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 07:40:46 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_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 16 -dit -name hc2cfdft_16 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 206 FP additions, 132 FP multiplications, Chris@10: * (or, 136 additions, 62 multiplications, 70 fused multiply/add), Chris@10: * 96 stack variables, 4 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_16(R *Rp, R *Ip, R *Rm, R *Im, 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: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 30); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 30, MAKE_VOLATILE_STRIDE(64, rs)) { Chris@10: E T4d, T4g; Chris@10: { Chris@10: E T1f, T2e, T3D, T1K, T2g, T1c, T3H, T2W, T2j, TR, T3E, T2R, T2l, T11, T3G; Chris@10: E T1v, T3p, T2s, Tl, T3o, T3w, T2G, T3z, T1Y, T23, T20, T2H, T21, T29, Tz; Chris@10: E T26, TE, TA, T2v, T2J, T27, Tv, T2u, TB, T22, T28; Chris@10: { Chris@10: E T1o, T1u, T2T, T2V; Chris@10: { Chris@10: E T1I, T1A, T16, T1C, T1H, T1G, T2U, T1z, T1b, T1x, T1w; Chris@10: { Chris@10: E T1d, T1e, T14, T15; Chris@10: T1d = Ip[0]; Chris@10: T1e = Im[0]; Chris@10: T14 = Ip[WS(rs, 4)]; Chris@10: T15 = Im[WS(rs, 4)]; Chris@10: { Chris@10: E T1F, T1D, T1E, T19, T1a; Chris@10: T1D = Rm[0]; Chris@10: T1I = T1d + T1e; Chris@10: T1f = T1d - T1e; Chris@10: T1E = Rp[0]; Chris@10: T1A = T14 + T15; Chris@10: T16 = T14 - T15; Chris@10: T1C = W[0]; Chris@10: T2e = T1E + T1D; Chris@10: T1F = T1D - T1E; Chris@10: T1H = W[1]; Chris@10: T19 = Rp[WS(rs, 4)]; Chris@10: T1a = Rm[WS(rs, 4)]; Chris@10: T1G = T1C * T1F; Chris@10: T2U = T1H * T1F; Chris@10: T1z = W[17]; Chris@10: T1b = T19 + T1a; Chris@10: T1x = T1a - T19; Chris@10: T1w = W[16]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2S, T1y, T13, T18; Chris@10: T2S = T1z * T1x; Chris@10: T1y = T1w * T1x; Chris@10: T13 = W[14]; Chris@10: T18 = W[15]; Chris@10: { Chris@10: E T1J, T1B, T2f, T17; Chris@10: T1J = FNMS(T1H, T1I, T1G); Chris@10: T1B = FNMS(T1z, T1A, T1y); Chris@10: T2f = T13 * T1b; Chris@10: T17 = T13 * T16; Chris@10: T2T = FMA(T1w, T1A, T2S); Chris@10: T3D = T1J - T1B; Chris@10: T1K = T1B + T1J; Chris@10: T2g = FMA(T18, T16, T2f); Chris@10: T1c = FNMS(T18, T1b, T17); Chris@10: T2V = FMA(T1C, T1I, T2U); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1n, TL, T1m, T1j, TQ, T1l, T2N, TV, T1t, T10, T1q, T1s, T1p, T1r, T2O; Chris@10: E T2Q; Chris@10: { Chris@10: E TO, TP, TJ, TK; Chris@10: TJ = Ip[WS(rs, 2)]; Chris@10: TK = Im[WS(rs, 2)]; Chris@10: TO = Rp[WS(rs, 2)]; Chris@10: T3H = T2V - T2T; Chris@10: T2W = T2T + T2V; Chris@10: T1n = TJ + TK; Chris@10: TL = TJ - TK; Chris@10: TP = Rm[WS(rs, 2)]; Chris@10: T1m = W[9]; Chris@10: T1j = W[8]; Chris@10: { Chris@10: E TT, T1k, TU, TY, TZ; Chris@10: TT = Ip[WS(rs, 6)]; Chris@10: TQ = TO + TP; Chris@10: T1k = TP - TO; Chris@10: TU = Im[WS(rs, 6)]; Chris@10: TY = Rp[WS(rs, 6)]; Chris@10: TZ = Rm[WS(rs, 6)]; Chris@10: T1l = T1j * T1k; Chris@10: T2N = T1m * T1k; Chris@10: TV = TT - TU; Chris@10: T1t = TT + TU; Chris@10: T10 = TY + TZ; Chris@10: T1q = TZ - TY; Chris@10: T1s = W[25]; Chris@10: T1p = W[24]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TN, T2P, T2i, TM, TI; Chris@10: TI = W[6]; Chris@10: TN = W[7]; Chris@10: T2P = T1s * T1q; Chris@10: T1r = T1p * T1q; Chris@10: T2i = TI * TQ; Chris@10: TM = TI * TL; Chris@10: T2O = FMA(T1j, T1n, T2N); Chris@10: T2Q = FMA(T1p, T1t, T2P); Chris@10: T2j = FMA(TN, TL, T2i); Chris@10: TR = FNMS(TN, TQ, TM); Chris@10: } Chris@10: { Chris@10: E TX, T2k, TW, TS; Chris@10: TS = W[22]; Chris@10: T3E = T2O - T2Q; Chris@10: T2R = T2O + T2Q; Chris@10: TX = W[23]; Chris@10: T2k = TS * T10; Chris@10: TW = TS * TV; Chris@10: T1o = FNMS(T1m, T1n, T1l); Chris@10: T1u = FNMS(T1s, T1t, T1r); Chris@10: T2l = FMA(TX, TV, T2k); Chris@10: T11 = FNMS(TX, T10, TW); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1Q, T1N, T2C, T1O, T1W, Te, T1T, Tj, Tf, T2q, T2E, T1U, Ta, T2p, Tg; Chris@10: E T1P, T1V; Chris@10: { Chris@10: E T4, T9, T5, T2o, Tb, T1S, T1, T1M, T6; Chris@10: { Chris@10: E T2, T3, T7, T8; Chris@10: T2 = Ip[WS(rs, 1)]; Chris@10: T3G = T1o - T1u; Chris@10: T1v = T1o + T1u; Chris@10: T3 = Im[WS(rs, 1)]; Chris@10: T7 = Rp[WS(rs, 1)]; Chris@10: T8 = Rm[WS(rs, 1)]; Chris@10: T1 = W[2]; Chris@10: T1Q = T2 + T3; Chris@10: T4 = T2 - T3; Chris@10: T1N = T7 - T8; Chris@10: T9 = T7 + T8; Chris@10: T1M = W[4]; Chris@10: T5 = T1 * T4; Chris@10: } Chris@10: { Chris@10: E Tc, Td, Th, Ti; Chris@10: Tc = Ip[WS(rs, 5)]; Chris@10: T2o = T1 * T9; Chris@10: T2C = T1M * T1Q; Chris@10: T1O = T1M * T1N; Chris@10: Td = Im[WS(rs, 5)]; Chris@10: Th = Rp[WS(rs, 5)]; Chris@10: Ti = Rm[WS(rs, 5)]; Chris@10: Tb = W[18]; Chris@10: T1W = Tc + Td; Chris@10: Te = Tc - Td; Chris@10: T1T = Th - Ti; Chris@10: Tj = Th + Ti; Chris@10: T1S = W[20]; Chris@10: Tf = Tb * Te; Chris@10: } Chris@10: T6 = W[3]; Chris@10: T2q = Tb * Tj; Chris@10: T2E = T1S * T1W; Chris@10: T1U = T1S * T1T; Chris@10: Ta = FNMS(T6, T9, T5); Chris@10: T2p = FMA(T6, T4, T2o); Chris@10: Tg = W[19]; Chris@10: T1P = W[5]; Chris@10: T1V = W[21]; Chris@10: } Chris@10: { Chris@10: E Tp, Tu, Tq, T2t, Tw, T25, Tm, T1Z, Tr; Chris@10: { Chris@10: E Tn, To, Ts, Tt, T2r, Tk; Chris@10: Tn = Ip[WS(rs, 7)]; Chris@10: T2r = FMA(Tg, Te, T2q); Chris@10: Tk = FNMS(Tg, Tj, Tf); Chris@10: { Chris@10: E T2D, T1R, T2F, T1X; Chris@10: T2D = FNMS(T1P, T1N, T2C); Chris@10: T1R = FMA(T1P, T1Q, T1O); Chris@10: T2F = FNMS(T1V, T1T, T2E); Chris@10: T1X = FMA(T1V, T1W, T1U); Chris@10: T3p = T2p - T2r; Chris@10: T2s = T2p + T2r; Chris@10: Tl = Ta + Tk; Chris@10: T3o = Ta - Tk; Chris@10: T3w = T2F - T2D; Chris@10: T2G = T2D + T2F; Chris@10: T3z = T1X - T1R; Chris@10: T1Y = T1R + T1X; Chris@10: To = Im[WS(rs, 7)]; Chris@10: } Chris@10: Ts = Rp[WS(rs, 7)]; Chris@10: Tt = Rm[WS(rs, 7)]; Chris@10: Tm = W[26]; Chris@10: T23 = Tn + To; Chris@10: Tp = Tn - To; Chris@10: T20 = Ts - Tt; Chris@10: Tu = Ts + Tt; Chris@10: T1Z = W[28]; Chris@10: Tq = Tm * Tp; Chris@10: } Chris@10: { Chris@10: E Tx, Ty, TC, TD; Chris@10: Tx = Ip[WS(rs, 3)]; Chris@10: T2t = Tm * Tu; Chris@10: T2H = T1Z * T23; Chris@10: T21 = T1Z * T20; Chris@10: Ty = Im[WS(rs, 3)]; Chris@10: TC = Rp[WS(rs, 3)]; Chris@10: TD = Rm[WS(rs, 3)]; Chris@10: Tw = W[10]; Chris@10: T29 = Tx + Ty; Chris@10: Tz = Tx - Ty; Chris@10: T26 = TC - TD; Chris@10: TE = TC + TD; Chris@10: T25 = W[12]; Chris@10: TA = Tw * Tz; Chris@10: } Chris@10: Tr = W[27]; Chris@10: T2v = Tw * TE; Chris@10: T2J = T25 * T29; Chris@10: T27 = T25 * T26; Chris@10: Tv = FNMS(Tr, Tu, Tq); Chris@10: T2u = FMA(Tr, Tp, T2t); Chris@10: TB = W[11]; Chris@10: T22 = W[29]; Chris@10: T28 = W[13]; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3r, T3s, T3A, T3x, T3M, T3l, T3L, T3m, T3f, T3i; Chris@10: { Chris@10: E T3c, TH, T36, T3g, T3h, T39, T32, T1h, T2A, T2d, T2h, T31, T2y, T30, T2Y; Chris@10: E T2m, T2B, T1i; Chris@10: { Chris@10: E T2x, T2M, T1L, T2c, T2X, T12, T1g; Chris@10: { Chris@10: E TG, T2b, T34, T2L, T2w, TF, T37, T38, T35; Chris@10: T2w = FMA(TB, Tz, T2v); Chris@10: TF = FNMS(TB, TE, TA); Chris@10: { Chris@10: E T2I, T24, T2K, T2a; Chris@10: T2I = FNMS(T22, T20, T2H); Chris@10: T24 = FMA(T22, T23, T21); Chris@10: T2K = FNMS(T28, T26, T2J); Chris@10: T2a = FMA(T28, T29, T27); Chris@10: T3r = T2u - T2w; Chris@10: T2x = T2u + T2w; Chris@10: TG = Tv + TF; Chris@10: T3s = Tv - TF; Chris@10: T2L = T2I + T2K; Chris@10: T3A = T2I - T2K; Chris@10: T3x = T2a - T24; Chris@10: T2b = T24 + T2a; Chris@10: } Chris@10: T2M = T2G + T2L; Chris@10: T34 = T2L - T2G; Chris@10: T37 = T1K - T1v; Chris@10: T1L = T1v + T1K; Chris@10: T2c = T1Y + T2b; Chris@10: T35 = T1Y - T2b; Chris@10: T3c = Tl - TG; Chris@10: TH = Tl + TG; Chris@10: T38 = T2W - T2R; Chris@10: T2X = T2R + T2W; Chris@10: T36 = T34 + T35; Chris@10: T3g = T34 - T35; Chris@10: T3M = TR - T11; Chris@10: T12 = TR + T11; Chris@10: T3h = T37 + T38; Chris@10: T39 = T37 - T38; Chris@10: T1g = T1c + T1f; Chris@10: T3l = T1f - T1c; Chris@10: } Chris@10: T32 = T1g - T12; Chris@10: T1h = T12 + T1g; Chris@10: T2A = T2c + T1L; Chris@10: T2d = T1L - T2c; Chris@10: T3L = T2e - T2g; Chris@10: T2h = T2e + T2g; Chris@10: T31 = T2x - T2s; Chris@10: T2y = T2s + T2x; Chris@10: T30 = T2M + T2X; Chris@10: T2Y = T2M - T2X; Chris@10: T2m = T2j + T2l; Chris@10: T3m = T2j - T2l; Chris@10: } Chris@10: T2B = T1h - TH; Chris@10: T1i = TH + T1h; Chris@10: { Chris@10: E T3e, T3d, T3j, T3k; Chris@10: { Chris@10: E T33, T3b, T2z, T2Z, T3a, T2n; Chris@10: T3f = T32 - T31; Chris@10: T33 = T31 + T32; Chris@10: T3b = T2h - T2m; Chris@10: T2n = T2h + T2m; Chris@10: Im[WS(rs, 7)] = KP500000000 * (T2d - T1i); Chris@10: Ip[0] = KP500000000 * (T1i + T2d); Chris@10: Im[WS(rs, 3)] = KP500000000 * (T2Y - T2B); Chris@10: Ip[WS(rs, 4)] = KP500000000 * (T2B + T2Y); Chris@10: T2z = T2n - T2y; Chris@10: T2Z = T2n + T2y; Chris@10: T3a = T36 + T39; Chris@10: T3e = T39 - T36; Chris@10: T3d = T3b - T3c; Chris@10: T3j = T3b + T3c; Chris@10: Rp[WS(rs, 4)] = KP500000000 * (T2z + T2A); Chris@10: Rm[WS(rs, 3)] = KP500000000 * (T2z - T2A); Chris@10: Rp[0] = KP500000000 * (T2Z + T30); Chris@10: Rm[WS(rs, 7)] = KP500000000 * (T2Z - T30); Chris@10: Im[WS(rs, 5)] = -(KP500000000 * (FNMS(KP707106781, T3a, T33))); Chris@10: Ip[WS(rs, 2)] = KP500000000 * (FMA(KP707106781, T3a, T33)); Chris@10: T3k = T3g + T3h; Chris@10: T3i = T3g - T3h; Chris@10: } Chris@10: Rp[WS(rs, 2)] = KP500000000 * (FMA(KP707106781, T3k, T3j)); Chris@10: Rm[WS(rs, 5)] = KP500000000 * (FNMS(KP707106781, T3k, T3j)); Chris@10: Rp[WS(rs, 6)] = KP500000000 * (FMA(KP707106781, T3e, T3d)); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (FNMS(KP707106781, T3e, T3d)); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3Z, T3n, T3F, T3I, T4e, T44, T4f, T47, T4a, T3u, T3U, T3C, T49, T3N, T40; Chris@10: E T3Q; Chris@10: { Chris@10: E T3y, T3B, T3O, T3q, T3t, T3P; Chris@10: { Chris@10: E T42, T43, T45, T46; Chris@10: T3y = T3w + T3x; Chris@10: T42 = T3w - T3x; Chris@10: Im[WS(rs, 1)] = -(KP500000000 * (FNMS(KP707106781, T3i, T3f))); Chris@10: Ip[WS(rs, 6)] = KP500000000 * (FMA(KP707106781, T3i, T3f)); Chris@10: T3Z = T3m + T3l; Chris@10: T3n = T3l - T3m; Chris@10: T43 = T3A - T3z; Chris@10: T3B = T3z + T3A; Chris@10: T3F = T3D - T3E; Chris@10: T45 = T3E + T3D; Chris@10: T46 = T3H - T3G; Chris@10: T3I = T3G + T3H; Chris@10: T3O = T3p + T3o; Chris@10: T3q = T3o - T3p; Chris@10: T4e = FNMS(KP414213562, T42, T43); Chris@10: T44 = FMA(KP414213562, T43, T42); Chris@10: T4f = FNMS(KP414213562, T45, T46); Chris@10: T47 = FMA(KP414213562, T46, T45); Chris@10: T3t = T3r + T3s; Chris@10: T3P = T3r - T3s; Chris@10: } Chris@10: T4a = T3q - T3t; Chris@10: T3u = T3q + T3t; Chris@10: T3U = FNMS(KP414213562, T3y, T3B); Chris@10: T3C = FMA(KP414213562, T3B, T3y); Chris@10: T49 = T3L - T3M; Chris@10: T3N = T3L + T3M; Chris@10: T40 = T3P - T3O; Chris@10: T3Q = T3O + T3P; Chris@10: } Chris@10: { Chris@10: E T3T, T3v, T3X, T3R, T3J, T3V; Chris@10: T3T = FNMS(KP707106781, T3u, T3n); Chris@10: T3v = FMA(KP707106781, T3u, T3n); Chris@10: T3X = FMA(KP707106781, T3Q, T3N); Chris@10: T3R = FNMS(KP707106781, T3Q, T3N); Chris@10: T3J = FNMS(KP414213562, T3I, T3F); Chris@10: T3V = FMA(KP414213562, T3F, T3I); Chris@10: { Chris@10: E T4c, T4b, T4h, T4i, T41, T48; Chris@10: T4d = FMA(KP707106781, T40, T3Z); Chris@10: T41 = FNMS(KP707106781, T40, T3Z); Chris@10: T48 = T44 - T47; Chris@10: T4c = T44 + T47; Chris@10: { Chris@10: E T3Y, T3W, T3K, T3S; Chris@10: T3Y = T3U + T3V; Chris@10: T3W = T3U - T3V; Chris@10: T3K = T3C + T3J; Chris@10: T3S = T3J - T3C; Chris@10: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP923879532, T3W, T3T))); Chris@10: Ip[WS(rs, 5)] = KP500000000 * (FMA(KP923879532, T3W, T3T)); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP923879532, T3Y, T3X)); Chris@10: Rm[WS(rs, 6)] = KP500000000 * (FNMS(KP923879532, T3Y, T3X)); Chris@10: Rp[WS(rs, 5)] = KP500000000 * (FMA(KP923879532, T3S, T3R)); Chris@10: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP923879532, T3S, T3R)); Chris@10: Im[WS(rs, 6)] = -(KP500000000 * (FNMS(KP923879532, T3K, T3v))); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP923879532, T3K, T3v)); Chris@10: Ip[WS(rs, 7)] = KP500000000 * (FMA(KP923879532, T48, T41)); Chris@10: Im[0] = -(KP500000000 * (FNMS(KP923879532, T48, T41))); Chris@10: } Chris@10: T4b = FMA(KP707106781, T4a, T49); Chris@10: T4h = FNMS(KP707106781, T4a, T49); Chris@10: T4i = T4e + T4f; Chris@10: T4g = T4e - T4f; Chris@10: Rm[0] = KP500000000 * (FMA(KP923879532, T4i, T4h)); Chris@10: Rp[WS(rs, 7)] = KP500000000 * (FNMS(KP923879532, T4i, T4h)); Chris@10: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP923879532, T4c, T4b)); Chris@10: Rm[WS(rs, 4)] = KP500000000 * (FNMS(KP923879532, T4c, T4b)); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Im[WS(rs, 4)] = -(KP500000000 * (FNMS(KP923879532, T4g, T4d))); Chris@10: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP923879532, T4g, T4d)); 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 hc2c_desc desc = { 16, "hc2cfdft_16", twinstr, &GENUS, {136, 62, 70, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_16, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 16 -dit -name hc2cfdft_16 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 206 FP additions, 100 FP multiplications, Chris@10: * (or, 168 additions, 62 multiplications, 38 fused multiply/add), Chris@10: * 61 stack variables, 4 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_16(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP461939766, +0.461939766255643378064091594698394143411208313); Chris@10: DK(KP191341716, +0.191341716182544885864229992015199433380672281); Chris@10: DK(KP353553390, +0.353553390593273762200422181052424519642417969); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 30); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 30, MAKE_VOLATILE_STRIDE(64, rs)) { Chris@10: E T19, T3h, T21, T2Y, T1o, T3d, T2s, T39, TW, T3i, T24, T2Z, T1z, T3c, T2p; Chris@10: E T3a, Tj, T2S, T28, T2R, T1L, T36, T2i, T32, TC, T2V, T2b, T2U, T1W, T35; Chris@10: E T2l, T33; Chris@10: { Chris@10: E T10, T1m, T14, T1k, T18, T1h, T1f, T1Z; Chris@10: { Chris@10: E TY, TZ, T12, T13; Chris@10: TY = Ip[WS(rs, 4)]; Chris@10: TZ = Im[WS(rs, 4)]; Chris@10: T10 = TY - TZ; Chris@10: T1m = TY + TZ; Chris@10: T12 = Rp[WS(rs, 4)]; Chris@10: T13 = Rm[WS(rs, 4)]; Chris@10: T14 = T12 + T13; Chris@10: T1k = T12 - T13; Chris@10: } Chris@10: { Chris@10: E T16, T17, T1d, T1e; Chris@10: T16 = Ip[0]; Chris@10: T17 = Im[0]; Chris@10: T18 = T16 - T17; Chris@10: T1h = T16 + T17; Chris@10: T1d = Rm[0]; Chris@10: T1e = Rp[0]; Chris@10: T1f = T1d - T1e; Chris@10: T1Z = T1e + T1d; Chris@10: } Chris@10: { Chris@10: E T15, T20, TX, T11; Chris@10: TX = W[14]; Chris@10: T11 = W[15]; Chris@10: T15 = FNMS(T11, T14, TX * T10); Chris@10: T20 = FMA(TX, T14, T11 * T10); Chris@10: T19 = T15 + T18; Chris@10: T3h = T1Z - T20; Chris@10: T21 = T1Z + T20; Chris@10: T2Y = T18 - T15; Chris@10: } Chris@10: { Chris@10: E T1i, T2r, T1n, T2q; Chris@10: { Chris@10: E T1c, T1g, T1j, T1l; Chris@10: T1c = W[0]; Chris@10: T1g = W[1]; Chris@10: T1i = FNMS(T1g, T1h, T1c * T1f); Chris@10: T2r = FMA(T1g, T1f, T1c * T1h); Chris@10: T1j = W[16]; Chris@10: T1l = W[17]; Chris@10: T1n = FMA(T1j, T1k, T1l * T1m); Chris@10: T2q = FNMS(T1l, T1k, T1j * T1m); Chris@10: } Chris@10: T1o = T1i - T1n; Chris@10: T3d = T2r - T2q; Chris@10: T2s = T2q + T2r; Chris@10: T39 = T1n + T1i; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TH, T1s, TL, T1q, TQ, T1x, TU, T1v; Chris@10: { Chris@10: E TF, TG, TJ, TK; Chris@10: TF = Ip[WS(rs, 2)]; Chris@10: TG = Im[WS(rs, 2)]; Chris@10: TH = TF - TG; Chris@10: T1s = TF + TG; Chris@10: TJ = Rp[WS(rs, 2)]; Chris@10: TK = Rm[WS(rs, 2)]; Chris@10: TL = TJ + TK; Chris@10: T1q = TJ - TK; Chris@10: } Chris@10: { Chris@10: E TO, TP, TS, TT; Chris@10: TO = Ip[WS(rs, 6)]; Chris@10: TP = Im[WS(rs, 6)]; Chris@10: TQ = TO - TP; Chris@10: T1x = TO + TP; Chris@10: TS = Rp[WS(rs, 6)]; Chris@10: TT = Rm[WS(rs, 6)]; Chris@10: TU = TS + TT; Chris@10: T1v = TS - TT; Chris@10: } Chris@10: { Chris@10: E TM, T22, TV, T23; Chris@10: { Chris@10: E TE, TI, TN, TR; Chris@10: TE = W[6]; Chris@10: TI = W[7]; Chris@10: TM = FNMS(TI, TL, TE * TH); Chris@10: T22 = FMA(TE, TL, TI * TH); Chris@10: TN = W[22]; Chris@10: TR = W[23]; Chris@10: TV = FNMS(TR, TU, TN * TQ); Chris@10: T23 = FMA(TN, TU, TR * TQ); Chris@10: } Chris@10: TW = TM + TV; Chris@10: T3i = TM - TV; Chris@10: T24 = T22 + T23; Chris@10: T2Z = T22 - T23; Chris@10: } Chris@10: { Chris@10: E T1t, T2n, T1y, T2o; Chris@10: { Chris@10: E T1p, T1r, T1u, T1w; Chris@10: T1p = W[8]; Chris@10: T1r = W[9]; Chris@10: T1t = FMA(T1p, T1q, T1r * T1s); Chris@10: T2n = FNMS(T1r, T1q, T1p * T1s); Chris@10: T1u = W[24]; Chris@10: T1w = W[25]; Chris@10: T1y = FMA(T1u, T1v, T1w * T1x); Chris@10: T2o = FNMS(T1w, T1v, T1u * T1x); Chris@10: } Chris@10: T1z = T1t + T1y; Chris@10: T3c = T1y - T1t; Chris@10: T2p = T2n + T2o; Chris@10: T3a = T2n - T2o; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T4, T1E, T8, T1C, Td, T1J, Th, T1H; Chris@10: { Chris@10: E T2, T3, T6, T7; Chris@10: T2 = Ip[WS(rs, 1)]; Chris@10: T3 = Im[WS(rs, 1)]; Chris@10: T4 = T2 - T3; Chris@10: T1E = T2 + T3; Chris@10: T6 = Rp[WS(rs, 1)]; Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = T6 + T7; Chris@10: T1C = T6 - T7; Chris@10: } Chris@10: { Chris@10: E Tb, Tc, Tf, Tg; Chris@10: Tb = Ip[WS(rs, 5)]; Chris@10: Tc = Im[WS(rs, 5)]; Chris@10: Td = Tb - Tc; Chris@10: T1J = Tb + Tc; Chris@10: Tf = Rp[WS(rs, 5)]; Chris@10: Tg = Rm[WS(rs, 5)]; Chris@10: Th = Tf + Tg; Chris@10: T1H = Tf - Tg; Chris@10: } Chris@10: { Chris@10: E T9, T26, Ti, T27; Chris@10: { Chris@10: E T1, T5, Ta, Te; Chris@10: T1 = W[2]; Chris@10: T5 = W[3]; Chris@10: T9 = FNMS(T5, T8, T1 * T4); Chris@10: T26 = FMA(T1, T8, T5 * T4); Chris@10: Ta = W[18]; Chris@10: Te = W[19]; Chris@10: Ti = FNMS(Te, Th, Ta * Td); Chris@10: T27 = FMA(Ta, Th, Te * Td); Chris@10: } Chris@10: Tj = T9 + Ti; Chris@10: T2S = T26 - T27; Chris@10: T28 = T26 + T27; Chris@10: T2R = T9 - Ti; Chris@10: } Chris@10: { Chris@10: E T1F, T2g, T1K, T2h; Chris@10: { Chris@10: E T1B, T1D, T1G, T1I; Chris@10: T1B = W[4]; Chris@10: T1D = W[5]; Chris@10: T1F = FMA(T1B, T1C, T1D * T1E); Chris@10: T2g = FNMS(T1D, T1C, T1B * T1E); Chris@10: T1G = W[20]; Chris@10: T1I = W[21]; Chris@10: T1K = FMA(T1G, T1H, T1I * T1J); Chris@10: T2h = FNMS(T1I, T1H, T1G * T1J); Chris@10: } Chris@10: T1L = T1F + T1K; Chris@10: T36 = T2g - T2h; Chris@10: T2i = T2g + T2h; Chris@10: T32 = T1K - T1F; Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tn, T1P, Tr, T1N, Tw, T1U, TA, T1S; Chris@10: { Chris@10: E Tl, Tm, Tp, Tq; Chris@10: Tl = Ip[WS(rs, 7)]; Chris@10: Tm = Im[WS(rs, 7)]; Chris@10: Tn = Tl - Tm; Chris@10: T1P = Tl + Tm; Chris@10: Tp = Rp[WS(rs, 7)]; Chris@10: Tq = Rm[WS(rs, 7)]; Chris@10: Tr = Tp + Tq; Chris@10: T1N = Tp - Tq; Chris@10: } Chris@10: { Chris@10: E Tu, Tv, Ty, Tz; Chris@10: Tu = Ip[WS(rs, 3)]; Chris@10: Tv = Im[WS(rs, 3)]; Chris@10: Tw = Tu - Tv; Chris@10: T1U = Tu + Tv; Chris@10: Ty = Rp[WS(rs, 3)]; Chris@10: Tz = Rm[WS(rs, 3)]; Chris@10: TA = Ty + Tz; Chris@10: T1S = Ty - Tz; Chris@10: } Chris@10: { Chris@10: E Ts, T29, TB, T2a; Chris@10: { Chris@10: E Tk, To, Tt, Tx; Chris@10: Tk = W[26]; Chris@10: To = W[27]; Chris@10: Ts = FNMS(To, Tr, Tk * Tn); Chris@10: T29 = FMA(Tk, Tr, To * Tn); Chris@10: Tt = W[10]; Chris@10: Tx = W[11]; Chris@10: TB = FNMS(Tx, TA, Tt * Tw); Chris@10: T2a = FMA(Tt, TA, Tx * Tw); Chris@10: } Chris@10: TC = Ts + TB; Chris@10: T2V = Ts - TB; Chris@10: T2b = T29 + T2a; Chris@10: T2U = T29 - T2a; Chris@10: } Chris@10: { Chris@10: E T1Q, T2j, T1V, T2k; Chris@10: { Chris@10: E T1M, T1O, T1R, T1T; Chris@10: T1M = W[28]; Chris@10: T1O = W[29]; Chris@10: T1Q = FMA(T1M, T1N, T1O * T1P); Chris@10: T2j = FNMS(T1O, T1N, T1M * T1P); Chris@10: T1R = W[12]; Chris@10: T1T = W[13]; Chris@10: T1V = FMA(T1R, T1S, T1T * T1U); Chris@10: T2k = FNMS(T1T, T1S, T1R * T1U); Chris@10: } Chris@10: T1W = T1Q + T1V; Chris@10: T35 = T1V - T1Q; Chris@10: T2l = T2j + T2k; Chris@10: T33 = T2j - T2k; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1b, T2f, T2u, T2w, T1Y, T2e, T2d, T2v; Chris@10: { Chris@10: E TD, T1a, T2m, T2t; Chris@10: TD = Tj + TC; Chris@10: T1a = TW + T19; Chris@10: T1b = TD + T1a; Chris@10: T2f = T1a - TD; Chris@10: T2m = T2i + T2l; Chris@10: T2t = T2p + T2s; Chris@10: T2u = T2m - T2t; Chris@10: T2w = T2m + T2t; Chris@10: } Chris@10: { Chris@10: E T1A, T1X, T25, T2c; Chris@10: T1A = T1o - T1z; Chris@10: T1X = T1L + T1W; Chris@10: T1Y = T1A - T1X; Chris@10: T2e = T1X + T1A; Chris@10: T25 = T21 + T24; Chris@10: T2c = T28 + T2b; Chris@10: T2d = T25 - T2c; Chris@10: T2v = T25 + T2c; Chris@10: } Chris@10: Ip[0] = KP500000000 * (T1b + T1Y); Chris@10: Rp[0] = KP500000000 * (T2v + T2w); Chris@10: Im[WS(rs, 7)] = KP500000000 * (T1Y - T1b); Chris@10: Rm[WS(rs, 7)] = KP500000000 * (T2v - T2w); Chris@10: Rm[WS(rs, 3)] = KP500000000 * (T2d - T2e); Chris@10: Im[WS(rs, 3)] = KP500000000 * (T2u - T2f); Chris@10: Rp[WS(rs, 4)] = KP500000000 * (T2d + T2e); Chris@10: Ip[WS(rs, 4)] = KP500000000 * (T2f + T2u); Chris@10: } Chris@10: { Chris@10: E T2z, T2L, T2J, T2P, T2C, T2M, T2F, T2N; Chris@10: { Chris@10: E T2x, T2y, T2H, T2I; Chris@10: T2x = T2b - T28; Chris@10: T2y = T19 - TW; Chris@10: T2z = KP500000000 * (T2x + T2y); Chris@10: T2L = KP500000000 * (T2y - T2x); Chris@10: T2H = T21 - T24; Chris@10: T2I = Tj - TC; Chris@10: T2J = KP500000000 * (T2H - T2I); Chris@10: T2P = KP500000000 * (T2H + T2I); Chris@10: } Chris@10: { Chris@10: E T2A, T2B, T2D, T2E; Chris@10: T2A = T2l - T2i; Chris@10: T2B = T1L - T1W; Chris@10: T2C = T2A + T2B; Chris@10: T2M = T2A - T2B; Chris@10: T2D = T1z + T1o; Chris@10: T2E = T2s - T2p; Chris@10: T2F = T2D - T2E; Chris@10: T2N = T2D + T2E; Chris@10: } Chris@10: { Chris@10: E T2G, T2Q, T2K, T2O; Chris@10: T2G = KP353553390 * (T2C + T2F); Chris@10: Ip[WS(rs, 2)] = T2z + T2G; Chris@10: Im[WS(rs, 5)] = T2G - T2z; Chris@10: T2Q = KP353553390 * (T2M + T2N); Chris@10: Rm[WS(rs, 5)] = T2P - T2Q; Chris@10: Rp[WS(rs, 2)] = T2P + T2Q; Chris@10: T2K = KP353553390 * (T2F - T2C); Chris@10: Rm[WS(rs, 1)] = T2J - T2K; Chris@10: Rp[WS(rs, 6)] = T2J + T2K; Chris@10: T2O = KP353553390 * (T2M - T2N); Chris@10: Ip[WS(rs, 6)] = T2L + T2O; Chris@10: Im[WS(rs, 1)] = T2O - T2L; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T30, T3w, T3F, T3j, T2X, T3G, T3D, T3L, T3m, T3v, T38, T3q, T3A, T3K, T3f; Chris@10: E T3r; Chris@10: { Chris@10: E T2T, T2W, T34, T37; Chris@10: T30 = KP500000000 * (T2Y - T2Z); Chris@10: T3w = KP500000000 * (T2Z + T2Y); Chris@10: T3F = KP500000000 * (T3h - T3i); Chris@10: T3j = KP500000000 * (T3h + T3i); Chris@10: T2T = T2R - T2S; Chris@10: T2W = T2U + T2V; Chris@10: T2X = KP353553390 * (T2T + T2W); Chris@10: T3G = KP353553390 * (T2T - T2W); Chris@10: { Chris@10: E T3B, T3C, T3k, T3l; Chris@10: T3B = T3a + T39; Chris@10: T3C = T3d - T3c; Chris@10: T3D = FNMS(KP461939766, T3C, KP191341716 * T3B); Chris@10: T3L = FMA(KP461939766, T3B, KP191341716 * T3C); Chris@10: T3k = T2S + T2R; Chris@10: T3l = T2U - T2V; Chris@10: T3m = KP353553390 * (T3k + T3l); Chris@10: T3v = KP353553390 * (T3l - T3k); Chris@10: } Chris@10: T34 = T32 + T33; Chris@10: T37 = T35 - T36; Chris@10: T38 = FMA(KP191341716, T34, KP461939766 * T37); Chris@10: T3q = FNMS(KP191341716, T37, KP461939766 * T34); Chris@10: { Chris@10: E T3y, T3z, T3b, T3e; Chris@10: T3y = T33 - T32; Chris@10: T3z = T36 + T35; Chris@10: T3A = FMA(KP461939766, T3y, KP191341716 * T3z); Chris@10: T3K = FNMS(KP461939766, T3z, KP191341716 * T3y); Chris@10: T3b = T39 - T3a; Chris@10: T3e = T3c + T3d; Chris@10: T3f = FNMS(KP191341716, T3e, KP461939766 * T3b); Chris@10: T3r = FMA(KP191341716, T3b, KP461939766 * T3e); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T31, T3g, T3t, T3u; Chris@10: T31 = T2X + T30; Chris@10: T3g = T38 + T3f; Chris@10: Ip[WS(rs, 1)] = T31 + T3g; Chris@10: Im[WS(rs, 6)] = T3g - T31; Chris@10: T3t = T3j + T3m; Chris@10: T3u = T3q + T3r; Chris@10: Rm[WS(rs, 6)] = T3t - T3u; Chris@10: Rp[WS(rs, 1)] = T3t + T3u; Chris@10: } Chris@10: { Chris@10: E T3n, T3o, T3p, T3s; Chris@10: T3n = T3j - T3m; Chris@10: T3o = T3f - T38; Chris@10: Rm[WS(rs, 2)] = T3n - T3o; Chris@10: Rp[WS(rs, 5)] = T3n + T3o; Chris@10: T3p = T30 - T2X; Chris@10: T3s = T3q - T3r; Chris@10: Ip[WS(rs, 5)] = T3p + T3s; Chris@10: Im[WS(rs, 2)] = T3s - T3p; Chris@10: } Chris@10: { Chris@10: E T3x, T3E, T3N, T3O; Chris@10: T3x = T3v + T3w; Chris@10: T3E = T3A + T3D; Chris@10: Ip[WS(rs, 3)] = T3x + T3E; Chris@10: Im[WS(rs, 4)] = T3E - T3x; Chris@10: T3N = T3F + T3G; Chris@10: T3O = T3K + T3L; Chris@10: Rm[WS(rs, 4)] = T3N - T3O; Chris@10: Rp[WS(rs, 3)] = T3N + T3O; Chris@10: } Chris@10: { Chris@10: E T3H, T3I, T3J, T3M; Chris@10: T3H = T3F - T3G; Chris@10: T3I = T3D - T3A; Chris@10: Rm[0] = T3H - T3I; Chris@10: Rp[WS(rs, 7)] = T3H + T3I; Chris@10: T3J = T3w - T3v; Chris@10: T3M = T3K - T3L; Chris@10: Ip[WS(rs, 7)] = T3J + T3M; Chris@10: Im[0] = T3M - T3J; Chris@10: } 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 hc2c_desc desc = { 16, "hc2cfdft_16", twinstr, &GENUS, {168, 62, 38, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_16, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */