/* * 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 * */ /* Complex DFTs of rank == 1 when the vector length vn is >= # processes. In this case, we don't need to use a six-step type algorithm, and can instead transpose the DFT dimension with the vector dimension to make the DFT local. */ #include "mpi-dft.h" #include "mpi-transpose.h" #include "dft/dft.h" typedef struct { solver super; int preserve_input; /* preserve input even if DESTROY_INPUT was passed */ rearrangement rearrange; } S; typedef struct { plan_mpi_dft super; plan *cldt_before, *cld, *cldt_after; INT roff, ioff; int preserve_input; rearrangement rearrange; } P; static void apply(const plan *ego_, R *I, R *O) { const P *ego = (const P *) ego_; plan_dft *cld; plan_rdft *cldt_before, *cldt_after; INT roff = ego->roff, ioff = ego->ioff; /* global transpose */ cldt_before = (plan_rdft *) ego->cldt_before; cldt_before->apply(ego->cldt_before, I, O); if (ego->preserve_input) I = O; /* 1d DFT(s) */ cld = (plan_dft *) ego->cld; cld->apply(ego->cld, O+roff, O+ioff, I+roff, I+ioff); /* global transpose */ cldt_after = (plan_rdft *) ego->cldt_after; cldt_after->apply(ego->cldt_after, I, O); } static int applicable(const S *ego, const problem *p_, const planner *plnr) { const problem_mpi_dft *p = (const problem_mpi_dft *) p_; int n_pes; MPI_Comm_size(p->comm, &n_pes); return (1 && p->sz->rnk == 1 && !(p->flags & ~RANK1_BIGVEC_ONLY) && (!ego->preserve_input || (!NO_DESTROY_INPUTP(plnr) && p->I != p->O)) && (p->vn >= n_pes /* TODO: relax this, using more memory? */ || (p->flags & RANK1_BIGVEC_ONLY)) && XM(rearrange_applicable)(ego->rearrange, p->sz->dims[0], p->vn, n_pes) && (!NO_SLOWP(plnr) /* slow if dft-serial is applicable */ || !XM(dft_serial_applicable)(p)) ); } static void awake(plan *ego_, enum wakefulness wakefulness) { P *ego = (P *) ego_; X(plan_awake)(ego->cldt_before, wakefulness); X(plan_awake)(ego->cld, wakefulness); X(plan_awake)(ego->cldt_after, wakefulness); } static void destroy(plan *ego_) { P *ego = (P *) ego_; X(plan_destroy_internal)(ego->cldt_after); X(plan_destroy_internal)(ego->cld); X(plan_destroy_internal)(ego->cldt_before); } static void print(const plan *ego_, printer *p) { const P *ego = (const P *) ego_; const char descrip[][16] = { "contig", "discontig", "square-after", "square-middle", "square-before" }; p->print(p, "(mpi-dft-rank1-bigvec/%s%s %(%p%) %(%p%) %(%p%))", descrip[ego->rearrange], ego->preserve_input==2 ?"/p":"", ego->cldt_before, ego->cld, ego->cldt_after); } static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr) { const S *ego = (const S *) ego_; const problem_mpi_dft *p; P *pln; plan *cld = 0, *cldt_before = 0, *cldt_after = 0; R *ri, *ii, *ro, *io, *I, *O; INT yblock, yb, nx, ny, vn; int my_pe, n_pes; static const plan_adt padt = { XM(dft_solve), awake, print, destroy }; UNUSED(ego); if (!applicable(ego, p_, plnr)) return (plan *) 0; p = (const problem_mpi_dft *) p_; MPI_Comm_rank(p->comm, &my_pe); MPI_Comm_size(p->comm, &n_pes); nx = p->sz->dims[0].n; if (!(ny = XM(rearrange_ny)(ego->rearrange, p->sz->dims[0],p->vn,n_pes))) return (plan *) 0; vn = p->vn / ny; A(ny * vn == p->vn); yblock = XM(default_block)(ny, n_pes); cldt_before = X(mkplan_d)(plnr, XM(mkproblem_transpose)( nx, ny, vn*2, I = p->I, O = p->O, p->sz->dims[0].b[IB], yblock, p->comm, 0)); if (XM(any_true)(!cldt_before, p->comm)) goto nada; if (ego->preserve_input || NO_DESTROY_INPUTP(plnr)) { I = O; } X(extract_reim)(p->sign, I, &ri, &ii); X(extract_reim)(p->sign, O, &ro, &io); yb = XM(block)(ny, yblock, my_pe); cld = X(mkplan_d)(plnr, X(mkproblem_dft_d)(X(mktensor_1d)(nx, vn*2, vn*2), X(mktensor_2d)(yb, vn*2*nx, vn*2*nx, vn, 2, 2), ro, io, ri, ii)); if (XM(any_true)(!cld, p->comm)) goto nada; cldt_after = X(mkplan_d)(plnr, XM(mkproblem_transpose)( ny, nx, vn*2, I, O, yblock, p->sz->dims[0].b[OB], p->comm, 0)); if (XM(any_true)(!cldt_after, p->comm)) goto nada; pln = MKPLAN_MPI_DFT(P, &padt, apply); pln->cldt_before = cldt_before; pln->cld = cld; pln->cldt_after = cldt_after; pln->preserve_input = ego->preserve_input ? 2 : NO_DESTROY_INPUTP(plnr); pln->roff = ro - p->O; pln->ioff = io - p->O; pln->rearrange = ego->rearrange; X(ops_add)(&cldt_before->ops, &cld->ops, &pln->super.super.ops); X(ops_add2)(&cldt_after->ops, &pln->super.super.ops); return &(pln->super.super); nada: X(plan_destroy_internal)(cldt_after); X(plan_destroy_internal)(cld); X(plan_destroy_internal)(cldt_before); return (plan *) 0; } static solver *mksolver(rearrangement rearrange, int preserve_input) { static const solver_adt sadt = { PROBLEM_MPI_DFT, mkplan, 0 }; S *slv = MKSOLVER(S, &sadt); slv->rearrange = rearrange; slv->preserve_input = preserve_input; return &(slv->super); } void XM(dft_rank1_bigvec_register)(planner *p) { rearrangement rearrange; int preserve_input; FORALL_REARRANGE(rearrange) for (preserve_input = 0; preserve_input <= 1; ++preserve_input) REGISTER_SOLVER(p, mksolver(rearrange, preserve_input)); }