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 07:39:54 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 -n 16 -dit -name hf_16 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 174 FP additions, 100 FP multiplications, Chris@10: * (or, 104 additions, 30 multiplications, 70 fused multiply/add), Chris@10: * 95 stack variables, 3 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_16(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DK(KP414213562, +0.414213562373095048801688724209698078569671875); Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 30); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 30, MAKE_VOLATILE_STRIDE(32, rs)) { Chris@10: E T2T, T2Q; Chris@10: { Chris@10: E T3A, T3o, T8, T1I, T2w, T35, T2k, T1s, T2p, T36, T2r, T1F, T3k, T1N, T3z; Chris@10: E Tl, T1U, T2W, T1P, Tz, T2g, T30, T25, T11, TB, TE, T2a, T31, T2h, T1e; Chris@10: E TC, T1X, TH, TK, TG, TD, TJ; Chris@10: { Chris@10: E Ta, Td, Tb, T1J, Tg, Tj, Tf, Tc, Ti; Chris@10: { Chris@10: E T1h, T1k, T1n, T2s, T1i, T1q, T1m, T1j, T1p; Chris@10: { Chris@10: E T1, T3n, T3, T6, T2, T5; Chris@10: T1 = cr[0]; Chris@10: T3n = ci[0]; Chris@10: T3 = cr[WS(rs, 8)]; Chris@10: T6 = ci[WS(rs, 8)]; Chris@10: T2 = W[14]; Chris@10: T5 = W[15]; Chris@10: { Chris@10: E T3l, T4, T1g, T3m, T7; Chris@10: T1h = cr[WS(rs, 15)]; Chris@10: T1k = ci[WS(rs, 15)]; Chris@10: T3l = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T1g = W[28]; Chris@10: T1n = cr[WS(rs, 7)]; Chris@10: T3m = FNMS(T5, T3, T3l); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: T2s = T1g * T1k; Chris@10: T1i = T1g * T1h; Chris@10: T3A = T3n - T3m; Chris@10: T3o = T3m + T3n; Chris@10: T8 = T1 + T7; Chris@10: T1I = T1 - T7; Chris@10: T1q = ci[WS(rs, 7)]; Chris@10: T1m = W[12]; Chris@10: } Chris@10: T1j = W[29]; Chris@10: T1p = W[13]; Chris@10: } Chris@10: { Chris@10: E T1u, T1x, T1v, T2l, T1A, T1D, T1z, T1w, T1C; Chris@10: { Chris@10: E T2t, T1l, T2v, T1r, T2u, T1o, T1t; Chris@10: T1u = cr[WS(rs, 3)]; Chris@10: T2u = T1m * T1q; Chris@10: T1o = T1m * T1n; Chris@10: T2t = FNMS(T1j, T1h, T2s); Chris@10: T1l = FMA(T1j, T1k, T1i); Chris@10: T2v = FNMS(T1p, T1n, T2u); Chris@10: T1r = FMA(T1p, T1q, T1o); Chris@10: T1x = ci[WS(rs, 3)]; Chris@10: T1t = W[4]; Chris@10: T2w = T2t - T2v; Chris@10: T35 = T2t + T2v; Chris@10: T2k = T1l - T1r; Chris@10: T1s = T1l + T1r; Chris@10: T1v = T1t * T1u; Chris@10: T2l = T1t * T1x; Chris@10: } Chris@10: T1A = cr[WS(rs, 11)]; Chris@10: T1D = ci[WS(rs, 11)]; Chris@10: T1z = W[20]; Chris@10: T1w = W[5]; Chris@10: T1C = W[21]; Chris@10: { Chris@10: E T2m, T1y, T2o, T1E, T2n, T1B, T9; Chris@10: Ta = cr[WS(rs, 4)]; Chris@10: T2n = T1z * T1D; Chris@10: T1B = T1z * T1A; Chris@10: T2m = FNMS(T1w, T1u, T2l); Chris@10: T1y = FMA(T1w, T1x, T1v); Chris@10: T2o = FNMS(T1C, T1A, T2n); Chris@10: T1E = FMA(T1C, T1D, T1B); Chris@10: Td = ci[WS(rs, 4)]; Chris@10: T9 = W[6]; Chris@10: T2p = T2m - T2o; Chris@10: T36 = T2m + T2o; Chris@10: T2r = T1E - T1y; Chris@10: T1F = T1y + T1E; Chris@10: Tb = T9 * Ta; Chris@10: T1J = T9 * Td; Chris@10: } Chris@10: Tg = cr[WS(rs, 12)]; Chris@10: Tj = ci[WS(rs, 12)]; Chris@10: Tf = W[22]; Chris@10: Tc = W[7]; Chris@10: Ti = W[23]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TQ, TT, TR, T2c, TW, TZ, TV, TS, TY; Chris@10: { Chris@10: E To, Tr, Tp, T1Q, Tu, Tx, Tt, Tq, Tw; Chris@10: { Chris@10: E T1K, Te, T1M, Tk, T1L, Th, Tn; Chris@10: To = cr[WS(rs, 2)]; Chris@10: T1L = Tf * Tj; Chris@10: Th = Tf * Tg; Chris@10: T1K = FNMS(Tc, Ta, T1J); Chris@10: Te = FMA(Tc, Td, Tb); Chris@10: T1M = FNMS(Ti, Tg, T1L); Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: Tr = ci[WS(rs, 2)]; Chris@10: Tn = W[2]; Chris@10: T3k = T1K + T1M; Chris@10: T1N = T1K - T1M; Chris@10: T3z = Te - Tk; Chris@10: Tl = Te + Tk; Chris@10: Tp = Tn * To; Chris@10: T1Q = Tn * Tr; Chris@10: } Chris@10: Tu = cr[WS(rs, 10)]; Chris@10: Tx = ci[WS(rs, 10)]; Chris@10: Tt = W[18]; Chris@10: Tq = W[3]; Chris@10: Tw = W[19]; Chris@10: { Chris@10: E T1R, Ts, T1T, Ty, T1S, Tv, TP; Chris@10: TQ = cr[WS(rs, 1)]; Chris@10: T1S = Tt * Tx; Chris@10: Tv = Tt * Tu; Chris@10: T1R = FNMS(Tq, To, T1Q); Chris@10: Ts = FMA(Tq, Tr, Tp); Chris@10: T1T = FNMS(Tw, Tu, T1S); Chris@10: Ty = FMA(Tw, Tx, Tv); Chris@10: TT = ci[WS(rs, 1)]; Chris@10: TP = W[0]; Chris@10: T1U = T1R - T1T; Chris@10: T2W = T1R + T1T; Chris@10: T1P = Ts - Ty; Chris@10: Tz = Ts + Ty; Chris@10: TR = TP * TQ; Chris@10: T2c = TP * TT; Chris@10: } Chris@10: TW = cr[WS(rs, 9)]; Chris@10: TZ = ci[WS(rs, 9)]; Chris@10: TV = W[16]; Chris@10: TS = W[1]; Chris@10: TY = W[17]; Chris@10: } Chris@10: { Chris@10: E T13, T16, T14, T26, T19, T1c, T18, T15, T1b; Chris@10: { Chris@10: E T2d, TU, T2f, T10, T2e, TX, T12; Chris@10: T13 = cr[WS(rs, 5)]; Chris@10: T2e = TV * TZ; Chris@10: TX = TV * TW; Chris@10: T2d = FNMS(TS, TQ, T2c); Chris@10: TU = FMA(TS, TT, TR); Chris@10: T2f = FNMS(TY, TW, T2e); Chris@10: T10 = FMA(TY, TZ, TX); Chris@10: T16 = ci[WS(rs, 5)]; Chris@10: T12 = W[8]; Chris@10: T2g = T2d - T2f; Chris@10: T30 = T2d + T2f; Chris@10: T25 = TU - T10; Chris@10: T11 = TU + T10; Chris@10: T14 = T12 * T13; Chris@10: T26 = T12 * T16; Chris@10: } Chris@10: T19 = cr[WS(rs, 13)]; Chris@10: T1c = ci[WS(rs, 13)]; Chris@10: T18 = W[24]; Chris@10: T15 = W[9]; Chris@10: T1b = W[25]; Chris@10: { Chris@10: E T27, T17, T29, T1d, T28, T1a, TA; Chris@10: TB = cr[WS(rs, 14)]; Chris@10: T28 = T18 * T1c; Chris@10: T1a = T18 * T19; Chris@10: T27 = FNMS(T15, T13, T26); Chris@10: T17 = FMA(T15, T16, T14); Chris@10: T29 = FNMS(T1b, T19, T28); Chris@10: T1d = FMA(T1b, T1c, T1a); Chris@10: TE = ci[WS(rs, 14)]; Chris@10: TA = W[26]; Chris@10: T2a = T27 - T29; Chris@10: T31 = T27 + T29; Chris@10: T2h = T17 - T1d; Chris@10: T1e = T17 + T1d; Chris@10: TC = TA * TB; Chris@10: T1X = TA * TE; Chris@10: } Chris@10: TH = cr[WS(rs, 6)]; Chris@10: TK = ci[WS(rs, 6)]; Chris@10: TG = W[10]; Chris@10: TD = W[27]; Chris@10: TJ = W[11]; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2U, T3u, T2Z, T21, T1W, T34, T2X, T37, T3t, T3q, T3e, T32, T3i, T3h; Chris@10: { Chris@10: E T3f, T3r, T1H, T3s, TO, T3g; Chris@10: { Chris@10: E Tm, T1Y, TF, T20, TL, T3p, T1Z, TI; Chris@10: T2U = T8 - Tl; Chris@10: Tm = T8 + Tl; Chris@10: T1Z = TG * TK; Chris@10: TI = TG * TH; Chris@10: T1Y = FNMS(TD, TB, T1X); Chris@10: TF = FMA(TD, TE, TC); Chris@10: T20 = FNMS(TJ, TH, T1Z); Chris@10: TL = FMA(TJ, TK, TI); Chris@10: T3p = T3k + T3o; Chris@10: T3u = T3o - T3k; Chris@10: { Chris@10: E T1f, TM, T1G, T3j, T2V, TN; Chris@10: T2Z = T11 - T1e; Chris@10: T1f = T11 + T1e; Chris@10: T21 = T1Y - T20; Chris@10: T2V = T1Y + T20; Chris@10: T1W = TF - TL; Chris@10: TM = TF + TL; Chris@10: T1G = T1s + T1F; Chris@10: T34 = T1s - T1F; Chris@10: T2X = T2V - T2W; Chris@10: T3j = T2W + T2V; Chris@10: T3f = T35 + T36; Chris@10: T37 = T35 - T36; Chris@10: T3t = Tz - TM; Chris@10: TN = Tz + TM; Chris@10: T3r = T1G - T1f; Chris@10: T1H = T1f + T1G; Chris@10: T3s = T3p - T3j; Chris@10: T3q = T3j + T3p; Chris@10: T3e = Tm - TN; Chris@10: TO = Tm + TN; Chris@10: T3g = T30 + T31; Chris@10: T32 = T30 - T31; Chris@10: } Chris@10: } Chris@10: cr[WS(rs, 12)] = T3r - T3s; Chris@10: ci[WS(rs, 11)] = T3r + T3s; Chris@10: ci[WS(rs, 7)] = TO - T1H; Chris@10: T3i = T3g + T3f; Chris@10: T3h = T3f - T3g; Chris@10: cr[0] = TO + T1H; Chris@10: } Chris@10: { Chris@10: E T3a, T2Y, T3x, T3v; Chris@10: ci[WS(rs, 15)] = T3i + T3q; Chris@10: cr[WS(rs, 8)] = T3i - T3q; Chris@10: ci[WS(rs, 3)] = T3e + T3h; Chris@10: cr[WS(rs, 4)] = T3e - T3h; Chris@10: T3a = T2U + T2X; Chris@10: T2Y = T2U - T2X; Chris@10: T3x = T3u - T3t; Chris@10: T3v = T3t + T3u; Chris@10: { Chris@10: E T2E, T1O, T3B, T3H, T2q, T2x, T3I, T23, T2R, T2O, T2J, T2K, T3C, T2H, T2B; Chris@10: E T2j; Chris@10: { Chris@10: E T2F, T1V, T22, T2G; Chris@10: { Chris@10: E T3b, T33, T3c, T38; Chris@10: T2E = T1I + T1N; Chris@10: T1O = T1I - T1N; Chris@10: T3b = T2Z - T32; Chris@10: T33 = T2Z + T32; Chris@10: T3c = T34 + T37; Chris@10: T38 = T34 - T37; Chris@10: T3B = T3z + T3A; Chris@10: T3H = T3A - T3z; Chris@10: { Chris@10: E T3d, T3y, T3w, T39; Chris@10: T3d = T3b + T3c; Chris@10: T3y = T3c - T3b; Chris@10: T3w = T38 - T33; Chris@10: T39 = T33 + T38; Chris@10: ci[WS(rs, 1)] = FMA(KP707106781, T3d, T3a); Chris@10: cr[WS(rs, 6)] = FNMS(KP707106781, T3d, T3a); Chris@10: ci[WS(rs, 13)] = FMA(KP707106781, T3y, T3x); Chris@10: cr[WS(rs, 10)] = FMS(KP707106781, T3y, T3x); Chris@10: ci[WS(rs, 9)] = FMA(KP707106781, T3w, T3v); Chris@10: cr[WS(rs, 14)] = FMS(KP707106781, T3w, T3v); Chris@10: cr[WS(rs, 2)] = FMA(KP707106781, T39, T2Y); Chris@10: ci[WS(rs, 5)] = FNMS(KP707106781, T39, T2Y); Chris@10: T2F = T1P + T1U; Chris@10: T1V = T1P - T1U; Chris@10: T22 = T1W + T21; Chris@10: T2G = T1W - T21; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2M, T2N, T2b, T2i; Chris@10: T2q = T2k - T2p; Chris@10: T2M = T2k + T2p; Chris@10: T2N = T2w + T2r; Chris@10: T2x = T2r - T2w; Chris@10: T3I = T22 - T1V; Chris@10: T23 = T1V + T22; Chris@10: T2R = FMA(KP414213562, T2M, T2N); Chris@10: T2O = FNMS(KP414213562, T2N, T2M); Chris@10: T2J = T25 + T2a; Chris@10: T2b = T25 - T2a; Chris@10: T2i = T2g + T2h; Chris@10: T2K = T2g - T2h; Chris@10: T3C = T2F - T2G; Chris@10: T2H = T2F + T2G; Chris@10: T2B = FMA(KP414213562, T2b, T2i); Chris@10: T2j = FNMS(KP414213562, T2i, T2b); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2A, T3G, T2P, T2D, T3E, T3F, T3D, T2I; Chris@10: { Chris@10: E T24, T2L, T2C, T2y, T3J, T3L, T3K, T2S, T2z, T3M; Chris@10: T2A = FNMS(KP707106781, T23, T1O); Chris@10: T24 = FMA(KP707106781, T23, T1O); Chris@10: T2S = FNMS(KP414213562, T2J, T2K); Chris@10: T2L = FMA(KP414213562, T2K, T2J); Chris@10: T2C = FMA(KP414213562, T2q, T2x); Chris@10: T2y = FNMS(KP414213562, T2x, T2q); Chris@10: T3J = FMA(KP707106781, T3I, T3H); Chris@10: T3L = FNMS(KP707106781, T3I, T3H); Chris@10: T2T = T2R - T2S; Chris@10: T3K = T2S + T2R; Chris@10: T3G = T2y - T2j; Chris@10: T2z = T2j + T2y; Chris@10: T3M = T2O - T2L; Chris@10: T2P = T2L + T2O; Chris@10: ci[WS(rs, 14)] = FMA(KP923879532, T3K, T3J); Chris@10: cr[WS(rs, 9)] = FMS(KP923879532, T3K, T3J); Chris@10: ci[0] = FMA(KP923879532, T2z, T24); Chris@10: cr[WS(rs, 7)] = FNMS(KP923879532, T2z, T24); Chris@10: cr[WS(rs, 13)] = FMS(KP923879532, T3M, T3L); Chris@10: ci[WS(rs, 10)] = FMA(KP923879532, T3M, T3L); Chris@10: T2D = T2B + T2C; Chris@10: T3E = T2C - T2B; Chris@10: } Chris@10: T2Q = FNMS(KP707106781, T2H, T2E); Chris@10: T2I = FMA(KP707106781, T2H, T2E); Chris@10: T3F = FNMS(KP707106781, T3C, T3B); Chris@10: T3D = FMA(KP707106781, T3C, T3B); Chris@10: cr[WS(rs, 3)] = FMA(KP923879532, T2D, T2A); Chris@10: ci[WS(rs, 4)] = FNMS(KP923879532, T2D, T2A); Chris@10: cr[WS(rs, 1)] = FMA(KP923879532, T2P, T2I); Chris@10: ci[WS(rs, 6)] = FNMS(KP923879532, T2P, T2I); Chris@10: ci[WS(rs, 8)] = FMA(KP923879532, T3E, T3D); Chris@10: cr[WS(rs, 15)] = FMS(KP923879532, T3E, T3D); Chris@10: ci[WS(rs, 12)] = FMA(KP923879532, T3G, T3F); Chris@10: cr[WS(rs, 