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 07:35:53 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 -n 20 -name t1_20 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 246 FP additions, 148 FP multiplications, Chris@10: * (or, 136 additions, 38 multiplications, 110 fused multiply/add), Chris@10: * 97 stack variables, 4 constants, and 80 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t1_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 * 38); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 38, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E T4P, T4Y, T50, T4U, T4S, T4T, T4Z, T4V; Chris@10: { Chris@10: E T4N, T4r, T8, T2i, T4n, T2n, T4O, Tl, T2v, T3v, T40, T4b, TN, T2b, T3F; Chris@10: E T3i, T2R, T3z, T3W, T4f, T27, T2f, T3J, T3a, T2K, T3y, T3T, T4e, T1G, T2e; Chris@10: E T3I, T33, T2C, T3w, T43, T4c, T1e, T2c, T3G, T3p; Chris@10: { Chris@10: E T1, T4q, T3, T6, T2, T5; Chris@10: T1 = ri[0]; Chris@10: T4q = ii[0]; Chris@10: T3 = ri[WS(rs, 10)]; Chris@10: T6 = ii[WS(rs, 10)]; Chris@10: T2 = W[18]; Chris@10: T5 = W[19]; Chris@10: { Chris@10: E Ta, Td, Tg, T2j, Tb, Tj, Tf, Tc, Ti; Chris@10: { Chris@10: E T4o, T4, T9, T4p, T7; Chris@10: Ta = ri[WS(rs, 5)]; Chris@10: Td = ii[WS(rs, 5)]; Chris@10: T4o = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T9 = W[8]; Chris@10: Tg = ri[WS(rs, 15)]; Chris@10: T4p = FNMS(T5, T3, T4o); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: T2j = T9 * Td; Chris@10: Tb = T9 * Ta; Chris@10: T4N = T4q - T4p; Chris@10: T4r = T4p + T4q; Chris@10: T8 = T1 + T7; Chris@10: T2i = T1 - T7; Chris@10: Tj = ii[WS(rs, 15)]; Chris@10: Tf = W[28]; Chris@10: } Chris@10: Tc = W[9]; Chris@10: Ti = W[29]; Chris@10: { Chris@10: E T3d, Ts, T2t, TL, TB, TE, TD, T3f, Ty, T2q, TC; Chris@10: { Chris@10: E TH, TK, TJ, T2s, TI; Chris@10: { Chris@10: E To, Tr, Tp, T3c, Tq, TG; Chris@10: { Chris@10: E T2k, Te, T2m, Tk, T2l, Th, Tn; Chris@10: To = ri[WS(rs, 4)]; Chris@10: T2l = Tf * Tj; Chris@10: Th = Tf * Tg; Chris@10: T2k = FNMS(Tc, Ta, T2j); Chris@10: Te = FMA(Tc, Td, Tb); Chris@10: T2m = FNMS(Ti, Tg, T2l); Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: Tr = ii[WS(rs, 4)]; Chris@10: Tn = W[6]; Chris@10: T4n = T2k + T2m; Chris@10: T2n = T2k - T2m; Chris@10: T4O = Te - Tk; Chris@10: Tl = Te + Tk; Chris@10: Tp = Tn * To; Chris@10: T3c = Tn * Tr; Chris@10: } Chris@10: Tq = W[7]; Chris@10: TH = ri[WS(rs, 19)]; Chris@10: TK = ii[WS(rs, 19)]; Chris@10: TG = W[36]; Chris@10: T3d = FNMS(Tq, To, T3c); Chris@10: Ts = FMA(Tq, Tr, Tp); Chris@10: TJ = W[37]; Chris@10: T2s = TG * TK; Chris@10: TI = TG * TH; Chris@10: } Chris@10: { Chris@10: E Tu, Tx, Tt, Tw, T3e, Tv, TA; Chris@10: Tu = ri[WS(rs, 14)]; Chris@10: Tx = ii[WS(rs, 14)]; Chris@10: T2t = FNMS(TJ, TH, T2s); Chris@10: TL = FMA(TJ, TK, TI); Chris@10: Tt = W[26]; Chris@10: Tw = W[27]; Chris@10: TB = ri[WS(rs, 9)]; Chris@10: TE = ii[WS(rs, 9)]; Chris@10: T3e = Tt * Tx; Chris@10: Tv = Tt * Tu; Chris@10: TA = W[16]; Chris@10: TD = W[17]; Chris@10: T3f = FNMS(Tw, Tu, T3e); Chris@10: Ty = FMA(Tw, Tx, Tv); Chris@10: T2q = TA * TE; Chris@10: TC = TA * TB; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3g, T3Y, Tz, T2p, T2r, TF; Chris@10: T3g = T3d - T3f; Chris@10: T3Y = T3d + T3f; Chris@10: Tz = Ts + Ty; Chris@10: T2p = Ts - Ty; Chris@10: T2r = FNMS(TD, TB, T2q); Chris@10: TF = FMA(TD, TE, TC); Chris@10: { Chris@10: E T3Z, T2u, T3h, TM; Chris@10: T3Z = T2r + T2t; Chris@10: T2u = T2r - T2t; Chris@10: T3h = TF - TL; Chris@10: TM = TF + TL; Chris@10: T2v = T2p - T2u; Chris@10: T3v = T2p + T2u; Chris@10: T40 = T3Y - T3Z; Chris@10: T4b = T3Y + T3Z; Chris@10: TN = Tz - TM; Chris@10: T2b = Tz + TM; Chris@10: T3F = T3g - T3h; Chris@10: T3i = T3g + T3h; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T35, T1M, T2P, T25, T1V, T1Y, T1X, T37, T1S, T2M, T1W; Chris@10: { Chris@10: E T21, T24, T23, T2O, T22; Chris@10: { Chris@10: E T1I, T1L, T1H, T1K, T34, T1J, T20; Chris@10: T1I = ri[WS(rs, 12)]; Chris@10: T1L = ii[WS(rs, 12)]; Chris@10: T1H = W[22]; Chris@10: T1K = W[23]; Chris@10: T21 = ri[WS(rs, 7)]; Chris@10: T24 = ii[WS(rs, 7)]; Chris@10: T34 = T1H * T1L; Chris@10: T1J = T1H * T1I; Chris@10: T20 = W[12]; Chris@10: T23 = W[13]; Chris@10: T35 = FNMS(T1K, T1I, T34); Chris@10: T1M = FMA(T1K, T1L, T1J); Chris@10: T2O = T20 * T24; Chris@10: T22 = T20 * T21; Chris@10: } Chris@10: { Chris@10: E T1O, T1R, T1N, T1Q, T36, T1P, T1U; Chris@10: T1O = ri[WS(rs, 2)]; Chris@10: T1R = ii[WS(rs, 2)]; Chris@10: T2P = FNMS(T23, T21, T2O); Chris@10: T25 = FMA(T23, T24, T22); Chris@10: T1N = W[2]; Chris@10: T1Q = W[3]; Chris@10: T1V = ri[WS(rs, 17)]; Chris@10: T1Y = ii[WS(rs, 17)]; Chris@10: T36 = T1N * T1R; Chris@10: T1P = T1N * T1O; Chris@10: T1U = W[32]; Chris@10: T1X = W[33]; Chris@10: T37 = FNMS(T1Q, T1O, T36); Chris@10: T1S = FMA(T1Q, T1R, T1P); Chris@10: T2M = T1U * T1Y; Chris@10: T1W = T1U * T1V; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T38, T3U, T1T, T2L, T2N, T1Z; Chris@10: T38 = T35 - T37; Chris@10: T3U = T35 + T37; Chris@10: T1T = T1M + T1S; Chris@10: T2L = T1M - T1S; Chris@10: T2N = FNMS(T1X, T1V, T2M); Chris@10: T1Z = FMA(T1X, T1Y, T1W); Chris@10: { Chris@10: E T3V, T2Q, T39, T26; Chris@10: T3V = T2N + T2P; Chris@10: T2Q = T2N - T2P; Chris@10: T39 = T1Z - T25; Chris@10: T26 = T1Z + T25; Chris@10: T2R = T2L - T2Q; Chris@10: T3z = T2L + T2Q; Chris@10: T3W = T3U - T3V; Chris@10: T4f = T3U + T3V; Chris@10: T27 = T1T - T26; Chris@10: T2f = T1T + T26; Chris@10: T3J = T38 - T39; Chris@10: T3a = T38 + T39; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2Y, T1l, T2I, T1E, T1u, T1x, T1w, T30, T1r, T2F, T1v; Chris@10: { Chris@10: E T1A, T1D, T1C, T2H, T1B; Chris@10: { Chris@10: E T1h, T1k, T1g, T1j, T2X, T1i, T1z; Chris@10: T1h = ri[WS(rs, 8)]; Chris@10: T1k = ii[WS(rs, 8)]; Chris@10: T1g = W[14]; Chris@10: T1j = W[15]; Chris@10: T1A = ri[WS(rs, 3)]; Chris@10: T1D = ii[WS(rs, 3)]; Chris@10: T2X = T1g * T1k; Chris@10: T1i = T1g * T1h; Chris@10: T1z = W[4]; Chris@10: T1C = W[5]; Chris@10: T2Y = FNMS(T1j, T1h, T2X); Chris@10: T1l = FMA(T1j, T1k, T1i); Chris@10: T2H = T1z * T1D; Chris@10: T1B = T1z * T1A; Chris@10: } Chris@10: { Chris@10: E T1n, T1q, T1m, T1p, T2Z, T1o, T1t; Chris@10: T1n = ri[WS(rs, 18)]; Chris@10: T1q = ii[WS(rs, 18)]; Chris@10: T2I = FNMS(T1C, T1A, T2H); Chris@10: T1E = FMA(T1C, T1D, T1B); Chris@10: T1m = W[34]; Chris@10: T1p = W[35]; Chris@10: T1u = ri[WS(rs, 13)]; Chris@10: T1x = ii[WS(rs, 13)]; Chris@10: T2Z = T1m * T1q; Chris@10: T1o = T1m * T1n; Chris@10: T1t = W[24]; Chris@10: T1w = W[25]; Chris@10: T30 = FNMS(T1p, T1n, T2Z); Chris@10: T1r = FMA(T1p, T1q, T1o); Chris@10: T2F = T1t * T1x; Chris@10: T1v = T1t * T1u; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T31, T3R, T1s, T2E, T2G, T1y; Chris@10: T31 = T2Y - T30; Chris@10: T3R = T2Y + T30; Chris@10: T1s = T1l + T1r; Chris@10: T2E = T1l - T1r; Chris@10: T2G = FNMS(T1w, T1u, T2F); Chris@10: T1y = FMA(T1w, T1x, T1v); Chris@10: { Chris@10: E T3S, T2J, T32, T1F; Chris@10: T3S = T2G + T2I; Chris@10: T2J = T2G - T2I; Chris@10: T32 = T1y - T1E; Chris@10: T1F = T1y + T1E; Chris@10: T2K = T2E - T2J; Chris@10: T3y = T2E + T2J; Chris@10: T3T = T3R - T3S; Chris@10: T4e = T3R + T3S; Chris@10: T1G = T1s - T1F; Chris@10: T2e = T1s + T1F; Chris@10: T3I = T31 - T32; Chris@10: T33 = T31 + T32; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3k, TT, T2A, T1c, T12, T15, T14, T3m, TZ, T2x, T13; Chris@10: { Chris@10: E T18, T1b, T1a, T2z, T19; Chris@10: { Chris@10: E TP, TS, TO, TR, T3j, TQ, T17; Chris@10: TP = ri[WS(rs, 16)]; Chris@10: TS = ii[WS(rs, 16)]; Chris@10: TO = W[30]; Chris@10: TR = W[31]; Chris@10: T18 = ri[WS(rs, 11)]; Chris@10: T1b = ii[WS(rs, 11)]; Chris@10: T3j = TO * TS; Chris@10: TQ = TO * TP; Chris@10: T17 = W[20]; Chris@10: T1a = W[21]; Chris@10: T3k = FNMS(TR, TP, T3j); Chris@10: TT = FMA(TR, TS, TQ); Chris@10: T2z = T17 * T1b; Chris@10: T19 = T17 * T18; Chris@10: } Chris@10: { Chris@10: E TV, TY, TU, TX, T3l, TW, T11; Chris@10: TV = ri[WS(rs, 6)]; Chris@10: TY = ii[WS(rs, 6)]; Chris@10: T2A = FNMS(T1a, T18, T2z); Chris@10: T1c = FMA(T1a, T1b, T19); Chris@10: TU = W[10]; Chris@10: TX = W[11]; Chris@10: T12 = ri[WS(rs, 1)]; Chris@10: T15 = ii[WS(rs, 1)]; Chris@10: T3l = TU * TY; Chris@10: TW = TU * TV; Chris@10: T11 = W[0]; Chris@10: T14 = W[1]; Chris@10: T3m = FNMS(TX, TV, T3l); Chris@10: TZ = FMA(TX, TY, TW); Chris@10: T2x = T11 * T15; Chris@10: T13 = T11 * T12; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3n, T41, T10, T2w, T2y, T16; Chris@10: T3n = T3k - T3m; Chris@10: T41 = T3k + T3m; Chris@10: T10 = TT + TZ; Chris@10: T2w = TT - TZ; Chris@10: T2y = FNMS(T14, T12, T2x); Chris@10: T16 = FMA(T14, T15, T13); Chris@10: { Chris@10: E T42, T2B, T3o, T1d; Chris@10: T42 = T2y + T2A; Chris@10: T2B = T2y - T2A; Chris@10: T3o = T16 - T1c; Chris@10: T1d = T16 + T1c; Chris@10: T2C = T2w - T2B; Chris@10: T3w = T2w + T2B; Chris@10: T43 = T41 - T42; Chris@10: T4c = T41 + T42; Chris@10: T1e = T10 - T1d; Chris@10: T2c = T10 + T1d; Chris@10: T3G = T3n - T3o; Chris@10: T3p = T3n + T3o; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T4s, T4k, T4l, T4h, T4j, T49, T4y, T4A, T48; Chris@10: { Chris@10: E T4D, T4C, T2a, T47, T45, T4B, T4M, T4K, T46, T3Q; Chris@10: { Chris@10: E Tm, T1f, T4J, T4I, T28, T3X, T44, T29, T3P, T3O; Chris@10: T4D = T3T + T3W; Chris@10: T3X = T3T - T3W; Chris@10: T44 = T40 - T43; Chris@10: T4C = T40 + T43; Chris@10: T2a = T8 + Tl; Chris@10: Tm = T8 - Tl; Chris@10: T1f = TN + T1e; Chris@10: T4J = TN - T1e; Chris@10: T4I = T1G - T27; Chris@10: T28 = T1G + T27; Chris@10: T47 = FMA(KP618033988, T3X, T44); Chris@10: T45 = FNMS(KP618033988, T44, T3X); Chris@10: T29 = T1f + T28; Chris@10: T3P = T1f - T28; Chris@10: T4B = T4r - T4n; Chris@10: T4s = T4n + T4r; Chris@10: ri[WS(rs, 10)] = Tm + T29; Chris@10: T3O = FNMS(KP250000000, T29, Tm); Chris@10: T4M = FMA(KP618033988, T4I, T4J); Chris@10: T4K = FNMS(KP618033988, T4J, T4I); Chris@10: T46 = FMA(KP559016994, T3P, T3O); Chris@10: T3Q = FNMS(KP559016994, T3P, T3O); Chris@10: } Chris@10: { Chris@10: E T2d, T4w, T4x, T2g, T2h; Chris@10: { Chris@10: E T4d, T4G, T4F, T4g, T4E, T4L, T4H; Chris@10: T4k = T4b + T4c; Chris@10: T4d = T4b - T4c; Chris@10: T4G = T4C - T4D; Chris@10: T4E = T4C + T4D; Chris@10: ri[WS(rs, 18)] = FMA(KP951056516, T45, T3Q); Chris@10: ri[WS(rs, 2)] = FNMS(KP951056516, T45, T3Q); Chris@10: ri[WS(rs, 6)] = FMA(KP951056516, T47, T46); Chris@10: ri[WS(rs, 14)] = FNMS(KP951056516, T47, T46); Chris@10: ii[WS(rs, 10)] = T4E + T4B; Chris@10: T4F = FNMS(KP250000000, T4E, T4B); Chris@10: T4g = T4e - T4f; Chris@10: T4l = T4e + T4f; Chris@10: T2d = T2b + T2c; Chris@10: T4w = T2b - T2c; Chris@10: T4L = FMA(KP559016994, T4G, T4F); Chris@10: T4H = FNMS(KP559016994, T4G, T4F); Chris@10: T4h = FMA(KP618033988, T4g, T4d); Chris@10: T4j = FNMS(KP618033988, T4d, T4g); Chris@10: ii[WS(rs, 18)] = FNMS(KP951056516, T4K, T4H); Chris@10: ii[WS(rs, 2)] = FMA(KP951056516, T4K, T4H); Chris@10: ii[WS(rs, 14)] = FMA(KP951056516, T4M, T4L); Chris@10: ii[WS(rs, 6)] = FNMS(KP951056516, T4M, T4L); Chris@10: T4x = T2e - T2f; Chris@10: T2g = T2e + T2f; Chris@10: } Chris@10: T2h = T2d + T2g; Chris@10: T49 = T2d - T2g; Chris@10: T4y = FMA(KP618033988, T4x, T4w); Chris@10: T4A = FNMS(KP618033988, T4w, T4x); Chris@10: ri[0] = T2a + T2h; Chris@10: T48 = FNMS(KP250000000, T2h, T2a); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3u, T51, T5a, T5c, T56, T54; Chris@10: { Chris@10: E T53, T52, T3t, T3r, T2o, T59, T58, T2T, T2V, T4u, T4t, T2U, T3s, T2W; Chris@10: { Chris@10: E T3b, T3q, T4i, T4a, T4m; Chris@10: T53 = T33 + T3a; Chris@10: T3b = T33 - T3a; Chris@10: T3q = T3i - T3p; Chris@10: T52 = T3i + T3p; Chris@10: T4i = FNMS(KP559016994, T49, T48); Chris@10: T4a = FMA(KP559016994, T49, T48); Chris@10: T4m = T4k + T4l; Chris@10: T4u = T4k - T4l; Chris@10: ri[WS(rs, 16)] = FMA(KP951056516, T4h, T4a); Chris@10: ri[WS(rs, 4)] = FNMS(KP951056516, T4h, T4a); Chris@10: ri[WS(rs, 8)] = FMA(KP951056516, T4j, T4i); Chris@10: ri[WS(rs, 12)] = FNMS(KP951056516, T4j, T4i); Chris@10: ii[0] = T4m + T4s; Chris@10: T4t = FNMS(KP250000000, T4m, T4s); Chris@10: T3t = FMA(KP618033988, T3b, T3q); Chris@10: T3r = FNMS(KP618033988, T3q, T3b); Chris@10: } Chris@10: T3u = T2i + T2n; Chris@10: T2o = T2i - T2n; Chris@10: { Chris@10: E T4v, T4z, T2D, T2S; Chris@10: T4v = FMA(KP559016994, T4u, T4t); Chris@10: T4z = FNMS(KP559016994, T4u, T4t); Chris@10: T2D = T2v + T2C; Chris@10: T59 = T2v - T2C; Chris@10: T58 = T2K - T2R; Chris@10: T2S = T2K + T2R; Chris@10: ii[WS(rs, 16)] = FNMS(KP951056516, T4y, T4v); Chris@10: ii[WS(rs, 4)] = FMA(KP951056516, T4y, T4v); Chris@10: ii[WS(rs, 12)] = FMA(KP951056516, T4A, T4z); Chris@10: ii[WS(rs, 8)] = FNMS(KP951056516, T4A, T4z); Chris@10: T2T = T2D + T2S; Chris@10: T2V = T2D - T2S; Chris@10: } Chris@10: ri[WS(rs, 15)] = T2o + T2T; Chris@10: T2U = FNMS(KP250000000, T2T, T2o); Chris@10: T51 = T4O + T4N; Chris@10: T4P = T4N - T4O; Chris@10: T5a = FNMS(KP618033988, T59, T58); Chris@10: T5c = FMA(KP618033988, T58, T59); Chris@10: T3s = FMA(KP559016994, T2V, T2U); Chris@10: T2W = FNMS(KP559016994, T2V, T2U); Chris@10: ri[WS(rs, 7)] = FNMS(KP951056516, T3r, T2W); Chris@10: ri[WS(rs, 3)] = FMA(KP951056516, T3r, T2W); Chris@10: ri[WS(rs, 19)] = FNMS(KP951056516, T3t, T3s); Chris@10: ri[WS(rs, 11)] = FMA(KP951056516, T3t, T3s); Chris@10: T56 = T52 - T53; Chris@10: T54 = T52 + T53; Chris@10: } Chris@10: { Chris@10: E T4Q, T4R, T3N, T3L, T4W, T4X, T3B, T3D, T3H, T3K, T55, T3C, T3M, T3E; Chris@10: T4Q = T3F + T3G; Chris@10: T3H = T3F - T3G; Chris@10: T3K = T3I - T3J; Chris@10: T4R = T3I + T3J; Chris@10: ii[WS(rs, 15)] = T54 + T51; Chris@10: T55 = FNMS(KP250000000, T54, T51); Chris@10: T3N = FNMS(KP618033988, T3H, T3K); Chris@10: T3L = FMA(KP618033988, T3K, T3H); Chris@10: { Chris@10: E T57, T5b, T3x, T3A; Chris@10: T57 = FNMS(KP559016994, T56, T55); Chris@10: T5b = FMA(KP559016994, T56, T55); Chris@10: T3x = T3v + T3w; Chris@10: T4W = T3v - T3w; Chris@10: T4X = T3y - T3z; Chris@10: T3A = T3y + T3z; Chris@10: ii[WS(rs, 7)] = FMA(KP951056516, T5a, T57); Chris@10: ii[WS(rs, 3)] = FNMS(KP951056516, T5a, T57); Chris@10: ii[WS(rs, 19)] = FMA(KP951056516, T5c, T5b); Chris@10: ii[WS(rs, 11)] = FNMS(KP951056516, T5c, T5b); Chris@10: T3B = T3x + T3A; Chris@10: T3D = T3x - T3A; Chris@10: } Chris@10: ri[WS(rs, 5)] = T3u + T3B; Chris@10: T3C = FNMS(KP250000000, T3B, T3u); Chris@10: T4Y = FMA(KP618033988, T4X, T4W); Chris@10: T50 = FNMS(KP618033988, T4W, T4X); Chris@10: T3M = FNMS(KP559016994, T3D, T3C); Chris@10: T3E = FMA(KP559016994, T3D, T3C); Chris@10: ri[WS(rs, 9)] = FNMS(KP951056516, T3L, T3E); Chris@10: ri[WS(rs, 1)] = FMA(KP951056516, T3L, T3E); Chris@10: ri[WS(rs, 17)] = FNMS(KP951056516, T3N, T3M); Chris@10: ri[WS(rs, 13)] = FMA(KP951056516, T3N, T3M); Chris@10: T4U = T4Q - T4R; Chris@10: T4S = T4Q + T4R; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: ii[WS(rs, 5)] = T4S + T4P; Chris@10: T4T = FNMS(KP250000000, T4S, T4P); Chris@10: T4Z = FNMS(KP559016994, T4U, T4T); Chris@10: T4V = FMA(KP559016994, T4U, T4T); Chris@10: ii[WS(rs, 9)] = FMA(KP951056516, T4Y, T4V); Chris@10: ii[WS(rs, 1)] = FNMS(KP951056516, T4Y, T4V); Chris@10: ii[WS(rs, 17)] = FMA(KP951056516, T50, T4Z); Chris@10: ii[WS(rs, 13)] = FNMS(KP951056516, T50, T4Z); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 0, 20}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, "t1_20", twinstr, &GENUS, {136, 38, 110, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t1_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1_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 -n 20 -name t1_20 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 246 FP additions, 124 FP multiplications, Chris@10: * (or, 184 additions, 62 multiplications, 62 fused multiply/add), Chris@10: * 85 stack variables, 4 constants, and 80 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t1_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 * 38); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 38, MAKE_VOLATILE_STRIDE(40, rs)) { Chris@10: E Tj, T1R, T4g, T4p, T2q, T37, T3Q, T42, T1r, T1O, T1P, T3i, T3l, T44, T3D; Chris@10: E T3E, T3K, T1V, T1W, T1X, T23, T28, T4r, T2W, T2X, T4c, T33, T34, T35, T2G; Chris@10: E T2L, T2M, TG, T13, T14, T3p, T3s, T43, T3A, T3B, T3J, T1S, T1T, T1U, T2e; Chris@10: E T2j, T4q, T2T, T2U, T4b, T30, T31, T32, T2v, T2A, T2B; Chris@10: { Chris@10: E T1, T3O, T6, T3N, Tc, T2n, Th, T2o; Chris@10: T1 = ri[0]; Chris@10: T3O = ii[0]; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = ri[WS(rs, 10)]; Chris@10: T5 = ii[WS(rs, 10)]; Chris@10: T2 = W[18]; Chris@10: T4 = W[19]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: T3N = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: { Chris@10: E T9, Tb, T8, Ta; Chris@10: T9 = ri[WS(rs, 5)]; Chris@10: Tb = ii[WS(rs, 5)]; Chris@10: T8 = W[8]; Chris@10: Ta = W[9]; Chris@10: Tc = FMA(T8, T9, Ta * Tb); Chris@10: T2n = FNMS(Ta, T9, T8 * Tb); Chris@10: } Chris@10: { Chris@10: E Te, Tg, Td, Tf; Chris@10: Te = ri[WS(rs, 15)]; Chris@10: Tg = ii[WS(rs, 15)]; Chris@10: Td = W[28]; Chris@10: Tf = W[29]; Chris@10: Th = FMA(Td, Te, Tf * Tg); Chris@10: T2o = FNMS(Tf, Te, Td * Tg); Chris@10: } Chris@10: { Chris@10: E T7, Ti, T4e, T4f; Chris@10: T7 = T1 + T6; Chris@10: Ti = Tc + Th; Chris@10: Tj = T7 - Ti; Chris@10: T1R = T7 + Ti; Chris@10: T4e = T3O - T3N; Chris@10: T4f = Tc - Th; Chris@10: T4g = T4e - T4f; Chris@10: T4p = T4f + T4e; Chris@10: } Chris@10: { Chris@10: E T2m, T2p, T3M, T3P; Chris@10: T2m = T1 - T6; Chris@10: T2p = T2n - T2o; Chris@10: T2q = T2m - T2p; Chris@10: T37 = T2m + T2p; Chris@10: T3M = T2n + T2o; Chris@10: T3P = T3N + T3O; Chris@10: T3Q = T3M + T3P; Chris@10: T42 = T3P - T3M; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1f, T3g, T21, T2C, T1N, T3k, T27, T2K, T1q, T3h, T22, T2F, T1C, T3j, T26; Chris@10: E T2H; Chris@10: { Chris@10: E T19, T1Z, T1e, T20; Chris@10: { Chris@10: E T16, T18, T15, T17; Chris@10: T16 = ri[WS(rs, 8)]; Chris@10: T18 = ii[WS(rs, 8)]; Chris@10: T15 = W[14]; Chris@10: T17 = W[15]; Chris@10: T19 = FMA(T15, T16, T17 * T18); Chris@10: T1Z = FNMS(T17, T16, T15 * T18); Chris@10: } Chris@10: { Chris@10: E T1b, T1d, T1a, T1c; Chris@10: T1b = ri[WS(rs, 18)]; Chris@10: T1d = ii[WS(rs, 18)]; Chris@10: T1a = W[34]; Chris@10: T1c = W[35]; Chris@10: T1e = FMA(T1a, T1b, T1c * T1d); Chris@10: T20 = FNMS(T1c, T1b, T1a * T1d); Chris@10: } Chris@10: T1f = T19 + T1e; Chris@10: T3g = T1Z + T20; Chris@10: T21 = T1Z - T20; Chris@10: T2C = T19 - T1e; Chris@10: } Chris@10: { Chris@10: E T1H, T2I, T1M, T2J; Chris@10: { Chris@10: E T1E, T1G, T1D, T1F; Chris@10: T1E = ri[WS(rs, 17)]; Chris@10: T1G = ii[WS(rs, 17)]; Chris@10: T1D = W[32]; Chris@10: T1F = W[33]; Chris@10: T1H = FMA(T1D, T1E, T1F * T1G); Chris@10: T2I = FNMS(T1F, T1E, T1D * T1G); Chris@10: } Chris@10: { Chris@10: E T1J, T1L, T1I, T1K; Chris@10: T1J = ri[WS(rs, 7)]; Chris@10: T1L = ii[WS(rs, 7)]; Chris@10: T1I = W[12]; Chris@10: T1K = W[13]; Chris@10: T1M = FMA(T1I, T1J, T1K * T1L); Chris@10: T2J = FNMS(T1K, T1J, T1I * T1L); Chris@10: } Chris@10: T1N = T1H + T1M; Chris@10: T3k = T2I + T2J; Chris@10: T27 = T1H - T1M; Chris@10: T2K = T2I - T2J; Chris@10: } Chris@10: { Chris@10: E T1k, T2D, T1p, T2E; Chris@10: { Chris@10: E T1h, T1j, T1g, T1i; Chris@10: T1h = ri[WS(rs, 13)]; Chris@10: T1j = ii[WS(rs, 13)]; Chris@10: T1g = W[24]; Chris@10: T1i = W[25]; Chris@10: T1k = FMA(T1g, T1h, T1i * T1j); Chris@10: T2D = FNMS(T1i, T1h, T1g * T1j); Chris@10: } Chris@10: { Chris@10: E T1m, T1o, T1l, T1n; Chris@10: T1m = ri[WS(rs, 3)]; Chris@10: T1o = ii[WS(rs, 3)]; Chris@10: T1l = W[4]; Chris@10: T1n = W[5]; Chris@10: T1p = FMA(T1l, T1m, T1n * T1o); Chris@10: T2E = FNMS(T1n, T1m, T1l * T1o); Chris@10: } Chris@10: T1q = T1k + T1p; Chris@10: T3h = T2D + T2E; Chris@10: T22 = T1k - T1p; Chris@10: T2F = T2D - T2E; Chris@10: } Chris@10: { Chris@10: E T1w, T24, T1B, T25; Chris@10: { Chris@10: E T1t, T1v, T1s, T1u; Chris@10: T1t = ri[WS(rs, 12)]; Chris@10: T1v = ii[WS(rs, 12)]; Chris@10: T1s = W[22]; Chris@10: T1u = W[23]; Chris@10: T1w = FMA(T1s, T1t, T1u * T1v); Chris@10: T24 = FNMS(T1u, T1t, T1s * T1v); Chris@10: } Chris@10: { Chris@10: E T1y, T1A, T1x, T1z; Chris@10: T1y = ri[WS(rs, 2)]; Chris@10: T1A = ii[WS(rs, 2)]; Chris@10: T1x = W[2]; Chris@10: T1z = W[3]; Chris@10: T1B = FMA(T1x, T1y, T1z * T1A); Chris@10: T25 = FNMS(T1z, T1y, T1x * T1A); Chris@10: } Chris@10: T1C = T1w + T1B; Chris@10: T3j = T24 + T25; Chris@10: T26 = T24 - T25; Chris@10: T2H = T1w - T1B; Chris@10: } Chris@10: T1r = T1f - T1q; Chris@10: T1O = T1C - T1N; Chris@10: T1P = T1r + T1O; Chris@10: T3i = T3g - T3h; Chris@10: T3l = T3j - T3k; Chris@10: T44 = T3i + T3l; Chris@10: T3D = T3g + T3h; Chris@10: T3E = T3j + T3k; Chris@10: T3K = T3D + T3E; Chris@10: T1V = T1f + T1q; Chris@10: T1W = T1C + T1N; Chris@10: T1X = T1V + T1W; Chris@10: T23 = T21 + T22; Chris@10: T28 = T26 + T27; Chris@10: T4r = T23 + T28; Chris@10: T2W = T21 - T22; Chris@10: T2X = T26 - T27; Chris@10: T4c = T2W + T2X; Chris@10: T33 = T2C + T2F; Chris@10: T34 = T2H + T2K; Chris@10: T35 = T33 + T34; Chris@10: T2G = T2C - T2F; Chris@10: T2L = T2H - T2K; Chris@10: T2M = T2G + T2L; Chris@10: } Chris@10: { Chris@10: E Tu, T3n, T2c, T2r, T12, T3r, T2i, T2z, TF, T3o, T2d, T2u, TR, T3q, T2h; Chris@10: E T2w; Chris@10: { Chris@10: E To, T2a, Tt, T2b; Chris@10: { Chris@10: E Tl, Tn, Tk, Tm; Chris@10: Tl = ri[WS(rs, 4)]; Chris@10: Tn = ii[WS(rs, 4)]; Chris@10: Tk = W[6]; Chris@10: Tm = W[7]; Chris@10: To = FMA(Tk, Tl, Tm * Tn); Chris@10: T2a = FNMS(Tm, Tl, Tk * Tn); Chris@10: } Chris@10: { Chris@10: E Tq, Ts, Tp, Tr; Chris@10: Tq = ri[WS(rs, 14)]; Chris@10: Ts = ii[WS(rs, 14)]; Chris@10: Tp = W[26]; Chris@10: Tr = W[27]; Chris@10: Tt = FMA(Tp, Tq, Tr * Ts); Chris@10: T2b = FNMS(Tr, Tq, Tp * Ts); Chris@10: } Chris@10: Tu = To + Tt; Chris@10: T3n = T2a + T2b; Chris@10: T2c = T2a - T2b; Chris@10: T2r = To - Tt; Chris@10: } Chris@10: { Chris@10: E TW, T2x, T11, T2y; Chris@10: { Chris@10: E TT, TV, TS, TU; Chris@10: TT = ri[WS(rs, 1)]; Chris@10: TV = ii[WS(rs, 1)]; Chris@10: TS = W[0]; Chris@10: TU = W[1]; Chris@10: TW = FMA(TS, TT, TU * TV); Chris@10: T2x = FNMS(TU, TT, TS * TV); Chris@10: } Chris@10: { Chris@10: E TY, T10, TX, TZ; Chris@10: TY = ri[WS(rs, 11)]; Chris@10: T10 = ii[WS(rs, 11)]; Chris@10: TX = W[20]; Chris@10: TZ = W[21]; Chris@10: T11 = FMA(TX, TY, TZ * T10); Chris@10: T2y = FNMS(TZ, TY, TX * T10); Chris@10: } Chris@10: T12 = TW + T11; Chris@10: T3r = T2x + T2y; Chris@10: T2i = TW - T11; Chris@10: T2z = T2x - T2y; Chris@10: } Chris@10: { Chris@10: E Tz, T2s, TE, T2t; Chris@10: { Chris@10: E Tw, Ty, Tv, Tx; Chris@10: Tw = ri[WS(rs, 9)]; Chris@10: Ty = ii[WS(rs, 9)]; Chris@10: Tv = W[16]; Chris@10: Tx = W[17]; Chris@10: Tz = FMA(Tv, Tw, Tx * Ty); Chris@10: T2s = FNMS(Tx, Tw, Tv * Ty); Chris@10: } Chris@10: { Chris@10: E TB, TD, TA, TC; Chris@10: TB = ri[WS(rs, 19)]; Chris@10: TD = ii[WS(rs, 19)]; Chris@10: TA = W[36]; Chris@10: TC = W[37]; Chris@10: TE = FMA(TA, TB, TC * TD); Chris@10: T2t = FNMS(TC, TB, TA * TD); Chris@10: } Chris@10: TF = Tz + TE; Chris@10: T3o = T2s + T2t; Chris@10: T2d = Tz - TE; Chris@10: T2u = T2s - T2t; Chris@10: } Chris@10: { Chris@10: E TL, T2f, TQ, T2g; Chris@10: { Chris@10: E TI, TK, TH, TJ; Chris@10: TI = ri[WS(rs, 16)]; Chris@10: TK = ii[WS(rs, 16)]; Chris@10: TH = W[30]; Chris@10: TJ = W[31]; Chris@10: TL = FMA(TH, TI, TJ * TK); Chris@10: T2f = FNMS(TJ, TI, TH * TK); Chris@10: } Chris@10: { Chris@10: E TN, TP, TM, TO; Chris@10: TN = ri[WS(rs, 6)]; Chris@10: TP = ii[WS(rs, 6)]; Chris@10: TM = W[10]; Chris@10: TO = W[11]; Chris@10: TQ = FMA(TM, TN, TO * TP); Chris@10: T2g = FNMS(TO, TN, TM * TP); Chris@10: } Chris@10: TR = TL + TQ; Chris@10: T3q = T2f + T2g; Chris@10: T2h = T2f - T2g; Chris@10: T2w = TL - TQ; Chris@10: } Chris@10: TG = Tu - TF; Chris@10: T13 = TR - T12; Chris@10: T14 = TG + T13; Chris@10: T3p = T3n - T3o; Chris@10: T3s = T3q - T3r; Chris@10: T43 = T3p + T3s; Chris@10: T3A = T3n + T3o; Chris@10: T3B = T3q + T3r; Chris@10: T3J = T3A + T3B; Chris@10: T1S = Tu + TF; Chris@10: T1T = TR + T12; Chris@10: T1U = T1S + T1T; Chris@10: T2e = T2c + T2d; Chris@10: T2j = T2h + T2i; Chris@10: T4q = T2e + T2j; Chris@10: T2T = T2c - T2d; Chris@10: T2U = T2h - T2i; Chris@10: T4b = T2T + T2U; Chris@10: T30 = T2r + T2u; Chris@10: T31 = T2w + T2z; Chris@10: T32 = T30 + T31; Chris@10: T2v = T2r - T2u; Chris@10: T2A = T2w - T2z; Chris@10: T2B = T2v + T2A; Chris@10: } Chris@10: { Chris@10: E T3e, T1Q, T3d, T3u, T3w, T3m, T3t, T3v, T3f; Chris@10: T3e = KP559016994 * (T14 - T1P); Chris@10: T1Q = T14 + T1P; Chris@10: T3d = FNMS(KP250000000, T1Q, Tj); Chris@10: T3m = T3i - T3l; Chris@10: T3t = T3p - T3s; Chris@10: T3u = FNMS(KP587785252, T3t, KP951056516 * T3m); Chris@10: T3w = FMA(KP951056516, T3t, KP587785252 * T3m); Chris@10: ri[WS(rs, 10)] = Tj + T1Q; Chris@10: T3v = T3e + T3d; Chris@10: ri[WS(rs, 14)] = T3v - T3w; Chris@10: ri[WS(rs, 6)] = T3v + T3w; Chris@10: T3f = T3d - T3e; Chris@10: ri[WS(rs, 2)] = T3f - T3u; Chris@10: ri[WS(rs, 18)] = T3f + T3u; Chris@10: } Chris@10: { Chris@10: E T47, T45, T46, T41, T4a, T3Z, T40, T49, T48; Chris@10: T47 = KP559016994 * (T43 - T44); Chris@10: T45 = T43 + T44; Chris@10: T46 = FNMS(KP250000000, T45, T42); Chris@10: T3Z = T1r - T1O; Chris@10: T40 = TG - T13; Chris@10: T41 = FNMS(KP587785252, T40, KP951056516 * T3Z); Chris@10: T4a = FMA(KP951056516, T40, KP587785252 * T3Z); Chris@10: ii[WS(rs, 10)] = T45 + T42; Chris@10: T49 = T47 + T46; Chris@10: ii[WS(rs, 6)] = T49 - T4a; Chris@10: ii[WS(rs, 14)] = T4a + T49; Chris@10: T48 = T46 - T47; Chris@10: ii[WS(rs, 2)] = T41 + T48; Chris@10: ii[WS(rs, 18)] = T48 - T41; Chris@10: } Chris@10: { Chris@10: E T3x, T1Y, T3y, T3G, T3I, T3C, T3F, T3H, T3z; Chris@10: T3x = KP559016994 * (T1U - T1X); Chris@10: T1Y = T1U + T1X; Chris@10: T3y = FNMS(KP250000000, T1Y, T1R); Chris@10: T3C = T3A - T3B; Chris@10: T3F = T3D - T3E; Chris@10: T3G = FMA(KP951056516, T3C, KP587785252 * T3F); Chris@10: T3I = FNMS(KP587785252, T3C, KP951056516 * T3F); Chris@10: ri[0] = T1R + T1Y; Chris@10: T3H = T3y - T3x; Chris@10: ri[WS(rs, 12)] = T3H - T3I; Chris@10: ri[WS(rs, 8)] = T3H + T3I; Chris@10: T3z = T3x + T3y; Chris@10: ri[WS(rs, 4)] = T3z - T3G; Chris@10: ri[WS(rs, 16)] = T3z + T3G; Chris@10: } Chris@10: { Chris@10: E T3U, T3L, T3V, T3T, T3Y, T3R, T3S, T3X, T3W; Chris@10: T3U = KP559016994 * (T3J - T3K); Chris@10: T3L = T3J + T3K; Chris@10: T3V = FNMS(KP250000000, T3L, T3Q); Chris@10: T3R = T1S - T1T; Chris@10: T3S = T1V - T1W; Chris@10: T3T = FMA(KP951056516, T3R, KP587785252 * T3S); Chris@10: T3Y = FNMS(KP587785252, T3R, KP951056516 * T3S); Chris@10: ii[0] = T3L + T3Q; Chris@10: T3X = T3V - T3U; Chris@10: ii[WS(rs, 8)] = T3X - T3Y; Chris@10: ii[WS(rs, 12)] = T3Y + T3X; Chris@10: T3W = T3U + T3V; Chris@10: ii[WS(rs, 4)] = T3T + T3W; Chris@10: ii[WS(rs, 16)] = T3W - T3T; Chris@10: } Chris@10: { Chris@10: E T2P, T2N, T2O, T2l, T2R, T29, T2k, T2S, T2Q; Chris@10: T2P = KP559016994 * (T2B - T2M); Chris@10: T2N = T2B + T2M; Chris@10: T2O = FNMS(KP250000000, T2N, T2q); Chris@10: T29 = T23 - T28; Chris@10: T2k = T2e - T2j; Chris@10: T2l = FNMS(KP587785252, T2k, KP951056516 * T29); Chris@10: T2R = FMA(KP951056516, T2k, KP587785252 * T29); Chris@10: ri[WS(rs, 15)] = T2q + T2N; Chris@10: T2S = T2P + T2O; Chris@10: ri[WS(rs, 11)] = T2R + T2S; Chris@10: ri[WS(rs, 19)] = T2S - T2R; Chris@10: T2Q = T2O - T2P; Chris@10: ri[WS(rs, 3)] = T2l + T2Q; Chris@10: ri[WS(rs, 7)] = T2Q - T2l; Chris@10: } Chris@10: { Chris@10: E T4u, T4s, T4t, T4y, T4A, T4w, T4x, T4z, T4v; Chris@10: T4u = KP559016994 * (T4q - T4r); Chris@10: T4s = T4q + T4r; Chris@10: T4t = FNMS(KP250000000, T4s, T4p); Chris@10: T4w = T2G - T2L; Chris@10: T4x = T2v - T2A; Chris@10: T4y = FNMS(KP587785252, T4x, KP951056516 * T4w); Chris@10: T4A = FMA(KP951056516, T4x, KP587785252 * T4w); Chris@10: ii[WS(rs, 15)] = T4s + T4p; Chris@10: T4z = T4u + T4t; Chris@10: ii[WS(rs, 11)] = T4z - T4A; Chris@10: ii[WS(rs, 19)] = T4A + T4z; Chris@10: T4v = T4t - T4u; Chris@10: ii[WS(rs, 3)] = T4v - T4y; Chris@10: ii[WS(rs, 7)] = T4y + T4v; Chris@10: } Chris@10: { Chris@10: E T36, T38, T39, T2Z, T3b, T2V, T2Y, T3c, T3a; Chris@10: T36 = KP559016994 * (T32 - T35); Chris@10: T38 = T32 + T35; Chris@10: T39 = FNMS(KP250000000, T38, T37); Chris@10: T2V = T2T - T2U; Chris@10: T2Y = T2W - T2X; Chris@10: T2Z = FMA(KP951056516, T2V, KP587785252 * T2Y); Chris@10: T3b = FNMS(KP587785252, T2V, KP951056516 * T2Y); Chris@10: ri[WS(rs, 5)] = T37 + T38; Chris@10: T3c = T39 - T36; Chris@10: ri[WS(rs, 13)] = T3b + T3c; Chris@10: ri[WS(rs, 17)] = T3c - T3b; Chris@10: T3a = T36 + T39; Chris@10: ri[WS(rs, 1)] = T2Z + T3a; Chris@10: ri[WS(rs, 9)] = T3a - T2Z; Chris@10: } Chris@10: { Chris@10: E T4d, T4h, T4i, T4m, T4o, T4k, T4l, T4n, T4j; Chris@10: T4d = KP559016994 * (T4b - T4c); Chris@10: T4h = T4b + T4c; Chris@10: T4i = FNMS(KP250000000, T4h, T4g); Chris@10: T4k = T30 - T31; Chris@10: T4l = T33 - T34; Chris@10: T4m = FMA(KP951056516, T4k, KP587785252 * T4l); Chris@10: T4o = FNMS(KP587785252, T4k, KP951056516 * T4l); Chris@10: ii[WS(rs, 5)] = T4h + T4g; Chris@10: T4n = T4i - T4d; Chris@10: ii[WS(rs, 13)] = T4n - T4o; Chris@10: ii[WS(rs, 17)] = T4o + T4n; Chris@10: T4j = T4d + T4i; Chris@10: ii[WS(rs, 1)] = T4j - T4m; Chris@10: ii[WS(rs, 9)] = T4m + T4j; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 0, 20}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, "t1_20", twinstr, &GENUS, {184, 62, 62, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t1_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1_20, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */