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 07:40:45 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 10 -dit -name hc2cfdft_10 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 122 FP additions, 92 FP multiplications, Chris@10: * (or, 68 additions, 38 multiplications, 54 fused multiply/add), Chris@10: * 94 stack variables, 5 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E T1x, T1I, T1T, T22, T20; Chris@10: { Chris@10: E T3, T1u, T1S, T2f, Td, T1w, T14, T1p, T1j, T1q, T1N, T2e, T1z, To, T2i; Chris@10: E T1H, TQ, T1n, Ty, T1B; Chris@10: { Chris@10: E T1h, TW, Tc, T1b, T1g, T1f, T1Q, TV, T7, TS, T1J, TU, Ts, T19, T18; Chris@10: E T15, Tx, T17, T1O, T1A, Tt, TD, Ti, TE, Tn, TA, T1F, TC, T1y, Tj; Chris@10: E T11, T12, TJ, TZ, TO, TY, TG, T1L, T1e, T1, T2; Chris@10: T1 = Ip[0]; Chris@10: T2 = Im[0]; Chris@10: { Chris@10: E Ta, Tb, T1c, T1d; Chris@10: Ta = Rp[WS(rs, 2)]; Chris@10: Tb = Rm[WS(rs, 2)]; Chris@10: T1c = Rm[0]; Chris@10: T1h = T1 + T2; Chris@10: T3 = T1 - T2; Chris@10: T1d = Rp[0]; Chris@10: TW = Ta + Tb; Chris@10: Tc = Ta - Tb; Chris@10: T1b = W[0]; Chris@10: T1u = T1d + T1c; Chris@10: T1e = T1c - T1d; Chris@10: T1g = W[1]; Chris@10: } Chris@10: { Chris@10: E T16, Tp, TT, T5, T6, TB, Tf; Chris@10: T5 = Ip[WS(rs, 2)]; Chris@10: T6 = Im[WS(rs, 2)]; Chris@10: T1f = T1b * T1e; Chris@10: T1Q = T1g * T1e; Chris@10: TV = W[7]; Chris@10: T7 = T5 + T6; Chris@10: TT = T5 - T6; Chris@10: TS = W[6]; Chris@10: { Chris@10: E Tv, Tw, Tq, Tr; Chris@10: Tq = Rm[WS(rs, 3)]; Chris@10: Tr = Rp[WS(rs, 3)]; Chris@10: T1J = TV * TT; Chris@10: TU = TS * TT; Chris@10: Tv = Ip[WS(rs, 3)]; Chris@10: Ts = Tq - Tr; Chris@10: T19 = Tr + Tq; Chris@10: Tw = Im[WS(rs, 3)]; Chris@10: T18 = W[11]; Chris@10: T15 = W[10]; Chris@10: Tx = Tv + Tw; Chris@10: T16 = Tv - Tw; Chris@10: Tp = W[12]; Chris@10: } Chris@10: { Chris@10: E Tg, Th, Tl, Tm; Chris@10: Tg = Ip[WS(rs, 1)]; Chris@10: T17 = T15 * T16; Chris@10: T1O = T18 * T16; Chris@10: T1A = Tp * Tx; Chris@10: Tt = Tp * Ts; Chris@10: Th = Im[WS(rs, 1)]; Chris@10: Tl = Rp[WS(rs, 1)]; Chris@10: Tm = Rm[WS(rs, 1)]; Chris@10: TD = W[5]; Chris@10: Ti = Tg - Th; Chris@10: TE = Tg + Th; Chris@10: Tn = Tl + Tm; Chris@10: TB = Tm - Tl; Chris@10: TA = W[4]; Chris@10: Tf = W[2]; Chris@10: T1F = TD * TB; Chris@10: } Chris@10: { Chris@10: E TH, TI, TM, TN; Chris@10: TH = Ip[WS(rs, 4)]; Chris@10: TC = TA * TB; Chris@10: T1y = Tf * Tn; Chris@10: Tj = Tf * Ti; Chris@10: TI = Im[WS(rs, 4)]; Chris@10: TM = Rp[WS(rs, 4)]; Chris@10: TN = Rm[WS(rs, 4)]; Chris@10: T11 = W[17]; Chris@10: T12 = TH + TI; Chris@10: TJ = TH - TI; Chris@10: TZ = TN - TM; Chris@10: TO = TM + TN; Chris@10: TY = W[16]; Chris@10: TG = W[14]; Chris@10: T1L = T11 * TZ; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T10, T1D, TK, T4, T9, T1P, T1R, T8, T1v; Chris@10: T10 = TY * TZ; Chris@10: T1D = TG * TO; Chris@10: TK = TG * TJ; Chris@10: T4 = W[9]; Chris@10: T9 = W[8]; Chris@10: T1P = FMA(T15, T19, T1O); Chris@10: T1R = FMA(T1b, T1h, T1Q); Chris@10: T8 = T4 * T7; Chris@10: T1v = T9 * T7; Chris@10: { Chris@10: E TX, T13, T1a, T1i; Chris@10: TX = FNMS(TV, TW, TU); Chris@10: T1S = T1P - T1R; Chris@10: T2f = T1P + T1R; Chris@10: Td = FMA(T9, Tc, T8); Chris@10: T1w = FNMS(T4, Tc, T1v); Chris@10: T13 = FNMS(T11, T12, T10); Chris@10: T1a = FNMS(T18, T19, T17); Chris@10: T1i = FNMS(T1g, T1h, T1f); Chris@10: { Chris@10: E T1K, T1M, TF, T1G, TL; Chris@10: T1K = FMA(TS, TW, T1J); Chris@10: T14 = TX + T13; Chris@10: T1p = T13 - TX; Chris@10: T1j = T1a + T1i; Chris@10: T1q = T1i - T1a; Chris@10: T1M = FMA(TY, T12, T1L); Chris@10: TF = FNMS(TD, TE, TC); Chris@10: T1G = FMA(TA, TE, T1F); Chris@10: TL = W[15]; Chris@10: T1N = T1K - T1M; Chris@10: T2e = T1K + T1M; Chris@10: { Chris@10: E Tk, T1E, TP, Tu; Chris@10: Tk = W[3]; Chris@10: T1E = FMA(TL, TJ, T1D); Chris@10: TP = FNMS(TL, TO, TK); Chris@10: Tu = W[13]; Chris@10: T1z = FMA(Tk, Ti, T1y); Chris@10: To = FNMS(Tk, Tn, Tj); Chris@10: T2i = T1G + T1E; Chris@10: T1H = T1E - T1G; Chris@10: TQ = TF + TP; Chris@10: T1n = TF - TP; Chris@10: Ty = FNMS(Tu, Tx, Tt); Chris@10: T1B = FMA(Tu, Ts, T1A); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2p, T1t, T1m, T1C, T2o, T2m, T2k, T2w, T2y, T2n, T2d, T2l; Chris@10: { Chris@10: E T2g, Te, T2h, T2u, T1k, TR, T2v, Tz; Chris@10: T2p = T2e + T2f; Chris@10: T2g = T2e - T2f; Chris@10: Te = T3 - Td; Chris@10: T1t = Td + T3; Chris@10: Tz = To + Ty; Chris@10: T1m = Ty - To; Chris@10: T2h = T1z + T1B; Chris@10: T1C = T1z - T1B; Chris@10: T2u = T14 - T1j; Chris@10: T1k = T14 + T1j; Chris@10: TR = Tz + TQ; Chris@10: T2v = Tz - TQ; Chris@10: { Chris@10: E T2c, T2b, T2j, T1l; Chris@10: T2j = T2h - T2i; Chris@10: T2o = T2h + T2i; Chris@10: T2c = TR - T1k; Chris@10: T1l = TR + T1k; Chris@10: T2m = FMA(KP618033988, T2g, T2j); Chris@10: T2k = FNMS(KP618033988, T2j, T2g); Chris@10: T2w = FNMS(KP618033988, T2v, T2u); Chris@10: T2y = FMA(KP618033988, T2u, T2v); Chris@10: Ip[0] = KP500000000 * (Te + T1l); Chris@10: T2b = FNMS(KP250000000, T1l, Te); Chris@10: T2n = T1u + T1w; Chris@10: T1x = T1u - T1w; Chris@10: T2d = FNMS(KP559016994, T2c, T2b); Chris@10: T2l = FMA(KP559016994, T2c, T2b); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1o, T1Y, T28, T2a, T1Z, T1r, T2t, T2x; Chris@10: { Chris@10: E T26, T2s, T2q, T27, T2r; Chris@10: T1I = T1C + T1H; Chris@10: T26 = T1H - T1C; Chris@10: Im[WS(rs, 1)] = -(KP500000000 * (FNMS(KP951056516, T2k, T2d))); Chris@10: Ip[WS(rs, 2)] = KP500000000 * (FMA(KP951056516, T2k, T2d)); Chris@10: Im[WS(rs, 3)] = -(KP500000000 * (FNMS(KP951056516, T2m, T2l))); Chris@10: Ip[WS(rs, 4)] = KP500000000 * (FMA(KP951056516, T2m, T2l)); Chris@10: T2s = T2o - T2p; Chris@10: T2q = T2o + T2p; Chris@10: T27 = T1S - T1N; Chris@10: T1T = T1N + T1S; Chris@10: T1o = T1m + T1n; Chris@10: T1Y = T1n - T1m; Chris@10: Rp[0] = KP500000000 * (T2n + T2q); Chris@10: T2r = FNMS(KP250000000, T2q, T2n); Chris@10: T28 = FMA(KP618033988, T27, T26); Chris@10: T2a = FNMS(KP618033988, T26, T27); Chris@10: T1Z = T1q - T1p; Chris@10: T1r = T1p + T1q; Chris@10: T2t = FNMS(KP559016994, T2s, T2r); Chris@10: T2x = FMA(KP559016994, T2s, T2r); Chris@10: } Chris@10: { Chris@10: E T24, T23, T1s, T25, T29; Chris@10: T1s = T1o + T1r; Chris@10: T24 = T1r - T1o; Chris@10: Rm[WS(rs, 1)] = KP500000000 * (FMA(KP951056516, T2w, T2t)); Chris@10: Rp[WS(rs, 2)] = KP500000000 * (FNMS(KP951056516, T2w, T2t)); Chris@10: Rm[WS(rs, 3)] = KP500000000 * (FMA(KP951056516, T2y, T2x)); Chris@10: Rp[WS(rs, 4)] = KP500000000 * (FNMS(KP951056516, T2y, T2x)); Chris@10: Im[WS(rs, 4)] = KP500000000 * (T1s - T1t); Chris@10: T23 = FMA(KP250000000, T1s, T1t); Chris@10: T25 = FMA(KP559016994, T24, T23); Chris@10: T29 = FNMS(KP559016994, T24, T23); Chris@10: T22 = FNMS(KP618033988, T1Y, T1Z); Chris@10: T20 = FMA(KP618033988, T1Z, T1Y); Chris@10: Im[0] = -(KP500000000 * (FNMS(KP951056516, T28, T25))); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP951056516, T28, T25)); Chris@10: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP951056516, T2a, T29))); Chris@10: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP951056516, T2a, T29)); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1U, T1W, T1V, T21, T1X; Chris@10: T1U = T1I + T1T; Chris@10: T1W = T1I - T1T; Chris@10: Rm[WS(rs, 4)] = KP500000000 * (T1x + T1U); Chris@10: T1V = FNMS(KP250000000, T1U, T1x); Chris@10: T21 = FNMS(KP559016994, T1W, T1V); Chris@10: T1X = FMA(KP559016994, T1W, T1V); Chris@10: Rm[0] = KP500000000 * (FNMS(KP951056516, T20, T1X)); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP951056516, T20, T1X)); Chris@10: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP951056516, T22, T21)); Chris@10: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP951056516, T22, T21)); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 10}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 10, "hc2cfdft_10", twinstr, &GENUS, {68, 38, 54, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_10) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_10, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 10 -dit -name hc2cfdft_10 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 122 FP additions, 68 FP multiplications, Chris@10: * (or, 92 additions, 38 multiplications, 30 fused multiply/add), Chris@10: * 62 stack variables, 5 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP293892626, +0.293892626146236564584352977319536384298826219); Chris@10: DK(KP475528258, +0.475528258147576786058219666689691071702849317); Chris@10: DK(KP125000000, +0.125000000000000000000000000000000000000000000); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP279508497, +0.279508497187473712051146708591409529430077295); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E Tw, TL, TM, T1W, T1X, T27, T1Z, T20, T26, TX, T1a, T1b, T1d, T1e, T1f; Chris@10: E T1q, T1t, T1u, T1x, T1A, T1B, T1g, T1h, T1i, Td, T25, T1k, T1F; Chris@10: { Chris@10: E T3, T1D, T19, T1z, T7, Tb, TR, T1v, Tm, T1o, TK, T1s, Tv, T1p, T12; Chris@10: E T1y, TF, T1r, TW, T1w; Chris@10: { Chris@10: E T1, T2, T18, T14, T15, T16, T13, T17; Chris@10: T1 = Ip[0]; Chris@10: T2 = Im[0]; Chris@10: T18 = T1 + T2; Chris@10: T14 = Rm[0]; Chris@10: T15 = Rp[0]; Chris@10: T16 = T14 - T15; Chris@10: T3 = T1 - T2; Chris@10: T1D = T15 + T14; Chris@10: T13 = W[0]; Chris@10: T17 = W[1]; Chris@10: T19 = FNMS(T17, T18, T13 * T16); Chris@10: T1z = FMA(T17, T16, T13 * T18); Chris@10: } Chris@10: { Chris@10: E T5, T6, TO, T9, Ta, TQ, TN, TP; Chris@10: T5 = Ip[WS(rs, 2)]; Chris@10: T6 = Im[WS(rs, 2)]; Chris@10: TO = T5 - T6; Chris@10: T9 = Rp[WS(rs, 2)]; Chris@10: Ta = Rm[WS(rs, 2)]; Chris@10: TQ = T9 + Ta; Chris@10: T7 = T5 + T6; Chris@10: Tb = T9 - Ta; Chris@10: TN = W[6]; Chris@10: TP = W[7]; Chris@10: TR = FNMS(TP, TQ, TN * TO); Chris@10: T1v = FMA(TP, TO, TN * TQ); Chris@10: } Chris@10: { Chris@10: E Th, TJ, Tl, TH; Chris@10: { Chris@10: E Tf, Tg, Tj, Tk; Chris@10: Tf = Ip[WS(rs, 1)]; Chris@10: Tg = Im[WS(rs, 1)]; Chris@10: Th = Tf - Tg; Chris@10: TJ = Tf + Tg; Chris@10: Tj = Rp[WS(rs, 1)]; Chris@10: Tk = Rm[WS(rs, 1)]; Chris@10: Tl = Tj + Tk; Chris@10: TH = Tj - Tk; Chris@10: } Chris@10: { Chris@10: E Te, Ti, TG, TI; Chris@10: Te = W[2]; Chris@10: Ti = W[3]; Chris@10: Tm = FNMS(Ti, Tl, Te * Th); Chris@10: T1o = FMA(Te, Tl, Ti * Th); Chris@10: TG = W[4]; Chris@10: TI = W[5]; Chris@10: TK = FMA(TG, TH, TI * TJ); Chris@10: T1s = FNMS(TI, TH, TG * TJ); Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tq, TZ, Tu, T11; Chris@10: { Chris@10: E To, Tp, Ts, Tt; Chris@10: To = Ip[WS(rs, 3)]; Chris@10: Tp = Im[WS(rs, 3)]; Chris@10: Tq = To + Tp; Chris@10: TZ = To - Tp; Chris@10: Ts = Rp[WS(rs, 3)]; Chris@10: Tt = Rm[WS(rs, 3)]; Chris@10: Tu = Ts - Tt; Chris@10: T11 = Ts + Tt; Chris@10: } Chris@10: { Chris@10: E Tn, Tr, TY, T10; Chris@10: Tn = W[13]; Chris@10: Tr = W[12]; Chris@10: Tv = FMA(Tn, Tq, Tr * Tu); Chris@10: T1p = FNMS(Tn, Tu, Tr * Tq); Chris@10: TY = W[10]; Chris@10: T10 = W[11]; Chris@10: T12 = FNMS(T10, T11, TY * TZ); Chris@10: T1y = FMA(T10, TZ, TY * T11); Chris@10: } Chris@10: } Chris@10: { Chris@10: E TA, TV, TE, TT; Chris@10: { Chris@10: E Ty, Tz, TC, TD; Chris@10: Ty = Ip[WS(rs, 4)]; Chris@10: Tz = Im[WS(rs, 4)]; Chris@10: TA = Ty - Tz; Chris@10: TV = Ty + Tz; Chris@10: TC = Rp[WS(rs, 4)]; Chris@10: TD = Rm[WS(rs, 4)]; Chris@10: TE = TC + TD; Chris@10: TT = TC - TD; Chris@10: } Chris@10: { Chris@10: E Tx, TB, TS, TU; Chris@10: Tx = W[14]; Chris@10: TB = W[15]; Chris@10: TF = FNMS(TB, TE, Tx * TA); Chris@10: T1r = FMA(Tx, TE, TB * TA); Chris@10: TS = W[16]; Chris@10: TU = W[17]; Chris@10: TW = FMA(TS, TT, TU * TV); Chris@10: T1w = FNMS(TU, TT, TS * TV); Chris@10: } Chris@10: } Chris@10: Tw = Tm - Tv; Chris@10: TL = TF - TK; Chris@10: TM = Tw + TL; Chris@10: T1W = T1v + T1w; Chris@10: T1X = T1y + T1z; Chris@10: T27 = T1W + T1X; Chris@10: T1Z = T1o + T1p; Chris@10: T20 = T1s + T1r; Chris@10: T26 = T1Z + T20; Chris@10: TX = TR - TW; Chris@10: T1a = T12 + T19; Chris@10: T1b = TX + T1a; Chris@10: T1d = T19 - T12; Chris@10: T1e = TR + TW; Chris@10: T1f = T1d - T1e; Chris@10: T1q = T1o - T1p; Chris@10: T1t = T1r - T1s; Chris@10: T1u = T1q + T1t; Chris@10: T1x = T1v - T1w; Chris@10: T1A = T1y - T1z; Chris@10: T1B = T1x + T1A; Chris@10: T1g = Tm + Tv; Chris@10: T1h = TK + TF; Chris@10: T1i = T1g + T1h; Chris@10: { Chris@10: E Tc, T1E, T4, T8; Chris@10: T4 = W[9]; Chris@10: T8 = W[8]; Chris@10: Tc = FMA(T4, T7, T8 * Tb); Chris@10: T1E = FNMS(T4, Tb, T8 * T7); Chris@10: Td = T3 - Tc; Chris@10: T25 = T1D + T1E; Chris@10: T1k = Tc + T3; Chris@10: T1F = T1D - T1E; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1U, T1c, T1T, T22, T24, T1Y, T21, T23, T1V; Chris@10: T1U = KP279508497 * (TM - T1b); Chris@10: T1c = TM + T1b; Chris@10: T1T = FNMS(KP125000000, T1c, KP500000000 * Td); Chris@10: T1Y = T1W - T1X; Chris@10: T21 = T1Z - T20; Chris@10: T22 = FNMS(KP293892626, T21, KP475528258 * T1Y); Chris@10: T24 = FMA(KP475528258, T21, KP293892626 * T1Y); Chris@10: Ip[0] = KP500000000 * (Td + T1c); Chris@10: T23 = T1U + T1T; Chris@10: Ip[WS(rs, 4)] = T23 + T24; Chris@10: Im[WS(rs, 3)] = T24 - T23; Chris@10: T1V = T1T - T1U; Chris@10: Ip[WS(rs, 2)] = T1V + T22; Chris@10: Im[WS(rs, 1)] = T22 - T1V; Chris@10: } Chris@10: { Chris@10: E T2a, T28, T29, T2e, T2g, T2c, T2d, T2f, T2b; Chris@10: T2a = KP279508497 * (T26 - T27); Chris@10: T28 = T26 + T27; Chris@10: T29 = FNMS(KP125000000, T28, KP500000000 * T25); Chris@10: T2c = TX - T1a; Chris@10: T2d = Tw - TL; Chris@10: T2e = FNMS(KP293892626, T2d, KP475528258 * T2c); Chris@10: T2g = FMA(KP475528258, T2d, KP293892626 * T2c); Chris@10: Rp[0] = KP500000000 * (T25 + T28); Chris@10: T2f = T2a + T29; Chris@10: Rp[WS(rs, 4)] = T2f - T2g; Chris@10: Rm[WS(rs, 3)] = T2g + T2f; Chris@10: T2b = T29 - T2a; Chris@10: Rp[WS(rs, 2)] = T2b - T2e; Chris@10: Rm[WS(rs, 1)] = T2e + T2b; Chris@10: } Chris@10: { Chris@10: E T1M, T1j, T1L, T1Q, T1S, T1O, T1P, T1R, T1N; Chris@10: T1M = KP279508497 * (T1i + T1f); Chris@10: T1j = T1f - T1i; Chris@10: T1L = FMA(KP500000000, T1k, KP125000000 * T1j); Chris@10: T1O = T1A - T1x; Chris@10: T1P = T1q - T1t; Chris@10: T1Q = FNMS(KP475528258, T1P, KP293892626 * T1O); Chris@10: T1S = FMA(KP293892626, T1P, KP475528258 * T1O); Chris@10: Im[WS(rs, 4)] = KP500000000 * (T1j - T1k); Chris@10: T1R = T1L - T1M; Chris@10: Ip[WS(rs, 3)] = T1R + T1S; Chris@10: Im[WS(rs, 2)] = T1S - T1R; Chris@10: T1N = T1L + T1M; Chris@10: Ip[WS(rs, 1)] = T1N + T1Q; Chris@10: Im[0] = T1Q - T1N; Chris@10: } Chris@10: { Chris@10: E T1C, T1G, T1H, T1n, T1J, T1l, T1m, T1K, T1I; Chris@10: T1C = KP279508497 * (T1u - T1B); Chris@10: T1G = T1u + T1B; Chris@10: T1H = FNMS(KP125000000, T1G, KP500000000 * T1F); Chris@10: T1l = T1g - T1h; Chris@10: T1m = T1e + T1d; Chris@10: T1n = FMA(KP475528258, T1l, KP293892626 * T1m); Chris@10: T1J = FNMS(KP293892626, T1l, KP475528258 * T1m); Chris@10: Rm[WS(rs, 4)] = KP500000000 * (T1F + T1G); Chris@10: T1K = T1H - T1C; Chris@10: Rp[WS(rs, 3)] = T1J + T1K; Chris@10: Rm[WS(rs, 2)] = T1K - T1J; Chris@10: T1I = T1C + T1H; Chris@10: Rp[WS(rs, 1)] = T1n + T1I; Chris@10: Rm[0] = T1I - T1n; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 10}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 10, "hc2cfdft_10", twinstr, &GENUS, {92, 38, 30, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_10) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_10, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */