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 07:42:05 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 -sign 1 -n 16 -dif -name hc2cbdft_16 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 206 FP additions, 100 FP multiplications, Chris@10: * (or, 136 additions, 30 multiplications, 70 fused multiply/add), Chris@10: * 97 stack variables, 3 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_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: { 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 T3w, T3z, T2Y, T3D, T3x, T3m, T3u, T3C, T3y, T3o, T3k, T3E, T3A; Chris@10: { Chris@10: E T20, Tf, T3Q, T32, T3V, T3f, T2a, TN, T2f, T1m, T3G, T2G, T3L, T2T, T26; Chris@10: E T1F, T3M, T2N, T3H, T2W, T25, Tu, T1n, T1o, T3R, T3i, T2g, T1a, T21, T1y; Chris@10: E T3W, T39; Chris@10: { Chris@10: E T2R, T1B, T2S, T1E; Chris@10: { Chris@10: E T1e, T3, T1C, TA, Tx, T6, T1D, T1h, Td, T1A, TL, T1k, Ta, TC, TF; Chris@10: E T1z; Chris@10: { Chris@10: E T4, T5, T1f, T1g; Chris@10: { Chris@10: E T1, T2, Ty, Tz; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 7)]; Chris@10: Ty = Ip[0]; Chris@10: Tz = Im[WS(rs, 7)]; Chris@10: T4 = Rp[WS(rs, 4)]; Chris@10: T1e = T1 - T2; Chris@10: T3 = T1 + T2; Chris@10: T1C = Ty - Tz; Chris@10: TA = Ty + Tz; Chris@10: T5 = Rm[WS(rs, 3)]; Chris@10: } Chris@10: T1f = Ip[WS(rs, 4)]; Chris@10: T1g = Im[WS(rs, 3)]; Chris@10: { Chris@10: E Tb, Tc, TI, TJ; Chris@10: Tb = Rm[WS(rs, 1)]; Chris@10: Tx = T4 - T5; Chris@10: T6 = T4 + T5; Chris@10: T1D = T1f - T1g; Chris@10: T1h = T1f + T1g; Chris@10: Tc = Rp[WS(rs, 6)]; Chris@10: TI = Im[WS(rs, 1)]; Chris@10: TJ = Ip[WS(rs, 6)]; Chris@10: { Chris@10: E T8, TH, TK, T9, TD, TE; Chris@10: T8 = Rp[WS(rs, 2)]; Chris@10: Td = Tb + Tc; Chris@10: TH = Tb - Tc; Chris@10: T1A = TJ - TI; Chris@10: TK = TI + TJ; Chris@10: T9 = Rm[WS(rs, 5)]; Chris@10: TD = Ip[WS(rs, 2)]; Chris@10: TE = Im[WS(rs, 5)]; Chris@10: TL = TH + TK; Chris@10: T1k = TH - TK; Chris@10: Ta = T8 + T9; Chris@10: TC = T8 - T9; Chris@10: TF = TD + TE; Chris@10: T1z = TD - TE; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2E, TB, T1l, T1i, T3d, T3e, TM, T2F; Chris@10: { Chris@10: E T7, TG, Te, T30, T31, T1j; Chris@10: T2E = T3 - T6; Chris@10: T7 = T3 + T6; Chris@10: T1j = TC - TF; Chris@10: TG = TC + TF; Chris@10: Te = Ta + Td; Chris@10: T2R = Ta - Td; Chris@10: TB = Tx + TA; Chris@10: T30 = TA - Tx; Chris@10: T31 = T1j - T1k; Chris@10: T1l = T1j + T1k; Chris@10: T1i = T1e - T1h; Chris@10: T3d = T1e + T1h; Chris@10: T20 = T7 - Te; Chris@10: Tf = T7 + Te; Chris@10: T3Q = FNMS(KP707106781, T31, T30); Chris@10: T32 = FMA(KP707106781, T31, T30); Chris@10: T3e = TG + TL; Chris@10: TM = TG - TL; Chris@10: } Chris@10: T3V = FMA(KP707106781, T3e, T3d); Chris@10: T3f = FNMS(KP707106781, T3e, T3d); Chris@10: T2a = FNMS(KP707106781, TM, TB); Chris@10: TN = FMA(KP707106781, TM, TB); Chris@10: T2F = T1A - T1z; Chris@10: T1B = T1z + T1A; Chris@10: T2f = FNMS(KP707106781, T1l, T1i); Chris@10: T1m = FMA(KP707106781, T1l, T1i); Chris@10: T3G = T2E - T2F; Chris@10: T2G = T2E + T2F; Chris@10: T2S = T1C - T1D; Chris@10: T1E = T1C + T1D; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T34, TS, T2H, Tm, T1u, T2I, T33, TX, Tq, T14, Tp, T1v, T12, Tr, T15; Chris@10: E T16; Chris@10: { Chris@10: E Tj, TT, Ti, T1s, TR, Tk, TU, TV; Chris@10: { Chris@10: E Tg, Th, TP, TQ; Chris@10: Tg = Rp[WS(rs, 1)]; Chris@10: T3L = T2S - T2R; Chris@10: T2T = T2R + T2S; Chris@10: T26 = T1E - T1B; Chris@10: T1F = T1B + T1E; Chris@10: Th = Rm[WS(rs, 6)]; Chris@10: TP = Ip[WS(rs, 1)]; Chris@10: TQ = Im[WS(rs, 6)]; Chris@10: Tj = Rp[WS(rs, 5)]; Chris@10: TT = Tg - Th; Chris@10: Ti = Tg + Th; Chris@10: T1s = TP - TQ; Chris@10: TR = TP + TQ; Chris@10: Tk = Rm[WS(rs, 2)]; Chris@10: TU = Ip[WS(rs, 5)]; Chris@10: TV = Im[WS(rs, 2)]; Chris@10: } Chris@10: { Chris@10: E Tn, To, T10, T11; Chris@10: Tn = Rm[0]; Chris@10: { Chris@10: E TO, Tl, T1t, TW; Chris@10: TO = Tj - Tk; Chris@10: Tl = Tj + Tk; Chris@10: T1t = TU - TV; Chris@10: TW = TU + TV; Chris@10: T34 = TR - TO; Chris@10: TS = TO + TR; Chris@10: T2H = Ti - Tl; Chris@10: Tm = Ti + Tl; Chris@10: T1u = T1s + T1t; Chris@10: T2I = T1s - T1t; Chris@10: T33 = TT + TW; Chris@10: TX = TT - TW; Chris@10: To = Rp[WS(rs, 7)]; Chris@10: } Chris@10: T10 = Im[0]; Chris@10: T11 = Ip[WS(rs, 7)]; Chris@10: Tq = Rp[WS(rs, 3)]; Chris@10: T14 = Tn - To; Chris@10: Tp = Tn + To; Chris@10: T1v = T11 - T10; Chris@10: T12 = T10 + T11; Chris@10: Tr = Rm[WS(rs, 4)]; Chris@10: T15 = Ip[WS(rs, 3)]; Chris@10: T16 = Im[WS(rs, 4)]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T13, T1x, T18, T35, T3g, T3h, T38, TY, T19; Chris@10: { Chris@10: E T2U, T2J, T37, Tt, T36, T2V, T2M, T2K, T2L; Chris@10: T2U = T2H + T2I; Chris@10: T2J = T2H - T2I; Chris@10: { Chris@10: E TZ, Ts, T1w, T17; Chris@10: TZ = Tq - Tr; Chris@10: Ts = Tq + Tr; Chris@10: T1w = T15 - T16; Chris@10: T17 = T15 + T16; Chris@10: T37 = TZ + T12; Chris@10: T13 = TZ - T12; Chris@10: T2K = Tp - Ts; Chris@10: Tt = Tp + Ts; Chris@10: T1x = T1v + T1w; Chris@10: T2L = T1v - T1w; Chris@10: T36 = T14 + T17; Chris@10: T18 = T14 - T17; Chris@10: } Chris@10: T2V = T2L - T2K; Chris@10: T2M = T2K + T2L; Chris@10: T3M = T2J - T2M; Chris@10: T2N = T2J + T2M; Chris@10: T3H = T2V - T2U; Chris@10: T2W = T2U + T2V; Chris@10: T35 = FMA(KP414213562, T34, T33); Chris@10: T3g = FNMS(KP414213562, T33, T34); Chris@10: T25 = Tm - Tt; Chris@10: Tu = Tm + Tt; Chris@10: T3h = FNMS(KP414213562, T36, T37); Chris@10: T38 = FMA(KP414213562, T37, T36); Chris@10: } Chris@10: T1n = FNMS(KP414213562, TS, TX); Chris@10: TY = FMA(KP414213562, TX, TS); Chris@10: T19 = FNMS(KP414213562, T18, T13); Chris@10: T1o = FMA(KP414213562, T13, T18); Chris@10: T3R = T3h - T3g; Chris@10: T3i = T3g + T3h; Chris@10: T2g = T19 - TY; Chris@10: T1a = TY + T19; Chris@10: T21 = T1x - T1u; Chris@10: T1y = T1u + T1x; Chris@10: T3W = T35 + T38; Chris@10: T39 = T35 - T38; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T27, T22, T2c, T2u, T2x, T2h, T2s, T2A, T2w, T2B, T2v; Chris@10: { Chris@10: E T1K, Tv, T1G, T1N, T1Q, T1b, T2b, T1p, Tw, T1d; Chris@10: T1K = Tf - Tu; Chris@10: Tv = Tf + Tu; Chris@10: T1G = T1y + T1F; Chris@10: T1N = T1F - T1y; Chris@10: T1Q = FNMS(KP923879532, T1a, TN); Chris@10: T1b = FMA(KP923879532, T1a, TN); Chris@10: T2b = T1n - T1o; Chris@10: T1p = T1n + T1o; Chris@10: Tw = W[0]; Chris@10: T1d = W[1]; Chris@10: { Chris@10: E T1T, T1O, T1W, T1S, T1X, T1R; Chris@10: { Chris@10: E T1J, T1M, T1L, T1V, T1P, T1q; Chris@10: T1T = FNMS(KP923879532, T1p, T1m); Chris@10: T1q = FMA(KP923879532, T1p, T1m); Chris@10: { Chris@10: E T1c, T1I, T1H, T1r; Chris@10: T1c = Tw * T1b; Chris@10: T1J = W[14]; Chris@10: T1H = Tw * T1q; Chris@10: T1r = FMA(T1d, T1q, T1c); Chris@10: T1M = W[15]; Chris@10: T1L = T1J * T1K; Chris@10: T1I = FNMS(T1d, T1b, T1H); Chris@10: Rm[0] = Tv + T1r; Chris@10: Rp[0] = Tv - T1r; Chris@10: T1V = T1M * T1K; Chris@10: Im[0] = T1I - T1G; Chris@10: Ip[0] = T1G + T1I; Chris@10: T1P = W[16]; Chris@10: } Chris@10: T1O = FNMS(T1M, T1N, T1L); Chris@10: T1W = FMA(T1J, T1N, T1V); Chris@10: T1S = W[17]; Chris@10: T1X = T1P * T1T; Chris@10: T1R = T1P * T1Q; Chris@10: } Chris@10: { Chris@10: E T2r, T2n, T2q, T2p, T2z, T2t, T2o, T1Y, T1U; Chris@10: T27 = T25 + T26; Chris@10: T2r = T26 - T25; Chris@10: T2o = T20 - T21; Chris@10: T22 = T20 + T21; Chris@10: T1Y = FNMS(T1S, T1Q, T1X); Chris@10: T1U = FMA(T1S, T1T, T1R); Chris@10: T2n = W[22]; Chris@10: T2q = W[23]; Chris@10: Im[WS(rs, 4)] = T1Y - T1W; Chris@10: Ip[WS(rs, 4)] = T1W + T1Y; Chris@10: Rm[WS(rs, 4)] = T1O + T1U; Chris@10: Rp[WS(rs, 4)] = T1O - T1U; Chris@10: T2p = T2n * T2o; Chris@10: T2z = T2q * T2o; Chris@10: T2c = FMA(KP923879532, T2b, T2a); Chris@10: T2u = FNMS(KP923879532, T2b, T2a); Chris@10: T2x = FNMS(KP923879532, T2g, T2f); Chris@10: T2h = FMA(KP923879532, T2g, T2f); Chris@10: T2t = W[24]; Chris@10: T2s = FNMS(T2q, T2r, T2p); Chris@10: T2A = FMA(T2n, T2r, T2z); Chris@10: T2w = W[25]; Chris@10: T2B = T2t * T2x; Chris@10: T2v = T2t * T2u; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T28, T2k, T2e, T2l, T2d; Chris@10: { Chris@10: E T1Z, T24, T23, T2j, T29, T2C, T2y; Chris@10: T2C = FNMS(T2w, T2u, T2B); Chris@10: T2y = FMA(T2w, T2x, T2v); Chris@10: T1Z = W[6]; Chris@10: T24 = W[7]; Chris@10: Im[WS(rs, 6)] = T2C - T2A; Chris@10: Ip[WS(rs, 6)] = T2A + T2C; Chris@10: Rm[WS(rs, 6)] = T2s + T2y; Chris@10: Rp[WS(rs, 6)] = T2s - T2y; Chris@10: T23 = T1Z * T22; Chris@10: T2j = T24 * T22; Chris@10: T29 = W[8]; Chris@10: T28 = FNMS(T24, T27, T23); Chris@10: T2k = FMA(T1Z, T27, T2j); Chris@10: T2e = W[9]; Chris@10: T2l = T29 * T2h; Chris@10: T2d = T29 * T2c; Chris@10: } Chris@10: { Chris@10: E T4a, T4d, T3O, T4h, T4b, T40, T48, T4g, T4c, T42, T3Y; Chris@10: { Chris@10: E T3N, T47, T43, T46, T3F, T45, T4f, T3K, T3J, T3S, T3X, T3Z, T49, T41, T3T; Chris@10: E T3U; Chris@10: { Chris@10: E T44, T3I, T2m, T2i, T3P; Chris@10: T44 = FNMS(KP707106781, T3H, T3G); Chris@10: T3I = FMA(KP707106781, T3H, T3G); Chris@10: T2m = FNMS(T2e, T2c, T2l); Chris@10: T2i = FMA(T2e, T2h, T2d); Chris@10: T3N = FMA(KP707106781, T3M, T3L); Chris@10: T47 = FNMS(KP707106781, T3M, T3L); Chris@10: Im[WS(rs, 2)] = T2m - T2k; Chris@10: Ip[WS(rs, 2)] = T2k + T2m; Chris@10: Rm[WS(rs, 2)] = T28 + T2i; Chris@10: Rp[WS(rs, 2)] = T28 - T2i; Chris@10: T43 = W[26]; Chris@10: T46 = W[27]; Chris@10: T3F = W[10]; Chris@10: T45 = T43 * T44; Chris@10: T4f = T46 * T44; Chris@10: T3K = W[11]; Chris@10: T3J = T3F * T3I; Chris@10: T4a = FNMS(KP923879532, T3R, T3Q); Chris@10: T3S = FMA(KP923879532, T3R, T3Q); Chris@10: T3X = FNMS(KP923879532, T3W, T3V); Chris@10: T4d = FMA(KP923879532, T3W, T3V); Chris@10: T3Z = T3K * T3I; Chris@10: T3P = W[12]; Chris@10: T49 = W[28]; Chris@10: T41 = T3P * T3X; Chris@10: T3T = T3P * T3S; Chris@10: } Chris@10: T3O = FNMS(T3K, T3N, T3J); Chris@10: T4h = T49 * T4d; Chris@10: T4b = T49 * T4a; Chris@10: T40 = FMA(T3F, T3N, T3Z); Chris@10: T3U = W[13]; Chris@10: T48 = FNMS(T46, T47, T45); Chris@10: T4g = FMA(T43, T47, T4f); Chris@10: T4c = W[29]; Chris@10: T42 = FNMS(T3U, T3S, T41); Chris@10: T3Y = FMA(T3U, T3X, T3T); Chris@10: } Chris@10: { Chris@10: E T3t, T2X, T3p, T3s, T2D, T3r, T3B, T2Q, T2P, T3a, T3j, T3l, T3v, T3n, T3b; Chris@10: E T3c; Chris@10: { Chris@10: E T2O, T3q, T4i, T4e, T2Z; Chris@10: T4i = FNMS(T4c, T4a, T4h); Chris@10: T4e = FMA(T4c, T4d, T4b); Chris@10: Im[WS(rs, 3)] = T42 - T40; Chris@10: Ip[WS(rs, 3)] = T40 + T42; Chris@10: Rm[WS(rs, 3)] = T3O + T3Y; Chris@10: Rp[WS(rs, 3)] = T3O - T3Y; Chris@10: Im[WS(rs, 7)] = T4i - T4g; Chris@10: Ip[WS(rs, 7)] = T4g + T4i; Chris@10: Rm[WS(rs, 7)] = T48 + T4e; Chris@10: Rp[WS(rs, 7)] = T48 - T4e; Chris@10: T3t = FNMS(KP707106781, T2W, T2T); Chris@10: T2X = FMA(KP707106781, T2W, T2T); Chris@10: T2O = FMA(KP707106781, T2N, T2G); Chris@10: T3q = FNMS(KP707106781, T2N, T2G); Chris@10: T3p = W[18]; Chris@10: T3s = W[19]; Chris@10: T2D = W[2]; Chris@10: T3r = T3p * T3q; Chris@10: T3B = T3s * T3q; Chris@10: T2Q = W[3]; Chris@10: T2P = T2D * T2O; Chris@10: T3a = FMA(KP923879532, T39, T32); Chris@10: T3w = FNMS(KP923879532, T39, T32); Chris@10: T3z = FMA(KP923879532, T3i, T3f); Chris@10: T3j = FNMS(KP923879532, T3i, T3f); Chris@10: T3l = T2Q * T2O; Chris@10: T2Z = W[4]; Chris@10: T3v = W[20]; Chris@10: T3n = T2Z * T3j; Chris@10: T3b = T2Z * T3a; Chris@10: } Chris@10: T2Y = FNMS(T2Q, T2X, T2P); Chris@10: T3D = T3v * T3z; Chris@10: T3x = T3v * T3w; Chris@10: T3m = FMA(T2D, T2X, T3l); Chris@10: T3c = W[5]; Chris@10: T3u = FNMS(T3s, T3t, T3r); Chris@10: T3C = FMA(T3p, T3t, T3B); Chris@10: T3y = W[21]; Chris@10: T3o = FNMS(T3c, T3a, T3n); Chris@10: T3k = FMA(T3c, T3j, T3b); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: T3E = FNMS(T3y, T3w, T3D); Chris@10: T3A = FMA(T3y, T3z, T3x); Chris@10: Im[WS(rs, 1)] = T3o - T3m; Chris@10: Ip[WS(rs, 1)] = T3m + T3o; Chris@10: Rm[WS(rs, 1)] = T2Y + T3k; Chris@10: Rp[WS(rs, 1)] = T2Y - T3k; Chris@10: Im[WS(rs, 5)] = T3E - T3C; Chris@10: Ip[WS(rs, 5)] = T3C + T3E; Chris@10: Rm[WS(rs, 5)] = T3u + T3A; Chris@10: Rp[WS(rs, 5)] = T3u - T3A; 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, "hc2cbdft_16", twinstr, &GENUS, {136, 30, 70, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_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 -sign 1 -n 16 -dif -name hc2cbdft_16 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 206 FP additions, 84 FP multiplications, Chris@10: * (or, 168 additions, 46 multiplications, 38 fused multiply/add), Chris@10: * 60 stack variables, 3 constants, and 64 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_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(KP382683432, +0.382683432365089771728459984030398866761344562); 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 30, MAKE_VOLATILE_STRIDE(64, rs)) { Chris@10: E TB, T2L, T30, T1n, Tf, T1U, T2H, T3p, T1E, T1Z, TM, T31, T2s, T3k, T1i; Chris@10: E T2M, Tu, T1Y, T2Q, T2X, T2T, T2Y, TY, T1d, T19, T1e, T2v, T2C, T2y, T2D; Chris@10: E T1x, T1V; Chris@10: { Chris@10: E T3, T1j, TA, T1B, T6, Tx, T1m, T1C, Ta, TC, TF, T1y, Td, TH, TK; Chris@10: E T1z; Chris@10: { Chris@10: E T1, T2, Ty, Tz; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 7)]; Chris@10: T3 = T1 + T2; Chris@10: T1j = T1 - T2; Chris@10: Ty = Ip[0]; Chris@10: Tz = Im[WS(rs, 7)]; Chris@10: TA = Ty + Tz; Chris@10: T1B = Ty - Tz; Chris@10: } Chris@10: { Chris@10: E T4, T5, T1k, T1l; Chris@10: T4 = Rp[WS(rs, 4)]; Chris@10: T5 = Rm[WS(rs, 3)]; Chris@10: T6 = T4 + T5; Chris@10: Tx = T4 - T5; Chris@10: T1k = Ip[WS(rs, 4)]; Chris@10: T1l = Im[WS(rs, 3)]; Chris@10: T1m = T1k + T1l; Chris@10: T1C = T1k - T1l; Chris@10: } Chris@10: { Chris@10: E T8, T9, TD, TE; Chris@10: T8 = Rp[WS(rs, 2)]; Chris@10: T9 = Rm[WS(rs, 5)]; Chris@10: Ta = T8 + T9; Chris@10: TC = T8 - T9; Chris@10: TD = Ip[WS(rs, 2)]; Chris@10: TE = Im[WS(rs, 5)]; Chris@10: TF = TD + TE; Chris@10: T1y = TD - TE; Chris@10: } Chris@10: { Chris@10: E Tb, Tc, TI, TJ; Chris@10: Tb = Rm[WS(rs, 1)]; Chris@10: Tc = Rp[WS(rs, 6)]; Chris@10: Td = Tb + Tc; Chris@10: TH = Tb - Tc; Chris@10: TI = Im[WS(rs, 1)]; Chris@10: TJ = Ip[WS(rs, 6)]; Chris@10: TK = TI + TJ; Chris@10: T1z = TJ - TI; Chris@10: } Chris@10: { Chris@10: E T7, Te, TG, TL; Chris@10: TB = Tx + TA; Chris@10: T2L = TA - Tx; Chris@10: T30 = T1j + T1m; Chris@10: T1n = T1j - T1m; Chris@10: T7 = T3 + T6; Chris@10: Te = Ta + Td; Chris@10: Tf = T7 + Te; Chris@10: T1U = T7 - Te; Chris@10: { Chris@10: E T2F, T2G, T1A, T1D; Chris@10: T2F = Ta - Td; Chris@10: T2G = T1B - T1C; Chris@10: T2H = T2F + T2G; Chris@10: T3p = T2G - T2F; Chris@10: T1A = T1y + T1z; Chris@10: T1D = T1B + T1C; Chris@10: T1E = T1A + T1D; Chris@10: T1Z = T1D - T1A; Chris@10: } Chris@10: TG = TC + TF; Chris@10: TL = TH + TK; Chris@10: TM = KP707106781 * (TG - TL); Chris@10: T31 = KP707106781 * (TG + TL); Chris@10: { Chris@10: E T2q, T2r, T1g, T1h; Chris@10: T2q = T3 - T6; Chris@10: T2r = T1z - T1y; Chris@10: T2s = T2q + T2r; Chris@10: T3k = T2q - T2r; Chris@10: T1g = TC - TF; Chris@10: T1h = TH - TK; Chris@10: T1i = KP707106781 * (T1g + T1h); Chris@10: T2M = KP707106781 * (T1g - T1h); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Ti, TT, TR, T1r, Tl, TO, TW, T1s, Tp, T14, T12, T1u, Ts, TZ, T17; Chris@10: E T1v; Chris@10: { Chris@10: E Tg, Th, TP, TQ; Chris@10: Tg = Rp[WS(rs, 1)]; Chris@10: Th = Rm[WS(rs, 6)]; Chris@10: Ti = Tg + Th; Chris@10: TT = Tg - Th; Chris@10: TP = Ip[WS(rs, 1)]; Chris@10: TQ = Im[WS(rs, 6)]; Chris@10: TR = TP + TQ; Chris@10: T1r = TP - TQ; Chris@10: } Chris@10: { Chris@10: E Tj, Tk, TU, TV; Chris@10: Tj = Rp[WS(rs, 5)]; Chris@10: Tk = Rm[WS(rs, 2)]; Chris@10: Tl = Tj + Tk; Chris@10: TO = Tj - Tk; Chris@10: TU = Ip[WS(rs, 5)]; Chris@10: TV = Im[WS(rs, 2)]; Chris@10: TW = TU + TV; Chris@10: T1s = TU - TV; Chris@10: } Chris@10: { Chris@10: E Tn, To, T10, T11; Chris@10: Tn = Rm[0]; Chris@10: To = Rp[WS(rs, 7)]; Chris@10: Tp = Tn + To; Chris@10: T14 = Tn - To; Chris@10: T10 = Im[0]; Chris@10: T11 = Ip[WS(rs, 7)]; Chris@10: T12 = T10 + T11; Chris@10: T1u = T11 - T10; Chris@10: } Chris@10: { Chris@10: E Tq, Tr, T15, T16; Chris@10: Tq = Rp[WS(rs, 3)]; Chris@10: Tr = Rm[WS(rs, 4)]; Chris@10: Ts = Tq + Tr; Chris@10: TZ = Tq - Tr; Chris@10: T15 = Ip[WS(rs, 3)]; Chris@10: T16 = Im[WS(rs, 4)]; Chris@10: T17 = T15 + T16; Chris@10: T1v = T15 - T16; Chris@10: } Chris@10: { Chris@10: E Tm, Tt, T2O, T2P; Chris@10: Tm = Ti + Tl; Chris@10: Tt = Tp + Ts; Chris@10: Tu = Tm + Tt; Chris@10: T1Y = Tm - Tt; Chris@10: T2O = TR - TO; Chris@10: T2P = TT + TW; Chris@10: T2Q = FMA(KP382683432, T2O, KP923879532 * T2P); Chris@10: T2X = FNMS(KP923879532, T2O, KP382683432 * T2P); Chris@10: } Chris@10: { Chris@10: E T2R, T2S, TS, TX; Chris@10: T2R = TZ + T12; Chris@10: T2S = T14 + T17; Chris@10: T2T = FMA(KP382683432, T2R, KP923879532 * T2S); Chris@10: T2Y = FNMS(KP923879532, T2R, KP382683432 * T2S); Chris@10: TS = TO + TR; Chris@10: TX = TT - TW; Chris@10: TY = FMA(KP923879532, TS, KP382683432 * TX); Chris@10: T1d = FNMS(KP382683432, TS, KP923879532 * TX); Chris@10: } Chris@10: { Chris@10: E T13, T18, T2t, T2u; Chris@10: T13 = TZ - T12; Chris@10: T18 = T14 - T17; Chris@10: T19 = FNMS(KP382683432, T18, KP923879532 * T13); Chris@10: T1e = FMA(KP382683432, T13, KP923879532 * T18); Chris@10: T2t = Ti - Tl; Chris@10: T2u = T1r - T1s; Chris@10: T2v = T2t - T2u; Chris@10: T2C = T2t + T2u; Chris@10: } Chris@10: { Chris@10: E T2w, T2x, T1t, T1w; Chris@10: T2w = Tp - Ts; Chris@10: T2x = T1u - T1v; Chris@10: T2y = T2w + T2x; Chris@10: T2D = T2x - T2w; Chris@10: T1t = T1r + T1s; Chris@10: T1w = T1u + T1v; Chris@10: T1x = T1t + T1w; Chris@10: T1V = T1w - T1t; Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tv, T1F, T1b, T1N, T1p, T1P, T1L, T1R; Chris@10: Tv = Tf + Tu; Chris@10: T1F = T1x + T1E; Chris@10: { Chris@10: E TN, T1a, T1f, T1o; Chris@10: TN = TB + TM; Chris@10: T1a = TY + T19; Chris@10: T1b = TN + T1a; Chris@10: T1N = TN - T1a; Chris@10: T1f = T1d + T1e; Chris@10: T1o = T1i + T1n; Chris@10: T1p = T1f + T1o; Chris@10: T1P = T1o - T1f; Chris@10: { Chris@10: E T1I, T1K, T1H, T1J; Chris@10: T1I = Tf - Tu; Chris@10: T1K = T1E - T1x; Chris@10: T1H = W[14]; Chris@10: T1J = W[15]; Chris@10: T1L = FNMS(T1J, T1K, T1H * T1I); Chris@10: T1R = FMA(T1J, T1I, T1H * T1K); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1q, T1G, Tw, T1c; Chris@10: Tw = W[0]; Chris@10: T1c = W[1]; Chris@10: T1q = FMA(Tw, T1b, T1c * T1p); Chris@10: T1G = FNMS(T1c, T1b, Tw * T1p); Chris@10: Rp[0] = Tv - T1q; Chris@10: Ip[0] = T1F + T1G; Chris@10: Rm[0] = Tv + T1q; Chris@10: Im[0] = T1G - T1F; Chris@10: } Chris@10: { Chris@10: E T1Q, T1S, T1M, T1O; Chris@10: T1M = W[16]; Chris@10: T1O = W[17]; Chris@10: T1Q = FMA(T1M, T1N, T1O * T1P); Chris@10: T1S = FNMS(T1O, T1N, T1M * T1P); Chris@10: Rp[WS(rs, 4)] = T1L - T1Q; Chris@10: Ip[WS(rs, 4)] = T1R + T1S; Chris@10: Rm[WS(rs, 4)] = T1L + T1Q; Chris@10: Im[WS(rs, 4)] = T1S - T1R; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T25, T2j, T29, T2l, T21, T2b, T2h, T2n; Chris@10: { Chris@10: E T23, T24, T27, T28; Chris@10: T23 = TB - TM; Chris@10: T24 = T1d - T1e; Chris@10: T25 = T23 + T24; Chris@10: T2j = T23 - T24; Chris@10: T27 = T19 - TY; Chris@10: T28 = T1n - T1i; Chris@10: T29 = T27 + T28; Chris@10: T2l = T28 - T27; Chris@10: } Chris@10: { Chris@10: E T1W, T20, T1T, T1X; Chris@10: T1W = T1U + T1V; Chris@10: T20 = T1Y + T1Z; Chris@10: T1T = W[6]; Chris@10: T1X = W[7]; Chris@10: T21 = FNMS(T1X, T20, T1T * T1W); Chris@10: T2b = FMA(T1X, T1W, T1T * T20); Chris@10: } Chris@10: { Chris@10: E T2e, T2g, T2d, T2f; Chris@10: T2e = T1U - T1V; Chris@10: T2g = T1Z - T1Y; Chris@10: T2d = W[22]; Chris@10: T2f = W[23]; Chris@10: T2h = FNMS(T2f, T2g, T2d * T2e); Chris@10: T2n = FMA(T2f, T2e, T2d * T2g); Chris@10: } Chris@10: { Chris@10: E T2a, T2c, T22, T26; Chris@10: T22 = W[8]; Chris@10: T26 = W[9]; Chris@10: T2a = FMA(T22, T25, T26 * T29); Chris@10: T2c = FNMS(T26, T25, T22 * T29); Chris@10: Rp[WS(rs, 2)] = T21 - T2a; Chris@10: Ip[WS(rs, 2)] = T2b + T2c; Chris@10: Rm[WS(rs, 2)] = T21 + T2a; Chris@10: Im[WS(rs, 2)] = T2c - T2b; Chris@10: } Chris@10: { Chris@10: E T2m, T2o, T2i, T2k; Chris@10: T2i = W[24]; Chris@10: T2k = W[25]; Chris@10: T2m = FMA(T2i, T2j, T2k * T2l); Chris@10: T2o = FNMS(T2k, T2j, T2i * T2l); Chris@10: Rp[WS(rs, 6)] = T2h - T2m; Chris@10: Ip[WS(rs, 6)] = T2n + T2o; Chris@10: Rm[WS(rs, 6)] = T2h + T2m; Chris@10: Im[WS(rs, 6)] = T2o - T2n; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2A, T38, T2I, T3a, T2V, T3d, T33, T3f, T2z, T2E; Chris@10: T2z = KP707106781 * (T2v + T2y); Chris@10: T2A = T2s + T2z; Chris@10: T38 = T2s - T2z; Chris@10: T2E = KP707106781 * (T2C + T2D); Chris@10: T2I = T2E + T2H; Chris@10: T3a = T2H - T2E; Chris@10: { Chris@10: E T2N, T2U, T2Z, T32; Chris@10: T2N = T2L + T2M; Chris@10: T2U = T2Q - T2T; Chris@10: T2V = T2N + T2U; Chris@10: T3d = T2N - T2U; Chris@10: T2Z = T2X + T2Y; Chris@10: T32 = T30 - T31; Chris@10: T33 = T2Z + T32; Chris@10: T3f = T32 - T2Z; Chris@10: } Chris@10: { Chris@10: E T2J, T35, T34, T36; Chris@10: { Chris@10: E T2p, T2B, T2K, T2W; Chris@10: T2p = W[2]; Chris@10: T2B = W[3]; Chris@10: T2J = FNMS(T2B, T2I, T2p * T2A); Chris@10: T35 = FMA(T2B, T2A, T2p * T2I); Chris@10: T2K = W[4]; Chris@10: T2W = W[5]; Chris@10: T34 = FMA(T2K, T2V, T2W * T33); Chris@10: T36 = FNMS(T2W, T2V, T2K * T33); Chris@10: } Chris@10: Rp[WS(rs, 1)] = T2J - T34; Chris@10: Ip[WS(rs, 1)] = T35 + T36; Chris@10: Rm[WS(rs, 1)] = T2J + T34; Chris@10: Im[WS(rs, 1)] = T36 - T35; Chris@10: } Chris@10: { Chris@10: E T3b, T3h, T3g, T3i; Chris@10: { Chris@10: E T37, T39, T3c, T3e; Chris@10: T37 = W[18]; Chris@10: T39 = W[19]; Chris@10: T3b = FNMS(T39, T3a, T37 * T38); Chris@10: T3h = FMA(T39, T38, T37 * T3a); Chris@10: T3c = W[20]; Chris@10: T3e = W[21]; Chris@10: T3g = FMA(T3c, T3d, T3e * T3f); Chris@10: T3i = FNMS(T3e, T3d, T3c * T3f); Chris@10: } Chris@10: Rp[WS(rs, 5)] = T3b - T3g; Chris@10: Ip[WS(rs, 5)] = T3h + T3i; Chris@10: Rm[WS(rs, 5)] = T3b + T3g; Chris@10: Im[WS(rs, 5)] = T3i - T3h; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T3m, T3E, T3q, T3G, T3v, T3J, T3z, T3L, T3l, T3o; Chris@10: T3l = KP707106781 * (T2D - T2C); Chris@10: T3m = T3k + T3l; Chris@10: T3E = T3k - T3l; Chris@10: T3o = KP707106781 * (T2v - T2y); Chris@10: T3q = T3o + T3p; Chris@10: T3G = T3p - T3o; Chris@10: { Chris@10: E T3t, T3u, T3x, T3y; Chris@10: T3t = T2L - T2M; Chris@10: T3u = T2X - T2Y; Chris@10: T3v = T3t + T3u; Chris@10: T3J = T3t - T3u; Chris@10: T3x = T31 + T30; Chris@10: T3y = T2Q + T2T; Chris@10: T3z = T3x - T3y; Chris@10: T3L = T3y + T3x; Chris@10: } Chris@10: { Chris@10: E T3r, T3B, T3A, T3C; Chris@10: { Chris@10: E T3j, T3n, T3s, T3w; Chris@10: T3j = W[10]; Chris@10: T3n = W[11]; Chris@10: T3r = FNMS(T3n, T3q, T3j * T3m); Chris@10: T3B = FMA(T3n, T3m, T3j * T3q); Chris@10: T3s = W[12]; Chris@10: T3w = W[13]; Chris@10: T3A = FMA(T3s, T3v, T3w * T3z); Chris@10: T3C = FNMS(T3w, T3v, T3s * T3z); Chris@10: } Chris@10: Rp[WS(rs, 3)] = T3r - T3A; Chris@10: Ip[WS(rs, 3)] = T3B + T3C; Chris@10: Rm[WS(rs, 3)] = T3r + T3A; Chris@10: Im[WS(rs, 3)] = T3C - T3B; Chris@10: } Chris@10: { Chris@10: E T3H, T3N, T3M, T3O; Chris@10: { Chris@10: E T3D, T3F, T3I, T3K; Chris@10: T3D = W[26]; Chris@10: T3F = W[27]; Chris@10: T3H = FNMS(T3F, T3G, T3D * T3E); Chris@10: T3N = FMA(T3F, T3E, T3D * T3G); Chris@10: T3I = W[28]; Chris@10: T3K = W[29]; Chris@10: T3M = FMA(T3I, T3J, T3K * T3L); Chris@10: T3O = FNMS(T3K, T3J, T3I * T3L); Chris@10: } Chris@10: Rp[WS(rs, 7)] = T3H - T3M; Chris@10: Ip[WS(rs, 7)] = T3N + T3O; Chris@10: Rm[WS(rs, 7)] = T3H + T3M; Chris@10: Im[WS(rs, 7)] = T3O - T3N; 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, "hc2cbdft_16", twinstr, &GENUS, {168, 46, 38, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_16) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_16, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */