Chris@10: /**************************************************************************/
Chris@10: /* NOTE to users: this is the FFTW self-test and benchmark program.
Chris@10:    It is probably NOT a good place to learn FFTW usage, since it has a
Chris@10:    lot of added complexity in order to exercise and test the full API,
Chris@10:    etcetera.  We suggest reading the manual. 
Chris@10: 
Chris@10:    (Some of the self-test code is split off into fftw-bench.c and
Chris@10:    hook.c.) */
Chris@10: /**************************************************************************/
Chris@10: 
Chris@10: #include <math.h>
Chris@10: #include <stdio.h>
Chris@10: #include <string.h>
Chris@10: #include "fftw-bench.h"
Chris@10: 
Chris@10: static const char *mkversion(void) { return FFTW(version); }
Chris@10: static const char *mkcc(void) { return FFTW(cc); }
Chris@10: static const char *mkcodelet_optim(void) { return FFTW(codelet_optim); }
Chris@10: 
Chris@10: BEGIN_BENCH_DOC
Chris@10: BENCH_DOC("name", "fftw3")
Chris@10: BENCH_DOCF("version", mkversion)
Chris@10: BENCH_DOCF("cc", mkcc)
Chris@10: BENCH_DOCF("codelet-optim", mkcodelet_optim)
Chris@10: END_BENCH_DOC 
Chris@10: 
Chris@10: static FFTW(iodim) *bench_tensor_to_fftw_iodim(bench_tensor *t)
Chris@10: {
Chris@10:      FFTW(iodim) *d;
Chris@10:      int i;
Chris@10: 
Chris@10:      BENCH_ASSERT(t->rnk >= 0);
Chris@10:      if (t->rnk == 0) return 0;
Chris@10:      
Chris@10:      d = (FFTW(iodim) *)bench_malloc(sizeof(FFTW(iodim)) * t->rnk);
Chris@10:      for (i = 0; i < t->rnk; ++i) {
Chris@10: 	  d[i].n = t->dims[i].n;
Chris@10: 	  d[i].is = t->dims[i].is;
Chris@10: 	  d[i].os = t->dims[i].os;
Chris@10:      }
Chris@10: 
Chris@10:      return d;
Chris@10: }
Chris@10: 
Chris@10: static void extract_reim_split(int sign, int size, bench_real *p,
Chris@10: 			       bench_real **r, bench_real **i)
Chris@10: {
Chris@10:      if (sign == FFTW_FORWARD) {
Chris@10:           *r = p + 0;
Chris@10:           *i = p + size;
Chris@10:      } else {
Chris@10:           *r = p + size;
Chris@10:           *i = p + 0;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static int sizeof_problem(bench_problem *p)
Chris@10: {
Chris@10:      return tensor_sz(p->sz) * tensor_sz(p->vecsz);
Chris@10: }
Chris@10: 
Chris@10: /* ouch */
Chris@10: static int expressible_as_api_many(bench_tensor *t)
Chris@10: {
Chris@10:      int i;
Chris@10: 
Chris@10:      BENCH_ASSERT(FINITE_RNK(t->rnk));
Chris@10: 
Chris@10:      i = t->rnk - 1;
Chris@10:      while (--i >= 0) {
Chris@10: 	  bench_iodim *d = t->dims + i;
Chris@10: 	  if (d[0].is % d[1].is) return 0;
Chris@10: 	  if (d[0].os % d[1].os) return 0;
Chris@10:      }
Chris@10:      return 1;
Chris@10: }
Chris@10: 
Chris@10: static int *mkn(bench_tensor *t)
Chris@10: {
Chris@10:      int *n = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@10:      int i;
Chris@10:      for (i = 0; i < t->rnk; ++i) 
Chris@10: 	  n[i] = t->dims[i].n;
Chris@10:      return n;
Chris@10: }
Chris@10: 
Chris@10: static void mknembed_many(bench_tensor *t, int **inembedp, int **onembedp)
Chris@10: {
Chris@10:      int i;
Chris@10:      bench_iodim *d;
Chris@10:      int *inembed = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@10:      int *onembed = (int *) bench_malloc(sizeof(int *) * t->rnk);
Chris@10: 
Chris@10:      BENCH_ASSERT(FINITE_RNK(t->rnk));
Chris@10:      *inembedp = inembed; *onembedp = onembed;
Chris@10: 
Chris@10:      i = t->rnk - 1;
Chris@10:      while (--i >= 0) {
Chris@10: 	  d = t->dims + i;
Chris@10: 	  inembed[i+1] = d[0].is / d[1].is;
Chris@10: 	  onembed[i+1] = d[0].os / d[1].os;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* try to use the most appropriate API function.  Big mess. */
Chris@10: 
Chris@10: static int imax(int a, int b) { return (a > b ? a : b); }
Chris@10: 
Chris@10: static int halfish_sizeof_problem(bench_problem *p)
Chris@10: {
Chris@10:      int n2 = sizeof_problem(p);
Chris@10:      if (FINITE_RNK(p->sz->rnk) && p->sz->rnk > 0)
Chris@10:           n2 = (n2 / imax(p->sz->dims[p->sz->rnk - 1].n, 1)) *
Chris@10:                (p->sz->dims[p->sz->rnk - 1].n / 2 + 1);
Chris@10:      return n2;
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_real_split(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      FFTW(plan) pln;
Chris@10:      bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@10:      FFTW(iodim) *dims, *howmany_dims;
Chris@10:      bench_real *ri, *ii, *ro, *io;
Chris@10:      int n2 = halfish_sizeof_problem(p);
Chris@10: 
Chris@10:      extract_reim_split(FFTW_FORWARD, n2, (bench_real *) p->in, &ri, &ii);
Chris@10:      extract_reim_split(FFTW_FORWARD, n2, (bench_real *) p->out, &ro, &io);
Chris@10: 
Chris@10:      dims = bench_tensor_to_fftw_iodim(sz);
Chris@10:      howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10:      if (p->sign < 0) {
Chris@10: 	  if (verbose > 2) printf("using plan_guru_split_dft_r2c\n");
Chris@10: 	  pln = FFTW(plan_guru_split_dft_r2c)(sz->rnk, dims,
Chris@10: 					vecsz->rnk, howmany_dims,
Chris@10: 					ri, ro, io, flags);
Chris@10:      }
Chris@10:      else {
Chris@10: 	  if (verbose > 2) printf("using plan_guru_split_dft_c2r\n");
Chris@10: 	  pln = FFTW(plan_guru_split_dft_c2r)(sz->rnk, dims,
Chris@10: 					vecsz->rnk, howmany_dims,
Chris@10: 					ri, ii, ro, flags);
Chris@10:      }
Chris@10:      bench_free(dims);
Chris@10:      bench_free(howmany_dims);
Chris@10:      return pln;
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_real_interleaved(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      FFTW(plan) pln;
Chris@10:      bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@10: 
Chris@10:      if (vecsz->rnk == 0 && tensor_unitstridep(sz) 
Chris@10: 	 && tensor_real_rowmajorp(sz, p->sign, p->in_place)) 
Chris@10: 	  goto api_simple;
Chris@10:      
Chris@10:      if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@10: 	  goto api_many;
Chris@10: 
Chris@10:      goto api_guru;
Chris@10: 
Chris@10:  api_simple:
Chris@10:      switch (sz->rnk) {
Chris@10: 	 case 1:
Chris@10: 	      if (p->sign < 0) {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_r2c_1d\n");
Chris@10: 		   return FFTW(plan_dft_r2c_1d)(sz->dims[0].n, 
Chris@10: 						(bench_real *) p->in, 
Chris@10: 						(bench_complex *) p->out,
Chris@10: 						flags);
Chris@10: 	      }
Chris@10: 	      else {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_c2r_1d\n");
Chris@10: 		   return FFTW(plan_dft_c2r_1d)(sz->dims[0].n, 
Chris@10: 						(bench_complex *) p->in, 
Chris@10: 						(bench_real *) p->out,
Chris@10: 						flags);
Chris@10: 	      }
Chris@10: 	      break;
Chris@10: 	 case 2:
Chris@10: 	      if (p->sign < 0) {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_r2c_2d\n");
Chris@10: 		   return FFTW(plan_dft_r2c_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@10: 						(bench_real *) p->in, 
Chris@10: 						(bench_complex *) p->out,
Chris@10: 						flags);
Chris@10: 	      }
Chris@10: 	      else {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_c2r_2d\n");
Chris@10: 		   return FFTW(plan_dft_c2r_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@10: 						(bench_complex *) p->in, 
Chris@10: 						(bench_real *) p->out,
Chris@10: 						flags);
Chris@10: 	      }
Chris@10: 	      break;
Chris@10: 	 case 3:
Chris@10: 	      if (p->sign < 0) {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_r2c_3d\n");
Chris@10: 		   return FFTW(plan_dft_r2c_3d)(
Chris@10: 			sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@10: 			(bench_real *) p->in, (bench_complex *) p->out,
Chris@10: 			flags);
Chris@10: 	      }
Chris@10: 	      else {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_c2r_3d\n");
Chris@10: 		   return FFTW(plan_dft_c2r_3d)(
Chris@10: 			sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@10: 			(bench_complex *) p->in, (bench_real *) p->out,
Chris@10: 			flags);
Chris@10: 	      }
Chris@10: 	      break;
Chris@10: 	 default: {
Chris@10: 	      int *n = mkn(sz);
Chris@10: 	      if (p->sign < 0) {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_r2c\n");
Chris@10: 		   pln = FFTW(plan_dft_r2c)(sz->rnk, n,
Chris@10: 					    (bench_real *) p->in, 
Chris@10: 					    (bench_complex *) p->out,
Chris@10: 					    flags);
Chris@10: 	      }
Chris@10: 	      else {
Chris@10: 		   if (verbose > 2) printf("using plan_dft_c2r\n");
Chris@10: 		   pln = FFTW(plan_dft_c2r)(sz->rnk, n,
Chris@10: 					    (bench_complex *) p->in, 
Chris@10: 					    (bench_real *) p->out,
Chris@10: 					    flags);
Chris@10: 	      }
Chris@10: 	      bench_free(n);
Chris@10: 	      return pln;
Chris@10: 	 }
Chris@10:      }
Chris@10: 
Chris@10:  api_many:
Chris@10:      {
Chris@10: 	  int *n, *inembed, *onembed;
Chris@10: 	  BENCH_ASSERT(vecsz->rnk == 1);
Chris@10: 	  n = mkn(sz);
Chris@10: 	  mknembed_many(sz, &inembed, &onembed);
Chris@10: 	  if (p->sign < 0) {
Chris@10: 	       if (verbose > 2) printf("using plan_many_dft_r2c\n");
Chris@10: 	       pln = FFTW(plan_many_dft_r2c)(
Chris@10: 		    sz->rnk, n, vecsz->dims[0].n, 
Chris@10: 		    (bench_real *) p->in, inembed,
Chris@10: 		    sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@10: 		    (bench_complex *) p->out, onembed,
Chris@10: 		    sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@10: 		    flags);
Chris@10: 	  }
Chris@10: 	  else {
Chris@10: 	       if (verbose > 2) printf("using plan_many_dft_c2r\n");
Chris@10: 	       pln = FFTW(plan_many_dft_c2r)(
Chris@10: 		    sz->rnk, n, vecsz->dims[0].n, 
Chris@10: 		    (bench_complex *) p->in, inembed,
Chris@10: 		    sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@10: 		    (bench_real *) p->out, onembed,
Chris@10: 		    sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@10: 		    flags);
Chris@10: 	  }
Chris@10: 	  bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@10: 	  return pln;
Chris@10:      }
Chris@10: 
Chris@10:  api_guru:
Chris@10:      {
Chris@10: 	  FFTW(iodim) *dims, *howmany_dims;
Chris@10: 
Chris@10: 	  if (p->sign < 0) {
Chris@10: 	       dims = bench_tensor_to_fftw_iodim(sz);
Chris@10: 	       howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10: 	       if (verbose > 2) printf("using plan_guru_dft_r2c\n");
Chris@10: 	       pln = FFTW(plan_guru_dft_r2c)(sz->rnk, dims,
Chris@10: 					     vecsz->rnk, howmany_dims,
Chris@10: 					     (bench_real *) p->in,
Chris@10: 					     (bench_complex *) p->out,
Chris@10: 					     flags);
Chris@10: 	  }
Chris@10: 	  else {
Chris@10: 	       dims = bench_tensor_to_fftw_iodim(sz);
Chris@10: 	       howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10: 	       if (verbose > 2) printf("using plan_guru_dft_c2r\n");
Chris@10: 	       pln = FFTW(plan_guru_dft_c2r)(sz->rnk, dims,
Chris@10: 					     vecsz->rnk, howmany_dims,
Chris@10: 					     (bench_complex *) p->in,
Chris@10: 					     (bench_real *) p->out,
Chris@10: 					     flags);
Chris@10: 	  }
Chris@10: 	  bench_free(dims);
Chris@10: 	  bench_free(howmany_dims);
Chris@10: 	  return pln;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_real(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      if (p->split)
Chris@10: 	  return mkplan_real_split(p, flags);
Chris@10:      else
Chris@10: 	  return mkplan_real_interleaved(p, flags);
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_complex_split(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      FFTW(plan) pln;
Chris@10:      bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@10:      FFTW(iodim) *dims, *howmany_dims;
Chris@10:      bench_real *ri, *ii, *ro, *io;
Chris@10: 
Chris@10:      extract_reim_split(p->sign, p->iphyssz, (bench_real *) p->in, &ri, &ii);
Chris@10:      extract_reim_split(p->sign, p->ophyssz, (bench_real *) p->out, &ro, &io);
Chris@10: 
Chris@10:      dims = bench_tensor_to_fftw_iodim(sz);
Chris@10:      howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10:      if (verbose > 2) printf("using plan_guru_split_dft\n");
Chris@10:      pln = FFTW(plan_guru_split_dft)(sz->rnk, dims,
Chris@10: 			       vecsz->rnk, howmany_dims,
Chris@10: 			       ri, ii, ro, io, flags);
Chris@10:      bench_free(dims);
Chris@10:      bench_free(howmany_dims);
Chris@10:      return pln;
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_complex_interleaved(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      FFTW(plan) pln;
Chris@10:      bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@10: 
Chris@10:      if (vecsz->rnk == 0 && tensor_unitstridep(sz) && tensor_rowmajorp(sz)) 
Chris@10: 	  goto api_simple;
Chris@10:      
Chris@10:      if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@10: 	  goto api_many;
Chris@10: 
Chris@10:      goto api_guru;
Chris@10: 
Chris@10:  api_simple:
Chris@10:      switch (sz->rnk) {
Chris@10: 	 case 1:
Chris@10: 	      if (verbose > 2) printf("using plan_dft_1d\n");
Chris@10: 	      return FFTW(plan_dft_1d)(sz->dims[0].n, 
Chris@10: 				       (bench_complex *) p->in,
Chris@10: 				       (bench_complex *) p->out, 
Chris@10: 				       p->sign, flags);
Chris@10: 	      break;
Chris@10: 	 case 2:
Chris@10: 	      if (verbose > 2) printf("using plan_dft_2d\n");
Chris@10: 	      return FFTW(plan_dft_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@10: 				       (bench_complex *) p->in,
Chris@10: 				       (bench_complex *) p->out, 
Chris@10: 				       p->sign, flags);
Chris@10: 	      break;
Chris@10: 	 case 3:
Chris@10: 	      if (verbose > 2) printf("using plan_dft_3d\n");
Chris@10: 	      return FFTW(plan_dft_3d)(
Chris@10: 		   sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@10: 		   (bench_complex *) p->in, (bench_complex *) p->out, 
Chris@10: 		   p->sign, flags);
Chris@10: 	      break;
Chris@10: 	 default: {
Chris@10: 	      int *n = mkn(sz);
Chris@10: 	      if (verbose > 2) printf("using plan_dft\n");
Chris@10: 	      pln = FFTW(plan_dft)(sz->rnk, n, 
Chris@10: 				   (bench_complex *) p->in, 
Chris@10: 				   (bench_complex *) p->out, p->sign, flags);
Chris@10: 	      bench_free(n);
Chris@10: 	      return pln;
Chris@10: 	 }
Chris@10:      }
Chris@10: 
Chris@10:  api_many:
Chris@10:      {
Chris@10: 	  int *n, *inembed, *onembed;
Chris@10: 	  BENCH_ASSERT(vecsz->rnk == 1);
Chris@10: 	  n = mkn(sz);
Chris@10: 	  mknembed_many(sz, &inembed, &onembed);
Chris@10: 	  if (verbose > 2) printf("using plan_many_dft\n");
Chris@10: 	  pln = FFTW(plan_many_dft)(
Chris@10: 	       sz->rnk, n, vecsz->dims[0].n, 
Chris@10: 	       (bench_complex *) p->in, 
Chris@10: 	       inembed, sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@10: 	       (bench_complex *) p->out,
Chris@10: 	       onembed, sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@10: 	       p->sign, flags);
Chris@10: 	  bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@10: 	  return pln;
Chris@10:      }
Chris@10: 
Chris@10:  api_guru:
Chris@10:      {
Chris@10: 	  FFTW(iodim) *dims, *howmany_dims;
Chris@10: 
Chris@10: 	  dims = bench_tensor_to_fftw_iodim(sz);
Chris@10: 	  howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10: 	  if (verbose > 2) printf("using plan_guru_dft\n");
Chris@10: 	  pln = FFTW(plan_guru_dft)(sz->rnk, dims,
Chris@10: 				    vecsz->rnk, howmany_dims,
Chris@10: 				    (bench_complex *) p->in,
Chris@10: 				    (bench_complex *) p->out,
Chris@10: 				    p->sign, flags);
Chris@10: 	  bench_free(dims);
Chris@10: 	  bench_free(howmany_dims);
Chris@10: 	  return pln;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_complex(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      if (p->split)
Chris@10: 	  return mkplan_complex_split(p, flags);
Chris@10:      else
Chris@10: 	  return mkplan_complex_interleaved(p, flags);
Chris@10: }
Chris@10: 
Chris@10: static FFTW(plan) mkplan_r2r(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      FFTW(plan) pln;
Chris@10:      bench_tensor *sz = p->sz, *vecsz = p->vecsz;
Chris@10:      FFTW(r2r_kind) *k;
Chris@10: 
Chris@10:      k = (FFTW(r2r_kind) *) bench_malloc(sizeof(FFTW(r2r_kind)) * sz->rnk);
Chris@10:      {
Chris@10: 	  int i;
Chris@10: 	  for (i = 0; i < sz->rnk; ++i)
Chris@10: 	       switch (p->k[i]) {
Chris@10: 		   case R2R_R2HC: k[i] = FFTW_R2HC; break;
Chris@10: 		   case R2R_HC2R: k[i] = FFTW_HC2R; break;
Chris@10: 		   case R2R_DHT: k[i] = FFTW_DHT; break;
Chris@10: 		   case R2R_REDFT00: k[i] = FFTW_REDFT00; break;
Chris@10: 		   case R2R_REDFT01: k[i] = FFTW_REDFT01; break;
Chris@10: 		   case R2R_REDFT10: k[i] = FFTW_REDFT10; break;
Chris@10: 		   case R2R_REDFT11: k[i] = FFTW_REDFT11; break;
Chris@10: 		   case R2R_RODFT00: k[i] = FFTW_RODFT00; break;
Chris@10: 		   case R2R_RODFT01: k[i] = FFTW_RODFT01; break;
Chris@10: 		   case R2R_RODFT10: k[i] = FFTW_RODFT10; break;
Chris@10: 		   case R2R_RODFT11: k[i] = FFTW_RODFT11; break;
Chris@10: 		   default: BENCH_ASSERT(0);
Chris@10: 	       }
Chris@10:      }
Chris@10: 
Chris@10:      if (vecsz->rnk == 0 && tensor_unitstridep(sz) && tensor_rowmajorp(sz)) 
Chris@10: 	  goto api_simple;
Chris@10:      
Chris@10:      if (vecsz->rnk == 1 && expressible_as_api_many(sz))
Chris@10: 	  goto api_many;
Chris@10: 
Chris@10:      goto api_guru;
Chris@10: 
Chris@10:  api_simple:
Chris@10:      switch (sz->rnk) {
Chris@10: 	 case 1:
Chris@10: 	      if (verbose > 2) printf("using plan_r2r_1d\n");
Chris@10: 	      pln = FFTW(plan_r2r_1d)(sz->dims[0].n, 
Chris@10: 				      (bench_real *) p->in,
Chris@10: 				      (bench_real *) p->out, 
Chris@10: 				      k[0], flags);
Chris@10: 	      goto done;
Chris@10: 	 case 2:
Chris@10: 	      if (verbose > 2) printf("using plan_r2r_2d\n");
Chris@10: 	      pln = FFTW(plan_r2r_2d)(sz->dims[0].n, sz->dims[1].n,
Chris@10: 				      (bench_real *) p->in,
Chris@10: 				      (bench_real *) p->out, 
Chris@10: 				      k[0], k[1], flags);
Chris@10: 	      goto done;
Chris@10: 	 case 3:
Chris@10: 	      if (verbose > 2) printf("using plan_r2r_3d\n");
Chris@10: 	      pln = FFTW(plan_r2r_3d)(
Chris@10: 		   sz->dims[0].n, sz->dims[1].n, sz->dims[2].n,
Chris@10: 		   (bench_real *) p->in, (bench_real *) p->out, 
Chris@10: 		   k[0], k[1], k[2], flags);
Chris@10: 	      goto done;
Chris@10: 	 default: {
Chris@10: 	      int *n = mkn(sz);
Chris@10: 	      if (verbose > 2) printf("using plan_r2r\n");
Chris@10: 	      pln = FFTW(plan_r2r)(sz->rnk, n,
Chris@10: 				   (bench_real *) p->in, (bench_real *) p->out,
Chris@10: 				   k, flags);
Chris@10: 	      bench_free(n);
Chris@10: 	      goto done;
Chris@10: 	 }
Chris@10:      }
Chris@10: 
Chris@10:  api_many:
Chris@10:      {
Chris@10: 	  int *n, *inembed, *onembed;
Chris@10: 	  BENCH_ASSERT(vecsz->rnk == 1);
Chris@10: 	  n = mkn(sz);
Chris@10: 	  mknembed_many(sz, &inembed, &onembed);
Chris@10: 	  if (verbose > 2) printf("using plan_many_r2r\n");
Chris@10: 	  pln = FFTW(plan_many_r2r)(
Chris@10: 	       sz->rnk, n, vecsz->dims[0].n, 
Chris@10: 	       (bench_real *) p->in,
Chris@10: 	       inembed, sz->dims[sz->rnk - 1].is, vecsz->dims[0].is,
Chris@10: 	       (bench_real *) p->out,
Chris@10: 	       onembed, sz->dims[sz->rnk - 1].os, vecsz->dims[0].os,
Chris@10: 	       k, flags);
Chris@10: 	  bench_free(n); bench_free(inembed); bench_free(onembed);
Chris@10: 	  goto done;
Chris@10:      }
Chris@10: 
Chris@10:  api_guru:
Chris@10:      {
Chris@10: 	  FFTW(iodim) *dims, *howmany_dims;
Chris@10: 
Chris@10: 	  dims = bench_tensor_to_fftw_iodim(sz);
Chris@10: 	  howmany_dims = bench_tensor_to_fftw_iodim(vecsz);
Chris@10: 	  if (verbose > 2) printf("using plan_guru_r2r\n");
Chris@10: 	  pln = FFTW(plan_guru_r2r)(sz->rnk, dims,
Chris@10: 				    vecsz->rnk, howmany_dims,
Chris@10: 				    (bench_real *) p->in, 
Chris@10: 				    (bench_real *) p->out, k, flags);
Chris@10: 	  bench_free(dims);
Chris@10: 	  bench_free(howmany_dims);
Chris@10: 	  goto done;
Chris@10:      }
Chris@10:      
Chris@10:  done:
Chris@10:      bench_free(k);
Chris@10:      return pln;
Chris@10: }
Chris@10: 
Chris@10: FFTW(plan) mkplan(bench_problem *p, unsigned flags)
Chris@10: {
Chris@10:      switch (p->kind) {
Chris@10: 	 case PROBLEM_COMPLEX:	  return mkplan_complex(p, flags);
Chris@10: 	 case PROBLEM_REAL:	  return mkplan_real(p, flags);
Chris@10: 	 case PROBLEM_R2R:        return mkplan_r2r(p, flags);
Chris@10: 	 default: BENCH_ASSERT(0); return 0;
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: void main_init(int *argc, char ***argv)
Chris@10: {
Chris@10:      UNUSED(argc);
Chris@10:      UNUSED(argv);
Chris@10: }
Chris@10: 
Chris@10: void initial_cleanup(void)
Chris@10: {
Chris@10: }
Chris@10: 
Chris@10: void final_cleanup(void)
Chris@10: {
Chris@10: }
Chris@10: 
Chris@10: int import_wisdom(FILE *f)
Chris@10: {
Chris@10:      return FFTW(import_wisdom_from_file)(f);
Chris@10: }
Chris@10: 
Chris@10: void export_wisdom(FILE *f)
Chris@10: {
Chris@10:      FFTW(export_wisdom_to_file)(f);
Chris@10: }