/* * Copyright (c) 2003, 2007-8 Matteo Frigo * Copyright (c) 2003, 2007-8 Massachusetts Institute of Technology * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * */ /* This file was automatically generated --- DO NOT EDIT */ /* Generated on Sun Jul 12 06:45:14 EDT 2009 */ #include "codelet-rdft.h" #ifdef HAVE_FMA /* Generated by: ../../../genfft/gen_hc2cdft -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cfdft_12 -include hc2cf.h */ /* * This function contains 142 FP additions, 92 FP multiplications, * (or, 96 additions, 46 multiplications, 46 fused multiply/add), * 71 stack variables, 2 constants, and 48 memory accesses */ #include "hc2cf.h" static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) { DK(KP866025403, +0.866025403784438646763723170752936183471402627); DK(KP500000000, +0.500000000000000000000000000000000000000000000); INT m; 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(rs)) { E T2z, T2M; { E To, T1E, T2H, T1m, T1W, Tl, T1J, T2i, T2K, T1B, T2I, T2e, T19, T2E, T2C; E T27, T1M, Tz, T2B, T1f, T1O, TJ, TT, T1Q; { E T2b, T1s, T1A, T2d; { E T1u, T1z, T1v, T2c, T1i, Te, T1l, Tj, Tf, T1H, T4, T1o, T1, T1r, T9; E T1n, T5; { E T1x, T1y, T1t, Tm, Tn; Tm = Ip[0]; Tn = Im[0]; T1x = Rp[0]; T1y = Rm[0]; T1t = W[0]; T1u = Tm + Tn; To = Tm - Tn; { E Th, Ti, Tb, Tc, Td; Tc = Ip[WS(rs, 4)]; T1z = T1x - T1y; T1E = T1x + T1y; Td = Im[WS(rs, 4)]; T1v = T1t * T1u; Th = Rp[WS(rs, 4)]; T2c = T1t * T1z; T1i = Tc + Td; Te = Tc - Td; Ti = Rm[WS(rs, 4)]; Tb = W[14]; { E T7, T8, T2, T3; T2 = Ip[WS(rs, 2)]; T1l = Th - Ti; Tj = Th + Ti; Tf = Tb * Te; T3 = Im[WS(rs, 2)]; T7 = Rp[WS(rs, 2)]; T1H = Tb * Tj; T8 = Rm[WS(rs, 2)]; T4 = T2 - T3; T1o = T2 + T3; T1 = W[6]; T1r = T7 - T8; T9 = T7 + T8; T1n = W[8]; T5 = T1 * T4; } } } { E T1F, T2a, T1p, T1h, T1k; T1F = T1 * T9; T2a = T1n * T1r; T1p = T1n * T1o; T1h = W[16]; T1k = W[17]; { E T1G, Ta, Tk, T1I, T1q, T1w; { E T6, Tg, T2G, T1j; T6 = W[7]; Tg = W[15]; T2G = T1h * T1l; T1j = T1h * T1i; T1G = FMA(T6, T4, T1F); Ta = FNMS(T6, T9, T5); T2H = FMA(T1k, T1i, T2G); T1m = FNMS(T1k, T1l, T1j); Tk = FNMS(Tg, Tj, Tf); T1I = FMA(Tg, Te, T1H); } T1q = W[9]; T1w = W[1]; T1W = Ta - Tk; Tl = Ta + Tk; T1J = T1G + T1I; T2i = T1I - T1G; T2b = FMA(T1q, T1o, T2a); T1s = FNMS(T1q, T1r, T1p); T1A = FNMS(T1w, T1z, T1v); T2d = FMA(T1w, T1u, T2c); } } } { E T11, Tt, T10, TX, Ty, TZ, T23, T1b, TN, TS, T1e, T1P, TO, T17, TD; E T16, T13, T14, TI, TA; { E Tw, Tx, Tr, Ts, TK; Tr = Ip[WS(rs, 3)]; Ts = Im[WS(rs, 3)]; T2K = T1s - T1A; T1B = T1s + T1A; T2I = T2b + T2d; T2e = T2b - T2d; Tw = Rp[WS(rs, 3)]; T11 = Tr + Ts; Tt = Tr - Ts; Tx = Rm[WS(rs, 3)]; T10 = W[12]; TX = W[13]; { E TL, TY, TM, TQ, TR; TL = Ip[WS(rs, 1)]; Ty = Tw + Tx; TY = Tx - Tw; TM = Im[WS(rs, 1)]; TQ = Rp[WS(rs, 1)]; TR = Rm[WS(rs, 1)]; TZ = TX * TY; T23 = T10 * TY; T1b = TL + TM; TN = TL - TM; TS = TQ + TR; T1e = TQ - TR; } TK = W[2]; { E TG, TH, TB, TC; TB = Ip[WS(rs, 5)]; TC = Im[WS(rs, 5)]; TG = Rp[WS(rs, 5)]; T1P = TK * TS; TO = TK * TN; T17 = TB + TC; TD = TB - TC; TH = Rm[WS(rs, 5)]; T16 = W[20]; T13 = W[21]; T14 = TH - TG; TI = TG + TH; TA = W[18]; } } { E T12, T1N, TE, T18, T24, T26, T25, T15; T12 = FMA(T10, T11, TZ); T15 = T13 * T14; T25 = T16 * T14; T1N = TA * TI; TE = TA * TD; T18 = FMA(T16, T17, T15); T24 = FNMS(TX, T11, T23); T26 = FNMS(T13, T17, T25); { E Tv, T1L, Tu, Tq; Tq = W[10]; T19 = T12 + T18; T2E = T18 - T12; Tv = W[11]; T2C = T24 + T26; T27 = T24 - T26; T1L = Tq * Ty; Tu = Tq * Tt; { E T1d, T2A, T1c, T1a, TF, TP; T1a = W[4]; T1d = W[5]; T1M = FMA(Tv, Tt, T1L); Tz = FNMS(Tv, Ty, Tu); T2A = T1a * T1e; T1c = T1a * T1b; TF = W[19]; TP = W[3]; T2B = FMA(T1d, T1b, T2A); T1f = FNMS(T1d, T1e, T1c); T1O = FMA(TF, TD, T1N); TJ = FNMS(TF, TI, TE); TT = FNMS(TP, TS, TO); T1Q = FMA(TP, TN, T1P); } } } } } { E T2h, T2D, T1Z, T2l, T2J, T22, T2k, T29, T30, T1U, T1V, T1Y, T2Z, T1T; { E T2Y, TW, T2V, T1D, T1K, T1S; { E Tp, T2W, TU, T1R, T2X, T1g, TV, T1C; T2h = FNMS(KP500000000, Tl, To); Tp = Tl + To; T2W = T2C - T2B; T2D = FMA(KP500000000, T2C, T2B); T1Z = TJ - TT; TU = TJ + TT; T1R = T1O + T1Q; T2l = T1Q - T1O; T2J = FNMS(KP500000000, T2I, T2H); T2X = T2H + T2I; T1g = T19 + T1f; T22 = FNMS(KP500000000, T19, T1f); T2k = FNMS(KP500000000, TU, Tz); TV = Tz + TU; T1C = T1m + T1B; T29 = FNMS(KP500000000, T1B, T1m); T2Y = T2W - T2X; T30 = T2W + T2X; TW = Tp - TV; T2V = TV + Tp; T1U = T1g + T1C; T1D = T1g - T1C; T1V = FNMS(KP500000000, T1J, T1E); T1K = T1E + T1J; T1S = T1M + T1R; T1Y = FNMS(KP500000000, T1R, T1M); } Ip[WS(rs, 3)] = KP500000000 * (TW + T1D); Im[WS(rs, 2)] = KP500000000 * (T1D - TW); Im[WS(rs, 5)] = KP500000000 * (T2Y - T2V); T2Z = T1K - T1S; T1T = T1K + T1S; Ip[0] = KP500000000 * (T2V + T2Y); } { E T2v, T1X, T2Q, T2F, T2R, T2L, T2w, T20, T2t, T28, T2p, T2j; Rm[WS(rs, 2)] = KP500000000 * (T2Z + T30); Rp[WS(rs, 3)] = KP500000000 * (T2Z - T30); Rp[0] = KP500000000 * (T1T + T1U); Rm[WS(rs, 5)] = KP500000000 * (T1T - T1U); T2v = FMA(KP866025403, T1W, T1V); T1X = FNMS(KP866025403, T1W, T1V); T2Q = FMA(KP866025403, T2E, T2D); T2F = FNMS(KP866025403, T2E, T2D); T2R = FMA(KP866025403, T2K, T2J); T2L = FNMS(KP866025403, T2K, T2J); T2w = FMA(KP866025403, T1Z, T1Y); T20 = FNMS(KP866025403, T1Z, T1Y); T2t = FMA(KP866025403, T27, T22); T28 = FNMS(KP866025403, T27, T22); T2p = FMA(KP866025403, T2i, T2h); T2j = FNMS(KP866025403, T2i, T2h); { E T2T, T2q, T2s, T2U; { E T21, T2f, T2S, T2n, T2P, T2m, T2o, T2g; T2T = T1X - T20; T21 = T1X + T20; T2q = FMA(KP866025403, T2l, T2k); T2m = FNMS(KP866025403, T2l, T2k); T2s = FMA(KP866025403, T2e, T29); T2f = FNMS(KP866025403, T2e, T29); T2S = T2Q + T2R; T2U = T2R - T2Q; T2n = T2j - T2m; T2P = T2m + T2j; T2o = T2f - T28; T2g = T28 + T2f; Im[WS(rs, 3)] = KP500000000 * (T2S - T2P); Ip[WS(rs, 2)] = KP500000000 * (T2P + T2S); Rm[WS(rs, 3)] = KP500000000 * (T21 + T2g); Rp[WS(rs, 2)] = KP500000000 * (T21 - T2g); Ip[WS(rs, 5)] = KP500000000 * (T2n + T2o); Im[0] = KP500000000 * (T2o - T2n); } { E T2y, T2x, T2N, T2O, T2r, T2u; T2z = T2q + T2p; T2r = T2p - T2q; T2u = T2s - T2t; T2y = T2t + T2s; T2x = T2v + T2w; T2N = T2v - T2w; Rp[WS(rs, 5)] = KP500000000 * (T2T + T2U); Rm[0] = KP500000000 * (T2T - T2U); Im[WS(rs, 4)] = KP500000000 * (T2u - T2r); Ip[WS(rs, 1)] = KP500000000 * (T2r + T2u); T2O = T2L - T2F; T2M = T2F + T2L; Rp[WS(rs, 1)] = KP500000000 * (T2N + T2O); Rm[WS(rs, 4)] = KP500000000 * (T2N - T2O); Rp[WS(rs, 4)] = KP500000000 * (T2x + T2y); Rm[WS(rs, 1)] = KP500000000 * (T2x - T2y); } } } } } Im[WS(rs, 1)] = -(KP500000000 * (T2z + T2M)); Ip[WS(rs, 4)] = KP500000000 * (T2z - T2M); } } static const tw_instr twinstr[] = { {TW_FULL, 1, 12}, {TW_NEXT, 1, 0} }; static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {96, 46, 46, 0} }; void X(codelet_hc2cfdft_12) (planner *p) { X(khc2c_register) (p, hc2cfdft_12, &desc, HC2C_VIA_DFT); } #else /* HAVE_FMA */ /* Generated by: ../../../genfft/gen_hc2cdft -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cfdft_12 -include hc2cf.h */ /* * This function contains 142 FP additions, 76 FP multiplications, * (or, 112 additions, 46 multiplications, 30 fused multiply/add), * 52 stack variables, 3 constants, and 48 memory accesses */ #include "hc2cf.h" static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) { DK(KP250000000, +0.250000000000000000000000000000000000000000000); DK(KP500000000, +0.500000000000000000000000000000000000000000000); DK(KP433012701, +0.433012701892219323381861585376468091735701313); INT m; 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(rs)) { E Tm, T1t, T1d, T2j, Tj, T1Y, T1w, T1G, T1q, T2q, T1U, T2k, Tw, T1y, T17; E T2g, TP, T21, T1B, T1J, T12, T2u, T1P, T2h; { E Tk, Tl, T1k, T1m, T1n, T1o, T4, T1f, T8, T1h, Th, T1c, Td, T1a, T19; E T1b; { E T2, T3, T6, T7; Tk = Ip[0]; Tl = Im[0]; T1k = Tk + Tl; T1m = Rp[0]; T1n = Rm[0]; T1o = T1m - T1n; T2 = Ip[WS(rs, 2)]; T3 = Im[WS(rs, 2)]; T4 = T2 - T3; T1f = T2 + T3; T6 = Rp[WS(rs, 2)]; T7 = Rm[WS(rs, 2)]; T8 = T6 + T7; T1h = T6 - T7; { E Tf, Tg, Tb, Tc; Tf = Rp[WS(rs, 4)]; Tg = Rm[WS(rs, 4)]; Th = Tf + Tg; T1c = Tf - Tg; Tb = Ip[WS(rs, 4)]; Tc = Im[WS(rs, 4)]; Td = Tb - Tc; T1a = Tb + Tc; } } Tm = Tk - Tl; T1t = T1m + T1n; T19 = W[16]; T1b = W[17]; T1d = FNMS(T1b, T1c, T19 * T1a); T2j = FMA(T19, T1c, T1b * T1a); { E T9, T1u, Ti, T1v; { E T1, T5, Ta, Te; T1 = W[6]; T5 = W[7]; T9 = FNMS(T5, T8, T1 * T4); T1u = FMA(T1, T8, T5 * T4); Ta = W[14]; Te = W[15]; Ti = FNMS(Te, Th, Ta * Td); T1v = FMA(Ta, Th, Te * Td); } Tj = T9 + Ti; T1Y = KP433012701 * (T1v - T1u); T1w = T1u + T1v; T1G = KP433012701 * (T9 - Ti); } { E T1i, T1S, T1p, T1T; { E T1e, T1g, T1j, T1l; T1e = W[8]; T1g = W[9]; T1i = FNMS(T1g, T1h, T1e * T1f); T1S = FMA(T1e, T1h, T1g * T1f); T1j = W[0]; T1l = W[1]; T1p = FNMS(T1l, T1o, T1j * T1k); T1T = FMA(T1j, T1o, T1l * T1k); } T1q = T1i + T1p; T2q = KP433012701 * (T1i - T1p); T1U = KP433012701 * (T1S - T1T); T2k = T1S + T1T; } } { E Tr, TT, Tv, TV, TA, TY, TE, T10, TN, T14, TJ, T16; { E Tp, Tq, TC, TD; Tp = Ip[WS(rs, 3)]; Tq = Im[WS(rs, 3)]; Tr = Tp - Tq; TT = Tp + Tq; { E Tt, Tu, Ty, Tz; Tt = Rp[WS(rs, 3)]; Tu = Rm[WS(rs, 3)]; Tv = Tt + Tu; TV = Tt - Tu; Ty = Ip[WS(rs, 5)]; Tz = Im[WS(rs, 5)]; TA = Ty - Tz; TY = Ty + Tz; } TC = Rp[WS(rs, 5)]; TD = Rm[WS(rs, 5)]; TE = TC + TD; T10 = TC - TD; { E TL, TM, TH, TI; TL = Rp[WS(rs, 1)]; TM = Rm[WS(rs, 1)]; TN = TL + TM; T14 = TM - TL; TH = Ip[WS(rs, 1)]; TI = Im[WS(rs, 1)]; TJ = TH - TI; T16 = TH + TI; } } { E To, Ts, T13, T15; To = W[10]; Ts = W[11]; Tw = FNMS(Ts, Tv, To * Tr); T1y = FMA(To, Tv, Ts * Tr); T13 = W[5]; T15 = W[4]; T17 = FMA(T13, T14, T15 * T16); T2g = FNMS(T13, T16, T15 * T14); } { E TF, T1z, TO, T1A; { E Tx, TB, TG, TK; Tx = W[18]; TB = W[19]; TF = FNMS(TB, TE, Tx * TA); T1z = FMA(Tx, TE, TB * TA); TG = W[2]; TK = W[3]; TO = FNMS(TK, TN, TG * TJ); T1A = FMA(TG, TN, TK * TJ); } TP = TF + TO; T21 = KP433012701 * (T1A - T1z); T1B = T1z + T1A; T1J = KP433012701 * (TF - TO); } { E TW, T1O, T11, T1N; { E TS, TU, TX, TZ; TS = W[12]; TU = W[13]; TW = FNMS(TU, TV, TS * TT); T1O = FMA(TS, TV, TU * TT); TX = W[20]; TZ = W[21]; T11 = FNMS(TZ, T10, TX * TY); T1N = FMA(TX, T10, TZ * TY); } T12 = TW + T11; T2u = KP433012701 * (T11 - TW); T1P = KP433012701 * (T1N - T1O); T2h = T1O + T1N; } } { E TR, T2f, T2m, T2o, T1s, T1E, T1D, T2n; { E Tn, TQ, T2i, T2l; Tn = Tj + Tm; TQ = Tw + TP; TR = Tn - TQ; T2f = TQ + Tn; T2i = T2g - T2h; T2l = T2j + T2k; T2m = T2i - T2l; T2o = T2i + T2l; } { E T18, T1r, T1x, T1C; T18 = T12 + T17; T1r = T1d + T1q; T1s = T18 - T1r; T1E = T18 + T1r; T1x = T1t + T1w; T1C = T1y + T1B; T1D = T1x + T1C; T2n = T1x - T1C; } Ip[WS(rs, 3)] = KP500000000 * (TR + T1s); Rp[WS(rs, 3)] = KP500000000 * (T2n - T2o); Im[WS(rs, 2)] = KP500000000 * (T1s - TR); Rm[WS(rs, 2)] = KP500000000 * (T2n + T2o); Rm[WS(rs, 5)] = KP500000000 * (T1D - T1E); Im[WS(rs, 5)] = KP500000000 * (T2m - T2f); Rp[0] = KP500000000 * (T1D + T1E); Ip[0] = KP500000000 * (T2f + T2m); } { E T1H, T2b, T2s, T2B, T2v, T2A, T1K, T2c, T1Q, T29, T1Z, T25, T22, T26, T1V; E T28; { E T1F, T2r, T2t, T1I; T1F = FNMS(KP250000000, T1w, KP500000000 * T1t); T1H = T1F - T1G; T2b = T1F + T1G; T2r = FNMS(KP500000000, T2j, KP250000000 * T2k); T2s = T2q - T2r; T2B = T2q + T2r; T2t = FMA(KP250000000, T2h, KP500000000 * T2g); T2v = T2t - T2u; T2A = T2u + T2t; T1I = FNMS(KP250000000, T1B, KP500000000 * T1y); T1K = T1I - T1J; T2c = T1I + T1J; } { E T1M, T1X, T20, T1R; T1M = FNMS(KP250000000, T12, KP500000000 * T17); T1Q = T1M - T1P; T29 = T1P + T1M; T1X = FNMS(KP250000000, Tj, KP500000000 * Tm); T1Z = T1X - T1Y; T25 = T1Y + T1X; T20 = FNMS(KP250000000, TP, KP500000000 * Tw); T22 = T20 - T21; T26 = T21 + T20; T1R = FNMS(KP250000000, T1q, KP500000000 * T1d); T1V = T1R - T1U; T28 = T1R + T1U; } { E T1L, T1W, T2p, T2w; T1L = T1H + T1K; T1W = T1Q + T1V; Rp[WS(rs, 2)] = T1L - T1W; Rm[WS(rs, 3)] = T1L + T1W; T2p = T22 + T1Z; T2w = T2s - T2v; Ip[WS(rs, 2)] = T2p + T2w; Im[WS(rs, 3)] = T2w - T2p; } { E T23, T24, T2x, T2y; T23 = T1Z - T22; T24 = T1V - T1Q; Ip[WS(rs, 5)] = T23 + T24; Im[0] = T24 - T23; T2x = T1H - T1K; T2y = T2v + T2s; Rm[0] = T2x - T2y; Rp[WS(rs, 5)] = T2x + T2y; } { E T27, T2a, T2z, T2C; T27 = T25 - T26; T2a = T28 - T29; Ip[WS(rs, 1)] = T27 + T2a; Im[WS(rs, 4)] = T2a - T27; T2z = T2b - T2c; T2C = T2A - T2B; Rm[WS(rs, 4)] = T2z - T2C; Rp[WS(rs, 1)] = T2z + T2C; } { E T2d, T2e, T2D, T2E; T2d = T2b + T2c; T2e = T29 + T28; Rm[WS(rs, 1)] = T2d - T2e; Rp[WS(rs, 4)] = T2d + T2e; T2D = T26 + T25; T2E = T2A + T2B; Ip[WS(rs, 4)] = T2D + T2E; Im[WS(rs, 1)] = T2E - T2D; } } } } static const tw_instr twinstr[] = { {TW_FULL, 1, 12}, {TW_NEXT, 1, 0} }; static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {112, 46, 30, 0} }; void X(codelet_hc2cfdft_12) (planner *p) { X(khc2c_register) (p, hc2cfdft_12, &desc, HC2C_VIA_DFT); } #endif /* HAVE_FMA */