i + i < r; ++i) {
Chris@10: 			     t->cexp(t, MULMOD(i, (j + (INT)p->v), n), W);
Chris@10: 			     W += 2;
Chris@10: 			}
Chris@10: 			break;
Chris@10: 		   }
Chris@10: 
Chris@10: 		   case TW_COS: {
Chris@10: 			R d[2];
Chris@10: 
Chris@10: 			A((j + (INT)p->v) * p->i < n);
Chris@10: 			A((j + (INT)p->v) * p->i > -n);
Chris@10: 			t->cexp(t, (j + (INT)p->v) * (INT)p->i, d);
Chris@10: 			*W++ = d[0];
Chris@10: 			break;
Chris@10: 		   }
Chris@10: 
Chris@10: 		   case TW_SIN: {
Chris@10: 			R d[2];
Chris@10: 
Chris@10: 			A((j + (INT)p->v) * p->i < n);
Chris@10: 			A((j + (INT)p->v) * p->i > -n);
Chris@10: 			t->cexp(t, (j + (INT)p->v) * (INT)p->i, d);
Chris@10: 			*W++ = d[1];
Chris@10: 			break;
Chris@10: 		   }
Chris@10: 
Chris@10: 		   case TW_CEXP:
Chris@10: 			A((j + (INT)p->v) * p->i < n);
Chris@10: 			A((j + (INT)p->v) * p->i > -n);
Chris@10: 			t->cexp(t, (j + (INT)p->v) * (INT)p->i, W);
Chris@10: 			W += 2;
Chris@10: 			break;
Chris@10: 	       }
Chris@10: 	  }
Chris@10:      }
Chris@10: 
Chris@10:      X(triggen_destroy)(t);
Chris@10:      return W0;
Chris@10: }
Chris@10: 
Chris@10: static void mktwiddle(enum wakefulness wakefulness,
Chris@10: 		      twid **pp, const tw_instr *instr, INT n, INT r, INT m)
Chris@10: {
Chris@10:      twid *p;
Chris@10:      INT h;
Chris@10: 
Chris@10:      if ((p = lookup(wakefulness, instr, n, r, m))) {
Chris@10:           ++p->refcnt;
Chris@10:      } else {
Chris@10: 	  p = (twid *) MALLOC(sizeof(twid), TWIDDLES);
Chris@10: 	  p->n = n;
Chris@10: 	  p->r = r;
Chris@10: 	  p->m = m;
Chris@10: 	  p->instr = instr;
Chris@10: 	  p->refcnt = 1;
Chris@10: 	  p->wakefulness = wakefulness;
Chris@10: 	  p->W = compute(wakefulness, instr, n, r, m);
Chris@10: 
Chris@10: 	  /* cons! onto twlist */
Chris@10: 	  h = hash(n, r);
Chris@10: 	  p->cdr = twlist[h];
Chris@10: 	  twlist[h] = p;
Chris@10:      }
Chris@10: 
Chris@10:      *pp = p;
Chris@10: }
Chris@10: 
Chris@10: static void twiddle_destroy(twid **pp)
Chris@10: {
Chris@10:      twid *p = *pp;
Chris@10:      twid **q;
Chris@10: 
Chris@10:      if ((--p->refcnt) == 0) {
Chris@10: 	  /* remove p from twiddle list */
Chris@10: 	  for (q = &twlist[hash(p->n, p->r)]; *q; q = &((*q)->cdr)) {
Chris@10: 	       if (*q == p) {
Chris@10: 		    *q = p->cdr;
Chris@10: 		    X(ifree)(p->W);
Chris@10: 		    X(ifree)(p);
Chris@10: 		    *pp = 0;
Chris@10: 		    return;
Chris@10: 	       }
Chris@10: 	  }
Chris@10: 	  A(0 /* can't happen */ );
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: 
Chris@10: void X(twiddle_awake)(enum wakefulness wakefulness, twid **pp, 
Chris@10: 		      const tw_instr *instr, INT n, INT r, INT m)
Chris@10: {
Chris@10:      switch (wakefulness) {
Chris@10: 	 case SLEEPY: 
Chris@10: 	      twiddle_destroy(pp);
Chris@10: 	      break;
Chris@10: 	 default:
Chris@10: 	      mktwiddle(wakefulness, pp, instr, n, r, m);
Chris@10: 	      break;
Chris@10:      }
Chris@10: }