cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:40:31 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cf_12 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 118 FP additions, 68 FP multiplications, cannam@95: * (or, 72 additions, 22 multiplications, 46 fused multiply/add), cannam@95: * 84 stack variables, 2 constants, and 48 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cf_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) { cannam@95: E T2n, T2u; cannam@95: { cannam@95: E T1, T2i, T2e, Tl, T1Y, T10, T1S, TG, T2f, T1s, T2s, Ty, T1Z, T1H, T21; cannam@95: E T1d, TI, TL, T2h, T1l, T2p, Te, TJ, T1w, TO, TR, TN, TK, TQ; cannam@95: { cannam@95: E TW, TZ, TY, T1X, TX; cannam@95: T1 = Rp[0]; cannam@95: T2i = Rm[0]; cannam@95: { cannam@95: E Th, Tk, Tg, Tj, T2d, Ti, TV; cannam@95: Th = Rp[WS(rs, 3)]; cannam@95: Tk = Rm[WS(rs, 3)]; cannam@95: Tg = W[10]; cannam@95: Tj = W[11]; cannam@95: TW = Ip[WS(rs, 4)]; cannam@95: TZ = Im[WS(rs, 4)]; cannam@95: T2d = Tg * Tk; cannam@95: Ti = Tg * Th; cannam@95: TV = W[16]; cannam@95: TY = W[17]; cannam@95: T2e = FNMS(Tj, Th, T2d); cannam@95: Tl = FMA(Tj, Tk, Ti); cannam@95: T1X = TV * TZ; cannam@95: TX = TV * TW; cannam@95: } cannam@95: { cannam@95: E Tn, Tq, Tt, T1o, To, Tw, Ts, Tp, Tv; cannam@95: { cannam@95: E TC, TF, TB, TE, T1R, TD, Tm; cannam@95: TC = Ip[WS(rs, 1)]; cannam@95: TF = Im[WS(rs, 1)]; cannam@95: T1Y = FNMS(TY, TW, T1X); cannam@95: T10 = FMA(TY, TZ, TX); cannam@95: TB = W[4]; cannam@95: TE = W[5]; cannam@95: Tn = Rp[WS(rs, 5)]; cannam@95: Tq = Rm[WS(rs, 5)]; cannam@95: T1R = TB * TF; cannam@95: TD = TB * TC; cannam@95: Tm = W[18]; cannam@95: Tt = Rp[WS(rs, 1)]; cannam@95: T1S = FNMS(TE, TC, T1R); cannam@95: TG = FMA(TE, TF, TD); cannam@95: T1o = Tm * Tq; cannam@95: To = Tm * Tn; cannam@95: Tw = Rm[WS(rs, 1)]; cannam@95: Ts = W[2]; cannam@95: Tp = W[19]; cannam@95: Tv = W[3]; cannam@95: } cannam@95: { cannam@95: E T12, T15, T13, T1D, T18, T1b, T17, T14, T1a; cannam@95: { cannam@95: E T1p, Tr, T1r, Tx, T1q, Tu, T11; cannam@95: T12 = Ip[0]; cannam@95: T1q = Ts * Tw; cannam@95: Tu = Ts * Tt; cannam@95: T1p = FNMS(Tp, Tn, T1o); cannam@95: Tr = FMA(Tp, Tq, To); cannam@95: T1r = FNMS(Tv, Tt, T1q); cannam@95: Tx = FMA(Tv, Tw, Tu); cannam@95: T15 = Im[0]; cannam@95: T11 = W[0]; cannam@95: T2f = T1p + T1r; cannam@95: T1s = T1p - T1r; cannam@95: T2s = Tx - Tr; cannam@95: Ty = Tr + Tx; cannam@95: T13 = T11 * T12; cannam@95: T1D = T11 * T15; cannam@95: } cannam@95: T18 = Ip[WS(rs, 2)]; cannam@95: T1b = Im[WS(rs, 2)]; cannam@95: T17 = W[8]; cannam@95: T14 = W[1]; cannam@95: T1a = W[9]; cannam@95: { cannam@95: E T3, T6, T4, T1h, T9, Tc, T8, T5, Tb; cannam@95: { cannam@95: E T1E, T16, T1G, T1c, T1F, T19, T2; cannam@95: T3 = Rp[WS(rs, 2)]; cannam@95: T1F = T17 * T1b; cannam@95: T19 = T17 * T18; cannam@95: T1E = FNMS(T14, T12, T1D); cannam@95: T16 = FMA(T14, T15, T13); cannam@95: T1G = FNMS(T1a, T18, T1F); cannam@95: T1c = FMA(T1a, T1b, T19); cannam@95: T6 = Rm[WS(rs, 2)]; cannam@95: T2 = W[6]; cannam@95: T1Z = T1E + T1G; cannam@95: T1H = T1E - T1G; cannam@95: T21 = T1c - T16; cannam@95: T1d = T16 + T1c; cannam@95: T4 = T2 * T3; cannam@95: T1h = T2 * T6; cannam@95: } cannam@95: T9 = Rp[WS(rs, 4)]; cannam@95: Tc = Rm[WS(rs, 4)]; cannam@95: T8 = W[14]; cannam@95: T5 = W[7]; cannam@95: Tb = W[15]; cannam@95: { cannam@95: E T1i, T7, T1k, Td, T1j, Ta, TH; cannam@95: TI = Ip[WS(rs, 3)]; cannam@95: T1j = T8 * Tc; cannam@95: Ta = T8 * T9; cannam@95: T1i = FNMS(T5, T3, T1h); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: T1k = FNMS(Tb, T9, T1j); cannam@95: Td = FMA(Tb, Tc, Ta); cannam@95: TL = Im[WS(rs, 3)]; cannam@95: TH = W[12]; cannam@95: T2h = T1i + T1k; cannam@95: T1l = T1i - T1k; cannam@95: T2p = Td - T7; cannam@95: Te = T7 + Td; cannam@95: TJ = TH * TI; cannam@95: T1w = TH * TL; cannam@95: } cannam@95: TO = Ip[WS(rs, 5)]; cannam@95: TR = Im[WS(rs, 5)]; cannam@95: TN = W[20]; cannam@95: TK = W[13]; cannam@95: TQ = W[21]; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1g, T1n, T2r, T1A, T1V, T28, TA, T2o, T1v, T1C, T1U, T29, T2m, T2k, T2l; cannam@95: E T1f, T2a, T20; cannam@95: { cannam@95: E T2g, T1T, TT, T2j, TU, T1e; cannam@95: { cannam@95: E Tf, T1x, TM, T1z, TS, Tz, T1y, TP; cannam@95: T1g = FNMS(KP500000000, Te, T1); cannam@95: Tf = T1 + Te; cannam@95: T1y = TN * TR; cannam@95: TP = TN * TO; cannam@95: T1x = FNMS(TK, TI, T1w); cannam@95: TM = FMA(TK, TL, TJ); cannam@95: T1z = FNMS(TQ, TO, T1y); cannam@95: TS = FMA(TQ, TR, TP); cannam@95: Tz = Tl + Ty; cannam@95: T1n = FNMS(KP500000000, Ty, Tl); cannam@95: T2r = FNMS(KP500000000, T2f, T2e); cannam@95: T2g = T2e + T2f; cannam@95: T1T = T1x + T1z; cannam@95: T1A = T1x - T1z; cannam@95: T1V = TS - TM; cannam@95: TT = TM + TS; cannam@95: T28 = Tf - Tz; cannam@95: TA = Tf + Tz; cannam@95: T2j = T2h + T2i; cannam@95: T2o = FNMS(KP500000000, T2h, T2i); cannam@95: } cannam@95: T1v = FNMS(KP500000000, TT, TG); cannam@95: TU = TG + TT; cannam@95: T1e = T10 + T1d; cannam@95: T1C = FNMS(KP500000000, T1d, T10); cannam@95: T1U = FNMS(KP500000000, T1T, T1S); cannam@95: T29 = T1S + T1T; cannam@95: T2m = T2j - T2g; cannam@95: T2k = T2g + T2j; cannam@95: T2l = TU - T1e; cannam@95: T1f = TU + T1e; cannam@95: T2a = T1Y + T1Z; cannam@95: T20 = FNMS(KP500000000, T1Z, T1Y); cannam@95: } cannam@95: { cannam@95: E T1m, T1K, T2z, T2q, T2y, T2t, T1L, T1t, T1B, T1N, T2c, T2b; cannam@95: Im[WS(rs, 2)] = T2l - T2m; cannam@95: Ip[WS(rs, 3)] = T2l + T2m; cannam@95: Rp[0] = TA + T1f; cannam@95: Rm[WS(rs, 5)] = TA - T1f; cannam@95: T2c = T29 + T2a; cannam@95: T2b = T29 - T2a; cannam@95: T1m = FNMS(KP866025403, T1l, T1g); cannam@95: T1K = FMA(KP866025403, T1l, T1g); cannam@95: Ip[0] = T2c + T2k; cannam@95: Im[WS(rs, 5)] = T2c - T2k; cannam@95: Rm[WS(rs, 2)] = T28 + T2b; cannam@95: Rp[WS(rs, 3)] = T28 - T2b; cannam@95: T2z = FNMS(KP866025403, T2p, T2o); cannam@95: T2q = FMA(KP866025403, T2p, T2o); cannam@95: T2y = FNMS(KP866025403, T2s, T2r); cannam@95: T2t = FMA(KP866025403, T2s, T2r); cannam@95: T1L = FMA(KP866025403, T1s, T1n); cannam@95: T1t = FNMS(KP866025403, T1s, T1n); cannam@95: T1B = FNMS(KP866025403, T1A, T1v); cannam@95: T1N = FMA(KP866025403, T1A, T1v); cannam@95: { cannam@95: E T1Q, T2C, T23, T24, T2B, T27, T2v, T2w; cannam@95: { cannam@95: E T1u, T25, T26, T1O, T1I, T2A, T2x, T1W, T22, T1M, T1J, T1P; cannam@95: T1Q = T1m - T1t; cannam@95: T1u = T1m + T1t; cannam@95: T25 = FMA(KP866025403, T1V, T1U); cannam@95: T1W = FNMS(KP866025403, T1V, T1U); cannam@95: T26 = FMA(KP866025403, T21, T20); cannam@95: T22 = FNMS(KP866025403, T21, T20); cannam@95: T1O = FMA(KP866025403, T1H, T1C); cannam@95: T1I = FNMS(KP866025403, T1H, T1C); cannam@95: T2A = T2y + T2z; cannam@95: T2C = T2z - T2y; cannam@95: T23 = T1W - T22; cannam@95: T2x = T1W + T22; cannam@95: T1M = T1K + T1L; cannam@95: T24 = T1K - T1L; cannam@95: T2B = T1I - T1B; cannam@95: T1J = T1B + T1I; cannam@95: T1P = T1N + T1O; cannam@95: T2n = T1O - T1N; cannam@95: Ip[WS(rs, 2)] = T2A - T2x; cannam@95: Im[WS(rs, 3)] = -(T2x + T2A); cannam@95: Rm[WS(rs, 3)] = T1u + T1J; cannam@95: Rp[WS(rs, 2)] = T1u - T1J; cannam@95: Rm[WS(rs, 1)] = T1M - T1P; cannam@95: Rp[WS(rs, 4)] = T1M + T1P; cannam@95: T27 = T25 - T26; cannam@95: T2v = T25 + T26; cannam@95: T2w = T2t + T2q; cannam@95: T2u = T2q - T2t; cannam@95: } cannam@95: Ip[WS(rs, 4)] = T2v + T2w; cannam@95: Im[WS(rs, 1)] = T2v - T2w; cannam@95: Rp[WS(rs, 5)] = T1Q + T23; cannam@95: Rm[0] = T1Q - T23; cannam@95: Ip[WS(rs, 5)] = T2B + T2C; cannam@95: Im[0] = T2B - T2C; cannam@95: Rp[WS(rs, 1)] = T24 + T27; cannam@95: Rm[WS(rs, 4)] = T24 - T27; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: Ip[WS(rs, 1)] = T2n + T2u; cannam@95: Im[WS(rs, 4)] = T2n - T2u; cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 12}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 12, "hc2cf_12", twinstr, &GENUS, {72, 22, 46, 0} }; cannam@95: cannam@95: void X(codelet_hc2cf_12) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cf_12, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cf_12 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 118 FP additions, 60 FP multiplications, cannam@95: * (or, 88 additions, 30 multiplications, 30 fused multiply/add), cannam@95: * 47 stack variables, 2 constants, and 48 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cf_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) { cannam@95: E T1, T1W, T18, T22, Tc, T15, T1V, T23, TR, T1E, T1o, T1D, T12, T1l, T1F; cannam@95: E T1G, Ti, T1S, T1d, T25, Tt, T1a, T1T, T26, TA, T1y, T1j, T1B, TL, T1g; cannam@95: E T1z, T1A; cannam@95: { cannam@95: E T6, T16, Tb, T17; cannam@95: T1 = Rp[0]; cannam@95: T1W = Rm[0]; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 2)]; cannam@95: T2 = W[6]; cannam@95: T4 = W[7]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: T16 = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: { cannam@95: E T8, Ta, T7, T9; cannam@95: T8 = Rp[WS(rs, 4)]; cannam@95: Ta = Rm[WS(rs, 4)]; cannam@95: T7 = W[14]; cannam@95: T9 = W[15]; cannam@95: Tb = FMA(T7, T8, T9 * Ta); cannam@95: T17 = FNMS(T9, T8, T7 * Ta); cannam@95: } cannam@95: T18 = KP866025403 * (T16 - T17); cannam@95: T22 = KP866025403 * (Tb - T6); cannam@95: Tc = T6 + Tb; cannam@95: T15 = FNMS(KP500000000, Tc, T1); cannam@95: T1V = T16 + T17; cannam@95: T23 = FNMS(KP500000000, T1V, T1W); cannam@95: } cannam@95: { cannam@95: E T11, T1n, TW, T1m; cannam@95: { cannam@95: E TO, TQ, TN, TP; cannam@95: TO = Ip[WS(rs, 4)]; cannam@95: TQ = Im[WS(rs, 4)]; cannam@95: TN = W[16]; cannam@95: TP = W[17]; cannam@95: TR = FMA(TN, TO, TP * TQ); cannam@95: T1E = FNMS(TP, TO, TN * TQ); cannam@95: } cannam@95: { cannam@95: E TY, T10, TX, TZ; cannam@95: TY = Ip[WS(rs, 2)]; cannam@95: T10 = Im[WS(rs, 2)]; cannam@95: TX = W[8]; cannam@95: TZ = W[9]; cannam@95: T11 = FMA(TX, TY, TZ * T10); cannam@95: T1n = FNMS(TZ, TY, TX * T10); cannam@95: } cannam@95: { cannam@95: E TT, TV, TS, TU; cannam@95: TT = Ip[0]; cannam@95: TV = Im[0]; cannam@95: TS = W[0]; cannam@95: TU = W[1]; cannam@95: TW = FMA(TS, TT, TU * TV); cannam@95: T1m = FNMS(TU, TT, TS * TV); cannam@95: } cannam@95: T1o = KP866025403 * (T1m - T1n); cannam@95: T1D = KP866025403 * (T11 - TW); cannam@95: T12 = TW + T11; cannam@95: T1l = FNMS(KP500000000, T12, TR); cannam@95: T1F = T1m + T1n; cannam@95: T1G = FNMS(KP500000000, T1F, T1E); cannam@95: } cannam@95: { cannam@95: E Ts, T1c, Tn, T1b; cannam@95: { cannam@95: E Tf, Th, Te, Tg; cannam@95: Tf = Rp[WS(rs, 3)]; cannam@95: Th = Rm[WS(rs, 3)]; cannam@95: Te = W[10]; cannam@95: Tg = W[11]; cannam@95: Ti = FMA(Te, Tf, Tg * Th); cannam@95: T1S = FNMS(Tg, Tf, Te * Th); cannam@95: } cannam@95: { cannam@95: E Tp, Tr, To, Tq; cannam@95: Tp = Rp[WS(rs, 1)]; cannam@95: Tr = Rm[WS(rs, 1)]; cannam@95: To = W[2]; cannam@95: Tq = W[3]; cannam@95: Ts = FMA(To, Tp, Tq * Tr); cannam@95: T1c = FNMS(Tq, Tp, To * Tr); cannam@95: } cannam@95: { cannam@95: E Tk, Tm, Tj, Tl; cannam@95: Tk = Rp[WS(rs, 5)]; cannam@95: Tm = Rm[WS(rs, 5)]; cannam@95: Tj = W[18]; cannam@95: Tl = W[19]; cannam@95: Tn = FMA(Tj, Tk, Tl * Tm); cannam@95: T1b = FNMS(Tl, Tk, Tj * Tm); cannam@95: } cannam@95: T1d = KP866025403 * (T1b - T1c); cannam@95: T25 = KP866025403 * (Ts - Tn); cannam@95: Tt = Tn + Ts; cannam@95: T1a = FNMS(KP500000000, Tt, Ti); cannam@95: T1T = T1b + T1c; cannam@95: T26 = FNMS(KP500000000, T1T, T1S); cannam@95: } cannam@95: { cannam@95: E TK, T1i, TF, T1h; cannam@95: { cannam@95: E Tx, Tz, Tw, Ty; cannam@95: Tx = Ip[WS(rs, 1)]; cannam@95: Tz = Im[WS(rs, 1)]; cannam@95: Tw = W[4]; cannam@95: Ty = W[5]; cannam@95: TA = FMA(Tw, Tx, Ty * Tz); cannam@95: T1y = FNMS(Ty, Tx, Tw * Tz); cannam@95: } cannam@95: { cannam@95: E TH, TJ, TG, TI; cannam@95: TH = Ip[WS(rs, 5)]; cannam@95: TJ = Im[WS(rs, 5)]; cannam@95: TG = W[20]; cannam@95: TI = W[21]; cannam@95: TK = FMA(TG, TH, TI * TJ); cannam@95: T1i = FNMS(TI, TH, TG * TJ); cannam@95: } cannam@95: { cannam@95: E TC, TE, TB, TD; cannam@95: TC = Ip[WS(rs, 3)]; cannam@95: TE = Im[WS(rs, 3)]; cannam@95: TB = W[12]; cannam@95: TD = W[13]; cannam@95: TF = FMA(TB, TC, TD * TE); cannam@95: T1h = FNMS(TD, TC, TB * TE); cannam@95: } cannam@95: T1j = KP866025403 * (T1h - T1i); cannam@95: T1B = KP866025403 * (TK - TF); cannam@95: TL = TF + TK; cannam@95: T1g = FNMS(KP500000000, TL, TA); cannam@95: T1z = T1h + T1i; cannam@95: T1A = FNMS(KP500000000, T1z, T1y); cannam@95: } cannam@95: { cannam@95: E Tv, T1N, T1Y, T20, T14, T1Z, T1Q, T1R; cannam@95: { cannam@95: E Td, Tu, T1U, T1X; cannam@95: Td = T1 + Tc; cannam@95: Tu = Ti + Tt; cannam@95: Tv = Td + Tu; cannam@95: T1N = Td - Tu; cannam@95: T1U = T1S + T1T; cannam@95: T1X = T1V + T1W; cannam@95: T1Y = T1U + T1X; cannam@95: T20 = T1X - T1U; cannam@95: } cannam@95: { cannam@95: E TM, T13, T1O, T1P; cannam@95: TM = TA + TL; cannam@95: T13 = TR + T12; cannam@95: T14 = TM + T13; cannam@95: T1Z = TM - T13; cannam@95: T1O = T1y + T1z; cannam@95: T1P = T1E + T1F; cannam@95: T1Q = T1O - T1P; cannam@95: T1R = T1O + T1P; cannam@95: } cannam@95: Rm[WS(rs, 5)] = Tv - T14; cannam@95: Im[WS(rs, 5)] = T1R - T1Y; cannam@95: Rp[0] = Tv + T14; cannam@95: Ip[0] = T1R + T1Y; cannam@95: Rp[WS(rs, 3)] = T1N - T1Q; cannam@95: Ip[WS(rs, 3)] = T1Z + T20; cannam@95: Rm[WS(rs, 2)] = T1N + T1Q; cannam@95: Im[WS(rs, 2)] = T1Z - T20; cannam@95: } cannam@95: { cannam@95: E T1t, T1J, T28, T2a, T1w, T21, T1M, T29; cannam@95: { cannam@95: E T1r, T1s, T24, T27; cannam@95: T1r = T15 + T18; cannam@95: T1s = T1a + T1d; cannam@95: T1t = T1r + T1s; cannam@95: T1J = T1r - T1s; cannam@95: T24 = T22 + T23; cannam@95: T27 = T25 + T26; cannam@95: T28 = T24 - T27; cannam@95: T2a = T27 + T24; cannam@95: } cannam@95: { cannam@95: E T1u, T1v, T1K, T1L; cannam@95: T1u = T1g + T1j; cannam@95: T1v = T1l + T1o; cannam@95: T1w = T1u + T1v; cannam@95: T21 = T1v - T1u; cannam@95: T1K = T1B + T1A; cannam@95: T1L = T1D + T1G; cannam@95: T1M = T1K - T1L; cannam@95: T29 = T1K + T1L; cannam@95: } cannam@95: Rm[WS(rs, 1)] = T1t - T1w; cannam@95: Im[WS(rs, 1)] = T29 - T2a; cannam@95: Rp[WS(rs, 4)] = T1t + T1w; cannam@95: Ip[WS(rs, 4)] = T29 + T2a; cannam@95: Rm[WS(rs, 4)] = T1J - T1M; cannam@95: Im[WS(rs, 4)] = T21 - T28; cannam@95: Rp[WS(rs, 1)] = T1J + T1M; cannam@95: Ip[WS(rs, 1)] = T21 + T28; cannam@95: } cannam@95: { cannam@95: E T1f, T1x, T2e, T2g, T1q, T2f, T1I, T2b; cannam@95: { cannam@95: E T19, T1e, T2c, T2d; cannam@95: T19 = T15 - T18; cannam@95: T1e = T1a - T1d; cannam@95: T1f = T19 + T1e; cannam@95: T1x = T19 - T1e; cannam@95: T2c = T26 - T25; cannam@95: T2d = T23 - T22; cannam@95: T2e = T2c + T2d; cannam@95: T2g = T2d - T2c; cannam@95: } cannam@95: { cannam@95: E T1k, T1p, T1C, T1H; cannam@95: T1k = T1g - T1j; cannam@95: T1p = T1l - T1o; cannam@95: T1q = T1k + T1p; cannam@95: T2f = T1p - T1k; cannam@95: T1C = T1A - T1B; cannam@95: T1H = T1D - T1G; cannam@95: T1I = T1C + T1H; cannam@95: T2b = T1H - T1C; cannam@95: } cannam@95: Rp[WS(rs, 2)] = T1f - T1q; cannam@95: Ip[WS(rs, 2)] = T2b + T2e; cannam@95: Rm[WS(rs, 3)] = T1f + T1q; cannam@95: Im[WS(rs, 3)] = T2b - T2e; cannam@95: Rm[0] = T1x - T1I; cannam@95: Im[0] = T2f - T2g; cannam@95: Rp[WS(rs, 5)] = T1x + T1I; cannam@95: Ip[WS(rs, 5)] = T2f + T2g; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 12}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 12, "hc2cf_12", twinstr, &GENUS, {88, 30, 30, 0} }; cannam@95: cannam@95: void X(codelet_hc2cf_12) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cf_12, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */