cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:35:51 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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 */ cannam@127: cannam@127: /* cannam@127: * This function contains 80 FP additions, 56 FP multiplications, cannam@127: * (or, 24 additions, 0 multiplications, 56 fused multiply/add), cannam@127: * 59 stack variables, 10 constants, and 36 memory accesses cannam@127: */ cannam@127: #include "n.h" cannam@127: cannam@127: static void n1_9(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DK(KP954188894, +0.954188894138671133499268364187245676532219158); cannam@127: DK(KP363970234, +0.363970234266202361351047882776834043890471784); cannam@127: DK(KP852868531, +0.852868531952443209628250963940074071936020296); cannam@127: DK(KP984807753, +0.984807753012208059366743024589523013670643252); cannam@127: DK(KP492403876, +0.492403876506104029683371512294761506835321626); cannam@127: DK(KP777861913, +0.777861913430206160028177977318626690410586096); cannam@127: DK(KP839099631, +0.839099631177280011763127298123181364687434283); cannam@127: DK(KP176326980, +0.176326980708464973471090386868618986121633062); cannam@127: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: { cannam@127: INT i; cannam@127: 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)) { cannam@127: E T17, TV, T14, TY, T11, T15; cannam@127: { cannam@127: E Tm, TM, TL, T5, Tl, T1f, Tb, Tt, Ta, T1c, TI, TX, TF, TW, Tc; cannam@127: E Td, Tp, Tq; cannam@127: { cannam@127: E T1, Th, Ti, Tj, T4, T2, T3; cannam@127: T1 = ri[0]; cannam@127: T2 = ri[WS(is, 3)]; cannam@127: T3 = ri[WS(is, 6)]; cannam@127: Th = ii[0]; cannam@127: Ti = ii[WS(is, 3)]; cannam@127: Tj = ii[WS(is, 6)]; cannam@127: T4 = T2 + T3; cannam@127: Tm = T3 - T2; cannam@127: { cannam@127: E T6, Tz, T7, T8, TA, TB, Tk; cannam@127: T6 = ri[WS(is, 1)]; cannam@127: TM = Ti - Tj; cannam@127: Tk = Ti + Tj; cannam@127: TL = FNMS(KP500000000, T4, T1); cannam@127: T5 = T1 + T4; cannam@127: Tz = ii[WS(is, 1)]; cannam@127: Tl = FNMS(KP500000000, Tk, Th); cannam@127: T1f = Th + Tk; cannam@127: T7 = ri[WS(is, 4)]; cannam@127: T8 = ri[WS(is, 7)]; cannam@127: TA = ii[WS(is, 4)]; cannam@127: TB = ii[WS(is, 7)]; cannam@127: { cannam@127: E TE, T9, TH, TC, TG, TD; cannam@127: Tb = ri[WS(is, 2)]; cannam@127: TE = T7 - T8; cannam@127: T9 = T7 + T8; cannam@127: TH = TB - TA; cannam@127: TC = TA + TB; cannam@127: Tt = ii[WS(is, 2)]; cannam@127: Ta = T6 + T9; cannam@127: TG = FNMS(KP500000000, T9, T6); cannam@127: T1c = Tz + TC; cannam@127: TD = FNMS(KP500000000, TC, Tz); cannam@127: TI = FNMS(KP866025403, TH, TG); cannam@127: TX = FMA(KP866025403, TH, TG); cannam@127: TF = FNMS(KP866025403, TE, TD); cannam@127: TW = FMA(KP866025403, TE, TD); cannam@127: Tc = ri[WS(is, 5)]; cannam@127: Td = ri[WS(is, 8)]; cannam@127: Tp = ii[WS(is, 5)]; cannam@127: Tq = ii[WS(is, 8)]; cannam@127: } cannam@127: } cannam@127: } cannam@127: { cannam@127: E Tn, TN, TZ, T10, TO, Ty, TJ, TP; cannam@127: { cannam@127: E Tw, Te, Tu, Tr; cannam@127: T17 = FNMS(KP866025403, Tm, Tl); cannam@127: Tn = FMA(KP866025403, Tm, Tl); cannam@127: Tw = Td - Tc; cannam@127: Te = Tc + Td; cannam@127: Tu = Tp + Tq; cannam@127: Tr = Tp - Tq; cannam@127: TN = FMA(KP866025403, TM, TL); cannam@127: TV = FNMS(KP866025403, TM, TL); cannam@127: { cannam@127: E Tf, To, T1d, Tv; cannam@127: Tf = Tb + Te; cannam@127: To = FNMS(KP500000000, Te, Tb); cannam@127: T1d = Tt + Tu; cannam@127: Tv = FNMS(KP500000000, Tu, Tt); cannam@127: { cannam@127: E Ts, Tg, T1i, Tx; cannam@127: Ts = FMA(KP866025403, Tr, To); cannam@127: TZ = FNMS(KP866025403, Tr, To); cannam@127: Tg = Ta + Tf; cannam@127: T1i = Tf - Ta; cannam@127: Tx = FMA(KP866025403, Tw, Tv); cannam@127: T10 = FNMS(KP866025403, Tw, Tv); cannam@127: { cannam@127: E T1e, T1g, T1b, T1h; cannam@127: T1e = T1c - T1d; cannam@127: T1g = T1c + T1d; cannam@127: ro[0] = T5 + Tg; cannam@127: T1b = FNMS(KP500000000, Tg, T5); cannam@127: io[0] = T1f + T1g; cannam@127: T1h = FNMS(KP500000000, T1g, T1f); cannam@127: TO = FMA(KP176326980, Ts, Tx); cannam@127: Ty = FNMS(KP176326980, Tx, Ts); cannam@127: ro[WS(os, 6)] = FNMS(KP866025403, T1e, T1b); cannam@127: ro[WS(os, 3)] = FMA(KP866025403, T1e, T1b); cannam@127: io[WS(os, 6)] = FNMS(KP866025403, T1i, T1h); cannam@127: io[WS(os, 3)] = FMA(KP866025403, T1i, T1h); cannam@127: TJ = FNMS(KP839099631, TI, TF); cannam@127: TP = FMA(KP839099631, TF, TI); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: { cannam@127: E TS, TK, TU, TQ, TT, TR; cannam@127: TS = FMA(KP777861913, TJ, Ty); cannam@127: TK = FNMS(KP777861913, TJ, Ty); cannam@127: TU = FNMS(KP777861913, TP, TO); cannam@127: TQ = FMA(KP777861913, TP, TO); cannam@127: TT = FMA(KP492403876, TK, Tn); cannam@127: io[WS(os, 1)] = FNMS(KP984807753, TK, Tn); cannam@127: TR = FNMS(KP492403876, TQ, TN); cannam@127: ro[WS(os, 1)] = FMA(KP984807753, TQ, TN); cannam@127: io[WS(os, 4)] = FMA(KP852868531, TU, TT); cannam@127: io[WS(os, 7)] = FNMS(KP852868531, TU, TT); cannam@127: ro[WS(os, 7)] = FNMS(KP852868531, TS, TR); cannam@127: ro[WS(os, 4)] = FMA(KP852868531, TS, TR); cannam@127: T14 = FNMS(KP176326980, TW, TX); cannam@127: TY = FMA(KP176326980, TX, TW); cannam@127: T11 = FNMS(KP363970234, T10, TZ); cannam@127: T15 = FMA(KP363970234, TZ, T10); cannam@127: } cannam@127: } cannam@127: } cannam@127: { cannam@127: E T12, T1a, T16, T18, T13, T19; cannam@127: T12 = FNMS(KP954188894, T11, TY); cannam@127: T1a = FMA(KP954188894, T11, TY); cannam@127: T16 = FNMS(KP954188894, T15, T14); cannam@127: T18 = FMA(KP954188894, T15, T14); cannam@127: T13 = FNMS(KP492403876, T12, TV); cannam@127: ro[WS(os, 2)] = FMA(KP984807753, T12, TV); cannam@127: T19 = FMA(KP492403876, T18, T17); cannam@127: io[WS(os, 2)] = FNMS(KP984807753, T18, T17); cannam@127: ro[WS(os, 8)] = FMA(KP852868531, T16, T13); cannam@127: ro[WS(os, 5)] = FNMS(KP852868531, T16, T13); cannam@127: io[WS(os, 8)] = FMA(KP852868531, T1a, T19); cannam@127: io[WS(os, 5)] = FNMS(KP852868531, T1a, T19); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 9, "n1_9", {24, 0, 56, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_n1_9) (planner *p) { cannam@127: X(kdft_register) (p, n1_9, &desc); cannam@127: } cannam@127: cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 9 -name n1_9 -include n.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 80 FP additions, 40 FP multiplications, cannam@127: * (or, 60 additions, 20 multiplications, 20 fused multiply/add), cannam@127: * 39 stack variables, 8 constants, and 36 memory accesses cannam@127: */ cannam@127: #include "n.h" cannam@127: cannam@127: static void n1_9(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DK(KP939692620, +0.939692620785908384054109277324731469936208134); cannam@127: DK(KP342020143, +0.342020143325668733044099614682259580763083368); cannam@127: DK(KP984807753, +0.984807753012208059366743024589523013670643252); cannam@127: DK(KP173648177, +0.173648177666930348851716626769314796000375677); cannam@127: DK(KP642787609, +0.642787609686539326322643409907263432907559884); cannam@127: DK(KP766044443, +0.766044443118978035202392650555416673935832457); cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@127: { cannam@127: INT i; cannam@127: 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)) { cannam@127: E T5, TO, Th, Tk, T1g, TR, Ta, T1c, Tq, TW, Tv, TX, Tf, T1d, TB; cannam@127: E T10, TG, TZ; cannam@127: { cannam@127: E T1, T2, T3, T4; cannam@127: T1 = ri[0]; cannam@127: T2 = ri[WS(is, 3)]; cannam@127: T3 = ri[WS(is, 6)]; cannam@127: T4 = T2 + T3; cannam@127: T5 = T1 + T4; cannam@127: TO = KP866025403 * (T3 - T2); cannam@127: Th = FNMS(KP500000000, T4, T1); cannam@127: } cannam@127: { cannam@127: E TP, Ti, Tj, TQ; cannam@127: TP = ii[0]; cannam@127: Ti = ii[WS(is, 3)]; cannam@127: Tj = ii[WS(is, 6)]; cannam@127: TQ = Ti + Tj; cannam@127: Tk = KP866025403 * (Ti - Tj); cannam@127: T1g = TP + TQ; cannam@127: TR = FNMS(KP500000000, TQ, TP); cannam@127: } cannam@127: { cannam@127: E T6, Ts, T9, Tr, Tp, Tt, Tm, Tu; cannam@127: T6 = ri[WS(is, 1)]; cannam@127: Ts = ii[WS(is, 1)]; cannam@127: { cannam@127: E T7, T8, Tn, To; cannam@127: T7 = ri[WS(is, 4)]; cannam@127: T8 = ri[WS(is, 7)]; cannam@127: T9 = T7 + T8; cannam@127: Tr = KP866025403 * (T8 - T7); cannam@127: Tn = ii[WS(is, 4)]; cannam@127: To = ii[WS(is, 7)]; cannam@127: Tp = KP866025403 * (Tn - To); cannam@127: Tt = Tn + To; cannam@127: } cannam@127: Ta = T6 + T9; cannam@127: T1c = Ts + Tt; cannam@127: Tm = FNMS(KP500000000, T9, T6); cannam@127: Tq = Tm + Tp; cannam@127: TW = Tm - Tp; cannam@127: Tu = FNMS(KP500000000, Tt, Ts); cannam@127: Tv = Tr + Tu; cannam@127: TX = Tu - Tr; cannam@127: } cannam@127: { cannam@127: E Tb, TD, Te, TC, TA, TE, Tx, TF; cannam@127: Tb = ri[WS(is, 2)]; cannam@127: TD = ii[WS(is, 2)]; cannam@127: { cannam@127: E Tc, Td, Ty, Tz; cannam@127: Tc = ri[WS(is, 5)]; cannam@127: Td = ri[WS(is, 8)]; cannam@127: Te = Tc + Td; cannam@127: TC = KP866025403 * (Td - Tc); cannam@127: Ty = ii[WS(is, 5)]; cannam@127: Tz = ii[WS(is, 8)]; cannam@127: TA = KP866025403 * (Ty - Tz); cannam@127: TE = Ty + Tz; cannam@127: } cannam@127: Tf = Tb + Te; cannam@127: T1d = TD + TE; cannam@127: Tx = FNMS(KP500000000, Te, Tb); cannam@127: TB = Tx + TA; cannam@127: T10 = Tx - TA; cannam@127: TF = FNMS(KP500000000, TE, TD); cannam@127: TG = TC + TF; cannam@127: TZ = TF - TC; cannam@127: } cannam@127: { cannam@127: E T1e, Tg, T1b, T1f, T1h, T1i; cannam@127: T1e = KP866025403 * (T1c - T1d); cannam@127: Tg = Ta + Tf; cannam@127: T1b = FNMS(KP500000000, Tg, T5); cannam@127: ro[0] = T5 + Tg; cannam@127: ro[WS(os, 3)] = T1b + T1e; cannam@127: ro[WS(os, 6)] = T1b - T1e; cannam@127: T1f = KP866025403 * (Tf - Ta); cannam@127: T1h = T1c + T1d; cannam@127: T1i = FNMS(KP500000000, T1h, T1g); cannam@127: io[WS(os, 3)] = T1f + T1i; cannam@127: io[0] = T1g + T1h; cannam@127: io[WS(os, 6)] = T1i - T1f; cannam@127: } cannam@127: { cannam@127: E Tl, TS, TI, TN, TM, TT, TJ, TU; cannam@127: Tl = Th + Tk; cannam@127: TS = TO + TR; cannam@127: { cannam@127: E Tw, TH, TK, TL; cannam@127: Tw = FMA(KP766044443, Tq, KP642787609 * Tv); cannam@127: TH = FMA(KP173648177, TB, KP984807753 * TG); cannam@127: TI = Tw + TH; cannam@127: TN = KP866025403 * (TH - Tw); cannam@127: TK = FNMS(KP642787609, Tq, KP766044443 * Tv); cannam@127: TL = FNMS(KP984807753, TB, KP173648177 * TG); cannam@127: TM = KP866025403 * (TK - TL); cannam@127: TT = TK + TL; cannam@127: } cannam@127: ro[WS(os, 1)] = Tl + TI; cannam@127: io[WS(os, 1)] = TS + TT; cannam@127: TJ = FNMS(KP500000000, TI, Tl); cannam@127: ro[WS(os, 7)] = TJ - TM; cannam@127: ro[WS(os, 4)] = TJ + TM; cannam@127: TU = FNMS(KP500000000, TT, TS); cannam@127: io[WS(os, 4)] = TN + TU; cannam@127: io[WS(os, 7)] = TU - TN; cannam@127: } cannam@127: { cannam@127: E TV, T14, T12, T13, T17, T1a, T18, T19; cannam@127: TV = Th - Tk; cannam@127: T14 = TR - TO; cannam@127: { cannam@127: E TY, T11, T15, T16; cannam@127: TY = FMA(KP173648177, TW, KP984807753 * TX); cannam@127: T11 = FNMS(KP939692620, T10, KP342020143 * TZ); cannam@127: T12 = TY + T11; cannam@127: T13 = KP866025403 * (T11 - TY); cannam@127: T15 = FNMS(KP984807753, TW, KP173648177 * TX); cannam@127: T16 = FMA(KP342020143, T10, KP939692620 * TZ); cannam@127: T17 = T15 - T16; cannam@127: T1a = KP866025403 * (T15 + T16); cannam@127: } cannam@127: ro[WS(os, 2)] = TV + T12; cannam@127: io[WS(os, 2)] = T14 + T17; cannam@127: T18 = FNMS(KP500000000, T17, T14); cannam@127: io[WS(os, 5)] = T13 + T18; cannam@127: io[WS(os, 8)] = T18 - T13; cannam@127: T19 = FNMS(KP500000000, T12, TV); cannam@127: ro[WS(os, 8)] = T19 - T1a; cannam@127: ro[WS(os, 5)] = T19 + T1a; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 9, "n1_9", {60, 20, 20, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void X(codelet_n1_9) (planner *p) { cannam@127: X(kdft_register) (p, n1_9, &desc); cannam@127: } cannam@127: cannam@127: #endif /* HAVE_FMA */