static void dofft(info *nfo, R *in, R *out) { cpyr(in, nfo->pckdsz, (R *) nfo->p->in, nfo->totalsz); after_problem_rcopy_from(nfo->p, (bench_real *)nfo->p->in); doit(1, nfo->p); after_problem_rcopy_to(nfo->p, (bench_real *)nfo->p->out); cpyr((R *) nfo->p->out, nfo->totalsz, out, nfo->pckdsz); }
void verify_r2r(bench_problem *p, int rounds, double tol, errors *e) { R *inA, *inB, *inC, *outA, *outB, *outC, *tmp; info nfo; int n, vecn, N; double impulse_amp = 1.0; dim_stuff *d; int i; if (rounds == 0) rounds = 20; /* default value */ n = tensor_sz(p->sz); vecn = tensor_sz(p->vecsz); N = n * vecn; d = (dim_stuff *) bench_malloc(sizeof(dim_stuff) * p->sz->rnk); for (i = 0; i < p->sz->rnk; ++i) { int n0, i0, k0; trigfun ti, ts; d[i].n = n0 = p->sz->dims[i].n; if (p->k[i] > R2R_DHT) n0 = 2 * (n0 + (p->k[i] == R2R_REDFT00 ? -1 : (p->k[i] == R2R_RODFT00 ? 1 : 0))); switch (p->k[i]) { case R2R_R2HC: i0 = k0 = 0; ti = realhalf; ts = coshalf; break; case R2R_DHT: i0 = k0 = 0; ti = unity; ts = cos00; break; case R2R_HC2R: i0 = k0 = 0; ti = unity; ts = cos00; break; case R2R_REDFT00: i0 = k0 = 0; ti = ts = cos00; break; case R2R_REDFT01: i0 = k0 = 0; ti = ts = cos01; break; case R2R_REDFT10: i0 = k0 = 0; ti = cos10; impulse_amp *= 2.0; ts = cos00; break; case R2R_REDFT11: i0 = k0 = 0; ti = cos11; impulse_amp *= 2.0; ts = cos01; break; case R2R_RODFT00: i0 = k0 = 1; ti = sin00; impulse_amp *= 2.0; ts = cos00; break; case R2R_RODFT01: i0 = 1; k0 = 0; ti = sin01; impulse_amp *= n == 1 ? 1.0 : 2.0; ts = cos01; break; case R2R_RODFT10: i0 = 0; k0 = 1; ti = sin10; impulse_amp *= 2.0; ts = cos00; break; case R2R_RODFT11: i0 = k0 = 0; ti = sin11; impulse_amp *= 2.0; ts = cos01; break; default: BENCH_ASSERT(0); return; } d[i].n0 = n0; d[i].i0 = i0; d[i].k0 = k0; d[i].ti = ti; d[i].ts = ts; } inA = (R *) bench_malloc(N * sizeof(R)); inB = (R *) bench_malloc(N * sizeof(R)); inC = (R *) bench_malloc(N * sizeof(R)); outA = (R *) bench_malloc(N * sizeof(R)); outB = (R *) bench_malloc(N * sizeof(R)); outC = (R *) bench_malloc(N * sizeof(R)); tmp = (R *) bench_malloc(N * sizeof(R)); nfo.p = p; nfo.probsz = p->sz; nfo.totalsz = tensor_append(p->vecsz, nfo.probsz); nfo.pckdsz = verify_pack(nfo.totalsz, 1); nfo.pckdvecsz = verify_pack(p->vecsz, tensor_sz(nfo.probsz)); e->i = rimpulse(d, impulse_amp, n, vecn, &nfo, inA, inB, inC, outA, outB, outC, tmp, rounds, tol); e->l = rlinear(N, &nfo, inA, inB, inC, outA, outB, outC, tmp, rounds,tol); e->s = t_shift(n, vecn, &nfo, inA, inB, outA, outB, tmp, rounds, tol, d); /* grr, verify-lib.c:preserves_input() only works for complex */ if (!p->in_place && !p->destroy_input) { bench_tensor *totalsz_swap, *pckdsz_swap; totalsz_swap = tensor_copy_swapio(nfo.totalsz); pckdsz_swap = tensor_copy_swapio(nfo.pckdsz); for (i = 0; i < rounds; ++i) { rarand(inA, N); dofft(&nfo, inA, outB); cpyr((R *) nfo.p->in, totalsz_swap, inB, pckdsz_swap); racmp(inB, inA, N, "preserves_input", 0.0); } tensor_destroy(totalsz_swap); tensor_destroy(pckdsz_swap); } tensor_destroy(nfo.totalsz); tensor_destroy(nfo.pckdsz); tensor_destroy(nfo.pckdvecsz); bench_free(tmp); bench_free(outC); bench_free(outB); bench_free(outA); bench_free(inC); bench_free(inB); bench_free(inA); bench_free(d); }
static void rdft2_apply(dofft_closure *k_, bench_complex *in, bench_complex *out) { dofft_rdft2_closure *k = (dofft_rdft2_closure *)k_; bench_problem *p = k->p; bench_tensor *totalsz, *pckdsz, *totalsz_swap, *pckdsz_swap; bench_tensor *probsz2, *totalsz2, *pckdsz2; bench_tensor *probsz2_swap, *totalsz2_swap, *pckdsz2_swap; bench_real *ri, *ii, *ro, *io; int n2, totalscale; totalsz = tensor_append(p->vecsz, p->sz); pckdsz = verify_pack(totalsz, 2); n2 = tensor_sz(totalsz); if (FINITE_RNK(p->sz->rnk) && p->sz->rnk > 0) n2 = (n2 / p->sz->dims[p->sz->rnk - 1].n) * (p->sz->dims[p->sz->rnk - 1].n / 2 + 1); ri = (bench_real *) p->in; ro = (bench_real *) p->out; if (FINITE_RNK(p->sz->rnk) && p->sz->rnk > 0 && n2 > 0) { probsz2 = tensor_copy_sub(p->sz, p->sz->rnk - 1, 1); totalsz2 = tensor_copy_sub(totalsz, 0, totalsz->rnk - 1); pckdsz2 = tensor_copy_sub(pckdsz, 0, pckdsz->rnk - 1); } else { probsz2 = mktensor(0); totalsz2 = tensor_copy(totalsz); pckdsz2 = tensor_copy(pckdsz); } totalsz_swap = tensor_copy_swapio(totalsz); pckdsz_swap = tensor_copy_swapio(pckdsz); totalsz2_swap = tensor_copy_swapio(totalsz2); pckdsz2_swap = tensor_copy_swapio(pckdsz2); probsz2_swap = tensor_copy_swapio(probsz2); /* confusion: the stride is the distance between complex elements when using interleaved format, but it is the distance between real elements when using split format */ if (p->split) { ii = p->ini ? (bench_real *) p->ini : ri + n2; io = p->outi ? (bench_real *) p->outi : ro + n2; totalscale = 1; } else { ii = p->ini ? (bench_real *) p->ini : ri + 1; io = p->outi ? (bench_real *) p->outi : ro + 1; totalscale = 2; } if (p->sign < 0) { /* R2HC */ int N, vN, i; cpyr(&c_re(in[0]), pckdsz, ri, totalsz); after_problem_rcopy_from(p, ri); doit(1, p); after_problem_hccopy_to(p, ro, io); if (k->k.recopy_input) cpyr(ri, totalsz_swap, &c_re(in[0]), pckdsz_swap); cpyhc2(ro, io, probsz2, totalsz2, totalscale, &c_re(out[0]), &c_im(out[0]), pckdsz2); N = tensor_sz(p->sz); vN = tensor_sz(p->vecsz); for (i = 0; i < vN; ++i) mkhermitian(out + i*N, p->sz->rnk, p->sz->dims, 1); } else { /* HC2R */ icpyhc2(ri, ii, probsz2, totalsz2, totalscale, &c_re(in[0]), &c_im(in[0]), pckdsz2); after_problem_hccopy_from(p, ri, ii); doit(1, p); after_problem_rcopy_to(p, ro); if (k->k.recopy_input) cpyhc2(ri, ii, probsz2_swap, totalsz2_swap, totalscale, &c_re(in[0]), &c_im(in[0]), pckdsz2_swap); mkreal(out, tensor_sz(pckdsz)); cpyr(ro, totalsz, &c_re(out[0]), pckdsz); } tensor_destroy(totalsz); tensor_destroy(pckdsz); tensor_destroy(totalsz_swap); tensor_destroy(pckdsz_swap); tensor_destroy(probsz2); tensor_destroy(totalsz2); tensor_destroy(pckdsz2); tensor_destroy(probsz2_swap); tensor_destroy(totalsz2_swap); tensor_destroy(pckdsz2_swap); }
static void dofft(info *nfo, R *in, R *out) { cpyr(in, nfo->pckdsz, (R *) nfo->p->in, nfo->totalsz); doit(1, nfo->p); cpyr((R *) nfo->p->out, nfo->totalsz, out, nfo->pckdsz); }