11)] = FMS(KP923879532, T3G, T3F); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: ci[WS(rs, 2)] = FMA(KP923879532, T2T, T2Q); Chris@10: cr[WS(rs, 5)] = FNMS(KP923879532, T2T, T2Q); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 16}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 16, "hf_16", twinstr, &GENUS, {104, 30, 70, 0} }; Chris@10: Chris@10: void X(codelet_hf_16) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_16, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 16 -dit -name hf_16 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 174 FP additions, 84 FP multiplications, Chris@10: * (or, 136 additions, 46 multiplications, 38 fused multiply/add), Chris@10: * 52 stack variables, 3 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_16(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP382683432, +0.382683432365089771728459984030398866761344562); Chris@10: DK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 30); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 30, MAKE_VOLATILE_STRIDE(32, rs)) { Chris@10: E T7, T38, T1t, T2U, Ti, T37, T1w, T2R, Tu, T2t, T1C, T2c, TF, T2s, T1H; Chris@10: E T2d, T1f, T1q, T2B, T2C, T2D, T2E, T1Z, T2k, T24, T2j, TS, T13, T2w, T2x; Chris@10: E T2y, T2z, T1O, T2h, T1T, T2g; Chris@10: { Chris@10: E T1, T2T, T6, T2S; Chris@10: T1 = cr[0]; Chris@10: T2T = ci[0]; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = cr[WS(rs, 8)]; Chris@10: T5 = ci[WS(rs, 8)]; Chris@10: T2 = W[14]; Chris@10: T4 = W[15]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: T2S = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: T7 = T1 + T6; Chris@10: T38 = T2T - T2S; Chris@10: T1t = T1 - T6; Chris@10: T2U = T2S + T2T; Chris@10: } Chris@10: { Chris@10: E Tc, T1u, Th, T1v; Chris@10: { Chris@10: E T9, Tb, T8, Ta; Chris@10: T9 = cr[WS(rs, 4)]; Chris@10: Tb = ci[WS(rs, 4)]; Chris@10: T8 = W[6]; Chris@10: Ta = W[7]; Chris@10: Tc = FMA(T8, T9, Ta * Tb); Chris@10: T1u = FNMS(Ta, T9, T8 * Tb); Chris@10: } Chris@10: { Chris@10: E Te, Tg, Td, Tf; Chris@10: Te = cr[WS(rs, 12)]; Chris@10: Tg = ci[WS(rs, 12)]; Chris@10: Td = W[22]; Chris@10: Tf = W[23]; Chris@10: Th = FMA(Td, Te, Tf * Tg); Chris@10: T1v = FNMS(Tf, Te, Td * Tg); Chris@10: } Chris@10: Ti = Tc + Th; Chris@10: T37 = Tc - Th; Chris@10: T1w = T1u - T1v; Chris@10: T2R = T1u + T1v; Chris@10: } Chris@10: { Chris@10: E To, T1z, Tt, T1A, T1y, T1B; Chris@10: { Chris@10: E Tl, Tn, Tk, Tm; Chris@10: Tl = cr[WS(rs, 2)]; Chris@10: Tn = ci[WS(rs, 2)]; Chris@10: Tk = W[2]; Chris@10: Tm = W[3]; Chris@10: To = FMA(Tk, Tl, Tm * Tn); Chris@10: T1z = FNMS(Tm, Tl, Tk * Tn); Chris@10: } Chris@10: { Chris@10: E Tq, Ts, Tp, Tr; Chris@10: Tq = cr[WS(rs, 10)]; Chris@10: Ts = ci[WS(rs, 10)]; Chris@10: Tp = W[18]; Chris@10: Tr = W[19]; Chris@10: Tt = FMA(Tp, Tq, Tr * Ts); Chris@10: T1A = FNMS(Tr, Tq, Tp * Ts); Chris@10: } Chris@10: Tu = To + Tt; Chris@10: T2t = T1z + T1A; Chris@10: T1y = To - Tt; Chris@10: T1B = T1z - T1A; Chris@10: T1C = T1y - T1B; Chris@10: T2c = T1y + T1B; Chris@10: } Chris@10: { Chris@10: E Tz, T1E, TE, T1F, T1D, T1G; Chris@10: { Chris@10: E Tw, Ty, Tv, Tx; Chris@10: Tw = cr[WS(rs, 14)]; Chris@10: Ty = ci[WS(rs, 14)]; Chris@10: Tv = W[26]; Chris@10: Tx = W[27]; Chris@10: Tz = FMA(Tv, Tw, Tx * Ty); Chris@10: T1E = FNMS(Tx, Tw, Tv * Ty); Chris@10: } Chris@10: { Chris@10: E TB, TD, TA, TC; Chris@10: TB = cr[WS(rs, 6)]; Chris@10: TD = ci[WS(rs, 6)]; Chris@10: TA = W[10]; Chris@10: TC = W[11]; Chris@10: TE = FMA(TA, TB, TC * TD); Chris@10: T1F = FNMS(TC, TB, TA * TD); Chris@10: } Chris@10: TF = Tz + TE; Chris@10: T2s = T1E + T1F; Chris@10: T1D = Tz - TE; Chris@10: T1G = T1E - T1F; Chris@10: T1H = T1D + T1G; Chris@10: T2d = T1D - T1G; Chris@10: } Chris@10: { Chris@10: E T19, T1V, T1p, T22, T1e, T1W, T1k, T21; Chris@10: { Chris@10: E T16, T18, T15, T17; Chris@10: T16 = cr[WS(rs, 15)]; Chris@10: T18 = ci[WS(rs, 15)]; Chris@10: T15 = W[28]; Chris@10: T17 = W[29]; Chris@10: T19 = FMA(T15, T16, T17 * T18); Chris@10: T1V = FNMS(T17, T16, T15 * T18); Chris@10: } Chris@10: { Chris@10: E T1m, T1o, T1l, T1n; Chris@10: T1m = cr[WS(rs, 11)]; Chris@10: T1o = ci[WS(rs, 11)]; Chris@10: T1l = W[20]; Chris@10: T1n = W[21]; Chris@10: T1p = FMA(T1l, T1m, T1n * T1o); Chris@10: T22 = FNMS(T1n, T1m, T1l * T1o); Chris@10: } Chris@10: { Chris@10: E T1b, T1d, T1a, T1c; Chris@10: T1b = cr[WS(rs, 7)]; Chris@10: T1d = ci[WS(rs, 7)]; Chris@10: T1a = W[12]; Chris@10: T1c = W[13]; Chris@10: T1e = FMA(T1a, T1b, T1c * T1d); Chris@10: T1W = FNMS(T1c, T1b, T1a * T1d); Chris@10: } Chris@10: { Chris@10: E T1h, T1j, T1g, T1i; Chris@10: T1h = cr[WS(rs, 3)]; Chris@10: T1j = ci[WS(rs, 3)]; Chris@10: T1g = W[4]; Chris@10: T1i = W[5]; Chris@10: T1k = FMA(T1g, T1h, T1i * T1j); Chris@10: T21 = FNMS(T1i, T1h, T1g * T1j); Chris@10: } Chris@10: T1f = T19 + T1e; Chris@10: T1q = T1k + T1p; Chris@10: T2B = T1f - T1q; Chris@10: T2C = T1V + T1W; Chris@10: T2D = T21 + T22; Chris@10: T2E = T2C - T2D; Chris@10: { Chris@10: E T1X, T1Y, T20, T23; Chris@10: T1X = T1V - T1W; Chris@10: T1Y = T1k - T1p; Chris@10: T1Z = T1X + T1Y; Chris@10: T2k = T1X - T1Y; Chris@10: T20 = T19 - T1e; Chris@10: T23 = T21 - T22; Chris@10: T24 = T20 - T23; Chris@10: T2j = T20 + T23; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TM, T1P, T12, T1M, TR, T1Q, TX, T1L; Chris@10: { Chris@10: E TJ, TL, TI, TK; Chris@10: TJ = cr[WS(rs, 1)]; Chris@10: TL = ci[WS(rs, 1)]; Chris@10: TI = W[0]; Chris@10: TK = W[1]; Chris@10: TM = FMA(TI, TJ, TK * TL); Chris@10: T1P = FNMS(TK, TJ, TI * TL); Chris@10: } Chris@10: { Chris@10: E TZ, T11, TY, T10; Chris@10: TZ = cr[WS(rs, 13)]; Chris@10: T11 = ci[WS(rs, 13)]; Chris@10: TY = W[24]; Chris@10: T10 = W[25]; Chris@10: T12 = FMA(TY, TZ, T10 * T11); Chris@10: T1M = FNMS(T10, TZ, TY * T11); Chris@10: } Chris@10: { Chris@10: E TO, TQ, TN, TP; Chris@10: TO = cr[WS(rs, 9)]; Chris@10: TQ = ci[WS(rs, 9)]; Chris@10: TN = W[16]; Chris@10: TP = W[17]; Chris@10: TR = FMA(TN, TO, TP * TQ); Chris@10: T1Q = FNMS(TP, TO, TN * TQ); Chris@10: } Chris@10: { Chris@10: E TU, TW, TT, TV; Chris@10: TU = cr[WS(rs, 5)]; Chris@10: TW = ci[WS(rs, 5)]; Chris@10: TT = W[8]; Chris@10: TV = W[9]; Chris@10: TX = FMA(TT, TU, TV * TW); Chris@10: T1L = FNMS(TV, TU, TT * TW); Chris@10: } Chris@10: TS = TM + TR; Chris@10: T13 = TX + T12; Chris@10: T2w = TS - T13; Chris@10: T2x = T1P + T1Q; Chris@10: T2y = T1L + T1M; Chris@10: T2z = T2x - T2y; Chris@10: { Chris@10: E T1K, T1N, T1R, T1S; Chris@10: T1K = TM - TR; Chris@10: T1N = T1L - T1M; Chris@10: T1O = T1K - T1N; Chris@10: T2h = T1K + T1N; Chris@10: T1R = T1P - T1Q; Chris@10: T1S = TX - T12; Chris@10: T1T = T1R + T1S; Chris@10: T2g = T1R - T1S; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1J, T27, T3a, T3c, T26, T3b, T2a, T35; Chris@10: { Chris@10: E T1x, T1I, T36, T39; Chris@10: T1x = T1t - T1w; Chris@10: T1I = KP707106781 * (T1C + T1H); Chris@10: T1J = T1x + T1I; Chris@10: T27 = T1x - T1I; Chris@10: T36 = KP707106781 * (T2c - T2d); Chris@10: T39 = T37 + T38; Chris@10: T3a = T36 + T39; Chris@10: T3c = T39 - T36; Chris@10: } Chris@10: { Chris@10: E T1U, T25, T28, T29; Chris@10: T1U = FNMS(KP382683432, T1T, KP923879532 * T1O); Chris@10: T25 = FMA(KP382683432, T1Z, KP923879532 * T24); Chris@10: T26 = T1U + T25; Chris@10: T3b = T25 - T1U; Chris@10: T28 = FMA(KP923879532, T1T, KP382683432 * T1O); Chris@10: T29 = FNMS(KP923879532, T1Z, KP382683432 * T24); Chris@10: T2a = T28 + T29; Chris@10: T35 = T29 - T28; Chris@10: } Chris@10: cr[WS(rs, 7)] = T1J - T26; Chris@10: cr[WS(rs, 11)] = T3b - T3c; Chris@10: ci[WS(rs, 12)] = T3b + T3c; Chris@10: ci[0] = T1J + T26; Chris@10: ci[WS(rs, 4)] = T27 - T2a; Chris@10: cr[WS(rs, 15)] = T35 - T3a; Chris@10: ci[WS(rs, 8)] = T35 + T3a; Chris@10: cr[WS(rs, 3)] = T27 + T2a; Chris@10: } Chris@10: { Chris@10: E TH, T2L, T2W, T2Y, T1s, T2X, T2O, T2P; Chris@10: { Chris@10: E Tj, TG, T2Q, T2V; Chris@10: Tj = T7 + Ti; Chris@10: TG = Tu + TF; Chris@10: TH = Tj + TG; Chris@10: T2L = Tj - TG; Chris@10: T2Q = T2t + T2s; Chris@10: T2V = T2R + T2U; Chris@10: T2W = T2Q + T2V; Chris@10: T2Y = T2V - T2Q; Chris@10: } Chris@10: { Chris@10: E T14, T1r, T2M, T2N; Chris@10: T14 = TS + T13; Chris@10: T1r = T1f + T1q; Chris@10: T1s = T14 + T1r; Chris@10: T2X = T1r - T14; Chris@10: T2M = T2C + T2D; Chris@10: T2N = T2x + T2y; Chris@10: T2O = T2M - T2N; Chris@10: T2P = T2N + T2M; Chris@10: } Chris@10: ci[WS(rs, 7)] = TH - T1s; Chris@10: cr[WS(rs, 12)] = T2X - T2Y; Chris@10: ci[WS(rs, 11)] = T2X + T2Y; Chris@10: cr[0] = TH + T1s; Chris@10: cr[WS(rs, 4)] = T2L - T2O; Chris@10: cr[WS(rs, 8)] = T2P - T2W; Chris@10: ci[WS(rs, 15)] = T2P + T2W; Chris@10: ci[WS(rs, 3)] = T2L + T2O; Chris@10: } Chris@10: { Chris@10: E T2f, T2n, T3g, T3i, T2m, T3h, T2q, T3d; Chris@10: { Chris@10: E T2b, T2e, T3e, T3f; Chris@10: T2b = T1t + T1w; Chris@10: T2e = KP707106781 * (T2c + T2d); Chris@10: T2f = T2b + T2e; Chris@10: T2n = T2b - T2e; Chris@10: T3e = KP707106781 * (T1H - T1C); Chris@10: T3f = T38 - T37; Chris@10: T3g = T3e + T3f; Chris@10: T3i = T3f - T3e; Chris@10: } Chris@10: { Chris@10: E T2i, T2l, T2o, T2p; Chris@10: T2i = FMA(KP382683432, T2g, KP923879532 * T2h); Chris@10: T2l = FNMS(KP382683432, T2k, KP923879532 * T2j); Chris@10: T2m = T2i + T2l; Chris@10: T3h = T2l - T2i; Chris@10: T2o = FNMS(KP923879532, T2g, KP382683432 * T2h); Chris@10: T2p = FMA(KP923879532, T2k, KP382683432 * T2j); Chris@10: T2q = T2o + T2p; Chris@10: T3d = T2p - T2o; Chris@10: } Chris@10: ci[WS(rs, 6)] = T2f - T2m; Chris@10: cr[WS(rs, 13)] = T3h - T3i; Chris@10: ci[WS(rs, 10)] = T3h + T3i; Chris@10: cr[WS(rs, 1)] = T2f + T2m; Chris@10: cr[WS(rs, 5)] = T2n - T2q; Chris@10: cr[WS(rs, 9)] = T3d - T3g; Chris@10: ci[WS(rs, 14)] = T3d + T3g; Chris@10: ci[WS(rs, 2)] = T2n + T2q; Chris@10: } Chris@10: { Chris@10: E T2v, T2H, T32, T34, T2G, T2Z, T2K, T33; Chris@10: { Chris@10: E T2r, T2u, T30, T31; Chris@10: T2r = T7 - Ti; Chris@10: T2u = T2s - T2t; Chris@10: T2v = T2r - T2u; Chris@10: T2H = T2r + T2u; Chris@10: T30 = Tu - TF; Chris@10: T31 = T2U - T2R; Chris@10: T32 = T30 + T31; Chris@10: T34 = T31 - T30; Chris@10: } Chris@10: { Chris@10: E T2A, T2F, T2I, T2J; Chris@10: T2A = T2w + T2z; Chris@10: T2F = T2B - T2E; Chris@10: T2G = KP707106781 * (T2A + T2F); Chris@10: T2Z = KP707106781 * (T2F - T2A); Chris@10: T2I = T2w - T2z; Chris@10: T2J = T2B + T2E; Chris@10: T2K = KP707106781 * (T2I + T2J); Chris@10: T33 = KP707106781 * (T2J - T2I); Chris@10: } Chris@10: ci[WS(rs, 5)] = T2v - T2G; Chris@10: cr[WS(rs, 10)] = T33 - T34; Chris@10: ci[WS(rs, 13)] = T33 + T34; Chris@10: cr[WS(rs, 2)] = T2v + T2G; Chris@10: cr[WS(rs, 6)] = T2H - T2K; Chris@10: cr[WS(rs, 14)] = T2Z - T32; Chris@10: ci[WS(rs, 9)] = T2Z + T32; Chris@10: ci[WS(rs, 1)] = T2H + T2K; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 16}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 16, "hf_16", twinstr, &GENUS, {136, 46, 38, 0} }; Chris@10: Chris@10: void X(codelet_hf_16) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_16, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */