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 12 -dit -name hc2cfdft_12 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 142 FP additions, 92 FP multiplications, Chris@10: * (or, 96 additions, 46 multiplications, 46 fused multiply/add), Chris@10: * 71 stack variables, 2 constants, and 48 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: E T2z, T2M; Chris@10: { Chris@10: E To, T1E, T2H, T1m, T1W, Tl, T1J, T2i, T2K, T1B, T2I, T2e, T19, T2E, T2C; Chris@10: E T27, T1M, Tz, T2B, T1f, T1O, TJ, TT, T1Q; Chris@10: { Chris@10: E T2b, T1s, T1A, T2d; Chris@10: { Chris@10: E T1u, T1z, T1v, T2c, T1i, Te, T1l, Tj, Tf, T1H, T4, T1o, T1, T1r, T9; Chris@10: E T1n, T5; Chris@10: { Chris@10: E T1x, T1y, T1t, Tm, Tn; Chris@10: Tm = Ip[0]; Chris@10: Tn = Im[0]; Chris@10: T1x = Rp[0]; Chris@10: T1y = Rm[0]; Chris@10: T1t = W[0]; Chris@10: T1u = Tm + Tn; Chris@10: To = Tm - Tn; Chris@10: { Chris@10: E Th, Ti, Tb, Tc, Td; Chris@10: Tc = Ip[WS(rs, 4)]; Chris@10: T1z = T1x - T1y; Chris@10: T1E = T1x + T1y; Chris@10: Td = Im[WS(rs, 4)]; Chris@10: T1v = T1t * T1u; Chris@10: Th = Rp[WS(rs, 4)]; Chris@10: T2c = T1t * T1z; Chris@10: T1i = Tc + Td; Chris@10: Te = Tc - Td; Chris@10: Ti = Rm[WS(rs, 4)]; Chris@10: Tb = W[14]; Chris@10: { Chris@10: E T7, T8, T2, T3; Chris@10: T2 = Ip[WS(rs, 2)]; Chris@10: T1l = Th - Ti; Chris@10: Tj = Th + Ti; Chris@10: Tf = Tb * Te; Chris@10: T3 = Im[WS(rs, 2)]; Chris@10: T7 = Rp[WS(rs, 2)]; Chris@10: T1H = Tb * Tj; Chris@10: T8 = Rm[WS(rs, 2)]; Chris@10: T4 = T2 - T3; Chris@10: T1o = T2 + T3; Chris@10: T1 = W[6]; Chris@10: T1r = T7 - T8; Chris@10: T9 = T7 + T8; Chris@10: T1n = W[8]; Chris@10: T5 = T1 * T4; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1F, T2a, T1p, T1h, T1k; Chris@10: T1F = T1 * T9; Chris@10: T2a = T1n * T1r; Chris@10: T1p = T1n * T1o; Chris@10: T1h = W[16]; Chris@10: T1k = W[17]; Chris@10: { Chris@10: E T1G, Ta, Tk, T1I, T1q, T1w; Chris@10: { Chris@10: E T6, Tg, T2G, T1j; Chris@10: T6 = W[7]; Chris@10: Tg = W[15]; Chris@10: T2G = T1h * T1l; Chris@10: T1j = T1h * T1i; Chris@10: T1G = FMA(T6, T4, T1F); Chris@10: Ta = FNMS(T6, T9, T5); Chris@10: T2H = FMA(T1k, T1i, T2G); Chris@10: T1m = FNMS(T1k, T1l, T1j); Chris@10: Tk = FNMS(Tg, Tj, Tf); Chris@10: T1I = FMA(Tg, Te, T1H); Chris@10: } Chris@10: T1q = W[9]; Chris@10: T1w = W[1]; Chris@10: T1W = Ta - Tk; Chris@10: Tl = Ta + Tk; Chris@10: T1J = T1G + T1I; Chris@10: T2i = T1I - T1G; Chris@10: T2b = FMA(T1q, T1o, T2a); Chris@10: T1s = FNMS(T1q, T1r, T1p); Chris@10: T1A = FNMS(T1w, T1z, T1v); Chris@10: T2d = FMA(T1w, T1u, T2c); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T11, Tt, T10, TX, Ty, TZ, T23, T1b, TN, TS, T1e, T1P, TO, T17, TD; Chris@10: E T16, T13, T14, TI, TA; Chris@10: { Chris@10: E Tw, Tx, Tr, Ts, TK; Chris@10: Tr = Ip[WS(rs, 3)]; Chris@10: Ts = Im[WS(rs, 3)]; Chris@10: T2K = T1s - T1A; Chris@10: T1B = T1s + T1A; Chris@10: T2I = T2b + T2d; Chris@10: T2e = T2b - T2d; Chris@10: Tw = Rp[WS(rs, 3)]; Chris@10: T11 = Tr + Ts; Chris@10: Tt = Tr - Ts; Chris@10: Tx = Rm[WS(rs, 3)]; Chris@10: T10 = W[12]; Chris@10: TX = W[13]; Chris@10: { Chris@10: E TL, TY, TM, TQ, TR; Chris@10: TL = Ip[WS(rs, 1)]; Chris@10: Ty = Tw + Tx; Chris@10: TY = Tx - Tw; Chris@10: TM = Im[WS(rs, 1)]; Chris@10: TQ = Rp[WS(rs, 1)]; Chris@10: TR = Rm[WS(rs, 1)]; Chris@10: TZ = TX * TY; Chris@10: T23 = T10 * TY; Chris@10: T1b = TL + TM; Chris@10: TN = TL - TM; Chris@10: TS = TQ + TR; Chris@10: T1e = TQ - TR; Chris@10: } Chris@10: TK = W[2]; Chris@10: { Chris@10: E TG, TH, TB, TC; Chris@10: TB = Ip[WS(rs, 5)]; Chris@10: TC = Im[WS(rs, 5)]; Chris@10: TG = Rp[WS(rs, 5)]; Chris@10: T1P = TK * TS; Chris@10: TO = TK * TN; Chris@10: T17 = TB + TC; Chris@10: TD = TB - TC; Chris@10: TH = Rm[WS(rs, 5)]; Chris@10: T16 = W[20]; Chris@10: T13 = W[21]; Chris@10: T14 = TH - TG; Chris@10: TI = TG + TH; Chris@10: TA = W[18]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T12, T1N, TE, T18, T24, T26, T25, T15; Chris@10: T12 = FMA(T10, T11, TZ); Chris@10: T15 = T13 * T14; Chris@10: T25 = T16 * T14; Chris@10: T1N = TA * TI; Chris@10: TE = TA * TD; Chris@10: T18 = FMA(T16, T17, T15); Chris@10: T24 = FNMS(TX, T11, T23); Chris@10: T26 = FNMS(T13, T17, T25); Chris@10: { Chris@10: E Tv, T1L, Tu, Tq; Chris@10: Tq = W[10]; Chris@10: T19 = T12 + T18; Chris@10: T2E = T18 - T12; Chris@10: Tv = W[11]; Chris@10: T2C = T24 + T26; Chris@10: T27 = T24 - T26; Chris@10: T1L = Tq * Ty; Chris@10: Tu = Tq * Tt; Chris@10: { Chris@10: E T1d, T2A, T1c, T1a, TF, TP; Chris@10: T1a = W[4]; Chris@10: T1d = W[5]; Chris@10: T1M = FMA(Tv, Tt, T1L); Chris@10: Tz = FNMS(Tv, Ty, Tu); Chris@10: T2A = T1a * T1e; Chris@10: T1c = T1a * T1b; Chris@10: TF = W[19]; Chris@10: TP = W[3]; Chris@10: T2B = FMA(T1d, T1b, T2A); Chris@10: T1f = FNMS(T1d, T1e, T1c); Chris@10: T1O = FMA(TF, TD, T1N); Chris@10: TJ = FNMS(TF, TI, TE); Chris@10: TT = FNMS(TP, TS, TO); Chris@10: T1Q = FMA(TP, TN, T1P); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2h, T2D, T1Z, T2l, T2J, T22, T2k, T29, T30, T1U, T1V, T1Y, T2Z, T1T; Chris@10: { Chris@10: E T2Y, TW, T2V, T1D, T1K, T1S; Chris@10: { Chris@10: E Tp, T2W, TU, T1R, T2X, T1g, TV, T1C; Chris@10: T2h = FNMS(KP500000000, Tl, To); Chris@10: Tp = Tl + To; Chris@10: T2W = T2C - T2B; Chris@10: T2D = FMA(KP500000000, T2C, T2B); Chris@10: T1Z = TJ - TT; Chris@10: TU = TJ + TT; Chris@10: T1R = T1O + T1Q; Chris@10: T2l = T1Q - T1O; Chris@10: T2J = FNMS(KP500000000, T2I, T2H); Chris@10: T2X = T2H + T2I; Chris@10: T1g = T19 + T1f; Chris@10: T22 = FNMS(KP500000000, T19, T1f); Chris@10: T2k = FNMS(KP500000000, TU, Tz); Chris@10: TV = Tz + TU; Chris@10: T1C = T1m + T1B; Chris@10: T29 = FNMS(KP500000000, T1B, T1m); Chris@10: T2Y = T2W - T2X; Chris@10: T30 = T2W + T2X; Chris@10: TW = Tp - TV; Chris@10: T2V = TV + Tp; Chris@10: T1U = T1g + T1C; Chris@10: T1D = T1g - T1C; Chris@10: T1V = FNMS(KP500000000, T1J, T1E); Chris@10: T1K = T1E + T1J; Chris@10: T1S = T1M + T1R; Chris@10: T1Y = FNMS(KP500000000, T1R, T1M); Chris@10: } Chris@10: Ip[WS(rs, 3)] = KP500000000 * (TW + T1D); Chris@10: Im[WS(rs, 2)] = KP500000000 * (T1D - TW); Chris@10: Im[WS(rs, 5)] = KP500000000 * (T2Y - T2V); Chris@10: T2Z = T1K - T1S; Chris@10: T1T = T1K + T1S; Chris@10: Ip[0] = KP500000000 * (T2V + T2Y); Chris@10: } Chris@10: { Chris@10: E T2v, T1X, T2Q, T2F, T2R, T2L, T2w, T20, T2t, T28, T2p, T2j; Chris@10: Rm[WS(rs, 2)] = KP500000000 * (T2Z + T30); Chris@10: Rp[WS(rs, 3)] = KP500000000 * (T2Z - T30); Chris@10: Rp[0] = KP500000000 * (T1T + T1U); Chris@10: Rm[WS(rs, 5)] = KP500000000 * (T1T - T1U); Chris@10: T2v = FMA(KP866025403, T1W, T1V); Chris@10: T1X = FNMS(KP866025403, T1W, T1V); Chris@10: T2Q = FMA(KP866025403, T2E, T2D); Chris@10: T2F = FNMS(KP866025403, T2E, T2D); Chris@10: T2R = FMA(KP866025403, T2K, T2J); Chris@10: T2L = FNMS(KP866025403, T2K, T2J); Chris@10: T2w = FMA(KP866025403, T1Z, T1Y); Chris@10: T20 = FNMS(KP866025403, T1Z, T1Y); Chris@10: T2t = FMA(KP866025403, T27, T22); Chris@10: T28 = FNMS(KP866025403, T27, T22); Chris@10: T2p = FMA(KP866025403, T2i, T2h); Chris@10: T2j = FNMS(KP866025403, T2i, T2h); Chris@10: { Chris@10: E T2T, T2q, T2s, T2U; Chris@10: { Chris@10: E T21, T2f, T2S, T2n, T2P, T2m, T2o, T2g; Chris@10: T2T = T1X - T20; Chris@10: T21 = T1X + T20; Chris@10: T2q = FMA(KP866025403, T2l, T2k); Chris@10: T2m = FNMS(KP866025403, T2l, T2k); Chris@10: T2s = FMA(KP866025403, T2e, T29); Chris@10: T2f = FNMS(KP866025403, T2e, T29); Chris@10: T2S = T2Q + T2R; Chris@10: T2U = T2R - T2Q; Chris@10: T2n = T2j - T2m; Chris@10: T2P = T2m + T2j; Chris@10: T2o = T2f - T28; Chris@10: T2g = T28 + T2f; Chris@10: Im[WS(rs, 3)] = KP500000000 * (T2S - T2P); Chris@10: Ip[WS(rs, 2)] = KP500000000 * (T2P + T2S); Chris@10: Rm[WS(rs, 3)] = KP500000000 * (T21 + T2g); Chris@10: Rp[WS(rs, 2)] = KP500000000 * (T21 - T2g); Chris@10: Ip[WS(rs, 5)] = KP500000000 * (T2n + T2o); Chris@10: Im[0] = KP500000000 * (T2o - T2n); Chris@10: } Chris@10: { Chris@10: E T2y, T2x, T2N, T2O, T2r, T2u; Chris@10: T2z = T2q + T2p; Chris@10: T2r = T2p - T2q; Chris@10: T2u = T2s - T2t; Chris@10: T2y = T2t + T2s; Chris@10: T2x = T2v + T2w; Chris@10: T2N = T2v - T2w; Chris@10: Rp[WS(rs, 5)] = KP500000000 * (T2T + T2U); Chris@10: Rm[0] = KP500000000 * (T2T - T2U); Chris@10: Im[WS(rs, 4)] = KP500000000 * (T2u - T2r); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (T2r + T2u); Chris@10: T2O = T2L - T2F; Chris@10: T2M = T2F + T2L; Chris@10: Rp[WS(rs, 1)] = KP500000000 * (T2N + T2O); Chris@10: Rm[WS(rs, 4)] = KP500000000 * (T2N - T2O); Chris@10: Rp[WS(rs, 4)] = KP500000000 * (T2x + T2y); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (T2x - T2y); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Im[WS(rs, 1)] = -(KP500000000 * (T2z + T2M)); Chris@10: Ip[WS(rs, 4)] = KP500000000 * (T2z - T2M); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 12}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {96, 46, 46, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_12, &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 12 -dit -name hc2cfdft_12 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 142 FP additions, 76 FP multiplications, Chris@10: * (or, 112 additions, 46 multiplications, 30 fused multiply/add), Chris@10: * 52 stack variables, 3 constants, and 48 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP433012701, +0.433012701892219323381861585376468091735701313); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: E Tm, T1t, T1d, T2j, Tj, T1Y, T1w, T1G, T1q, T2q, T1U, T2k, Tw, T1y, T17; Chris@10: E T2g, TP, T21, T1B, T1J, T12, T2u, T1P, T2h; Chris@10: { Chris@10: E Tk, Tl, T1k, T1m, T1n, T1o, T4, T1f, T8, T1h, Th, T1c, Td, T1a, T19; Chris@10: E T1b; Chris@10: { Chris@10: E T2, T3, T6, T7; Chris@10: Tk = Ip[0]; Chris@10: Tl = Im[0]; Chris@10: T1k = Tk + Tl; Chris@10: T1m = Rp[0]; Chris@10: T1n = Rm[0]; Chris@10: T1o = T1m - T1n; Chris@10: T2 = Ip[WS(rs, 2)]; Chris@10: T3 = Im[WS(rs, 2)]; Chris@10: T4 = T2 - T3; Chris@10: T1f = T2 + T3; Chris@10: T6 = Rp[WS(rs, 2)]; Chris@10: T7 = Rm[WS(rs, 2)]; Chris@10: T8 = T6 + T7; Chris@10: T1h = T6 - T7; Chris@10: { Chris@10: E Tf, Tg, Tb, Tc; Chris@10: Tf = Rp[WS(rs, 4)]; Chris@10: Tg = Rm[WS(rs, 4)]; Chris@10: Th = Tf + Tg; Chris@10: T1c = Tf - Tg; Chris@10: Tb = Ip[WS(rs, 4)]; Chris@10: Tc = Im[WS(rs, 4)]; Chris@10: Td = Tb - Tc; Chris@10: T1a = Tb + Tc; Chris@10: } Chris@10: } Chris@10: Tm = Tk - Tl; Chris@10: T1t = T1m + T1n; Chris@10: T19 = W[16]; Chris@10: T1b = W[17]; Chris@10: T1d = FNMS(T1b, T1c, T19 * T1a); Chris@10: T2j = FMA(T19, T1c, T1b * T1a); Chris@10: { Chris@10: E T9, T1u, Ti, T1v; Chris@10: { Chris@10: E T1, T5, Ta, Te; Chris@10: T1 = W[6]; Chris@10: T5 = W[7]; Chris@10: T9 = FNMS(T5, T8, T1 * T4); Chris@10: T1u = FMA(T1, T8, T5 * T4); Chris@10: Ta = W[14]; Chris@10: Te = W[15]; Chris@10: Ti = FNMS(Te, Th, Ta * Td); Chris@10: T1v = FMA(Ta, Th, Te * Td); Chris@10: } Chris@10: Tj = T9 + Ti; Chris@10: T1Y = KP433012701 * (T1v - T1u); Chris@10: T1w = T1u + T1v; Chris@10: T1G = KP433012701 * (T9 - Ti); Chris@10: } Chris@10: { Chris@10: E T1i, T1S, T1p, T1T; Chris@10: { Chris@10: E T1e, T1g, T1j, T1l; Chris@10: T1e = W[8]; Chris@10: T1g = W[9]; Chris@10: T1i = FNMS(T1g, T1h, T1e * T1f); Chris@10: T1S = FMA(T1e, T1h, T1g * T1f); Chris@10: T1j = W[0]; Chris@10: T1l = W[1]; Chris@10: T1p = FNMS(T1l, T1o, T1j * T1k); Chris@10: T1T = FMA(T1j, T1o, T1l * T1k); Chris@10: } Chris@10: T1q = T1i + T1p; Chris@10: T2q = KP433012701 * (T1i - T1p); Chris@10: T1U = KP433012701 * (T1S - T1T); Chris@10: T2k = T1S + T1T; Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tr, TT, Tv, TV, TA, TY, TE, T10, TN, T14, TJ, T16; Chris@10: { Chris@10: E Tp, Tq, TC, TD; Chris@10: Tp = Ip[WS(rs, 3)]; Chris@10: Tq = Im[WS(rs, 3)]; Chris@10: Tr = Tp - Tq; Chris@10: TT = Tp + Tq; Chris@10: { Chris@10: E Tt, Tu, Ty, Tz; Chris@10: Tt = Rp[WS(rs, 3)]; Chris@10: Tu = Rm[WS(rs, 3)]; Chris@10: Tv = Tt + Tu; Chris@10: TV = Tt - Tu; Chris@10: Ty = Ip[WS(rs, 5)]; Chris@10: Tz = Im[WS(rs, 5)]; Chris@10: TA = Ty - Tz; Chris@10: TY = Ty + Tz; Chris@10: } Chris@10: TC = Rp[WS(rs, 5)]; Chris@10: TD = Rm[WS(rs, 5)]; Chris@10: TE = TC + TD; Chris@10: T10 = TC - TD; Chris@10: { Chris@10: E TL, TM, TH, TI; Chris@10: TL = Rp[WS(rs, 1)]; Chris@10: TM = Rm[WS(rs, 1)]; Chris@10: TN = TL + TM; Chris@10: T14 = TM - TL; Chris@10: TH = Ip[WS(rs, 1)]; Chris@10: TI = Im[WS(rs, 1)]; Chris@10: TJ = TH - TI; Chris@10: T16 = TH + TI; Chris@10: } Chris@10: } Chris@10: { Chris@10: E To, Ts, T13, T15; Chris@10: To = W[10]; Chris@10: Ts = W[11]; Chris@10: Tw = FNMS(Ts, Tv, To * Tr); Chris@10: T1y = FMA(To, Tv, Ts * Tr); Chris@10: T13 = W[5]; Chris@10: T15 = W[4]; Chris@10: T17 = FMA(T13, T14, T15 * T16); Chris@10: T2g = FNMS(T13, T16, T15 * T14); Chris@10: } Chris@10: { Chris@10: E TF, T1z, TO, T1A; Chris@10: { Chris@10: E Tx, TB, TG, TK; Chris@10: Tx = W[18]; Chris@10: TB = W[19]; Chris@10: TF = FNMS(TB, TE, Tx * TA); Chris@10: T1z = FMA(Tx, TE, TB * TA); Chris@10: TG = W[2]; Chris@10: TK = W[3]; Chris@10: TO = FNMS(TK, TN, TG * TJ); Chris@10: T1A = FMA(TG, TN, TK * TJ); Chris@10: } Chris@10: TP = TF + TO; Chris@10: T21 = KP433012701 * (T1A - T1z); Chris@10: T1B = T1z + T1A; Chris@10: T1J = KP433012701 * (TF - TO); Chris@10: } Chris@10: { Chris@10: E TW, T1O, T11, T1N; Chris@10: { Chris@10: E TS, TU, TX, TZ; Chris@10: TS = W[12]; Chris@10: TU = W[13]; Chris@10: TW = FNMS(TU, TV, TS * TT); Chris@10: T1O = FMA(TS, TV, TU * TT); Chris@10: TX = W[20]; Chris@10: TZ = W[21]; Chris@10: T11 = FNMS(TZ, T10, TX * TY); Chris@10: T1N = FMA(TX, T10, TZ * TY); Chris@10: } Chris@10: T12 = TW + T11; Chris@10: T2u = KP433012701 * (T11 - TW); Chris@10: T1P = KP433012701 * (T1N - T1O); Chris@10: T2h = T1O + T1N; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TR, T2f, T2m, T2o, T1s, T1E, T1D, T2n; Chris@10: { Chris@10: E Tn, TQ, T2i, T2l; Chris@10: Tn = Tj + Tm; Chris@10: TQ = Tw + TP; Chris@10: TR = Tn - TQ; Chris@10: T2f = TQ + Tn; Chris@10: T2i = T2g - T2h; Chris@10: T2l = T2j + T2k; Chris@10: T2m = T2i - T2l; Chris@10: T2o = T2i + T2l; Chris@10: } Chris@10: { Chris@10: E T18, T1r, T1x, T1C; Chris@10: T18 = T12 + T17; Chris@10: T1r = T1d + T1q; Chris@10: T1s = T18 - T1r; Chris@10: T1E = T18 + T1r; Chris@10: T1x = T1t + T1w; Chris@10: T1C = T1y + T1B; Chris@10: T1D = T1x + T1C; Chris@10: T2n = T1x - T1C; Chris@10: } Chris@10: Ip[WS(rs, 3)] = KP500000000 * (TR + T1s); Chris@10: Rp[WS(rs, 3)] = KP500000000 * (T2n - T2o); Chris@10: Im[WS(rs, 2)] = KP500000000 * (T1s - TR); Chris@10: Rm[WS(rs, 2)] = KP500000000 * (T2n + T2o); Chris@10: Rm[WS(rs, 5)] = KP500000000 * (T1D - T1E); Chris@10: Im[WS(rs, 5)] = KP500000000 * (T2m - T2f); Chris@10: Rp[0] = KP500000000 * (T1D + T1E); Chris@10: Ip[0] = KP500000000 * (T2f + T2m); Chris@10: } Chris@10: { Chris@10: E T1H, T2b, T2s, T2B, T2v, T2A, T1K, T2c, T1Q, T29, T1Z, T25, T22, T26, T1V; Chris@10: E T28; Chris@10: { Chris@10: E T1F, T2r, T2t, T1I; Chris@10: T1F = FNMS(KP250000000, T1w, KP500000000 * T1t); Chris@10: T1H = T1F - T1G; Chris@10: T2b = T1F + T1G; Chris@10: T2r = FNMS(KP500000000, T2j, KP250000000 * T2k); Chris@10: T2s = T2q - T2r; Chris@10: T2B = T2q + T2r; Chris@10: T2t = FMA(KP250000000, T2h, KP500000000 * T2g); Chris@10: T2v = T2t - T2u; Chris@10: T2A = T2u + T2t; Chris@10: T1I = FNMS(KP250000000, T1B, KP500000000 * T1y); Chris@10: T1K = T1I - T1J; Chris@10: T2c = T1I + T1J; Chris@10: } Chris@10: { Chris@10: E T1M, T1X, T20, T1R; Chris@10: T1M = FNMS(KP250000000, T12, KP500000000 * T17); Chris@10: T1Q = T1M - T1P; Chris@10: T29 = T1P + T1M; Chris@10: T1X = FNMS(KP250000000, Tj, KP500000000 * Tm); Chris@10: T1Z = T1X - T1Y; Chris@10: T25 = T1Y + T1X; Chris@10: T20 = FNMS(KP250000000, TP, KP500000000 * Tw); Chris@10: T22 = T20 - T21; Chris@10: T26 = T21 + T20; Chris@10: T1R = FNMS(KP250000000, T1q, KP500000000 * T1d); Chris@10: T1V = T1R - T1U; Chris@10: T28 = T1R + T1U; Chris@10: } Chris@10: { Chris@10: E T1L, T1W, T2p, T2w; Chris@10: T1L = T1H + T1K; Chris@10: T1W = T1Q + T1V; Chris@10: Rp[WS(rs, 2)] = T1L - T1W; Chris@10: Rm[WS(rs, 3)] = T1L + T1W; Chris@10: T2p = T22 + T1Z; Chris@10: T2w = T2s - T2v; Chris@10: Ip[WS(rs, 2)] = T2p + T2w; Chris@10: Im[WS(rs, 3)] = T2w - T2p; Chris@10: } Chris@10: { Chris@10: E T23, T24, T2x, T2y; Chris@10: T23 = T1Z - T22; Chris@10: T24 = T1V - T1Q; Chris@10: Ip[WS(rs, 5)] = T23 + T24; Chris@10: Im[0] = T24 - T23; Chris@10: T2x = T1H - T1K; Chris@10: T2y = T2v + T2s; Chris@10: Rm[0] = T2x - T2y; Chris@10: Rp[WS(rs, 5)] = T2x + T2y; Chris@10: } Chris@10: { Chris@10: E T27, T2a, T2z, T2C; Chris@10: T27 = T25 - T26; Chris@10: T2a = T28 - T29; Chris@10: Ip[WS(rs, 1)] = T27 + T2a; Chris@10: Im[WS(rs, 4)] = T2a - T27; Chris@10: T2z = T2b - T2c; Chris@10: T2C = T2A - T2B; Chris@10: Rm[WS(rs, 4)] = T2z - T2C; Chris@10: Rp[WS(rs, 1)] = T2z + T2C; Chris@10: } Chris@10: { Chris@10: E T2d, T2e, T2D, T2E; Chris@10: T2d = T2b + T2c; Chris@10: T2e = T29 + T28; Chris@10: Rm[WS(rs, 1)] = T2d - T2e; Chris@10: Rp[WS(rs, 4)] = T2d + T2e; Chris@10: T2D = T26 + T25; Chris@10: T2E = T2A + T2B; Chris@10: Ip[WS(rs, 4)] = T2D + T2E; Chris@10: Im[WS(rs, 1)] = T2E - T2D; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 12}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {112, 46, 30, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_12, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */