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: #include "verify.h"
Chris@10: 
Chris@10: /* copy A into B, using output stride of A and input stride of B */
Chris@10: typedef struct {
Chris@10:      dotens2_closure k;
Chris@10:      R *ra; R *ia;
Chris@10:      R *rb; R *ib;
Chris@10:      int scalea, scaleb;
Chris@10: } cpy_closure;
Chris@10: 
Chris@10: static void cpy0(dotens2_closure *k_, 
Chris@10: 		 int indxa, int ondxa, int indxb, int ondxb)
Chris@10: {
Chris@10:      cpy_closure *k = (cpy_closure *)k_;
Chris@10:      k->rb[indxb * k->scaleb] = k->ra[ondxa * k->scalea];
Chris@10:      k->ib[indxb * k->scaleb] = k->ia[ondxa * k->scalea];
Chris@10:      UNUSED(indxa); UNUSED(ondxb);
Chris@10: }
Chris@10: 
Chris@10: static void cpy(R *ra, R *ia, const bench_tensor *sza, int scalea,
Chris@10: 		R *rb, R *ib, const bench_tensor *szb, int scaleb)
Chris@10: {
Chris@10:      cpy_closure k;
Chris@10:      k.k.apply = cpy0;
Chris@10:      k.ra = ra; k.ia = ia; k.rb = rb; k.ib = ib;
Chris@10:      k.scalea = scalea; k.scaleb = scaleb;
Chris@10:      bench_dotens2(sza, szb, &k.k);
Chris@10: }
Chris@10: 
Chris@10: typedef struct {
Chris@10:      dofft_closure k;
Chris@10:      bench_problem *p;
Chris@10: } dofft_dft_closure;
Chris@10: 
Chris@10: static void dft_apply(dofft_closure *k_, bench_complex *in, bench_complex *out)
Chris@10: {
Chris@10:      dofft_dft_closure *k = (dofft_dft_closure *)k_;
Chris@10:      bench_problem *p = k->p;
Chris@10:      bench_tensor *totalsz, *pckdsz;
Chris@10:      bench_tensor *totalsz_swap, *pckdsz_swap;
Chris@10:      bench_real *ri, *ii, *ro, *io;
Chris@10:      int totalscale;
Chris@10: 
Chris@10:      totalsz = tensor_append(p->vecsz, p->sz);
Chris@10:      pckdsz = verify_pack(totalsz, 2);
Chris@10:      ri = (bench_real *) p->in;
Chris@10:      ro = (bench_real *) p->out;
Chris@10: 
Chris@10:      totalsz_swap = tensor_copy_swapio(totalsz);
Chris@10:      pckdsz_swap = tensor_copy_swapio(pckdsz);
Chris@10: 
Chris@10:      /* confusion: the stride is the distance between complex elements
Chris@10: 	when using interleaved format, but it is the distance between
Chris@10: 	real elements when using split format */
Chris@10:      if (p->split) {
Chris@10: 	  ii = p->ini ? (bench_real *) p->ini : ri + p->iphyssz;
Chris@10: 	  io = p->outi ? (bench_real *) p->outi : ro + p->ophyssz;
Chris@10: 	  totalscale = 1;
Chris@10:      } else {
Chris@10: 	  ii = p->ini ? (bench_real *) p->ini : ri + 1;
Chris@10: 	  io = p->outi ? (bench_real *) p->outi : ro + 1;
Chris@10: 	  totalscale = 2;
Chris@10:      }
Chris@10: 
Chris@10:      cpy(&c_re(in[0]), &c_im(in[0]), pckdsz, 1,
Chris@10: 	    ri, ii, totalsz, totalscale);
Chris@10:      after_problem_ccopy_from(p, ri, ii);
Chris@10:      doit(1, p);
Chris@10:      after_problem_ccopy_to(p, ro, io);
Chris@10:      if (k->k.recopy_input)
Chris@10: 	  cpy(ri, ii, totalsz_swap, totalscale,
Chris@10: 	      &c_re(in[0]), &c_im(in[0]), pckdsz_swap, 1);
Chris@10:      cpy(ro, io, totalsz, totalscale,
Chris@10: 	 &c_re(out[0]), &c_im(out[0]), pckdsz, 1);
Chris@10: 
Chris@10:      tensor_destroy(totalsz);
Chris@10:      tensor_destroy(pckdsz);
Chris@10:      tensor_destroy(totalsz_swap);
Chris@10:      tensor_destroy(pckdsz_swap);
Chris@10: }
Chris@10: 
Chris@10: void verify_dft(bench_problem *p, int rounds, double tol, errors *e)
Chris@10: {
Chris@10:      C *inA, *inB, *inC, *outA, *outB, *outC, *tmp;
Chris@10:      int n, vecn, N;
Chris@10:      dofft_dft_closure k;
Chris@10: 
Chris@10:      BENCH_ASSERT(p->kind == PROBLEM_COMPLEX);
Chris@10: 
Chris@10:      k.k.apply = dft_apply;
Chris@10:      k.k.recopy_input = 0;
Chris@10:      k.p = p;
Chris@10: 
Chris@10:      if (rounds == 0)
Chris@10: 	  rounds = 20;  /* default value */
Chris@10: 
Chris@10:      n = tensor_sz(p->sz);
Chris@10:      vecn = tensor_sz(p->vecsz);
Chris@10:      N = n * vecn;
Chris@10: 
Chris@10:      inA = (C *) bench_malloc(N * sizeof(C));
Chris@10:      inB = (C *) bench_malloc(N * sizeof(C));
Chris@10:      inC = (C *) bench_malloc(N * sizeof(C));
Chris@10:      outA = (C *) bench_malloc(N * sizeof(C));
Chris@10:      outB = (C *) bench_malloc(N * sizeof(C));
Chris@10:      outC = (C *) bench_malloc(N * sizeof(C));
Chris@10:      tmp = (C *) bench_malloc(N * sizeof(C));
Chris@10: 
Chris@10:      e->i = impulse(&k.k, n, vecn, inA, inB, inC, outA, outB, outC, 
Chris@10: 		    tmp, rounds, tol);
Chris@10:      e->l = linear(&k.k, 0, N, inA, inB, inC, outA, outB, outC,
Chris@10: 		   tmp, rounds, tol);
Chris@10: 
Chris@10:      e->s = 0.0;
Chris@10:      e->s = dmax(e->s, tf_shift(&k.k, 0, p->sz, n, vecn, p->sign,
Chris@10: 				inA, inB, outA, outB, 
Chris@10: 				tmp, rounds, tol, TIME_SHIFT));
Chris@10:      e->s = dmax(e->s, tf_shift(&k.k, 0, p->sz, n, vecn, p->sign,
Chris@10: 				inA, inB, outA, outB, 
Chris@10: 				tmp, rounds, tol, FREQ_SHIFT));
Chris@10: 
Chris@10:      if (!p->in_place && !p->destroy_input)
Chris@10: 	  preserves_input(&k.k, 0, N, inA, inB, outB, rounds);
Chris@10: 
Chris@10:      bench_free(tmp);
Chris@10:      bench_free(outC);
Chris@10:      bench_free(outB);
Chris@10:      bench_free(outA);
Chris@10:      bench_free(inC);
Chris@10:      bench_free(inB);
Chris@10:      bench_free(inA);
Chris@10: }
Chris@10: 
Chris@10: 
Chris@10: void accuracy_dft(bench_problem *p, int rounds, int impulse_rounds,
Chris@10: 		  double t[6])
Chris@10: {
Chris@10:      dofft_dft_closure k;
Chris@10:      int n;
Chris@10:      C *a, *b;
Chris@10: 
Chris@10:      BENCH_ASSERT(p->kind == PROBLEM_COMPLEX);
Chris@10:      BENCH_ASSERT(p->sz->rnk == 1);
Chris@10:      BENCH_ASSERT(p->vecsz->rnk == 0);
Chris@10: 
Chris@10:      k.k.apply = dft_apply;
Chris@10:      k.k.recopy_input = 0;
Chris@10:      k.p = p;
Chris@10:      n = tensor_sz(p->sz);
Chris@10: 
Chris@10:      a = (C *) bench_malloc(n * sizeof(C));
Chris@10:      b = (C *) bench_malloc(n * sizeof(C));
Chris@10:      accuracy_test(&k.k, 0, p->sign, n, a, b, rounds, impulse_rounds, t);
Chris@10:      bench_free(b);
Chris@10:      bench_free(a);
Chris@10: }