/* * Copyright (c) 2003, 2007-14 Matteo Frigo * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * */ /* express a hc2hc problem in terms of rdft + multiplication by twiddle factors */ #include "hc2hc.h" typedef hc2hc_solver S; typedef struct { plan_hc2hc super; INT r, m, s, vl, vs, mstart1, mcount1; plan *cld0; plan *cld; twid *td; } P; /**************************************************************/ static void mktwiddle(P *ego, enum wakefulness wakefulness) { static const tw_instr tw[] = { { TW_HALF, 0, 0 }, { TW_NEXT, 1, 0 } }; /* note that R and M are swapped, to allow for sequential access both to data and twiddles */ X(twiddle_awake)(wakefulness, &ego->td, tw, ego->r * ego->m, ego->m, ego->r); } static void bytwiddle(const P *ego, R *IO, R sign) { INT i, j, k; INT r = ego->r, m = ego->m, s = ego->s, vl = ego->vl, vs = ego->vs; INT ms = m * s; INT mstart1 = ego->mstart1, mcount1 = ego->mcount1; INT wrem = 2 * ((m-1)/2 - mcount1); for (i = 0; i < vl; ++i, IO += vs) { const R *W = ego->td->W; A(m % 2 == 1); for (k = 1, W += (m - 1) + 2*(mstart1-1); k < r; ++k) { /* pr := IO + (j + mstart1) * s + k * ms */ R *pr = IO + mstart1 * s + k * ms; /* pi := IO + (m - j - mstart1) * s + k * ms */ R *pi = IO - mstart1 * s + (k + 1) * ms; for (j = 0; j < mcount1; ++j, pr += s, pi -= s) { E xr = *pr; E xi = *pi; E wr = W[0]; E wi = sign * W[1]; *pr = xr * wr - xi * wi; *pi = xi * wr + xr * wi; W += 2; } W += wrem; } } } static void swapri(R *IO, INT r, INT m, INT s, INT jstart, INT jend) { INT k; INT ms = m * s; INT js = jstart * s; for (k = 0; k + k < r; ++k) { /* pr := IO + (m - j) * s + k * ms */ R *pr = IO + (k + 1) * ms - js; /* pi := IO + (m - j) * s + (r - 1 - k) * ms */ R *pi = IO + (r - k) * ms - js; INT j; for (j = jstart; j < jend; j += 1, pr -= s, pi -= s) { R t = *pr; *pr = *pi; *pi = t; } } } static void reorder_dit(const P *ego, R *IO) { INT i, k; INT r = ego->r, m = ego->m, s = ego->s, vl = ego->vl, vs = ego->vs; INT ms = m * s; INT mstart1 = ego->mstart1, mend1 = mstart1 + ego->mcount1; for (i = 0; i < vl; ++i, IO += vs) { for (k = 1; k + k < r; ++k) { R *p0 = IO + k * ms; R *p1 = IO + (r - k) * ms; INT j; for (j = mstart1; j < mend1; ++j) { E rp, ip, im, rm; rp = p0[j * s]; im = p1[ms - j * s]; rm = p1[j * s]; ip = p0[ms - j * s]; p0[j * s] = rp - im; p1[ms - j * s] = rp + im; p1[j * s] = rm - ip; p0[ms - j * s] = ip + rm; } } swapri(IO, r, m, s, mstart1, mend1); } } static void reorder_dif(const P *ego, R *IO) { INT i, k; INT r = ego->r, m = ego->m, s = ego->s, vl = ego->vl, vs = ego->vs; INT ms = m * s; INT mstart1 = ego->mstart1, mend1 = mstart1 + ego->mcount1; for (i = 0; i < vl; ++i, IO += vs) { swapri(IO, r, m, s, mstart1, mend1); for (k = 1; k + k < r; ++k) { R *p0 = IO + k * ms; R *p1 = IO + (r - k) * ms; const R half = K(0.5); INT j; for (j = mstart1; j < mend1; ++j) { E rp, ip, im, rm; rp = half * p0[j * s]; im = half * p1[ms - j * s]; rm = half * p1[j * s]; ip = half * p0[ms - j * s]; p0[j * s] = rp + im; p1[ms - j * s] = im - rp; p1[j * s] = rm + ip; p0[ms - j * s] = ip - rm; } } } } static int applicable(rdft_kind kind, INT r, INT m, const planner *plnr) { return (1 && (kind == R2HC || kind == HC2R) && (m % 2) && (r % 2) && !NO_SLOWP(plnr) ); } /**************************************************************/ static void apply_dit(const plan *ego_, R *IO) { const P *ego = (const P *) ego_; INT start; plan_rdft *cld, *cld0; bytwiddle(ego, IO, K(-1.0)); cld0 = (plan_rdft *) ego->cld0; cld0->apply(ego->cld0, IO, IO); start = ego->mstart1 * ego->s; cld = (plan_rdft *) ego->cld; cld->apply(ego->cld, IO + start, IO + start); reorder_dit(ego, IO); } static void apply_dif(const plan *ego_, R *IO) { const P *ego = (const P *) ego_; INT start; plan_rdft *cld, *cld0; reorder_dif(ego, IO); cld0 = (plan_rdft *) ego->cld0; cld0->apply(ego->cld0, IO, IO); start = ego->mstart1 * ego->s; cld = (plan_rdft *) ego->cld; cld->apply(ego->cld, IO + start, IO + start); bytwiddle(ego, IO, K(1.0)); } static void awake(plan *ego_, enum wakefulness wakefulness) { P *ego = (P *) ego_; X(plan_awake)(ego->cld0, wakefulness); X(plan_awake)(ego->cld, wakefulness); mktwiddle(ego, wakefulness); } static void destroy(plan *ego_) { P *ego = (P *) ego_; X(plan_destroy_internal)(ego->cld); X(plan_destroy_internal)(ego->cld0); } static void print(const plan *ego_, printer *p) { const P *ego = (const P *) ego_; p->print(p, "(hc2hc-generic-%s-%D-%D%v%(%p%)%(%p%))", ego->super.apply == apply_dit ? "dit" : "dif", ego->r, ego->m, ego->vl, ego->cld0, ego->cld); } static plan *mkcldw(const hc2hc_solver *ego_, rdft_kind kind, INT r, INT m, INT s, INT vl, INT vs, INT mstart, INT mcount, R *IO, planner *plnr) { P *pln; plan *cld0 = 0, *cld = 0; INT mstart1, mcount1, mstride; static const plan_adt padt = { 0, awake, print, destroy }; UNUSED(ego_); A(mstart >= 0 && mcount > 0 && mstart + mcount <= (m+2)/2); if (!applicable(kind, r, m, plnr)) return (plan *)0; A(m % 2); mstart1 = mstart + (mstart == 0); mcount1 = mcount - (mstart == 0); mstride = m - (mstart + mcount - 1) - mstart1; /* 0th (DC) transform (vl of these), if mstart == 0 */ cld0 = X(mkplan_d)(plnr, X(mkproblem_rdft_1_d)( mstart == 0 ? X(mktensor_1d)(r, m * s, m * s) : X(mktensor_0d)(), X(mktensor_1d)(vl, vs, vs), IO, IO, kind) ); if (!cld0) goto nada; /* twiddle transforms: there are 2 x mcount1 x vl of these (where 2 corresponds to the real and imaginary parts) ... the 2 x mcount1 loops are combined if mstart=0 and mcount=(m+2)/2. */ cld = X(mkplan_d)(plnr, X(mkproblem_rdft_1_d)( X(mktensor_1d)(r, m * s, m * s), X(mktensor_3d)(2, mstride * s, mstride * s, mcount1, s, s, vl, vs, vs), IO + s * mstart1, IO + s * mstart1, kind) ); if (!cld) goto nada; pln = MKPLAN_HC2HC(P, &padt, (kind == R2HC) ? apply_dit : apply_dif); pln->cld = cld; pln->cld0 = cld0; pln->r = r; pln->m = m; pln->s = s; pln->vl = vl; pln->vs = vs; pln->td = 0; pln->mstart1 = mstart1; pln->mcount1 = mcount1; { double n0 = 0.5 * (r - 1) * (2 * mcount1) * vl; pln->super.super.ops = cld->ops; pln->super.super.ops.mul += (kind == R2HC ? 5.0 : 7.0) * n0; pln->super.super.ops.add += 4.0 * n0; pln->super.super.ops.other += 11.0 * n0; } return &(pln->super.super); nada: X(plan_destroy_internal)(cld); X(plan_destroy_internal)(cld0); return (plan *) 0; } static void regsolver(planner *plnr, INT r) { S *slv = (S *)X(mksolver_hc2hc)(sizeof(S), r, mkcldw); REGISTER_SOLVER(plnr, &(slv->super)); if (X(mksolver_hc2hc_hook)) { slv = (S *)X(mksolver_hc2hc_hook)(sizeof(S), r, mkcldw); REGISTER_SOLVER(plnr, &(slv->super)); } } void X(hc2hc_generic_register)(planner *p) { regsolver(p, 0); }