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: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:35:42 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 9 -name n1_9 -include n.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 80 FP additions, 56 FP multiplications, Chris@10: * (or, 24 additions, 0 multiplications, 56 fused multiply/add), Chris@10: * 59 stack variables, 10 constants, and 36 memory accesses Chris@10: */ Chris@10: #include "n.h" Chris@10: Chris@10: static void n1_9(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP954188894, +0.954188894138671133499268364187245676532219158); Chris@10: DK(KP363970234, +0.363970234266202361351047882776834043890471784); Chris@10: DK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@10: DK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@10: DK(KP492403876, +0.492403876506104029683371512294761506835321626); Chris@10: DK(KP777861913, +0.777861913430206160028177977318626690410586096); Chris@10: DK(KP839099631, +0.839099631177280011763127298123181364687434283); Chris@10: DK(KP176326980, +0.176326980708464973471090386868618986121633062); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(36, is), MAKE_VOLATILE_STRIDE(36, os)) { Chris@10: E T17, TV, T14, TY, T11, T15; Chris@10: { Chris@10: E Tm, TM, TL, T5, Tl, T1f, Tb, Tt, Ta, T1c, TI, TX, TF, TW, Tc; Chris@10: E Td, Tp, Tq; Chris@10: { Chris@10: E T1, Th, Ti, Tj, T4, T2, T3; Chris@10: T1 = ri[0]; Chris@10: T2 = ri[WS(is, 3)]; Chris@10: T3 = ri[WS(is, 6)]; Chris@10: Th = ii[0]; Chris@10: Ti = ii[WS(is, 3)]; Chris@10: Tj = ii[WS(is, 6)]; Chris@10: T4 = T2 + T3; Chris@10: Tm = T3 - T2; Chris@10: { Chris@10: E T6, Tz, T7, T8, TA, TB, Tk; Chris@10: T6 = ri[WS(is, 1)]; Chris@10: TM = Ti - Tj; Chris@10: Tk = Ti + Tj; Chris@10: TL = FNMS(KP500000000, T4, T1); Chris@10: T5 = T1 + T4; Chris@10: Tz = ii[WS(is, 1)]; Chris@10: Tl = FNMS(KP500000000, Tk, Th); Chris@10: T1f = Th + Tk; Chris@10: T7 = ri[WS(is, 4)]; Chris@10: T8 = ri[WS(is, 7)]; Chris@10: TA = ii[WS(is, 4)]; Chris@10: TB = ii[WS(is, 7)]; Chris@10: { Chris@10: E TE, T9, TH, TC, TG, TD; Chris@10: Tb = ri[WS(is, 2)]; Chris@10: TE = T7 - T8; Chris@10: T9 = T7 + T8; Chris@10: TH = TB - TA; Chris@10: TC = TA + TB; Chris@10: Tt = ii[WS(is, 2)]; Chris@10: Ta = T6 + T9; Chris@10: TG = FNMS(KP500000000, T9, T6); Chris@10: T1c = Tz + TC; Chris@10: TD = FNMS(KP500000000, TC, Tz); Chris@10: TI = FNMS(KP866025403, TH, TG); Chris@10: TX = FMA(KP866025403, TH, TG); Chris@10: TF = FNMS(KP866025403, TE, TD); Chris@10: TW = FMA(KP866025403, TE, TD); Chris@10: Tc = ri[WS(is, 5)]; Chris@10: Td = ri[WS(is, 8)]; Chris@10: Tp = ii[WS(is, 5)]; Chris@10: Tq = ii[WS(is, 8)]; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tn, TN, TZ, T10, TO, Ty, TJ, TP; Chris@10: { Chris@10: E Tw, Te, Tu, Tr; Chris@10: T17 = FNMS(KP866025403, Tm, Tl); Chris@10: Tn = FMA(KP866025403, Tm, Tl); Chris@10: Tw = Td - Tc; Chris@10: Te = Tc + Td; Chris@10: Tu = Tp + Tq; Chris@10: Tr = Tp - Tq; Chris@10: TN = FMA(KP866025403, TM, TL); Chris@10: TV = FNMS(KP866025403, TM, TL); Chris@10: { Chris@10: E Tf, To, T1d, Tv; Chris@10: Tf = Tb + Te; Chris@10: To = FNMS(KP500000000, Te, Tb); Chris@10: T1d = Tt + Tu; Chris@10: Tv = FNMS(KP500000000, Tu, Tt); Chris@10: { Chris@10: E Ts, Tg, T1i, Tx; Chris@10: Ts = FMA(KP866025403, Tr, To); Chris@10: TZ = FNMS(KP866025403, Tr, To); Chris@10: Tg = Ta + Tf; Chris@10: T1i = Tf - Ta; Chris@10: Tx = FMA(KP866025403, Tw, Tv); Chris@10: T10 = FNMS(KP866025403, Tw, Tv); Chris@10: { Chris@10: E T1e, T1g, T1b, T1h; Chris@10: T1e = T1c - T1d; Chris@10: T1g = T1c + T1d; Chris@10: ro[0] = T5 + Tg; Chris@10: T1b = FNMS(KP500000000, Tg, T5); Chris@10: io[0] = T1f + T1g; Chris@10: T1h = FNMS(KP500000000, T1g, T1f); Chris@10: TO = FMA(KP176326980, Ts, Tx); Chris@10: Ty = FNMS(KP176326980, Tx, Ts); Chris@10: ro[WS(os, 6)] = FNMS(KP866025403, T1e, T1b); Chris@10: ro[WS(os, 3)] = FMA(KP866025403, T1e, T1b); Chris@10: io[WS(os, 6)] = FNMS(KP866025403, T1i, T1h); Chris@10: io[WS(os, 3)] = FMA(KP866025403, T1i, T1h); Chris@10: TJ = FNMS(KP839099631, TI, TF); Chris@10: TP = FMA(KP839099631, TF, TI); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E TS, TK, TU, TQ, TT, TR; Chris@10: TS = FMA(KP777861913, TJ, Ty); Chris@10: TK = FNMS(KP777861913, TJ, Ty); Chris@10: TU = FNMS(KP777861913, TP, TO); Chris@10: TQ = FMA(KP777861913, TP, TO); Chris@10: TT = FMA(KP492403876, TK, Tn); Chris@10: io[WS(os, 1)] = FNMS(KP984807753, TK, Tn); Chris@10: TR = FNMS(KP492403876, TQ, TN); Chris@10: ro[WS(os, 1)] = FMA(KP984807753, TQ, TN); Chris@10: io[WS(os, 4)] = FMA(KP852868531, TU, TT); Chris@10: io[WS(os, 7)] = FNMS(KP852868531, TU, TT); Chris@10: ro[WS(os, 7)] = FNMS(KP852868531, TS, TR); Chris@10: ro[WS(os, 4)] = FMA(KP852868531, TS, TR); Chris@10: T14 = FNMS(KP176326980, TW, TX); Chris@10: TY = FMA(KP176326980, TX, TW); Chris@10: T11 = FNMS(KP363970234, T10, TZ); Chris@10: T15 = FMA(KP363970234, TZ, T10); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T12, T1a, T16, T18, T13, T19; Chris@10: T12 = FNMS(KP954188894, T11, TY); Chris@10: T1a = FMA(KP954188894, T11, TY); Chris@10: T16 = FNMS(KP954188894, T15, T14); Chris@10: T18 = FMA(KP954188894, T15, T14); Chris@10: T13 = FNMS(KP492403876, T12, TV); Chris@10: ro[WS(os, 2)] = FMA(KP984807753, T12, TV); Chris@10: T19 = FMA(KP492403876, T18, T17); Chris@10: io[WS(os, 2)] = FNMS(KP984807753, T18, T17); Chris@10: ro[WS(os, 8)] = FMA(KP852868531, T16, T13); Chris@10: ro[WS(os, 5)] = FNMS(KP852868531, T16, T13); Chris@10: io[WS(os, 8)] = FMA(KP852868531, T1a, T19); Chris@10: io[WS(os, 5)] = FNMS(KP852868531, T1a, T19); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 9, "n1_9", {24, 0, 56, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_n1_9) (planner *p) { Chris@10: X(kdft_register) (p, n1_9, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 9 -name n1_9 -include n.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 80 FP additions, 40 FP multiplications, Chris@10: * (or, 60 additions, 20 multiplications, 20 fused multiply/add), Chris@10: * 39 stack variables, 8 constants, and 36 memory accesses Chris@10: */ Chris@10: #include "n.h" Chris@10: Chris@10: static void n1_9(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@10: DK(KP342020143, +0.342020143325668733044099614682259580763083368); Chris@10: DK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@10: DK(KP173648177, +0.173648177666930348851716626769314796000375677); Chris@10: DK(KP642787609, +0.642787609686539326322643409907263432907559884); Chris@10: DK(KP766044443, +0.766044443118978035202392650555416673935832457); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(36, is), MAKE_VOLATILE_STRIDE(36, os)) { Chris@10: E T5, TO, Th, Tk, T1g, TR, Ta, T1c, Tq, TW, Tv, TX, Tf, T1d, TB; Chris@10: E T10, TG, TZ; Chris@10: { Chris@10: E T1, T2, T3, T4; Chris@10: T1 = ri[0]; Chris@10: T2 = ri[WS(is, 3)]; Chris@10: T3 = ri[WS(is, 6)]; Chris@10: T4 = T2 + T3; Chris@10: T5 = T1 + T4; Chris@10: TO = KP866025403 * (T3 - T2); Chris@10: Th = FNMS(KP500000000, T4, T1); Chris@10: } Chris@10: { Chris@10: E TP, Ti, Tj, TQ; Chris@10: TP = ii[0]; Chris@10: Ti = ii[WS(is, 3)]; Chris@10: Tj = ii[WS(is, 6)]; Chris@10: TQ = Ti + Tj; Chris@10: Tk = KP866025403 * (Ti - Tj); Chris@10: T1g = TP + TQ; Chris@10: TR = FNMS(KP500000000, TQ, TP); Chris@10: } Chris@10: { Chris@10: E T6, Ts, T9, Tr, Tp, Tt, Tm, Tu; Chris@10: T6 = ri[WS(is, 1)]; Chris@10: Ts = ii[WS(is, 1)]; Chris@10: { Chris@10: E T7, T8, Tn, To; Chris@10: T7 = ri[WS(is, 4)]; Chris@10: T8 = ri[WS(is, 7)]; Chris@10: T9 = T7 + T8; Chris@10: Tr = KP866025403 * (T8 - T7); Chris@10: Tn = ii[WS(is, 4)]; Chris@10: To = ii[WS(is, 7)]; Chris@10: Tp = KP866025403 * (Tn - To); Chris@10: Tt = Tn + To; Chris@10: } Chris@10: Ta = T6 + T9; Chris@10: T1c = Ts + Tt; Chris@10: Tm = FNMS(KP500000000, T9, T6); Chris@10: Tq = Tm + Tp; Chris@10: TW = Tm - Tp; Chris@10: Tu = FNMS(KP500000000, Tt, Ts); Chris@10: Tv = Tr + Tu; Chris@10: TX = Tu - Tr; Chris@10: } Chris@10: { Chris@10: E Tb, TD, Te, TC, TA, TE, Tx, TF; Chris@10: Tb = ri[WS(is, 2)]; Chris@10: TD = ii[WS(is, 2)]; Chris@10: { Chris@10: E Tc, Td, Ty, Tz; Chris@10: Tc = ri[WS(is, 5)]; Chris@10: Td = ri[WS(is, 8)]; Chris@10: Te = Tc + Td; Chris@10: TC = KP866025403 * (Td - Tc); Chris@10: Ty = ii[WS(is, 5)]; Chris@10: Tz = ii[WS(is, 8)]; Chris@10: TA = KP866025403 * (Ty - Tz); Chris@10: TE = Ty + Tz; Chris@10: } Chris@10: Tf = Tb + Te; Chris@10: T1d = TD + TE; Chris@10: Tx = FNMS(KP500000000, Te, Tb); Chris@10: TB = Tx + TA; Chris@10: T10 = Tx - TA; Chris@10: TF = FNMS(KP500000000, TE, TD); Chris@10: TG = TC + TF; Chris@10: TZ = TF - TC; Chris@10: } Chris@10: { Chris@10: E T1e, Tg, T1b, T1f, T1h, T1i; Chris@10: T1e = KP866025403 * (T1c - T1d); Chris@10: Tg = Ta + Tf; Chris@10: T1b = FNMS(KP500000000, Tg, T5); Chris@10: ro[0] = T5 + Tg; Chris@10: ro[WS(os, 3)] = T1b + T1e; Chris@10: ro[WS(os, 6)] = T1b - T1e; Chris@10: T1f = KP866025403 * (Tf - Ta); Chris@10: T1h = T1c + T1d; Chris@10: T1i = FNMS(KP500000000, T1h, T1g); Chris@10: io[WS(os, 3)] = T1f + T1i; Chris@10: io[0] = T1g + T1h; Chris@10: io[WS(os, 6)] = T1i - T1f; Chris@10: } Chris@10: { Chris@10: E Tl, TS, TI, TN, TM, TT, TJ, TU; Chris@10: Tl = Th + Tk; Chris@10: TS = TO + TR; Chris@10: { Chris@10: E Tw, TH, TK, TL; Chris@10: Tw = FMA(KP766044443, Tq, KP642787609 * Tv); Chris@10: TH = FMA(KP173648177, TB, KP984807753 * TG); Chris@10: TI = Tw + TH; Chris@10: TN = KP866025403 * (TH - Tw); Chris@10: TK = FNMS(KP642787609, Tq, KP766044443 * Tv); Chris@10: TL = FNMS(KP984807753, TB, KP173648177 * TG); Chris@10: TM = KP866025403 * (TK - TL); Chris@10: TT = TK + TL; Chris@10: } Chris@10: ro[WS(os, 1)] = Tl + TI; Chris@10: io[WS(os, 1)] = TS + TT; Chris@10: TJ = FNMS(KP500000000, TI, Tl); Chris@10: ro[WS(os, 7)] = TJ - TM; Chris@10: ro[WS(os, 4)] = TJ + TM; Chris@10: TU = FNMS(KP500000000, TT, TS); Chris@10: io[WS(os, 4)] = TN + TU; Chris@10: io[WS(os, 7)] = TU - TN; Chris@10: } Chris@10: { Chris@10: E TV, T14, T12, T13, T17, T1a, T18, T19; Chris@10: TV = Th - Tk; Chris@10: T14 = TR - TO; Chris@10: { Chris@10: E TY, T11, T15, T16; Chris@10: TY = FMA(KP173648177, TW, KP984807753 * TX); Chris@10: T11 = FNMS(KP939692620, T10, KP342020143 * TZ); Chris@10: T12 = TY + T11; Chris@10: T13 = KP866025403 * (T11 - TY); Chris@10: T15 = FNMS(KP984807753, TW, KP173648177 * TX); Chris@10: T16 = FMA(KP342020143, T10, KP939692620 * TZ); Chris@10: T17 = T15 - T16; Chris@10: T1a = KP866025403 * (T15 + T16); Chris@10: } Chris@10: ro[WS(os, 2)] = TV + T12; Chris@10: io[WS(os, 2)] = T14 + T17; Chris@10: T18 = FNMS(KP500000000, T17, T14); Chris@10: io[WS(os, 5)] = T13 + T18; Chris@10: io[WS(os, 8)] = T18 - T13; Chris@10: T19 = FNMS(KP500000000, T12, TV); Chris@10: ro[WS(os, 8)] = T19 - T1a; Chris@10: ro[WS(os, 5)] = T19 + T1a; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 9, "n1_9", {60, 20, 20, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_n1_9) (planner *p) { Chris@10: X(kdft_register) (p, n1_9, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */