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 07:41:14 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_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 15 -dif -name hb_15 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 184 FP additions, 140 FP multiplications, Chris@10: * (or, 72 additions, 28 multiplications, 112 fused multiply/add), Chris@10: * 93 stack variables, 6 constants, and 60 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_15(R *cr, R *ci, 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: 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) * 28); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 28, MAKE_VOLATILE_STRIDE(30, rs)) { Chris@10: E T3v, T3u, T3r, T3w, T3t; Chris@10: { Chris@10: E T5, T11, T1C, T2U, T2f, T3f, T19, T18, TS, TH, T14, T16, T3g, T3a, Ts; Chris@10: E Tv, T37, T3h, T28, T2h, T1M, T21, T2g, T3n, T2X, T1P, T30, T3m, T1J, T2m; Chris@10: { Chris@10: E T1, TX, T2, T3, TY, TZ; Chris@10: T1 = cr[0]; Chris@10: TX = ci[WS(rs, 14)]; Chris@10: T2 = cr[WS(rs, 5)]; Chris@10: T3 = ci[WS(rs, 4)]; Chris@10: TY = ci[WS(rs, 9)]; Chris@10: TZ = cr[WS(rs, 10)]; Chris@10: { Chris@10: E T1W, T23, T1D, Ta, Tl, T1K, T1Z, T1H, T1G, Tf, TR, T1Y, T26, TI, T1O; Chris@10: E T1N, Tq, TG, T25, Tx, Ty, Tz, TL, T1E; Chris@10: { Chris@10: E Tb, TQ, TN, TO, Te; Chris@10: { Chris@10: E T6, Th, Ti, Tj, T9, Tc, Td, Tk; Chris@10: { Chris@10: E T7, T8, T2e, T4; Chris@10: T6 = cr[WS(rs, 3)]; Chris@10: T2e = T2 - T3; Chris@10: T4 = T2 + T3; Chris@10: { Chris@10: E T1B, T10, T1A, T2d; Chris@10: T1B = TY + TZ; Chris@10: T10 = TY - TZ; Chris@10: T7 = ci[WS(rs, 6)]; Chris@10: T5 = T1 + T4; Chris@10: T1A = FNMS(KP500000000, T4, T1); Chris@10: T11 = TX + T10; Chris@10: T2d = FNMS(KP500000000, T10, TX); Chris@10: T1C = FNMS(KP866025403, T1B, T1A); Chris@10: T2U = FMA(KP866025403, T1B, T1A); Chris@10: T2f = FMA(KP866025403, T2e, T2d); Chris@10: T3f = FNMS(KP866025403, T2e, T2d); Chris@10: T8 = ci[WS(rs, 1)]; Chris@10: } Chris@10: Th = cr[WS(rs, 6)]; Chris@10: Ti = ci[WS(rs, 3)]; Chris@10: Tj = cr[WS(rs, 1)]; Chris@10: T9 = T7 + T8; Chris@10: T1W = T7 - T8; Chris@10: } Chris@10: Tb = ci[WS(rs, 2)]; Chris@10: T23 = Ti - Tj; Chris@10: Tk = Ti + Tj; Chris@10: T1D = FNMS(KP500000000, T9, T6); Chris@10: Ta = T6 + T9; Chris@10: Tc = cr[WS(rs, 2)]; Chris@10: Tl = Th + Tk; Chris@10: T1K = FNMS(KP500000000, Tk, Th); Chris@10: Td = cr[WS(rs, 7)]; Chris@10: TQ = cr[WS(rs, 12)]; Chris@10: TN = ci[WS(rs, 12)]; Chris@10: TO = ci[WS(rs, 7)]; Chris@10: Te = Tc + Td; Chris@10: T1Z = Tc - Td; Chris@10: } Chris@10: { Chris@10: E Tm, TF, TC, TD, Tp, Tn, To, TP, TJ, TK, TE; Chris@10: Tm = ci[WS(rs, 5)]; Chris@10: T1H = TO - TN; Chris@10: TP = TN + TO; Chris@10: T1G = FNMS(KP500000000, Te, Tb); Chris@10: Tf = Tb + Te; Chris@10: Tn = ci[0]; Chris@10: TR = TP - TQ; Chris@10: T1Y = FMA(KP500000000, TP, TQ); Chris@10: To = cr[WS(rs, 4)]; Chris@10: TF = cr[WS(rs, 9)]; Chris@10: TC = ci[WS(rs, 10)]; Chris@10: TD = cr[WS(rs, 14)]; Chris@10: Tp = Tn + To; Chris@10: T26 = Tn - To; Chris@10: TI = ci[WS(rs, 11)]; Chris@10: T1O = TC + TD; Chris@10: TE = TC - TD; Chris@10: T1N = FNMS(KP500000000, Tp, Tm); Chris@10: Tq = Tm + Tp; Chris@10: TJ = cr[WS(rs, 8)]; Chris@10: TG = TE - TF; Chris@10: T25 = FMA(KP500000000, TE, TF); Chris@10: TK = cr[WS(rs, 13)]; Chris@10: Tx = ci[WS(rs, 8)]; Chris@10: Ty = ci[WS(rs, 13)]; Chris@10: Tz = cr[WS(rs, 11)]; Chris@10: TL = TJ + TK; Chris@10: T1E = TJ - TK; Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tg, T1L, Tr, T22, T12, T1X, T38, T13, T39, T20; Chris@10: { Chris@10: E TA, T1V, TM, TB; Chris@10: Tg = Ta + Tf; Chris@10: T19 = Ta - Tf; Chris@10: T1L = Ty + Tz; Chris@10: TA = Ty - Tz; Chris@10: T1V = FMA(KP500000000, TL, TI); Chris@10: TM = TI - TL; Chris@10: T18 = Tl - Tq; Chris@10: Tr = Tl + Tq; Chris@10: TB = Tx + TA; Chris@10: T22 = FNMS(KP500000000, TA, Tx); Chris@10: T12 = TM + TR; Chris@10: TS = TM - TR; Chris@10: T1X = FMA(KP866025403, T1W, T1V); Chris@10: T38 = FNMS(KP866025403, T1W, T1V); Chris@10: T13 = TB + TG; Chris@10: TH = TB - TG; Chris@10: T39 = FMA(KP866025403, T1Z, T1Y); Chris@10: T20 = FNMS(KP866025403, T1Z, T1Y); Chris@10: } Chris@10: { Chris@10: E T35, T24, T27, T36; Chris@10: T14 = T12 + T13; Chris@10: T16 = T12 - T13; Chris@10: T3g = T38 - T39; Chris@10: T3a = T38 + T39; Chris@10: T35 = FNMS(KP866025403, T23, T22); Chris@10: T24 = FMA(KP866025403, T23, T22); Chris@10: Ts = Tg + Tr; Chris@10: Tv = Tg - Tr; Chris@10: T27 = FNMS(KP866025403, T26, T25); Chris@10: T36 = FMA(KP866025403, T26, T25); Chris@10: T37 = T35 + T36; Chris@10: T3h = T35 - T36; Chris@10: T28 = T24 + T27; Chris@10: T2h = T24 - T27; Chris@10: { Chris@10: E T1F, T1I, T2Y, T2Z, T2V, T2W; Chris@10: T2V = FNMS(KP866025403, T1E, T1D); Chris@10: T1F = FMA(KP866025403, T1E, T1D); Chris@10: T1I = FMA(KP866025403, T1H, T1G); Chris@10: T2W = FNMS(KP866025403, T1H, T1G); Chris@10: T2Y = FNMS(KP866025403, T1L, T1K); Chris@10: T1M = FMA(KP866025403, T1L, T1K); Chris@10: T21 = T1X + T20; Chris@10: T2g = T1X - T20; Chris@10: T3n = T2V - T2W; Chris@10: T2X = T2V + T2W; Chris@10: T2Z = FNMS(KP866025403, T1O, T1N); Chris@10: T1P = FMA(KP866025403, T1O, T1N); Chris@10: T30 = T2Y + T2Z; Chris@10: T3m = T2Y - T2Z; Chris@10: T1J = T1F + T1I; Chris@10: T2m = T1F - T1I; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T31, T33, T2n, T1Q; Chris@10: cr[0] = T5 + Ts; Chris@10: T31 = T2X + T30; Chris@10: T33 = T2X - T30; Chris@10: T2n = T1M - T1P; Chris@10: T1Q = T1M + T1P; Chris@10: ci[0] = T11 + T14; Chris@10: { Chris@10: E T1T, T1R, T1r, T1o, T1n; Chris@10: { Chris@10: E T1q, T1a, TT, T1l, Tu, T17, T1p, T15; Chris@10: T1q = FMA(KP618033988, T18, T19); Chris@10: T1a = FNMS(KP618033988, T19, T18); Chris@10: T1T = T1J - T1Q; Chris@10: T1R = T1J + T1Q; Chris@10: T15 = FNMS(KP250000000, T14, T11); Chris@10: TT = FNMS(KP618033988, TS, TH); Chris@10: T1l = FMA(KP618033988, TH, TS); Chris@10: Tu = FNMS(KP250000000, Ts, T5); Chris@10: T17 = FNMS(KP559016994, T16, T15); Chris@10: T1p = FMA(KP559016994, T16, T15); Chris@10: { Chris@10: E T1h, T1m, T1e, T1x, T1w, T1v, T1g, T1d; Chris@10: { Chris@10: E TW, T1b, Tt, T1u, TU, T1k, Tw; Chris@10: TW = W[5]; Chris@10: T1k = FMA(KP559016994, Tv, Tu); Chris@10: Tw = FNMS(KP559016994, Tv, Tu); Chris@10: T1b = FMA(KP951056516, T1a, T17); Chris@10: T1h = FNMS(KP951056516, T1a, T17); Chris@10: Tt = W[4]; Chris@10: T1m = FNMS(KP951056516, T1l, T1k); Chris@10: T1u = FMA(KP951056516, T1l, T1k); Chris@10: T1e = FMA(KP951056516, TT, Tw); Chris@10: TU = FNMS(KP951056516, TT, Tw); Chris@10: { Chris@10: E T1t, TV, T1c, T1y; Chris@10: T1x = FNMS(KP951056516, T1q, T1p); Chris@10: T1r = FMA(KP951056516, T1q, T1p); Chris@10: T1w = W[17]; Chris@10: T1t = W[16]; Chris@10: TV = Tt * TU; Chris@10: T1c = TW * TU; Chris@10: T1y = T1w * T1u; Chris@10: T1v = T1t * T1u; Chris@10: cr[WS(rs, 3)] = FNMS(TW, T1b, TV); Chris@10: ci[WS(rs, 3)] = FMA(Tt, T1b, T1c); Chris@10: ci[WS(rs, 9)] = FMA(T1t, T1x, T1y); Chris@10: } Chris@10: } Chris@10: cr[WS(rs, 9)] = FNMS(T1w, T1x, T1v); Chris@10: T1g = W[23]; Chris@10: T1d = W[22]; Chris@10: { Chris@10: E T1j, T1s, T1i, T1f; Chris@10: T1o = W[11]; Chris@10: T1i = T1g * T1e; Chris@10: T1f = T1d * T1e; Chris@10: T1j = W[10]; Chris@10: T1s = T1o * T1m; Chris@10: ci[WS(rs, 12)] = FMA(T1d, T1h, T1i); Chris@10: cr[WS(rs, 12)] = FNMS(T1g, T1h, T1f); Chris@10: T1n = T1j * T1m; Chris@10: ci[WS(rs, 6)] = FMA(T1j, T1r, T1s); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2v, T2u, T2r, T2w, T2t; Chris@10: { Chris@10: E T1S, T2N, T2o, T2E, T2Q, T2P, T2k, T2S, T29, T2z, T2R, T2j, T2O, T2i; Chris@10: cr[WS(rs, 6)] = FNMS(T1o, T1r, T1n); Chris@10: T1S = FNMS(KP250000000, T1R, T1C); Chris@10: T2O = T1C + T1R; Chris@10: T2N = W[18]; Chris@10: T2o = FMA(KP618033988, T2n, T2m); Chris@10: T2E = FNMS(KP618033988, T2m, T2n); Chris@10: T2Q = W[19]; Chris@10: T2P = T2N * T2O; Chris@10: T2i = T2g + T2h; Chris@10: T2k = T2g - T2h; Chris@10: T2S = T2Q * T2O; Chris@10: T29 = FMA(KP618033988, T28, T21); Chris@10: T2z = FNMS(KP618033988, T21, T28); Chris@10: T2R = T2f + T2i; Chris@10: T2j = FNMS(KP250000000, T2i, T2f); Chris@10: { Chris@10: E T2D, T2p, T2I, T2A, T2a, T2s, T2c, T1z, T2l, T1U, T2y; Chris@10: cr[WS(rs, 10)] = FNMS(T2Q, T2R, T2P); Chris@10: T2l = FMA(KP559016994, T2k, T2j); Chris@10: T2D = FNMS(KP559016994, T2k, T2j); Chris@10: T1U = FMA(KP559016994, T1T, T1S); Chris@10: T2y = FNMS(KP559016994, T1T, T1S); Chris@10: ci[WS(rs, 10)] = FMA(T2N, T2R, T2S); Chris@10: T2p = FMA(KP951056516, T2o, T2l); Chris@10: T2v = FNMS(KP951056516, T2o, T2l); Chris@10: T2I = FNMS(KP951056516, T2z, T2y); Chris@10: T2A = FMA(KP951056516, T2z, T2y); Chris@10: T2a = FNMS(KP951056516, T29, T1U); Chris@10: T2s = FMA(KP951056516, T29, T1U); Chris@10: T2c = W[1]; Chris@10: T1z = W[0]; Chris@10: { Chris@10: E T2F, T2L, T2K, T2J; Chris@10: { Chris@10: E T2H, T2M, T2q, T2b; Chris@10: T2F = FNMS(KP951056516, T2E, T2D); Chris@10: T2L = FMA(KP951056516, T2E, T2D); Chris@10: T2K = W[25]; Chris@10: T2q = T2c * T2a; Chris@10: T2b = T1z * T2a; Chris@10: T2H = W[24]; Chris@10: T2M = T2K * T2I; Chris@10: ci[WS(rs, 1)] = FMA(T1z, T2p, T2q); Chris@10: cr[WS(rs, 1)] = FNMS(T2c, T2p, T2b); Chris@10: T2J = T2H * T2I; Chris@10: ci[WS(rs, 13)] = FMA(T2H, T2L, T2M); Chris@10: } Chris@10: { Chris@10: E T2C, T2x, T2G, T2B; Chris@10: T2C = W[13]; Chris@10: cr[WS(rs, 13)] = FNMS(T2K, T2L, T2J); Chris@10: T2x = W[12]; Chris@10: T2G = T2C * T2A; Chris@10: T2u = W[7]; Chris@10: T2B = T2x * T2A; Chris@10: T2r = W[6]; Chris@10: ci[WS(rs, 7)] = FMA(T2x, T2F, T2G); Chris@10: T2w = T2u * T2s; Chris@10: cr[WS(rs, 7)] = FNMS(T2C, T2F, T2B); Chris@10: T2t = T2r * T2s; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T32, T3N, T3E, T3o, T3Q, T3P, T3k, T3S, T3z, T3b, T3j, T3R, T3O, T3i; Chris@10: ci[WS(rs, 4)] = FMA(T2r, T2v, T2w); Chris@10: cr[WS(rs, 4)] = FNMS(T2u, T2v, T2t); Chris@10: T3O = T2U + T31; Chris@10: T32 = FNMS(KP250000000, T31, T2U); Chris@10: T3N = W[8]; Chris@10: T3E = FMA(KP618033988, T3m, T3n); Chris@10: T3o = FNMS(KP618033988, T3n, T3m); Chris@10: T3Q = W[9]; Chris@10: T3P = T3N * T3O; Chris@10: T3k = T3g - T3h; Chris@10: T3i = T3g + T3h; Chris@10: T3S = T3Q * T3O; Chris@10: T3z = FMA(KP618033988, T37, T3a); Chris@10: T3b = FNMS(KP618033988, T3a, T37); Chris@10: T3j = FNMS(KP250000000, T3i, T3f); Chris@10: T3R = T3f + T3i; Chris@10: { Chris@10: E T3D, T3p, T3A, T3I, T3s, T3c, T3e, T2T, T3l, T3y, T34; Chris@10: cr[WS(rs, 5)] = FNMS(T3Q, T3R, T3P); Chris@10: T3D = FMA(KP559016994, T3k, T3j); Chris@10: T3l = FNMS(KP559016994, T3k, T3j); Chris@10: T3y = FMA(KP559016994, T33, T32); Chris@10: T34 = FNMS(KP559016994, T33, T32); Chris@10: ci[WS(rs, 5)] = FMA(T3N, T3R, T3S); Chris@10: T3v = FMA(KP951056516, T3o, T3l); Chris@10: T3p = FNMS(KP951056516, T3o, T3l); Chris@10: T3A = FNMS(KP951056516, T3z, T3y); Chris@10: T3I = FMA(KP951056516, T3z, T3y); Chris@10: T3s = FNMS(KP951056516, T3b, T34); Chris@10: T3c = FMA(KP951056516, T3b, T34); Chris@10: T3e = W[3]; Chris@10: T2T = W[2]; Chris@10: { Chris@10: E T3L, T3F, T3K, T3J; Chris@10: { Chris@10: E T3H, T3M, T3q, T3d; Chris@10: T3L = FNMS(KP951056516, T3E, T3D); Chris@10: T3F = FMA(KP951056516, T3E, T3D); Chris@10: T3K = W[27]; Chris@10: T3q = T3e * T3c; Chris@10: T3d = T2T * T3c; Chris@10: T3H = W[26]; Chris@10: T3M = T3K * T3I; Chris@10: ci[WS(rs, 2)] = FMA(T2T, T3p, T3q); Chris@10: cr[WS(rs, 2)] = FNMS(T3e, T3p, T3d); Chris@10: T3J = T3H * T3I; Chris@10: ci[WS(rs, 14)] = FMA(T3H, T3L, T3M); Chris@10: } Chris@10: { Chris@10: E T3C, T3x, T3G, T3B; Chris@10: T3C = W[21]; Chris@10: cr[WS(rs, 14)] = FNMS(T3K, T3L, T3J); Chris@10: T3x = W[20]; Chris@10: T3G = T3C * T3A; Chris@10: T3u = W[15]; Chris@10: T3B = T3x * T3A; Chris@10: T3r = W[14]; Chris@10: ci[WS(rs, 11)] = FMA(T3x, T3F, T3G); Chris@10: T3w = T3u * T3s; Chris@10: cr[WS(rs, 11)] = FNMS(T3C, T3F, T3B); Chris@10: T3t = T3r * T3s; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: ci[WS(rs, 8)] = FMA(T3r, T3v, T3w); Chris@10: cr[WS(rs, 8)] = FNMS(T3u, T3v, T3t); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 15}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 15, "hb_15", twinstr, &GENUS, {72, 28, 112, 0} }; Chris@10: Chris@10: void X(codelet_hb_15) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_15, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 15 -dif -name hb_15 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 184 FP additions, 112 FP multiplications, Chris@10: * (or, 128 additions, 56 multiplications, 56 fused multiply/add), Chris@10: * 75 stack variables, 6 constants, and 60 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_15(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 28); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 28, MAKE_VOLATILE_STRIDE(30, rs)) { Chris@10: E T5, T10, T1J, T2C, T2c, T2M, TH, T18, T17, TS, T2Q, T2R, T2S, Tg, Tr; Chris@10: E Ts, T11, T12, T13, T2N, T2O, T2P, T1u, T1x, T1y, T1W, T1Z, T28, T1P, T1S; Chris@10: E T27, T1B, T1E, T1F, T2G, T2H, T2I, T2D, T2E, T2F; Chris@10: { Chris@10: E T1, TW, T4, T2a, TZ, T1I, T1H, T2b; Chris@10: T1 = cr[0]; Chris@10: TW = ci[WS(rs, 14)]; Chris@10: { Chris@10: E T2, T3, TX, TY; Chris@10: T2 = cr[WS(rs, 5)]; Chris@10: T3 = ci[WS(rs, 4)]; Chris@10: T4 = T2 + T3; Chris@10: T2a = KP866025403 * (T2 - T3); Chris@10: TX = ci[WS(rs, 9)]; Chris@10: TY = cr[WS(rs, 10)]; Chris@10: TZ = TX - TY; Chris@10: T1I = KP866025403 * (TX + TY); Chris@10: } Chris@10: T5 = T1 + T4; Chris@10: T10 = TW + TZ; Chris@10: T1H = FNMS(KP500000000, T4, T1); Chris@10: T1J = T1H - T1I; Chris@10: T2C = T1H + T1I; Chris@10: T2b = FNMS(KP500000000, TZ, TW); Chris@10: T2c = T2a + T2b; Chris@10: T2M = T2b - T2a; Chris@10: } Chris@10: { Chris@10: E Ta, T1N, T1s, Tl, T1U, T1z, Tf, T1Q, T1v, TG, T1R, T1w, Tq, T1X, T1C; Chris@10: E TM, T1V, T1A, TB, T1O, T1t, TR, T1Y, T1D; Chris@10: { Chris@10: E T6, T7, T8, T9; Chris@10: T6 = cr[WS(rs, 3)]; Chris@10: T7 = ci[WS(rs, 6)]; Chris@10: T8 = ci[WS(rs, 1)]; Chris@10: T9 = T7 + T8; Chris@10: Ta = T6 + T9; Chris@10: T1N = KP866025403 * (T7 - T8); Chris@10: T1s = FNMS(KP500000000, T9, T6); Chris@10: } Chris@10: { Chris@10: E Th, Ti, Tj, Tk; Chris@10: Th = cr[WS(rs, 6)]; Chris@10: Ti = ci[WS(rs, 3)]; Chris@10: Tj = cr[WS(rs, 1)]; Chris@10: Tk = Ti + Tj; Chris@10: Tl = Th + Tk; Chris@10: T1U = KP866025403 * (Ti - Tj); Chris@10: T1z = FNMS(KP500000000, Tk, Th); Chris@10: } Chris@10: { Chris@10: E Tb, Tc, Td, Te; Chris@10: Tb = ci[WS(rs, 2)]; Chris@10: Tc = cr[WS(rs, 2)]; Chris@10: Td = cr[WS(rs, 7)]; Chris@10: Te = Tc + Td; Chris@10: Tf = Tb + Te; Chris@10: T1Q = KP866025403 * (Tc - Td); Chris@10: T1v = FNMS(KP500000000, Te, Tb); Chris@10: } Chris@10: { Chris@10: E TF, TC, TD, TE; Chris@10: TF = cr[WS(rs, 12)]; Chris@10: TC = ci[WS(rs, 12)]; Chris@10: TD = ci[WS(rs, 7)]; Chris@10: TE = TC + TD; Chris@10: TG = TE - TF; Chris@10: T1R = FMA(KP500000000, TE, TF); Chris@10: T1w = KP866025403 * (TD - TC); Chris@10: } Chris@10: { Chris@10: E Tm, Tn, To, Tp; Chris@10: Tm = ci[WS(rs, 5)]; Chris@10: Tn = ci[0]; Chris@10: To = cr[WS(rs, 4)]; Chris@10: Tp = Tn + To; Chris@10: Tq = Tm + Tp; Chris@10: T1X = KP866025403 * (Tn - To); Chris@10: T1C = FNMS(KP500000000, Tp, Tm); Chris@10: } Chris@10: { Chris@10: E TI, TJ, TK, TL; Chris@10: TI = ci[WS(rs, 8)]; Chris@10: TJ = ci[WS(rs, 13)]; Chris@10: TK = cr[WS(rs, 11)]; Chris@10: TL = TJ - TK; Chris@10: TM = TI + TL; Chris@10: T1V = FNMS(KP500000000, TL, TI); Chris@10: T1A = KP866025403 * (TJ + TK); Chris@10: } Chris@10: { Chris@10: E Tx, Ty, Tz, TA; Chris@10: Tx = ci[WS(rs, 11)]; Chris@10: Ty = cr[WS(rs, 8)]; Chris@10: Tz = cr[WS(rs, 13)]; Chris@10: TA = Ty + Tz; Chris@10: TB = Tx - TA; Chris@10: T1O = FMA(KP500000000, TA, Tx); Chris@10: T1t = KP866025403 * (Ty - Tz); Chris@10: } Chris@10: { Chris@10: E TQ, TN, TO, TP; Chris@10: TQ = cr[WS(rs, 9)]; Chris@10: TN = ci[WS(rs, 10)]; Chris@10: TO = cr[WS(rs, 14)]; Chris@10: TP = TN - TO; Chris@10: TR = TP - TQ; Chris@10: T1Y = FMA(KP500000000, TP, TQ); Chris@10: T1D = KP866025403 * (TN + TO); Chris@10: } Chris@10: TH = TB - TG; Chris@10: T18 = Tl - Tq; Chris@10: T17 = Ta - Tf; Chris@10: TS = TM - TR; Chris@10: T2Q = T1V - T1U; Chris@10: T2R = T1X + T1Y; Chris@10: T2S = T2Q - T2R; Chris@10: Tg = Ta + Tf; Chris@10: Tr = Tl + Tq; Chris@10: Ts = Tg + Tr; Chris@10: T11 = TB + TG; Chris@10: T12 = TM + TR; Chris@10: T13 = T11 + T12; Chris@10: T2N = T1O - T1N; Chris@10: T2O = T1Q + T1R; Chris@10: T2P = T2N - T2O; Chris@10: T1u = T1s + T1t; Chris@10: T1x = T1v + T1w; Chris@10: T1y = T1u + T1x; Chris@10: T1W = T1U + T1V; Chris@10: T1Z = T1X - T1Y; Chris@10: T28 = T1W + T1Z; Chris@10: T1P = T1N + T1O; Chris@10: T1S = T1Q - T1R; Chris@10: T27 = T1P + T1S; Chris@10: T1B = T1z + T1A; Chris@10: T1E = T1C + T1D; Chris@10: T1F = T1B + T1E; Chris@10: T2G = T1z - T1A; Chris@10: T2H = T1C - T1D; Chris@10: T2I = T2G + T2H; Chris@10: T2D = T1s - T1t; Chris@10: T2E = T1v - T1w; Chris@10: T2F = T2D + T2E; Chris@10: } Chris@10: cr[0] = T5 + Ts; Chris@10: ci[0] = T10 + T13; Chris@10: { Chris@10: E TT, T19, T1k, T1h, T16, T1l, Tw, T1g; Chris@10: TT = FNMS(KP951056516, TS, KP587785252 * TH); Chris@10: T19 = FNMS(KP951056516, T18, KP587785252 * T17); Chris@10: T1k = FMA(KP951056516, T17, KP587785252 * T18); Chris@10: T1h = FMA(KP951056516, TH, KP587785252 * TS); Chris@10: { Chris@10: E T14, T15, Tu, Tv; Chris@10: T14 = FNMS(KP250000000, T13, T10); Chris@10: T15 = KP559016994 * (T11 - T12); Chris@10: T16 = T14 - T15; Chris@10: T1l = T15 + T14; Chris@10: Tu = FNMS(KP250000000, Ts, T5); Chris@10: Tv = KP559016994 * (Tg - Tr); Chris@10: Tw = Tu - Tv; Chris@10: T1g = Tv + Tu; Chris@10: } Chris@10: { Chris@10: E TU, T1a, Tt, TV; Chris@10: TU = Tw + TT; Chris@10: T1a = T16 - T19; Chris@10: Tt = W[4]; Chris@10: TV = W[5]; Chris@10: cr[WS(rs, 3)] = FNMS(TV, T1a, Tt * TU); Chris@10: ci[WS(rs, 3)] = FMA(TV, TU, Tt * T1a); Chris@10: } Chris@10: { Chris@10: E T1o, T1q, T1n, T1p; Chris@10: T1o = T1g + T1h; Chris@10: T1q = T1l - T1k; Chris@10: T1n = W[16]; Chris@10: T1p = W[17]; Chris@10: cr[WS(rs, 9)] = FNMS(T1p, T1q, T1n * T1o); Chris@10: ci[WS(rs, 9)] = FMA(T1p, T1o, T1n * T1q); Chris@10: } Chris@10: { Chris@10: E T1c, T1e, T1b, T1d; Chris@10: T1c = Tw - TT; Chris@10: T1e = T19 + T16; Chris@10: T1b = W[22]; Chris@10: T1d = W[23]; Chris@10: cr[WS(rs, 12)] = FNMS(T1d, T1e, T1b * T1c); Chris@10: ci[WS(rs, 12)] = FMA(T1d, T1c, T1b * T1e); Chris@10: } Chris@10: { Chris@10: E T1i, T1m, T1f, T1j; Chris@10: T1i = T1g - T1h; Chris@10: T1m = T1k + T1l; Chris@10: T1f = W[10]; Chris@10: T1j = W[11]; Chris@10: cr[WS(rs, 6)] = FNMS(T1j, T1m, T1f * T1i); Chris@10: ci[WS(rs, 6)] = FMA(T1j, T1i, T1f * T1m); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T21, T2n, T26, T2q, T1M, T2y, T2m, T2f, T2A, T2r, T2x, T2z; Chris@10: { Chris@10: E T1T, T20, T24, T25; Chris@10: T1T = T1P - T1S; Chris@10: T20 = T1W - T1Z; Chris@10: T21 = FMA(KP951056516, T1T, KP587785252 * T20); Chris@10: T2n = FNMS(KP951056516, T20, KP587785252 * T1T); Chris@10: T24 = T1u - T1x; Chris@10: T25 = T1B - T1E; Chris@10: T26 = FMA(KP951056516, T24, KP587785252 * T25); Chris@10: T2q = FNMS(KP951056516, T25, KP587785252 * T24); Chris@10: } Chris@10: { Chris@10: E T1G, T1K, T1L, T29, T2d, T2e; Chris@10: T1G = KP559016994 * (T1y - T1F); Chris@10: T1K = T1y + T1F; Chris@10: T1L = FNMS(KP250000000, T1K, T1J); Chris@10: T1M = T1G + T1L; Chris@10: T2y = T1J + T1K; Chris@10: T2m = T1L - T1G; Chris@10: T29 = KP559016994 * (T27 - T28); Chris@10: T2d = T27 + T28; Chris@10: T2e = FNMS(KP250000000, T2d, T2c); Chris@10: T2f = T29 + T2e; Chris@10: T2A = T2c + T2d; Chris@10: T2r = T2e - T29; Chris@10: } Chris@10: T2x = W[18]; Chris@10: T2z = W[19]; Chris@10: cr[WS(rs, 10)] = FNMS(T2z, T2A, T2x * T2y); Chris@10: ci[WS(rs, 10)] = FMA(T2z, T2y, T2x * T2A); Chris@10: { Chris@10: E T2u, T2w, T2t, T2v; Chris@10: T2u = T2m + T2n; Chris@10: T2w = T2r - T2q; Chris@10: T2t = W[24]; Chris@10: T2v = W[25]; Chris@10: cr[WS(rs, 13)] = FNMS(T2v, T2w, T2t * T2u); Chris@10: ci[WS(rs, 13)] = FMA(T2v, T2u, T2t * T2w); Chris@10: } Chris@10: { Chris@10: E T22, T2g, T1r, T23; Chris@10: T22 = T1M - T21; Chris@10: T2g = T26 + T2f; Chris@10: T1r = W[0]; Chris@10: T23 = W[1]; Chris@10: cr[WS(rs, 1)] = FNMS(T23, T2g, T1r * T22); Chris@10: ci[WS(rs, 1)] = FMA(T23, T22, T1r * T2g); Chris@10: } Chris@10: { Chris@10: E T2i, T2k, T2h, T2j; Chris@10: T2i = T1M + T21; Chris@10: T2k = T2f - T26; Chris@10: T2h = W[6]; Chris@10: T2j = W[7]; Chris@10: cr[WS(rs, 4)] = FNMS(T2j, T2k, T2h * T2i); Chris@10: ci[WS(rs, 4)] = FMA(T2j, T2i, T2h * T2k); Chris@10: } Chris@10: { Chris@10: E T2o, T2s, T2l, T2p; Chris@10: T2o = T2m - T2n; Chris@10: T2s = T2q + T2r; Chris@10: T2l = W[12]; Chris@10: T2p = W[13]; Chris@10: cr[WS(rs, 7)] = FNMS(T2p, T2s, T2l * T2o); Chris@10: ci[WS(rs, 7)] = FMA(T2p, T2o, T2l * T2s); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T31, T3h, T36, T3k, T2K, T3g, T2Y, T2U, T3l, T39, T2B, T2L; Chris@10: { Chris@10: E T2Z, T30, T34, T35; Chris@10: T2Z = T2N + T2O; Chris@10: T30 = T2Q + T2R; Chris@10: T31 = FNMS(KP951056516, T30, KP587785252 * T2Z); Chris@10: T3h = FMA(KP951056516, T2Z, KP587785252 * T30); Chris@10: T34 = T2D - T2E; Chris@10: T35 = T2G - T2H; Chris@10: T36 = FNMS(KP951056516, T35, KP587785252 * T34); Chris@10: T3k = FMA(KP951056516, T34, KP587785252 * T35); Chris@10: } Chris@10: { Chris@10: E T2X, T2J, T2W, T38, T2T, T37; Chris@10: T2X = KP559016994 * (T2F - T2I); Chris@10: T2J = T2F + T2I; Chris@10: T2W = FNMS(KP250000000, T2J, T2C); Chris@10: T2K = T2C + T2J; Chris@10: T3g = T2X + T2W; Chris@10: T2Y = T2W - T2X; Chris@10: T38 = KP559016994 * (T2P - T2S); Chris@10: T2T = T2P + T2S; Chris@10: T37 = FNMS(KP250000000, T2T, T2M); Chris@10: T2U = T2M + T2T; Chris@10: T3l = T38 + T37; Chris@10: T39 = T37 - T38; Chris@10: } Chris@10: T2B = W[8]; Chris@10: T2L = W[9]; Chris@10: cr[WS(rs, 5)] = FNMS(T2L, T2U, T2B * T2K); Chris@10: ci[WS(rs, 5)] = FMA(T2L, T2K, T2B * T2U); Chris@10: { Chris@10: E T3o, T3q, T3n, T3p; Chris@10: T3o = T3g + T3h; Chris@10: T3q = T3l - T3k; Chris@10: T3n = W[26]; Chris@10: T3p = W[27]; Chris@10: cr[WS(rs, 14)] = FNMS(T3p, T3q, T3n * T3o); Chris@10: ci[WS(rs, 14)] = FMA(T3n, T3q, T3p * T3o); Chris@10: } Chris@10: { Chris@10: E T32, T3a, T2V, T33; Chris@10: T32 = T2Y - T31; Chris@10: T3a = T36 + T39; Chris@10: T2V = W[2]; Chris@10: T33 = W[3]; Chris@10: cr[WS(rs, 2)] = FNMS(T33, T3a, T2V * T32); Chris@10: ci[WS(rs, 2)] = FMA(T2V, T3a, T33 * T32); Chris@10: } Chris@10: { Chris@10: E T3c, T3e, T3b, T3d; Chris@10: T3c = T2Y + T31; Chris@10: T3e = T39 - T36; Chris@10: T3b = W[14]; Chris@10: T3d = W[15]; Chris@10: cr[WS(rs, 8)] = FNMS(T3d, T3e, T3b * T3c); Chris@10: ci[WS(rs, 8)] = FMA(T3b, T3e, T3d * T3c); Chris@10: } Chris@10: { Chris@10: E T3i, T3m, T3f, T3j; Chris@10: T3i = T3g - T3h; Chris@10: T3m = T3k + T3l; Chris@10: T3f = W[20]; Chris@10: T3j = W[21]; Chris@10: cr[WS(rs, 11)] = FNMS(T3j, T3m, T3f * T3i); Chris@10: ci[WS(rs, 11)] = FMA(T3f, T3m, T3j * T3i); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 15}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 15, "hb_15", twinstr, &GENUS, {128, 56, 56, 0} }; Chris@10: Chris@10: void X(codelet_hb_15) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_15, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */