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 07:36:09 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 20 -name t2_20 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 276 FP additions, 198 FP multiplications, Chris@10: * (or, 136 additions, 58 multiplications, 140 fused multiply/add), Chris@10: * 142 stack variables, 4 constants, and 80 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t2_20(R *ri, R *ii, 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(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + (mb * 8); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 8, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E T59, T5i, T5k, T5e, T5c, T5d, T5j, T5f; Chris@10: { Chris@10: E T2, Th, Tf, T6, T5, Tl, T1p, T1n, Ti, T3, Tt, Tv, T24, T1f, T1D; Chris@10: E Tb, T1P, Tm, T21, T1b, T7, T1A, Tw, T1H, T13, TA, T1L, T17, T1S, Tq; Chris@10: E T1o, T2g, T1t, T2c, TO, TK; Chris@10: { Chris@10: E T1e, Ta, Tk, Tg; Chris@10: T2 = W[0]; Chris@10: Th = W[3]; Chris@10: Tf = W[2]; Chris@10: T6 = W[5]; Chris@10: T5 = W[1]; Chris@10: Tk = T2 * Th; Chris@10: Tg = T2 * Tf; Chris@10: T1e = Tf * T6; Chris@10: Ta = T2 * T6; Chris@10: Tl = FMA(T5, Tf, Tk); Chris@10: T1p = FNMS(T5, Tf, Tk); Chris@10: T1n = FMA(T5, Th, Tg); Chris@10: Ti = FNMS(T5, Th, Tg); Chris@10: T3 = W[4]; Chris@10: Tt = W[6]; Chris@10: Tv = W[7]; Chris@10: { Chris@10: E Tp, Tj, TN, TJ; Chris@10: Tp = Ti * T6; Chris@10: T24 = FMA(Th, T3, T1e); Chris@10: T1f = FNMS(Th, T3, T1e); Chris@10: T1D = FNMS(T5, T3, Ta); Chris@10: Tb = FMA(T5, T3, Ta); Chris@10: Tj = Ti * T3; Chris@10: { Chris@10: E T1a, T4, Tu, T1G; Chris@10: T1a = Tf * T3; Chris@10: T4 = T2 * T3; Chris@10: Tu = Ti * Tt; Chris@10: T1G = T2 * Tt; Chris@10: { Chris@10: E T12, Tz, T1K, T16; Chris@10: T12 = Tf * Tt; Chris@10: Tz = Ti * Tv; Chris@10: T1K = T2 * Tv; Chris@10: T16 = Tf * Tv; Chris@10: T1P = FNMS(Tl, T6, Tj); Chris@10: Tm = FMA(Tl, T6, Tj); Chris@10: T21 = FNMS(Th, T6, T1a); Chris@10: T1b = FMA(Th, T6, T1a); Chris@10: T7 = FNMS(T5, T6, T4); Chris@10: T1A = FMA(T5, T6, T4); Chris@10: Tw = FMA(Tl, Tv, Tu); Chris@10: T1H = FMA(T5, Tv, T1G); Chris@10: T13 = FMA(Th, Tv, T12); Chris@10: TA = FNMS(Tl, Tt, Tz); Chris@10: T1L = FNMS(T5, Tt, T1K); Chris@10: T17 = FNMS(Th, Tt, T16); Chris@10: T1S = FMA(Tl, T3, Tp); Chris@10: Tq = FNMS(Tl, T3, Tp); Chris@10: } Chris@10: } Chris@10: T1o = T1n * T3; Chris@10: T2g = T1n * Tv; Chris@10: TN = Tm * Tv; Chris@10: TJ = Tm * Tt; Chris@10: T1t = T1n * T6; Chris@10: T2c = T1n * Tt; Chris@10: TO = FNMS(Tq, Tt, TN); Chris@10: TK = FMA(Tq, Tv, TJ); Chris@10: } Chris@10: } Chris@10: { Chris@10: E Te, T2C, T4L, T57, T58, TD, T2H, T4H, T3C, T3Z, T11, T2v, T2P, T3P, T4k; Chris@10: E T4v, T3u, T43, T2r, T2z, T3b, T3T, T4g, T4z, T3n, T42, T20, T2y, T34, T3S; Chris@10: E T4d, T4y, T1c, T19, T1d, T3E, T1w, T2U, T1g, T1j, T1l; Chris@10: { Chris@10: E T2d, T2h, T2k, T1q, T1u, T2n, TL, TI, TM, T3x, TZ, T2N, TP, TS, TU; Chris@10: { Chris@10: E T1, T4K, T8, T9, Tc; Chris@10: T1 = ri[0]; Chris@10: T4K = ii[0]; Chris@10: T8 = ri[WS(rs, 10)]; Chris@10: T2d = FMA(T1p, Tv, T2c); Chris@10: T2h = FNMS(T1p, Tt, T2g); Chris@10: T2k = FMA(T1p, T6, T1o); Chris@10: T1q = FNMS(T1p, T6, T1o); Chris@10: T1u = FMA(T1p, T3, T1t); Chris@10: T2n = FNMS(T1p, T3, T1t); Chris@10: T9 = T7 * T8; Chris@10: Tc = ii[WS(rs, 10)]; Chris@10: { Chris@10: E Tx, Ts, T2F, TC, T2E; Chris@10: { Chris@10: E Tn, Tr, To, T2D, T4J, Ty, TB, Td, T4I; Chris@10: Tn = ri[WS(rs, 5)]; Chris@10: Tr = ii[WS(rs, 5)]; Chris@10: Tx = ri[WS(rs, 15)]; Chris@10: Td = FMA(Tb, Tc, T9); Chris@10: T4I = T7 * Tc; Chris@10: To = Tm * Tn; Chris@10: T2D = Tm * Tr; Chris@10: Te = T1 + Td; Chris@10: T2C = T1 - Td; Chris@10: T4J = FNMS(Tb, T8, T4I); Chris@10: Ty = Tw * Tx; Chris@10: TB = ii[WS(rs, 15)]; Chris@10: Ts = FMA(Tq, Tr, To); Chris@10: T4L = T4J + T4K; Chris@10: T57 = T4K - T4J; Chris@10: T2F = Tw * TB; Chris@10: TC = FMA(TA, TB, Ty); Chris@10: T2E = FNMS(Tq, Tn, T2D); Chris@10: } Chris@10: { Chris@10: E TF, TG, TH, TW, TY, T2G, T3w, TX, T2M; Chris@10: TF = ri[WS(rs, 4)]; Chris@10: T2G = FNMS(TA, Tx, T2F); Chris@10: T58 = Ts - TC; Chris@10: TD = Ts + TC; Chris@10: TG = Ti * TF; Chris@10: T2H = T2E - T2G; Chris@10: T4H = T2E + T2G; Chris@10: TH = ii[WS(rs, 4)]; Chris@10: TW = ri[WS(rs, 19)]; Chris@10: TY = ii[WS(rs, 19)]; Chris@10: TL = ri[WS(rs, 14)]; Chris@10: TI = FMA(Tl, TH, TG); Chris@10: T3w = Ti * TH; Chris@10: TX = Tt * TW; Chris@10: T2M = Tt * TY; Chris@10: TM = TK * TL; Chris@10: T3x = FNMS(Tl, TF, T3w); Chris@10: TZ = FMA(Tv, TY, TX); Chris@10: T2N = FNMS(Tv, TW, T2M); Chris@10: TP = ii[WS(rs, 14)]; Chris@10: TS = ri[WS(rs, 9)]; Chris@10: TU = ii[WS(rs, 9)]; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T27, T26, T28, T3p, T2p, T39, T29, T2e, T2i; Chris@10: { Chris@10: E T22, T23, T25, T2l, T2o, T3o, T2m, T38; Chris@10: { Chris@10: E TR, T2J, T3z, TV, T2L, T4i, T3A; Chris@10: T22 = ri[WS(rs, 12)]; Chris@10: { Chris@10: E TQ, T3y, TT, T2K; Chris@10: TQ = FMA(TO, TP, TM); Chris@10: T3y = TK * TP; Chris@10: TT = T3 * TS; Chris@10: T2K = T3 * TU; Chris@10: TR = TI + TQ; Chris@10: T2J = TI - TQ; Chris@10: T3z = FNMS(TO, TL, T3y); Chris@10: TV = FMA(T6, TU, TT); Chris@10: T2L = FNMS(T6, TS, T2K); Chris@10: T23 = T21 * T22; Chris@10: } Chris@10: T4i = T3x + T3z; Chris@10: T3A = T3x - T3z; Chris@10: { Chris@10: E T10, T3B, T4j, T2O; Chris@10: T10 = TV + TZ; Chris@10: T3B = TV - TZ; Chris@10: T4j = T2L + T2N; Chris@10: T2O = T2L - T2N; Chris@10: T3C = T3A + T3B; Chris@10: T3Z = T3A - T3B; Chris@10: T11 = TR - T10; Chris@10: T2v = TR + T10; Chris@10: T2P = T2J - T2O; Chris@10: T3P = T2J + T2O; Chris@10: T4k = T4i - T4j; Chris@10: T4v = T4i + T4j; Chris@10: T25 = ii[WS(rs, 12)]; Chris@10: } Chris@10: } Chris@10: T2l = ri[WS(rs, 7)]; Chris@10: T2o = ii[WS(rs, 7)]; Chris@10: T27 = ri[WS(rs, 2)]; Chris@10: T26 = FMA(T24, T25, T23); Chris@10: T3o = T21 * T25; Chris@10: T2m = T2k * T2l; Chris@10: T38 = T2k * T2o; Chris@10: T28 = T1n * T27; Chris@10: T3p = FNMS(T24, T22, T3o); Chris@10: T2p = FMA(T2n, T2o, T2m); Chris@10: T39 = FNMS(T2n, T2l, T38); Chris@10: T29 = ii[WS(rs, 2)]; Chris@10: T2e = ri[WS(rs, 17)]; Chris@10: T2i = ii[WS(rs, 17)]; Chris@10: } Chris@10: { Chris@10: E T1I, T1F, T1J, T3i, T1Y, T32, T1M, T1Q, T1T; Chris@10: { Chris@10: E T1B, T1C, T1E, T1V, T1X, T3h, T1W, T31; Chris@10: { Chris@10: E T2b, T35, T3r, T2j, T37, T4e, T3s; Chris@10: T1B = ri[WS(rs, 8)]; Chris@10: { Chris@10: E T2a, T3q, T2f, T36; Chris@10: T2a = FMA(T1p, T29, T28); Chris@10: T3q = T1n * T29; Chris@10: T2f = T2d * T2e; Chris@10: T36 = T2d * T2i; Chris@10: T2b = T26 + T2a; Chris@10: T35 = T26 - T2a; Chris@10: T3r = FNMS(T1p, T27, T3q); Chris@10: T2j = FMA(T2h, T2i, T2f); Chris@10: T37 = FNMS(T2h, T2e, T36); Chris@10: T1C = T1A * T1B; Chris@10: } Chris@10: T4e = T3p + T3r; Chris@10: T3s = T3p - T3r; Chris@10: { Chris@10: E T2q, T3t, T4f, T3a; Chris@10: T2q = T2j + T2p; Chris@10: T3t = T2j - T2p; Chris@10: T4f = T37 + T39; Chris@10: T3a = T37 - T39; Chris@10: T3u = T3s + T3t; Chris@10: T43 = T3s - T3t; Chris@10: T2r = T2b - T2q; Chris@10: T2z = T2b + T2q; Chris@10: T3b = T35 - T3a; Chris@10: T3T = T35 + T3a; Chris@10: T4g = T4e - T4f; Chris@10: T4z = T4e + T4f; Chris@10: T1E = ii[WS(rs, 8)]; Chris@10: } Chris@10: } Chris@10: T1V = ri[WS(rs, 3)]; Chris@10: T1X = ii[WS(rs, 3)]; Chris@10: T1I = ri[WS(rs, 18)]; Chris@10: T1F = FMA(T1D, T1E, T1C); Chris@10: T3h = T1A * T1E; Chris@10: T1W = Tf * T1V; Chris@10: T31 = Tf * T1X; Chris@10: T1J = T1H * T1I; Chris@10: T3i = FNMS(T1D, T1B, T3h); Chris@10: T1Y = FMA(Th, T1X, T1W); Chris@10: T32 = FNMS(Th, T1V, T31); Chris@10: T1M = ii[WS(rs, 18)]; Chris@10: T1Q = ri[WS(rs, 13)]; Chris@10: T1T = ii[WS(rs, 13)]; Chris@10: } Chris@10: { Chris@10: E T14, T15, T18, T1r, T1v, T3D, T1s, T2T; Chris@10: { Chris@10: E T1O, T2Y, T3k, T1U, T30, T4b, T3l; Chris@10: T14 = ri[WS(rs, 16)]; Chris@10: { Chris@10: E T1N, T3j, T1R, T2Z; Chris@10: T1N = FMA(T1L, T1M, T1J); Chris@10: T3j = T1H * T1M; Chris@10: T1R = T1P * T1Q; Chris@10: T2Z = T1P * T1T; Chris@10: T1O = T1F + T1N; Chris@10: T2Y = T1F - T1N; Chris@10: T3k = FNMS(T1L, T1I, T3j); Chris@10: T1U = FMA(T1S, T1T, T1R); Chris@10: T30 = FNMS(T1S, T1Q, T2Z); Chris@10: T15 = T13 * T14; Chris@10: } Chris@10: T4b = T3i + T3k; Chris@10: T3l = T3i - T3k; Chris@10: { Chris@10: E T1Z, T3m, T4c, T33; Chris@10: T1Z = T1U + T1Y; Chris@10: T3m = T1U - T1Y; Chris@10: T4c = T30 + T32; Chris@10: T33 = T30 - T32; Chris@10: T3n = T3l + T3m; Chris@10: T42 = T3l - T3m; Chris@10: T20 = T1O - T1Z; Chris@10: T2y = T1O + T1Z; Chris@10: T34 = T2Y - T33; Chris@10: T3S = T2Y + T33; Chris@10: T4d = T4b - T4c; Chris@10: T4y = T4b + T4c; Chris@10: T18 = ii[WS(rs, 16)]; Chris@10: } Chris@10: } Chris@10: T1r = ri[WS(rs, 11)]; Chris@10: T1v = ii[WS(rs, 11)]; Chris@10: T1c = ri[WS(rs, 6)]; Chris@10: T19 = FMA(T17, T18, T15); Chris@10: T3D = T13 * T18; Chris@10: T1s = T1q * T1r; Chris@10: T2T = T1q * T1v; Chris@10: T1d = T1b * T1c; Chris@10: T3E = FNMS(T17, T14, T3D); Chris@10: T1w = FMA(T1u, T1v, T1s); Chris@10: T2U = FNMS(T1u, T1r, T2T); Chris@10: T1g = ii[WS(rs, 6)]; Chris@10: T1j = ri[WS(rs, 1)]; Chris@10: T1l = ii[WS(rs, 1)]; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3J, T40, T2W, T3Q, T4M, T4E, T4F, T4U, T4S; Chris@10: { Chris@10: E T4X, T2u, T2w, T4w, T4W, T4r, T4p, T54, T56, T4V, T4a, T4q; Chris@10: { Chris@10: E T4h, TE, T4n, T53, T1z, T2s, T52; Chris@10: { Chris@10: E T1i, T2Q, T3G, T1m, T2S, T4l, T3H; Chris@10: T4h = T4d - T4g; Chris@10: T4X = T4d + T4g; Chris@10: { Chris@10: E T1h, T3F, T1k, T2R; Chris@10: T1h = FMA(T1f, T1g, T1d); Chris@10: T3F = T1b * T1g; Chris@10: T1k = T2 * T1j; Chris@10: T2R = T2 * T1l; Chris@10: T1i = T19 + T1h; Chris@10: T2Q = T19 - T1h; Chris@10: T3G = FNMS(T1f, T1c, T3F); Chris@10: T1m = FMA(T5, T1l, T1k); Chris@10: T2S = FNMS(T5, T1j, T2R); Chris@10: } Chris@10: TE = Te - TD; Chris@10: T2u = Te + TD; Chris@10: T4l = T3E + T3G; Chris@10: T3H = T3E - T3G; Chris@10: { Chris@10: E T1x, T3I, T4m, T2V, T1y; Chris@10: T1x = T1m + T1w; Chris@10: T3I = T1m - T1w; Chris@10: T4m = T2S + T2U; Chris@10: T2V = T2S - T2U; Chris@10: T3J = T3H + T3I; Chris@10: T40 = T3H - T3I; Chris@10: T1y = T1i - T1x; Chris@10: T2w = T1i + T1x; Chris@10: T2W = T2Q - T2V; Chris@10: T3Q = T2Q + T2V; Chris@10: T4n = T4l - T4m; Chris@10: T4w = T4l + T4m; Chris@10: T53 = T11 - T1y; Chris@10: T1z = T11 + T1y; Chris@10: T2s = T20 + T2r; Chris@10: T52 = T20 - T2r; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T49, T48, T4o, T2t; Chris@10: T4o = T4k - T4n; Chris@10: T4W = T4k + T4n; Chris@10: T49 = T1z - T2s; Chris@10: T2t = T1z + T2s; Chris@10: T4r = FMA(KP618033988, T4h, T4o); Chris@10: T4p = FNMS(KP618033988, T4o, T4h); Chris@10: T54 = FNMS(KP618033988, T53, T52); Chris@10: T56 = FMA(KP618033988, T52, T53); Chris@10: ri[WS(rs, 10)] = TE + T2t; Chris@10: T48 = FNMS(KP250000000, T2t, TE); Chris@10: T4V = T4L - T4H; Chris@10: T4M = T4H + T4L; Chris@10: T4a = FNMS(KP559016994, T49, T48); Chris@10: T4q = FMA(KP559016994, T49, T48); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2x, T4Q, T4B, T4D, T4R, T2A, T51, T55; Chris@10: { Chris@10: E T4x, T50, T4Y, T4A, T4Z; Chris@10: T4E = T4v + T4w; Chris@10: T4x = T4v - T4w; Chris@10: ri[WS(rs, 18)] = FMA(KP951056516, T4p, T4a); Chris@10: ri[WS(rs, 2)] = FNMS(KP951056516, T4p, T4a); Chris@10: ri[WS(rs, 6)] = FMA(KP951056516, T4r, T4q); Chris@10: ri[WS(rs, 14)] = FNMS(KP951056516, T4r, T4q); Chris@10: T50 = T4W - T4X; Chris@10: T4Y = T4W + T4X; Chris@10: T4A = T4y - T4z; Chris@10: T4F = T4y + T4z; Chris@10: T2x = T2v + T2w; Chris@10: T4Q = T2v - T2w; Chris@10: ii[WS(rs, 10)] = T4Y + T4V; Chris@10: T4Z = FNMS(KP250000000, T4Y, T4V); Chris@10: T4B = FMA(KP618033988, T4A, T4x); Chris@10: T4D = FNMS(KP618033988, T4x, T4A); Chris@10: T4R = T2y - T2z; Chris@10: T2A = T2y + T2z; Chris@10: T51 = FNMS(KP559016994, T50, T4Z); Chris@10: T55 = FMA(KP559016994, T50, T4Z); Chris@10: } Chris@10: { Chris@10: E T4t, T4s, T2B, T4u, T4C; Chris@10: T2B = T2x + T2A; Chris@10: T4t = T2x - T2A; Chris@10: ii[WS(rs, 18)] = FNMS(KP951056516, T54, T51); Chris@10: ii[WS(rs, 2)] = FMA(KP951056516, T54, T51); Chris@10: ii[WS(rs, 14)] = FMA(KP951056516, T56, T55); Chris@10: ii[WS(rs, 6)] = FNMS(KP951056516, T56, T55); Chris@10: ri[0] = T2u + T2B; Chris@10: T4s = FNMS(KP250000000, T2B, T2u); Chris@10: T4u = FMA(KP559016994, T4t, T4s); Chris@10: T4C = FNMS(KP559016994, T4t, T4s); Chris@10: T4U = FNMS(KP618033988, T4Q, T4R); Chris@10: T4S = FMA(KP618033988, T4R, T4Q); Chris@10: ri[WS(rs, 16)] = FMA(KP951056516, T4B, T4u); Chris@10: ri[WS(rs, 4)] = FNMS(KP951056516, T4B, T4u); Chris@10: ri[WS(rs, 8)] = FMA(KP951056516, T4D, T4C); Chris@10: ri[WS(rs, 12)] = FNMS(KP951056516, T4D, T4C); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3O, T5u, T5w, T5l, T5q, T5o; Chris@10: { Chris@10: E T5n, T5m, T2I, T4O, T3N, T3L, T2X, T5t, T4N, T5s, T3c, T3v, T3K, T4G; Chris@10: T5n = T3n + T3u; Chris@10: T3v = T3n - T3u; Chris@10: T3K = T3C - T3J; Chris@10: T5m = T3C + T3J; Chris@10: T3O = T2C + T2H; Chris@10: T2I = T2C - T2H; Chris@10: T4O = T4E - T4F; Chris@10: T4G = T4E + T4F; Chris@10: T3N = FMA(KP618033988, T3v, T3K); Chris@10: T3L = FNMS(KP618033988, T3K, T3v); Chris@10: T2X = T2P + T2W; Chris@10: T5t = T2P - T2W; Chris@10: ii[0] = T4G + T4M; Chris@10: T4N = FNMS(KP250000000, T4G, T4M); Chris@10: T5s = T34 - T3b; Chris@10: T3c = T34 + T3b; Chris@10: { Chris@10: E T3f, T3e, T4P, T4T, T3d, T3M, T3g; Chris@10: T4P = FMA(KP559016994, T4O, T4N); Chris@10: T4T = FNMS(KP559016994, T4O, T4N); Chris@10: T3f = T2X - T3c; Chris@10: T3d = T2X + T3c; Chris@10: ii[WS(rs, 16)] = FNMS(KP951056516, T4S, T4P); Chris@10: ii[WS(rs, 4)] = FMA(KP951056516, T4S, T4P); Chris@10: ii[WS(rs, 12)] = FMA(KP951056516, T4U, T4T); Chris@10: ii[WS(rs, 8)] = FNMS(KP951056516, T4U, T4T); Chris@10: ri[WS(rs, 15)] = T2I + T3d; Chris@10: T3e = FNMS(KP250000000, T3d, T2I); Chris@10: T5u = FNMS(KP618033988, T5t, T5s); Chris@10: T5w = FMA(KP618033988, T5s, T5t); Chris@10: T5l = T58 + T57; Chris@10: T59 = T57 - T58; Chris@10: T3M = FMA(KP559016994, T3f, T3e); Chris@10: T3g = FNMS(KP559016994, T3f, T3e); Chris@10: ri[WS(rs, 7)] = FNMS(KP951056516, T3L, T3g); Chris@10: ri[WS(rs, 3)] = FMA(KP951056516, T3L, T3g); Chris@10: ri[WS(rs, 19)] = FNMS(KP951056516, T3N, T3M); Chris@10: ri[WS(rs, 11)] = FMA(KP951056516, T3N, T3M); Chris@10: T5q = T5m - T5n; Chris@10: T5o = T5m + T5n; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T5a, T5b, T47, T45, T5g, T5h, T3V, T3X, T41, T44, T5p, T3W, T46, T3Y; Chris@10: T5a = T3Z + T40; Chris@10: T41 = T3Z - T40; Chris@10: T44 = T42 - T43; Chris@10: T5b = T42 + T43; Chris@10: ii[WS(rs, 15)] = T5o + T5l; Chris@10: T5p = FNMS(KP250000000, T5o, T5l); Chris@10: T47 = FNMS(KP618033988, T41, T44); Chris@10: T45 = FMA(KP618033988, T44, T41); Chris@10: { Chris@10: E T5r, T5v, T3R, T3U; Chris@10: T5r = FNMS(KP559016994, T5q, T5p); Chris@10: T5v = FMA(KP559016994, T5q, T5p); Chris@10: T3R = T3P + T3Q; Chris@10: T5g = T3P - T3Q; Chris@10: T5h = T3S - T3T; Chris@10: T3U = T3S + T3T; Chris@10: ii[WS(rs, 7)] = FMA(KP951056516, T5u, T5r); Chris@10: ii[WS(rs, 3)] = FNMS(KP951056516, T5u, T5r); Chris@10: ii[WS(rs, 19)] = FMA(KP951056516, T5w, T5v); Chris@10: ii[WS(rs, 11)] = FNMS(KP951056516, T5w, T5v); Chris@10: T3V = T3R + T3U; Chris@10: T3X = T3R - T3U; Chris@10: } Chris@10: ri[WS(rs, 5)] = T3O + T3V; Chris@10: T3W = FNMS(KP250000000, T3V, T3O); Chris@10: T5i = FMA(KP618033988, T5h, T5g); Chris@10: T5k = FNMS(KP618033988, T5g, T5h); Chris@10: T46 = FNMS(KP559016994, T3X, T3W); Chris@10: T3Y = FMA(KP559016994, T3X, T3W); Chris@10: ri[WS(rs, 9)] = FNMS(KP951056516, T45, T3Y); Chris@10: ri[WS(rs, 1)] = FMA(KP951056516, T45, T3Y); Chris@10: ri[WS(rs, 17)] = FNMS(KP951056516, T47, T46); Chris@10: ri[WS(rs, 13)] = FMA(KP951056516, T47, T46); Chris@10: T5e = T5a - T5b; Chris@10: T5c = T5a + T5b; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: ii[WS(rs, 5)] = T5c + T59; Chris@10: T5d = FNMS(KP250000000, T5c, T59); Chris@10: T5j = FNMS(KP559016994, T5e, T5d); Chris@10: T5f = FMA(KP559016994, T5e, T5d); Chris@10: ii[WS(rs, 9)] = FMA(KP951056516, T5i, T5f); Chris@10: ii[WS(rs, 1)] = FNMS(KP951056516, T5i, T5f); Chris@10: ii[WS(rs, 17)] = FMA(KP951056516, T5k, T5j); Chris@10: ii[WS(rs, 13)] = FNMS(KP951056516, T5k, T5j); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 0, 1}, Chris@10: {TW_CEXP, 0, 3}, Chris@10: {TW_CEXP, 0, 9}, Chris@10: {TW_CEXP, 0, 19}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, "t2_20", twinstr, &GENUS, {136, 58, 140, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t2_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2_20, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 20 -name t2_20 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 276 FP additions, 164 FP multiplications, Chris@10: * (or, 204 additions, 92 multiplications, 72 fused multiply/add), Chris@10: * 123 stack variables, 4 constants, and 80 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t2_20(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + (mb * 8); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 8, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E T2, T5, Tg, Ti, Tk, To, T1h, T1f, T6, T3, T8, T14, T1Q, Tc, T1O; Chris@10: E T1v, T18, T1t, T1n, T24, T1j, T22, Tq, Tu, T1E, T1G, Tx, Ty, Tz, TJ; Chris@10: E T1Z, TB, T1X, T1A, TZ, TL, T1y, TX; Chris@10: { Chris@10: E T7, T16, Ta, T13, T4, T17, Tb, T12; Chris@10: { Chris@10: E Th, Tn, Tj, Tm; Chris@10: T2 = W[0]; Chris@10: T5 = W[1]; Chris@10: Tg = W[2]; Chris@10: Ti = W[3]; Chris@10: Th = T2 * Tg; Chris@10: Tn = T5 * Tg; Chris@10: Tj = T5 * Ti; Chris@10: Tm = T2 * Ti; Chris@10: Tk = Th - Tj; Chris@10: To = Tm + Tn; Chris@10: T1h = Tm - Tn; Chris@10: T1f = Th + Tj; Chris@10: T6 = W[5]; Chris@10: T7 = T5 * T6; Chris@10: T16 = Tg * T6; Chris@10: Ta = T2 * T6; Chris@10: T13 = Ti * T6; Chris@10: T3 = W[4]; Chris@10: T4 = T2 * T3; Chris@10: T17 = Ti * T3; Chris@10: Tb = T5 * T3; Chris@10: T12 = Tg * T3; Chris@10: } Chris@10: T8 = T4 - T7; Chris@10: T14 = T12 + T13; Chris@10: T1Q = T16 + T17; Chris@10: Tc = Ta + Tb; Chris@10: T1O = T12 - T13; Chris@10: T1v = Ta - Tb; Chris@10: T18 = T16 - T17; Chris@10: T1t = T4 + T7; Chris@10: { Chris@10: E T1l, T1m, T1g, T1i; Chris@10: T1l = T1f * T6; Chris@10: T1m = T1h * T3; Chris@10: T1n = T1l + T1m; Chris@10: T24 = T1l - T1m; Chris@10: T1g = T1f * T3; Chris@10: T1i = T1h * T6; Chris@10: T1j = T1g - T1i; Chris@10: T22 = T1g + T1i; Chris@10: { Chris@10: E Tl, Tp, Ts, Tt; Chris@10: Tl = Tk * T3; Chris@10: Tp = To * T6; Chris@10: Tq = Tl + Tp; Chris@10: Ts = Tk * T6; Chris@10: Tt = To * T3; Chris@10: Tu = Ts - Tt; Chris@10: T1E = Tl - Tp; Chris@10: T1G = Ts + Tt; Chris@10: Tx = W[6]; Chris@10: Ty = W[7]; Chris@10: Tz = FMA(Tk, Tx, To * Ty); Chris@10: TJ = FMA(Tq, Tx, Tu * Ty); Chris@10: T1Z = FNMS(T1h, Tx, T1f * Ty); Chris@10: TB = FNMS(To, Tx, Tk * Ty); Chris@10: T1X = FMA(T1f, Tx, T1h * Ty); Chris@10: T1A = FNMS(T5, Tx, T2 * Ty); Chris@10: TZ = FNMS(Ti, Tx, Tg * Ty); Chris@10: TL = FNMS(Tu, Tx, Tq * Ty); Chris@10: T1y = FMA(T2, Tx, T5 * Ty); Chris@10: TX = FMA(Tg, Tx, Ti * Ty); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E TF, T2b, T4A, T4J, T2K, T3r, T4a, T4m, T1N, T28, T29, T3C, T3F, T4o, T3X; Chris@10: E T3Y, T44, T2f, T2g, T2h, T2n, T2s, T4L, T3g, T3h, T4w, T3n, T3o, T3p, T30; Chris@10: E T35, T36, TW, T1r, T1s, T3J, T3M, T4n, T3U, T3V, T43, T2c, T2d, T2e, T2y; Chris@10: E T2D, T4K, T3d, T3e, T4v, T3k, T3l, T3m, T2P, T2U, T2V; Chris@10: { Chris@10: E T1, T48, Te, T47, Tw, T2H, TD, T2I, T9, Td; Chris@10: T1 = ri[0]; Chris@10: T48 = ii[0]; Chris@10: T9 = ri[WS(rs, 10)]; Chris@10: Td = ii[WS(rs, 10)]; Chris@10: Te = FMA(T8, T9, Tc * Td); Chris@10: T47 = FNMS(Tc, T9, T8 * Td); Chris@10: { Chris@10: E Tr, Tv, TA, TC; Chris@10: Tr = ri[WS(rs, 5)]; Chris@10: Tv = ii[WS(rs, 5)]; Chris@10: Tw = FMA(Tq, Tr, Tu * Tv); Chris@10: T2H = FNMS(Tu, Tr, Tq * Tv); Chris@10: TA = ri[WS(rs, 15)]; Chris@10: TC = ii[WS(rs, 15)]; Chris@10: TD = FMA(Tz, TA, TB * TC); Chris@10: T2I = FNMS(TB, TA, Tz * TC); Chris@10: } Chris@10: { Chris@10: E Tf, TE, T4y, T4z; Chris@10: Tf = T1 + Te; Chris@10: TE = Tw + TD; Chris@10: TF = Tf - TE; Chris@10: T2b = Tf + TE; Chris@10: T4y = T48 - T47; Chris@10: T4z = Tw - TD; Chris@10: T4A = T4y - T4z; Chris@10: T4J = T4z + T4y; Chris@10: } Chris@10: { Chris@10: E T2G, T2J, T46, T49; Chris@10: T2G = T1 - Te; Chris@10: T2J = T2H - T2I; Chris@10: T2K = T2G - T2J; Chris@10: T3r = T2G + T2J; Chris@10: T46 = T2H + T2I; Chris@10: T49 = T47 + T48; Chris@10: T4a = T46 + T49; Chris@10: T4m = T49 - T46; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1D, T3A, T2l, T2W, T27, T3E, T2r, T34, T1M, T3B, T2m, T2Z, T1W, T3D, T2q; Chris@10: E T31; Chris@10: { Chris@10: E T1x, T2j, T1C, T2k; Chris@10: { Chris@10: E T1u, T1w, T1z, T1B; Chris@10: T1u = ri[WS(rs, 8)]; Chris@10: T1w = ii[WS(rs, 8)]; Chris@10: T1x = FMA(T1t, T1u, T1v * T1w); Chris@10: T2j = FNMS(T1v, T1u, T1t * T1w); Chris@10: T1z = ri[WS(rs, 18)]; Chris@10: T1B = ii[WS(rs, 18)]; Chris@10: T1C = FMA(T1y, T1z, T1A * T1B); Chris@10: T2k = FNMS(T1A, T1z, T1y * T1B); Chris@10: } Chris@10: T1D = T1x + T1C; Chris@10: T3A = T2j + T2k; Chris@10: T2l = T2j - T2k; Chris@10: T2W = T1x - T1C; Chris@10: } Chris@10: { Chris@10: E T21, T32, T26, T33; Chris@10: { Chris@10: E T1Y, T20, T23, T25; Chris@10: T1Y = ri[WS(rs, 17)]; Chris@10: T20 = ii[WS(rs, 17)]; Chris@10: T21 = FMA(T1X, T1Y, T1Z * T20); Chris@10: T32 = FNMS(T1Z, T1Y, T1X * T20); Chris@10: T23 = ri[WS(rs, 7)]; Chris@10: T25 = ii[WS(rs, 7)]; Chris@10: T26 = FMA(T22, T23, T24 * T25); Chris@10: T33 = FNMS(T24, T23, T22 * T25); Chris@10: } Chris@10: T27 = T21 + T26; Chris@10: T3E = T32 + T33; Chris@10: T2r = T21 - T26; Chris@10: T34 = T32 - T33; Chris@10: } Chris@10: { Chris@10: E T1I, T2X, T1L, T2Y; Chris@10: { Chris@10: E T1F, T1H, T1J, T1K; Chris@10: T1F = ri[WS(rs, 13)]; Chris@10: T1H = ii[WS(rs, 13)]; Chris@10: T1I = FMA(T1E, T1F, T1G * T1H); Chris@10: T2X = FNMS(T1G, T1F, T1E * T1H); Chris@10: T1J = ri[WS(rs, 3)]; Chris@10: T1K = ii[WS(rs, 3)]; Chris@10: T1L = FMA(Tg, T1J, Ti * T1K); Chris@10: T2Y = FNMS(Ti, T1J, Tg * T1K); Chris@10: } Chris@10: T1M = T1I + T1L; Chris@10: T3B = T2X + T2Y; Chris@10: T2m = T1I - T1L; Chris@10: T2Z = T2X - T2Y; Chris@10: } Chris@10: { Chris@10: E T1S, T2o, T1V, T2p; Chris@10: { Chris@10: E T1P, T1R, T1T, T1U; Chris@10: T1P = ri[WS(rs, 12)]; Chris@10: T1R = ii[WS(rs, 12)]; Chris@10: T1S = FMA(T1O, T1P, T1Q * T1R); Chris@10: T2o = FNMS(T1Q, T1P, T1O * T1R); Chris@10: T1T = ri[WS(rs, 2)]; Chris@10: T1U = ii[WS(rs, 2)]; Chris@10: T1V = FMA(T1f, T1T, T1h * T1U); Chris@10: T2p = FNMS(T1h, T1T, T1f * T1U); Chris@10: } Chris@10: T1W = T1S + T1V; Chris@10: T3D = T2o + T2p; Chris@10: T2q = T2o - T2p; Chris@10: T31 = T1S - T1V; Chris@10: } Chris@10: T1N = T1D - T1M; Chris@10: T28 = T1W - T27; Chris@10: T29 = T1N + T28; Chris@10: T3C = T3A - T3B; Chris@10: T3F = T3D - T3E; Chris@10: T4o = T3C + T3F; Chris@10: T3X = T3A + T3B; Chris@10: T3Y = T3D + T3E; Chris@10: T44 = T3X + T3Y; Chris@10: T2f = T1D + T1M; Chris@10: T2g = T1W + T27; Chris@10: T2h = T2f + T2g; Chris@10: T2n = T2l + T2m; Chris@10: T2s = T2q + T2r; Chris@10: T4L = T2n + T2s; Chris@10: T3g = T2l - T2m; Chris@10: T3h = T2q - T2r; Chris@10: T4w = T3g + T3h; Chris@10: T3n = T2W + T2Z; Chris@10: T3o = T31 + T34; Chris@10: T3p = T3n + T3o; Chris@10: T30 = T2W - T2Z; Chris@10: T35 = T31 - T34; Chris@10: T36 = T30 + T35; Chris@10: } Chris@10: { Chris@10: E TO, T3H, T2w, T2L, T1q, T3L, T2C, T2T, TV, T3I, T2x, T2O, T1b, T3K, T2B; Chris@10: E T2Q; Chris@10: { Chris@10: E TI, T2u, TN, T2v; Chris@10: { Chris@10: E TG, TH, TK, TM; Chris@10: TG = ri[WS(rs, 4)]; Chris@10: TH = ii[WS(rs, 4)]; Chris@10: TI = FMA(Tk, TG, To * TH); Chris@10: T2u = FNMS(To, TG, Tk * TH); Chris@10: TK = ri[WS(rs, 14)]; Chris@10: TM = ii[WS(rs, 14)]; Chris@10: TN = FMA(TJ, TK, TL * TM); Chris@10: T2v = FNMS(TL, TK, TJ * TM); Chris@10: } Chris@10: TO = TI + TN; Chris@10: T3H = T2u + T2v; Chris@10: T2w = T2u - T2v; Chris@10: T2L = TI - TN; Chris@10: } Chris@10: { Chris@10: E T1e, T2R, T1p, T2S; Chris@10: { Chris@10: E T1c, T1d, T1k, T1o; Chris@10: T1c = ri[WS(rs, 1)]; Chris@10: T1d = ii[WS(rs, 1)]; Chris@10: T1e = FMA(T2, T1c, T5 * T1d); Chris@10: T2R = FNMS(T5, T1c, T2 * T1d); Chris@10: T1k = ri[WS(rs, 11)]; Chris@10: T1o = ii[WS(rs, 11)]; Chris@10: T1p = FMA(T1j, T1k, T1n * T1o); Chris@10: T2S = FNMS(T1n, T1k, T1j * T1o); Chris@10: } Chris@10: T1q = T1e + T1p; Chris@10: T3L = T2R + T2S; Chris@10: T2C = T1e - T1p; Chris@10: T2T = T2R - T2S; Chris@10: } Chris@10: { Chris@10: E TR, T2M, TU, T2N; Chris@10: { Chris@10: E TP, TQ, TS, TT; Chris@10: TP = ri[WS(rs, 9)]; Chris@10: TQ = ii[WS(rs, 9)]; Chris@10: TR = FMA(T3, TP, T6 * TQ); Chris@10: T2M = FNMS(T6, TP, T3 * TQ); Chris@10: TS = ri[WS(rs, 19)]; Chris@10: TT = ii[WS(rs, 19)]; Chris@10: TU = FMA(Tx, TS, Ty * TT); Chris@10: T2N = FNMS(Ty, TS, Tx * TT); Chris@10: } Chris@10: TV = TR + TU; Chris@10: T3I = T2M + T2N; Chris@10: T2x = TR - TU; Chris@10: T2O = T2M - T2N; Chris@10: } Chris@10: { Chris@10: E T11, T2z, T1a, T2A; Chris@10: { Chris@10: E TY, T10, T15, T19; Chris@10: TY = ri[WS(rs, 16)]; Chris@10: T10 = ii[WS(rs, 16)]; Chris@10: T11 = FMA(TX, TY, TZ * T10); Chris@10: T2z = FNMS(TZ, TY, TX * T10); Chris@10: T15 = ri[WS(rs, 6)]; Chris@10: T19 = ii[WS(rs, 6)]; Chris@10: T1a = FMA(T14, T15, T18 * T19); Chris@10: T2A = FNMS(T18, T15, T14 * T19); Chris@10: } Chris@10: T1b = T11 + T1a; Chris@10: T3K = T2z + T2A; Chris@10: T2B = T2z - T2A; Chris@10: T2Q = T11 - T1a; Chris@10: } Chris@10: TW = TO - TV; Chris@10: T1r = T1b - T1q; Chris@10: T1s = TW + T1r; Chris@10: T3J = T3H - T3I; Chris@10: T3M = T3K - T3L; Chris@10: T4n = T3J + T3M; Chris@10: T3U = T3H + T3I; Chris@10: T3V = T3K + T3L; Chris@10: T43 = T3U + T3V; Chris@10: T2c = TO + TV; Chris@10: T2d = T1b + T1q; Chris@10: T2e = T2c + T2d; Chris@10: T2y = T2w + T2x; Chris@10: T2D = T2B + T2C; Chris@10: T4K = T2y + T2D; Chris@10: T3d = T2w - T2x; Chris@10: T3e = T2B - T2C; Chris@10: T4v = T3d + T3e; Chris@10: T3k = T2L + T2O; Chris@10: T3l = T2Q + T2T; Chris@10: T3m = T3k + T3l; Chris@10: T2P = T2L - T2O; Chris@10: T2U = T2Q - T2T; Chris@10: T2V = T2P + T2U; Chris@10: } Chris@10: { Chris@10: E T3y, T2a, T3x, T3O, T3Q, T3G, T3N, T3P, T3z; Chris@10: T3y = KP559016994 * (T1s - T29); Chris@10: T2a = T1s + T29; Chris@10: T3x = FNMS(KP250000000, T2a, TF); Chris@10: T3G = T3C - T3F; Chris@10: T3N = T3J - T3M; Chris@10: T3O = FNMS(KP587785252, T3N, KP951056516 * T3G); Chris@10: T3Q = FMA(KP951056516, T3N, KP587785252 * T3G); Chris@10: ri[WS(rs, 10)] = TF + T2a; Chris@10: T3P = T3y + T3x; Chris@10: ri[WS(rs, 14)] = T3P - T3Q; Chris@10: ri[WS(rs, 6)] = T3P + T3Q; Chris@10: T3z = T3x - T3y; Chris@10: ri[WS(rs, 2)] = T3z - T3O; Chris@10: ri[WS(rs, 18)] = T3z + T3O; Chris@10: } Chris@10: { Chris@10: E T4r, T4p, T4q, T4l, T4u, T4j, T4k, T4t, T4s; Chris@10: T4r = KP559016994 * (T4n - T4o); Chris@10: T4p = T4n + T4o; Chris@10: T4q = FNMS(KP250000000, T4p, T4m); Chris@10: T4j = T1N - T28; Chris@10: T4k = TW - T1r; Chris@10: T4l = FNMS(KP587785252, T4k, KP951056516 * T4j); Chris@10: T4u = FMA(KP951056516, T4k, KP587785252 * T4j); Chris@10: ii[WS(rs, 10)] = T4p + T4m; Chris@10: T4t = T4r + T4q; Chris@10: ii[WS(rs, 6)] = T4t - T4u; Chris@10: ii[WS(rs, 14)] = T4u + T4t; Chris@10: T4s = T4q - T4r; Chris@10: ii[WS(rs, 2)] = T4l + T4s; Chris@10: ii[WS(rs, 18)] = T4s - T4l; Chris@10: } Chris@10: { Chris@10: E T3R, T2i, T3S, T40, T42, T3W, T3Z, T41, T3T; Chris@10: T3R = KP559016994 * (T2e - T2h); Chris@10: T2i = T2e + T2h; Chris@10: T3S = FNMS(KP250000000, T2i, T2b); Chris@10: T3W = T3U - T3V; Chris@10: T3Z = T3X - T3Y; Chris@10: T40 = FMA(KP951056516, T3W, KP587785252 * T3Z); Chris@10: T42 = FNMS(KP587785252, T3W, KP951056516 * T3Z); Chris@10: ri[0] = T2b + T2i; Chris@10: T41 = T3S - T3R; Chris@10: ri[WS(rs, 12)] = T41 - T42; Chris@10: ri[WS(rs, 8)] = T41 + T42; Chris@10: T3T = T3R + T3S; Chris@10: ri[WS(rs, 4)] = T3T - T40; Chris@10: ri[WS(rs, 16)] = T3T + T40; Chris@10: } Chris@10: { Chris@10: E T4e, T45, T4f, T4d, T4i, T4b, T4c, T4h, T4g; Chris@10: T4e = KP559016994 * (T43 - T44); Chris@10: T45 = T43 + T44; Chris@10: T4f = FNMS(KP250000000, T45, T4a); Chris@10: T4b = T2c - T2d; Chris@10: T4c = T2f - T2g; Chris@10: T4d = FMA(KP951056516, T4b, KP587785252 * T4c); Chris@10: T4i = FNMS(KP587785252, T4b, KP951056516 * T4c); Chris@10: ii[0] = T45 + T4a; Chris@10: T4h = T4f - T4e; Chris@10: ii[WS(rs, 8)] = T4h - T4i; Chris@10: ii[WS(rs, 12)] = T4i + T4h; Chris@10: T4g = T4e + T4f; Chris@10: ii[WS(rs, 4)] = T4d + T4g; Chris@10: ii[WS(rs, 16)] = T4g - T4d; Chris@10: } Chris@10: { Chris@10: E T39, T37, T38, T2F, T3b, T2t, T2E, T3c, T3a; Chris@10: T39 = KP559016994 * (T2V - T36); Chris@10: T37 = T2V + T36; Chris@10: T38 = FNMS(KP250000000, T37, T2K); Chris@10: T2t = T2n - T2s; Chris@10: T2E = T2y - T2D; Chris@10: T2F = FNMS(KP587785252, T2E, KP951056516 * T2t); Chris@10: T3b = FMA(KP951056516, T2E, KP587785252 * T2t); Chris@10: ri[WS(rs, 15)] = T2K + T37; Chris@10: T3c = T39 + T38; Chris@10: ri[WS(rs, 11)] = T3b + T3c; Chris@10: ri[WS(rs, 19)] = T3c - T3b; Chris@10: T3a = T38 - T39; Chris@10: ri[WS(rs, 3)] = T2F + T3a; Chris@10: ri[WS(rs, 7)] = T3a - T2F; Chris@10: } Chris@10: { Chris@10: E T4O, T4M, T4N, T4S, T4U, T4Q, T4R, T4T, T4P; Chris@10: T4O = KP559016994 * (T4K - T4L); Chris@10: T4M = T4K + T4L; Chris@10: T4N = FNMS(KP250000000, T4M, T4J); Chris@10: T4Q = T30 - T35; Chris@10: T4R = T2P - T2U; Chris@10: T4S = FNMS(KP587785252, T4R, KP951056516 * T4Q); Chris@10: T4U = FMA(KP951056516, T4R, KP587785252 * T4Q); Chris@10: ii[WS(rs, 15)] = T4M + T4J; Chris@10: T4T = T4O + T4N; Chris@10: ii[WS(rs, 11)] = T4T - T4U; Chris@10: ii[WS(rs, 19)] = T4U + T4T; Chris@10: T4P = T4N - T4O; Chris@10: ii[WS(rs, 3)] = T4P - T4S; Chris@10: ii[WS(rs, 7)] = T4S + T4P; Chris@10: } Chris@10: { Chris@10: E T3q, T3s, T3t, T3j, T3v, T3f, T3i, T3w, T3u; Chris@10: T3q = KP559016994 * (T3m - T3p); Chris@10: T3s = T3m + T3p; Chris@10: T3t = FNMS(KP250000000, T3s, T3r); Chris@10: T3f = T3d - T3e; Chris@10: T3i = T3g - T3h; Chris@10: T3j = FMA(KP951056516, T3f, KP587785252 * T3i); Chris@10: T3v = FNMS(KP587785252, T3f, KP951056516 * T3i); Chris@10: ri[WS(rs, 5)] = T3r + T3s; Chris@10: T3w = T3t - T3q; Chris@10: ri[WS(rs, 13)] = T3v + T3w; Chris@10: ri[WS(rs, 17)] = T3w - T3v; Chris@10: T3u = T3q + T3t; Chris@10: ri[WS(rs, 1)] = T3j + T3u; Chris@10: ri[WS(rs, 9)] = T3u - T3j; Chris@10: } Chris@10: { Chris@10: E T4x, T4B, T4C, T4G, T4I, T4E, T4F, T4H, T4D; Chris@10: T4x = KP559016994 * (T4v - T4w); Chris@10: T4B = T4v + T4w; Chris@10: T4C = FNMS(KP250000000, T4B, T4A); Chris@10: T4E = T3k - T3l; Chris@10: T4F = T3n - T3o; Chris@10: T4G = FMA(KP951056516, T4E, KP587785252 * T4F); Chris@10: T4I = FNMS(KP587785252, T4E, KP951056516 * T4F); Chris@10: ii[WS(rs, 5)] = T4B + T4A; Chris@10: T4H = T4C - T4x; Chris@10: ii[WS(rs, 13)] = T4H - T4I; Chris@10: ii[WS(rs, 17)] = T4I + T4H; Chris@10: T4D = T4x + T4C; Chris@10: ii[WS(rs, 1)] = T4D - T4G; Chris@10: ii[WS(rs, 9)] = T4G + T4D; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 0, 1}, Chris@10: {TW_CEXP, 0, 3}, Chris@10: {TW_CEXP, 0, 9}, Chris@10: {TW_CEXP, 0, 19}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, "t2_20", twinstr, &GENUS, {204, 92, 72, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t2_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2_20, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */