Chris@10: /*
Chris@10:  * Copyright (c) 2001 Matteo Frigo
Chris@10:  * Copyright (c) 2001 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: #include "config.h"
Chris@10: #include "bench.h"
Chris@10: #include <stdio.h>
Chris@10: #include <stdlib.h>
Chris@10: #include <string.h>
Chris@10: #include <ctype.h>
Chris@10: 
Chris@10: int always_pad_real = 0; /* by default, only pad in-place case */
Chris@10: 
Chris@10: typedef enum {
Chris@10:      SAME, PADDED, HALFISH
Chris@10: } n_transform;
Chris@10: 
Chris@10: /* funny transformations for last dimension of PROBLEM_REAL */
Chris@10: static int transform_n(int n, n_transform nt)
Chris@10: {
Chris@10:      switch (nt) {
Chris@10: 	 case SAME: return n;
Chris@10: 	 case PADDED: return 2*(n/2+1);
Chris@10: 	 case HALFISH: return (n/2+1);
Chris@10: 	 default: BENCH_ASSERT(0); return 0;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* do what I mean */
Chris@10: static bench_tensor *dwim(bench_tensor *t, bench_iodim **last_iodim,
Chris@10: 			  n_transform nti, n_transform nto,
Chris@10: 			  bench_iodim *dt)
Chris@10: {
Chris@10:      int i;
Chris@10:      bench_iodim *d, *d1;
Chris@10: 
Chris@10:      if (!FINITE_RNK(t->rnk) || t->rnk < 1)
Chris@10: 	  return t;
Chris@10: 
Chris@10:      i = t->rnk;
Chris@10:      d1 = *last_iodim;
Chris@10: 
Chris@10:      while (--i >= 0) {
Chris@10: 	  d = t->dims + i;
Chris@10: 	  if (!d->is) 
Chris@10: 	       d->is = d1->is * transform_n(d1->n, d1==dt ? nti : SAME); 
Chris@10: 	  if (!d->os) 
Chris@10: 	       d->os = d1->os * transform_n(d1->n, d1==dt ? nto : SAME); 
Chris@10: 	  d1 = d;
Chris@10:      }
Chris@10: 
Chris@10:      *last_iodim = d1;
Chris@10:      return t;
Chris@10: }
Chris@10: 
Chris@10: static void transpose_tensor(bench_tensor *t)
Chris@10: {
Chris@10:      if (!FINITE_RNK(t->rnk) || t->rnk < 2)
Chris@10:           return;
Chris@10: 
Chris@10:      t->dims[0].os = t->dims[1].os;
Chris@10:      t->dims[1].os = t->dims[0].os * t->dims[0].n;
Chris@10: }
Chris@10: 
Chris@10: static const char *parseint(const char *s, int *n)
Chris@10: {
Chris@10:      int sign = 1;
Chris@10: 
Chris@10:      *n = 0;
Chris@10: 
Chris@10:      if (*s == '-') { 
Chris@10: 	  sign = -1;
Chris@10: 	  ++s;
Chris@10:      } else if (*s == '+') { 
Chris@10: 	  sign = +1; 
Chris@10: 	  ++s; 
Chris@10:      }
Chris@10: 
Chris@10:      BENCH_ASSERT(isdigit(*s));
Chris@10:      while (isdigit(*s)) {
Chris@10: 	  *n = *n * 10 + (*s - '0');
Chris@10: 	  ++s;
Chris@10:      }
Chris@10:      
Chris@10:      *n *= sign;
Chris@10: 
Chris@10:      if (*s == 'k' || *s == 'K') {
Chris@10: 	  *n *= 1024;
Chris@10: 	  ++s;
Chris@10:      }
Chris@10: 
Chris@10:      if (*s == 'm' || *s == 'M') {
Chris@10: 	  *n *= 1024 * 1024;
Chris@10: 	  ++s;
Chris@10:      }
Chris@10: 
Chris@10:      return s;
Chris@10: }
Chris@10: 
Chris@10: struct dimlist { bench_iodim car; r2r_kind_t k; struct dimlist *cdr; };
Chris@10: 
Chris@10: static const char *parsetensor(const char *s, bench_tensor **tp,
Chris@10: 			       r2r_kind_t **k)
Chris@10: {
Chris@10:      struct dimlist *l = 0, *m;
Chris@10:      bench_tensor *t;
Chris@10:      int rnk = 0;
Chris@10: 
Chris@10:  L1:
Chris@10:      m = (struct dimlist *)bench_malloc(sizeof(struct dimlist));
Chris@10:      /* nconc onto l */
Chris@10:      m->cdr = l; l = m;
Chris@10:      ++rnk; 
Chris@10: 
Chris@10:      s = parseint(s, &m->car.n);
Chris@10: 
Chris@10:      if (*s == ':') {
Chris@10: 	  /* read input stride */
Chris@10: 	  ++s;
Chris@10: 	  s = parseint(s, &m->car.is);
Chris@10: 	  if (*s == ':') {
Chris@10: 	       /* read output stride */
Chris@10: 	       ++s;
Chris@10: 	       s = parseint(s, &m->car.os);
Chris@10: 	  } else {
Chris@10: 	       /* default */
Chris@10: 	       m->car.os = m->car.is;
Chris@10: 	  }
Chris@10:      } else {
Chris@10: 	  m->car.is = 0;
Chris@10: 	  m->car.os = 0;
Chris@10:      }
Chris@10: 
Chris@10:      if (*s == 'f' || *s == 'F') {
Chris@10: 	  m->k = R2R_R2HC;
Chris@10: 	  ++s;
Chris@10:      }
Chris@10:      else if (*s == 'b' || *s == 'B') {
Chris@10: 	  m->k = R2R_HC2R;
Chris@10: 	  ++s;
Chris@10:      }
Chris@10:      else if (*s == 'h' || *s == 'H') {
Chris@10: 	  m->k = R2R_DHT;
Chris@10: 	  ++s;
Chris@10:      }
Chris@10:      else if (*s == 'e' || *s == 'E' || *s == 'o' || *s == 'O') {
Chris@10: 	  char c = *(s++);
Chris@10: 	  int ab;
Chris@10: 
Chris@10: 	  s = parseint(s, &ab);
Chris@10: 
Chris@10: 	  if (c == 'e' || c == 'E') {
Chris@10: 	       if (ab == 0)
Chris@10: 		    m->k = R2R_REDFT00;
Chris@10: 	       else if (ab == 1)
Chris@10: 		    m->k = R2R_REDFT01;
Chris@10: 	       else if (ab == 10)
Chris@10: 		    m->k = R2R_REDFT10;
Chris@10: 	       else if (ab == 11)
Chris@10: 		    m->k = R2R_REDFT11;
Chris@10: 	       else
Chris@10: 		    BENCH_ASSERT(0);
Chris@10: 	  }
Chris@10: 	  else {
Chris@10: 	       if (ab == 0)
Chris@10: 		    m->k = R2R_RODFT00;
Chris@10: 	       else if (ab == 1)
Chris@10: 		    m->k = R2R_RODFT01;
Chris@10: 	       else if (ab == 10)
Chris@10: 		    m->k = R2R_RODFT10;
Chris@10: 	       else if (ab == 11)
Chris@10: 		    m->k = R2R_RODFT11;
Chris@10: 	       else
Chris@10: 		    BENCH_ASSERT(0);
Chris@10: 	  }
Chris@10:      }
Chris@10:      else
Chris@10: 	  m->k = R2R_R2HC;
Chris@10: 
Chris@10:      if (*s == 'x' || *s == 'X') {
Chris@10: 	  ++s;
Chris@10: 	  goto L1;
Chris@10:      }
Chris@10:      
Chris@10:      /* now we have a dimlist.  Build bench_tensor, etc. */
Chris@10: 
Chris@10:      if (k && rnk > 0) {
Chris@10: 	  int i;
Chris@10: 	  *k = (r2r_kind_t *) bench_malloc(sizeof(r2r_kind_t) * rnk);
Chris@10: 	  for (m = l, i = rnk - 1; i >= 0; --i, m = m->cdr) {
Chris@10: 	       BENCH_ASSERT(m);
Chris@10: 	       (*k)[i] = m->k;
Chris@10: 	  }
Chris@10:      }
Chris@10: 
Chris@10:      t = mktensor(rnk);
Chris@10:      while (--rnk >= 0) {
Chris@10: 	  bench_iodim *d = t->dims + rnk;
Chris@10: 	  BENCH_ASSERT(l);
Chris@10: 	  m = l; l = m->cdr;
Chris@10: 	  d->n = m->car.n;
Chris@10: 	  d->is = m->car.is;
Chris@10: 	  d->os = m->car.os;
Chris@10: 	  bench_free(m);
Chris@10:      }
Chris@10: 
Chris@10:      *tp = t;
Chris@10:      return s;
Chris@10: }
Chris@10: 
Chris@10: /* parse a problem description, return a problem */
Chris@10: bench_problem *problem_parse(const char *s)
Chris@10: {
Chris@10:      bench_problem *p;
Chris@10:      bench_iodim last_iodim0 = {1,1,1}, *last_iodim = &last_iodim0;
Chris@10:      bench_iodim *sz_last_iodim;
Chris@10:      bench_tensor *sz;
Chris@10:      n_transform nti = SAME, nto = SAME;
Chris@10:      int transpose = 0;
Chris@10: 
Chris@10:      p = (bench_problem *) bench_malloc(sizeof(bench_problem));
Chris@10:      p->kind = PROBLEM_COMPLEX;
Chris@10:      p->k = 0;
Chris@10:      p->sign = -1;
Chris@10:      p->in = p->out = 0;
Chris@10:      p->inphys = p->outphys = 0;
Chris@10:      p->iphyssz = p->ophyssz = 0;
Chris@10:      p->in_place = 0;
Chris@10:      p->destroy_input = 0;
Chris@10:      p->split = 0;
Chris@10:      p->userinfo = 0;
Chris@10:      p->scrambled_in = p->scrambled_out = 0;
Chris@10:      p->sz = p->vecsz = 0;
Chris@10:      p->ini = p->outi = 0;
Chris@10:      p->pstring = (char *) bench_malloc(sizeof(char) * (strlen(s) + 1));
Chris@10:      strcpy(p->pstring, s);
Chris@10: 
Chris@10:  L1:
Chris@10:      switch (tolower(*s)) {
Chris@10: 	 case 'i': p->in_place = 1; ++s; goto L1;
Chris@10: 	 case 'o': p->in_place = 0; ++s; goto L1;
Chris@10: 	 case 'd': p->destroy_input = 1; ++s; goto L1;
Chris@10: 	 case '/': p->split = 1; ++s; goto L1;
Chris@10: 	 case 'f': 
Chris@10: 	 case '-': p->sign = -1; ++s; goto L1;
Chris@10: 	 case 'b': 
Chris@10: 	 case '+': p->sign = 1; ++s; goto L1;
Chris@10: 	 case 'r': p->kind = PROBLEM_REAL; ++s; goto L1;
Chris@10: 	 case 'c': p->kind = PROBLEM_COMPLEX; ++s; goto L1;
Chris@10: 	 case 'k': p->kind = PROBLEM_R2R; ++s; goto L1;
Chris@10: 	 case 't': transpose = 1; ++s; goto L1;
Chris@10: 	      
Chris@10: 	 /* hack for MPI: */
Chris@10: 	 case '[': p->scrambled_in = 1; ++s; goto L1;
Chris@10: 	 case ']': p->scrambled_out = 1; ++s; goto L1;
Chris@10: 
Chris@10: 	 default : ;
Chris@10:      }
Chris@10: 
Chris@10:      s = parsetensor(s, &sz, p->kind == PROBLEM_R2R ? &p->k : 0);
Chris@10: 
Chris@10:      if (p->kind == PROBLEM_REAL) {
Chris@10: 	  if (p->sign < 0) {
Chris@10: 	       nti = p->in_place || always_pad_real ? PADDED : SAME;
Chris@10: 	       nto = HALFISH;
Chris@10: 	  }
Chris@10: 	  else {
Chris@10: 	       nti = HALFISH;
Chris@10: 	       nto = p->in_place || always_pad_real ? PADDED : SAME;
Chris@10: 	  }
Chris@10:      }
Chris@10: 
Chris@10:      sz_last_iodim = sz->dims + sz->rnk - 1;
Chris@10:      if (*s == '*') { /* "external" vector */
Chris@10: 	  ++s;
Chris@10: 	  p->sz = dwim(sz, &last_iodim, nti, nto, sz_last_iodim);
Chris@10: 	  s = parsetensor(s, &sz, 0);
Chris@10: 	  p->vecsz = dwim(sz, &last_iodim, nti, nto, sz_last_iodim);
Chris@10:      } else if (*s == 'v' || *s == 'V') { /* "internal" vector */
Chris@10: 	  bench_tensor *vecsz;
Chris@10: 	  ++s;
Chris@10: 	  s = parsetensor(s, &vecsz, 0);
Chris@10: 	  p->vecsz = dwim(vecsz, &last_iodim, nti, nto, sz_last_iodim);
Chris@10: 	  p->sz = dwim(sz, &last_iodim, nti, nto, sz_last_iodim);
Chris@10:      } else {
Chris@10: 	  p->sz = dwim(sz, &last_iodim, nti, nto, sz_last_iodim);
Chris@10: 	  p->vecsz = mktensor(0);
Chris@10:      }
Chris@10: 
Chris@10:      if (transpose) {
Chris@10: 	  transpose_tensor(p->sz);
Chris@10: 	  transpose_tensor(p->vecsz);
Chris@10:      }
Chris@10: 
Chris@10:      if (!p->in_place)
Chris@10: 	  p->out = ((bench_real *) p->in) + (1 << 20);  /* whatever */
Chris@10: 
Chris@10:      BENCH_ASSERT(p->sz && p->vecsz);
Chris@10:      BENCH_ASSERT(!*s);
Chris@10:      return p;
Chris@10: }
Chris@10: 
Chris@10: void problem_destroy(bench_problem *p)
Chris@10: {
Chris@10:      BENCH_ASSERT(p);
Chris@10:      problem_free(p);
Chris@10:      bench_free0(p->k);
Chris@10:      bench_free0(p->pstring);
Chris@10:      bench_free(p);
Chris@10: }
Chris@10: