cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:07:11 EDT 2018 */ cannam@167: cannam@167: #include "rdft/codelet-rdft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cfdft_12 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 142 FP additions, 92 FP multiplications, cannam@167: * (or, 96 additions, 46 multiplications, 46 fused multiply/add), cannam@167: * 65 stack variables, 2 constants, and 48 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) { cannam@167: E To, T1E, T1m, T2H, Ta, T1G, Tk, T1I, Tl, T1J, T1s, T2b, T1A, T2d, T1B; cannam@167: E T2I, T12, T18, T19, T24, T26, T2C, Tz, T1M, T1f, T2B, TJ, T1O, TT, T1Q; cannam@167: E TU, T1R; cannam@167: { cannam@167: E Tm, Tn, T1u, T1x, T1y, T1z, T1v, T2c, Te, Tj, T1i, T1l, Tf, T1H, T4; cannam@167: E T1o, T9, T1r, T5, T1F, T1p, T2a, T1t, T1, T1n; cannam@167: Tm = Ip[0]; cannam@167: Tn = Im[0]; cannam@167: T1u = Tm + Tn; cannam@167: T1x = Rp[0]; cannam@167: T1y = Rm[0]; cannam@167: T1z = T1x - T1y; cannam@167: T1t = W[0]; cannam@167: T1v = T1t * T1u; cannam@167: T2c = T1t * T1z; cannam@167: { cannam@167: E Tc, Td, Th, Ti, Tb; cannam@167: Tc = Ip[WS(rs, 4)]; cannam@167: Td = Im[WS(rs, 4)]; cannam@167: Te = Tc - Td; cannam@167: Th = Rp[WS(rs, 4)]; cannam@167: Ti = Rm[WS(rs, 4)]; cannam@167: Tj = Th + Ti; cannam@167: T1i = Tc + Td; cannam@167: T1l = Th - Ti; cannam@167: Tb = W[14]; cannam@167: Tf = Tb * Te; cannam@167: T1H = Tb * Tj; cannam@167: } cannam@167: { cannam@167: E T2, T3, T7, T8; cannam@167: T2 = Ip[WS(rs, 2)]; cannam@167: T3 = Im[WS(rs, 2)]; cannam@167: T4 = T2 - T3; cannam@167: T1o = T2 + T3; cannam@167: T7 = Rp[WS(rs, 2)]; cannam@167: T8 = Rm[WS(rs, 2)]; cannam@167: T9 = T7 + T8; cannam@167: T1r = T7 - T8; cannam@167: } cannam@167: T1 = W[6]; cannam@167: T5 = T1 * T4; cannam@167: T1F = T1 * T9; cannam@167: T1n = W[8]; cannam@167: T1p = T1n * T1o; cannam@167: T2a = T1n * T1r; cannam@167: To = Tm - Tn; cannam@167: T1E = T1x + T1y; cannam@167: { cannam@167: E T1j, T2G, T1h, T1k; cannam@167: T1h = W[16]; cannam@167: T1j = T1h * T1i; cannam@167: T2G = T1h * T1l; cannam@167: T1k = W[17]; cannam@167: T1m = FNMS(T1k, T1l, T1j); cannam@167: T2H = FMA(T1k, T1i, T2G); cannam@167: } cannam@167: { cannam@167: E T6, Tg, T1q, T1w; cannam@167: T6 = W[7]; cannam@167: Ta = FNMS(T6, T9, T5); cannam@167: T1G = FMA(T6, T4, T1F); cannam@167: Tg = W[15]; cannam@167: Tk = FNMS(Tg, Tj, Tf); cannam@167: T1I = FMA(Tg, Te, T1H); cannam@167: Tl = Ta + Tk; cannam@167: T1J = T1G + T1I; cannam@167: T1q = W[9]; cannam@167: T1s = FNMS(T1q, T1r, T1p); cannam@167: T2b = FMA(T1q, T1o, T2a); cannam@167: T1w = W[1]; cannam@167: T1A = FNMS(T1w, T1z, T1v); cannam@167: T2d = FMA(T1w, T1u, T2c); cannam@167: T1B = T1s + T1A; cannam@167: T2I = T2b + T2d; cannam@167: } cannam@167: } cannam@167: { cannam@167: E Tt, T11, Ty, T10, T23, TX, TZ, TN, TS, T1b, T1e, TO, T1P, TD, TI; cannam@167: E T17, T16, T25, T13, T15, TE, T1N, TF, TP; cannam@167: { cannam@167: E Tr, Ts, Tw, Tx, TY; cannam@167: Tr = Ip[WS(rs, 3)]; cannam@167: Ts = Im[WS(rs, 3)]; cannam@167: Tt = Tr - Ts; cannam@167: T11 = Tr + Ts; cannam@167: Tw = Rp[WS(rs, 3)]; cannam@167: Tx = Rm[WS(rs, 3)]; cannam@167: TY = Tx - Tw; cannam@167: Ty = Tw + Tx; cannam@167: T10 = W[12]; cannam@167: T23 = T10 * TY; cannam@167: TX = W[13]; cannam@167: TZ = TX * TY; cannam@167: } cannam@167: { cannam@167: E TL, TM, TQ, TR, TK; cannam@167: TL = Ip[WS(rs, 1)]; cannam@167: TM = Im[WS(rs, 1)]; cannam@167: TN = TL - TM; cannam@167: TQ = Rp[WS(rs, 1)]; cannam@167: TR = Rm[WS(rs, 1)]; cannam@167: TS = TQ + TR; cannam@167: T1b = TL + TM; cannam@167: T1e = TQ - TR; cannam@167: TK = W[2]; cannam@167: TO = TK * TN; cannam@167: T1P = TK * TS; cannam@167: } cannam@167: { cannam@167: E TB, TC, T14, TG, TH, TA; cannam@167: TB = Ip[WS(rs, 5)]; cannam@167: TC = Im[WS(rs, 5)]; cannam@167: TD = TB - TC; cannam@167: TG = Rp[WS(rs, 5)]; cannam@167: TH = Rm[WS(rs, 5)]; cannam@167: TI = TG + TH; cannam@167: T14 = TH - TG; cannam@167: T17 = TB + TC; cannam@167: T16 = W[20]; cannam@167: T25 = T16 * T14; cannam@167: T13 = W[21]; cannam@167: T15 = T13 * T14; cannam@167: TA = W[18]; cannam@167: TE = TA * TD; cannam@167: T1N = TA * TI; cannam@167: } cannam@167: T12 = FMA(T10, T11, TZ); cannam@167: T18 = FMA(T16, T17, T15); cannam@167: T19 = T12 + T18; cannam@167: T24 = FNMS(TX, T11, T23); cannam@167: T26 = FNMS(T13, T17, T25); cannam@167: T2C = T24 + T26; cannam@167: { cannam@167: E Tu, T1L, Tq, Tv; cannam@167: Tq = W[10]; cannam@167: Tu = Tq * Tt; cannam@167: T1L = Tq * Ty; cannam@167: Tv = W[11]; cannam@167: Tz = FNMS(Tv, Ty, Tu); cannam@167: T1M = FMA(Tv, Tt, T1L); cannam@167: } cannam@167: { cannam@167: E T1c, T2A, T1a, T1d; cannam@167: T1a = W[4]; cannam@167: T1c = T1a * T1b; cannam@167: T2A = T1a * T1e; cannam@167: T1d = W[5]; cannam@167: T1f = FNMS(T1d, T1e, T1c); cannam@167: T2B = FMA(T1d, T1b, T2A); cannam@167: } cannam@167: TF = W[19]; cannam@167: TJ = FNMS(TF, TI, TE); cannam@167: T1O = FMA(TF, TD, T1N); cannam@167: TP = W[3]; cannam@167: TT = FNMS(TP, TS, TO); cannam@167: T1Q = FMA(TP, TN, T1P); cannam@167: TU = TJ + TT; cannam@167: T1R = T1O + T1Q; cannam@167: } cannam@167: { cannam@167: E TW, T2V, T2Y, T30, T1D, T1U, T1T, T2Z; cannam@167: { cannam@167: E Tp, TV, T2W, T2X; cannam@167: Tp = Tl + To; cannam@167: TV = Tz + TU; cannam@167: TW = Tp - TV; cannam@167: T2V = TV + Tp; cannam@167: T2W = T2C - T2B; cannam@167: T2X = T2H + T2I; cannam@167: T2Y = T2W - T2X; cannam@167: T30 = T2W + T2X; cannam@167: } cannam@167: { cannam@167: E T1g, T1C, T1K, T1S; cannam@167: T1g = T19 + T1f; cannam@167: T1C = T1m + T1B; cannam@167: T1D = T1g - T1C; cannam@167: T1U = T1g + T1C; cannam@167: T1K = T1E + T1J; cannam@167: T1S = T1M + T1R; cannam@167: T1T = T1K + T1S; cannam@167: T2Z = T1K - T1S; cannam@167: } cannam@167: Ip[WS(rs, 3)] = KP500000000 * (TW + T1D); cannam@167: Rp[WS(rs, 3)] = KP500000000 * (T2Z - T30); cannam@167: Im[WS(rs, 2)] = KP500000000 * (T1D - TW); cannam@167: Rm[WS(rs, 2)] = KP500000000 * (T2Z + T30); cannam@167: Rm[WS(rs, 5)] = KP500000000 * (T1T - T1U); cannam@167: Im[WS(rs, 5)] = KP500000000 * (T2Y - T2V); cannam@167: Rp[0] = KP500000000 * (T1T + T1U); cannam@167: Ip[0] = KP500000000 * (T2V + T2Y); cannam@167: } cannam@167: { cannam@167: E T1X, T2v, T2F, T2Q, T2L, T2R, T20, T2w, T28, T2t, T2j, T2p, T2m, T2q, T2f; cannam@167: E T2s; cannam@167: { cannam@167: E T1V, T1W, T2D, T2E; cannam@167: T1V = FNMS(KP500000000, T1J, T1E); cannam@167: T1W = Ta - Tk; cannam@167: T1X = FNMS(KP866025403, T1W, T1V); cannam@167: T2v = FMA(KP866025403, T1W, T1V); cannam@167: T2D = FMA(KP500000000, T2C, T2B); cannam@167: T2E = T18 - T12; cannam@167: T2F = FNMS(KP866025403, T2E, T2D); cannam@167: T2Q = FMA(KP866025403, T2E, T2D); cannam@167: } cannam@167: { cannam@167: E T2J, T2K, T1Y, T1Z; cannam@167: T2J = FNMS(KP500000000, T2I, T2H); cannam@167: T2K = T1s - T1A; cannam@167: T2L = FNMS(KP866025403, T2K, T2J); cannam@167: T2R = FMA(KP866025403, T2K, T2J); cannam@167: T1Y = FNMS(KP500000000, T1R, T1M); cannam@167: T1Z = TJ - TT; cannam@167: T20 = FNMS(KP866025403, T1Z, T1Y); cannam@167: T2w = FMA(KP866025403, T1Z, T1Y); cannam@167: } cannam@167: { cannam@167: E T22, T27, T2h, T2i; cannam@167: T22 = FNMS(KP500000000, T19, T1f); cannam@167: T27 = T24 - T26; cannam@167: T28 = FNMS(KP866025403, T27, T22); cannam@167: T2t = FMA(KP866025403, T27, T22); cannam@167: T2h = FNMS(KP500000000, Tl, To); cannam@167: T2i = T1I - T1G; cannam@167: T2j = FNMS(KP866025403, T2i, T2h); cannam@167: T2p = FMA(KP866025403, T2i, T2h); cannam@167: } cannam@167: { cannam@167: E T2k, T2l, T29, T2e; cannam@167: T2k = FNMS(KP500000000, TU, Tz); cannam@167: T2l = T1Q - T1O; cannam@167: T2m = FNMS(KP866025403, T2l, T2k); cannam@167: T2q = FMA(KP866025403, T2l, T2k); cannam@167: T29 = FNMS(KP500000000, T1B, T1m); cannam@167: T2e = T2b - T2d; cannam@167: T2f = FNMS(KP866025403, T2e, T29); cannam@167: T2s = FMA(KP866025403, T2e, T29); cannam@167: } cannam@167: { cannam@167: E T21, T2g, T2P, T2S; cannam@167: T21 = T1X + T20; cannam@167: T2g = T28 + T2f; cannam@167: Rp[WS(rs, 2)] = KP500000000 * (T21 - T2g); cannam@167: Rm[WS(rs, 3)] = KP500000000 * (T21 + T2g); cannam@167: T2P = T2m + T2j; cannam@167: T2S = T2Q + T2R; cannam@167: Ip[WS(rs, 2)] = KP500000000 * (T2P + T2S); cannam@167: Im[WS(rs, 3)] = KP500000000 * (T2S - T2P); cannam@167: } cannam@167: { cannam@167: E T2n, T2o, T2T, T2U; cannam@167: T2n = T2j - T2m; cannam@167: T2o = T2f - T28; cannam@167: Ip[WS(rs, 5)] = KP500000000 * (T2n + T2o); cannam@167: Im[0] = KP500000000 * (T2o - T2n); cannam@167: T2T = T1X - T20; cannam@167: T2U = T2R - T2Q; cannam@167: Rm[0] = KP500000000 * (T2T - T2U); cannam@167: Rp[WS(rs, 5)] = KP500000000 * (T2T + T2U); cannam@167: } cannam@167: { cannam@167: E T2r, T2u, T2N, T2O; cannam@167: T2r = T2p - T2q; cannam@167: T2u = T2s - T2t; cannam@167: Ip[WS(rs, 1)] = KP500000000 * (T2r + T2u); cannam@167: Im[WS(rs, 4)] = KP500000000 * (T2u - T2r); cannam@167: T2N = T2v - T2w; cannam@167: T2O = T2L - T2F; cannam@167: Rm[WS(rs, 4)] = KP500000000 * (T2N - T2O); cannam@167: Rp[WS(rs, 1)] = KP500000000 * (T2N + T2O); cannam@167: } cannam@167: { cannam@167: E T2x, T2y, T2z, T2M; cannam@167: T2x = T2v + T2w; cannam@167: T2y = T2t + T2s; cannam@167: Rm[WS(rs, 1)] = KP500000000 * (T2x - T2y); cannam@167: Rp[WS(rs, 4)] = KP500000000 * (T2x + T2y); cannam@167: T2z = T2q + T2p; cannam@167: T2M = T2F + T2L; cannam@167: Ip[WS(rs, 4)] = KP500000000 * (T2z - T2M); cannam@167: Im[WS(rs, 1)] = -(KP500000000 * (T2z + T2M)); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 12}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {96, 46, 46, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft_12) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft_12, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 12 -dit -name hc2cfdft_12 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 142 FP additions, 76 FP multiplications, cannam@167: * (or, 112 additions, 46 multiplications, 30 fused multiply/add), cannam@167: * 52 stack variables, 3 constants, and 48 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: DK(KP433012701, +0.433012701892219323381861585376468091735701313); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 22); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 22, MAKE_VOLATILE_STRIDE(48, rs)) { cannam@167: E Tm, T1t, T1d, T2j, Tj, T1Y, T1w, T1G, T1q, T2q, T1U, T2k, Tw, T1y, T17; cannam@167: E T2g, TP, T21, T1B, T1J, T12, T2u, T1P, T2h; cannam@167: { cannam@167: E Tk, Tl, T1k, T1m, T1n, T1o, T4, T1f, T8, T1h, Th, T1c, Td, T1a, T19; cannam@167: E T1b; cannam@167: { cannam@167: E T2, T3, T6, T7; cannam@167: Tk = Ip[0]; cannam@167: Tl = Im[0]; cannam@167: T1k = Tk + Tl; cannam@167: T1m = Rp[0]; cannam@167: T1n = Rm[0]; cannam@167: T1o = T1m - T1n; cannam@167: T2 = Ip[WS(rs, 2)]; cannam@167: T3 = Im[WS(rs, 2)]; cannam@167: T4 = T2 - T3; cannam@167: T1f = T2 + T3; cannam@167: T6 = Rp[WS(rs, 2)]; cannam@167: T7 = Rm[WS(rs, 2)]; cannam@167: T8 = T6 + T7; cannam@167: T1h = T6 - T7; cannam@167: { cannam@167: E Tf, Tg, Tb, Tc; cannam@167: Tf = Rp[WS(rs, 4)]; cannam@167: Tg = Rm[WS(rs, 4)]; cannam@167: Th = Tf + Tg; cannam@167: T1c = Tf - Tg; cannam@167: Tb = Ip[WS(rs, 4)]; cannam@167: Tc = Im[WS(rs, 4)]; cannam@167: Td = Tb - Tc; cannam@167: T1a = Tb + Tc; cannam@167: } cannam@167: } cannam@167: Tm = Tk - Tl; cannam@167: T1t = T1m + T1n; cannam@167: T19 = W[16]; cannam@167: T1b = W[17]; cannam@167: T1d = FNMS(T1b, T1c, T19 * T1a); cannam@167: T2j = FMA(T19, T1c, T1b * T1a); cannam@167: { cannam@167: E T9, T1u, Ti, T1v; cannam@167: { cannam@167: E T1, T5, Ta, Te; cannam@167: T1 = W[6]; cannam@167: T5 = W[7]; cannam@167: T9 = FNMS(T5, T8, T1 * T4); cannam@167: T1u = FMA(T1, T8, T5 * T4); cannam@167: Ta = W[14]; cannam@167: Te = W[15]; cannam@167: Ti = FNMS(Te, Th, Ta * Td); cannam@167: T1v = FMA(Ta, Th, Te * Td); cannam@167: } cannam@167: Tj = T9 + Ti; cannam@167: T1Y = KP433012701 * (T1v - T1u); cannam@167: T1w = T1u + T1v; cannam@167: T1G = KP433012701 * (T9 - Ti); cannam@167: } cannam@167: { cannam@167: E T1i, T1S, T1p, T1T; cannam@167: { cannam@167: E T1e, T1g, T1j, T1l; cannam@167: T1e = W[8]; cannam@167: T1g = W[9]; cannam@167: T1i = FNMS(T1g, T1h, T1e * T1f); cannam@167: T1S = FMA(T1e, T1h, T1g * T1f); cannam@167: T1j = W[0]; cannam@167: T1l = W[1]; cannam@167: T1p = FNMS(T1l, T1o, T1j * T1k); cannam@167: T1T = FMA(T1j, T1o, T1l * T1k); cannam@167: } cannam@167: T1q = T1i + T1p; cannam@167: T2q = KP433012701 * (T1i - T1p); cannam@167: T1U = KP433012701 * (T1S - T1T); cannam@167: T2k = T1S + T1T; cannam@167: } cannam@167: } cannam@167: { cannam@167: E Tr, TT, Tv, TV, TA, TY, TE, T10, TN, T14, TJ, T16; cannam@167: { cannam@167: E Tp, Tq, TC, TD; cannam@167: Tp = Ip[WS(rs, 3)]; cannam@167: Tq = Im[WS(rs, 3)]; cannam@167: Tr = Tp - Tq; cannam@167: TT = Tp + Tq; cannam@167: { cannam@167: E Tt, Tu, Ty, Tz; cannam@167: Tt = Rp[WS(rs, 3)]; cannam@167: Tu = Rm[WS(rs, 3)]; cannam@167: Tv = Tt + Tu; cannam@167: TV = Tt - Tu; cannam@167: Ty = Ip[WS(rs, 5)]; cannam@167: Tz = Im[WS(rs, 5)]; cannam@167: TA = Ty - Tz; cannam@167: TY = Ty + Tz; cannam@167: } cannam@167: TC = Rp[WS(rs, 5)]; cannam@167: TD = Rm[WS(rs, 5)]; cannam@167: TE = TC + TD; cannam@167: T10 = TC - TD; cannam@167: { cannam@167: E TL, TM, TH, TI; cannam@167: TL = Rp[WS(rs, 1)]; cannam@167: TM = Rm[WS(rs, 1)]; cannam@167: TN = TL + TM; cannam@167: T14 = TM - TL; cannam@167: TH = Ip[WS(rs, 1)]; cannam@167: TI = Im[WS(rs, 1)]; cannam@167: TJ = TH - TI; cannam@167: T16 = TH + TI; cannam@167: } cannam@167: } cannam@167: { cannam@167: E To, Ts, T13, T15; cannam@167: To = W[10]; cannam@167: Ts = W[11]; cannam@167: Tw = FNMS(Ts, Tv, To * Tr); cannam@167: T1y = FMA(To, Tv, Ts * Tr); cannam@167: T13 = W[5]; cannam@167: T15 = W[4]; cannam@167: T17 = FMA(T13, T14, T15 * T16); cannam@167: T2g = FNMS(T13, T16, T15 * T14); cannam@167: } cannam@167: { cannam@167: E TF, T1z, TO, T1A; cannam@167: { cannam@167: E Tx, TB, TG, TK; cannam@167: Tx = W[18]; cannam@167: TB = W[19]; cannam@167: TF = FNMS(TB, TE, Tx * TA); cannam@167: T1z = FMA(Tx, TE, TB * TA); cannam@167: TG = W[2]; cannam@167: TK = W[3]; cannam@167: TO = FNMS(TK, TN, TG * TJ); cannam@167: T1A = FMA(TG, TN, TK * TJ); cannam@167: } cannam@167: TP = TF + TO; cannam@167: T21 = KP433012701 * (T1A - T1z); cannam@167: T1B = T1z + T1A; cannam@167: T1J = KP433012701 * (TF - TO); cannam@167: } cannam@167: { cannam@167: E TW, T1O, T11, T1N; cannam@167: { cannam@167: E TS, TU, TX, TZ; cannam@167: TS = W[12]; cannam@167: TU = W[13]; cannam@167: TW = FNMS(TU, TV, TS * TT); cannam@167: T1O = FMA(TS, TV, TU * TT); cannam@167: TX = W[20]; cannam@167: TZ = W[21]; cannam@167: T11 = FNMS(TZ, T10, TX * TY); cannam@167: T1N = FMA(TX, T10, TZ * TY); cannam@167: } cannam@167: T12 = TW + T11; cannam@167: T2u = KP433012701 * (T11 - TW); cannam@167: T1P = KP433012701 * (T1N - T1O); cannam@167: T2h = T1O + T1N; cannam@167: } cannam@167: } cannam@167: { cannam@167: E TR, T2f, T2m, T2o, T1s, T1E, T1D, T2n; cannam@167: { cannam@167: E Tn, TQ, T2i, T2l; cannam@167: Tn = Tj + Tm; cannam@167: TQ = Tw + TP; cannam@167: TR = Tn - TQ; cannam@167: T2f = TQ + Tn; cannam@167: T2i = T2g - T2h; cannam@167: T2l = T2j + T2k; cannam@167: T2m = T2i - T2l; cannam@167: T2o = T2i + T2l; cannam@167: } cannam@167: { cannam@167: E T18, T1r, T1x, T1C; cannam@167: T18 = T12 + T17; cannam@167: T1r = T1d + T1q; cannam@167: T1s = T18 - T1r; cannam@167: T1E = T18 + T1r; cannam@167: T1x = T1t + T1w; cannam@167: T1C = T1y + T1B; cannam@167: T1D = T1x + T1C; cannam@167: T2n = T1x - T1C; cannam@167: } cannam@167: Ip[WS(rs, 3)] = KP500000000 * (TR + T1s); cannam@167: Rp[WS(rs, 3)] = KP500000000 * (T2n - T2o); cannam@167: Im[WS(rs, 2)] = KP500000000 * (T1s - TR); cannam@167: Rm[WS(rs, 2)] = KP500000000 * (T2n + T2o); cannam@167: Rm[WS(rs, 5)] = KP500000000 * (T1D - T1E); cannam@167: Im[WS(rs, 5)] = KP500000000 * (T2m - T2f); cannam@167: Rp[0] = KP500000000 * (T1D + T1E); cannam@167: Ip[0] = KP500000000 * (T2f + T2m); cannam@167: } cannam@167: { cannam@167: E T1H, T2b, T2s, T2B, T2v, T2A, T1K, T2c, T1Q, T29, T1Z, T25, T22, T26, T1V; cannam@167: E T28; cannam@167: { cannam@167: E T1F, T2r, T2t, T1I; cannam@167: T1F = FNMS(KP250000000, T1w, KP500000000 * T1t); cannam@167: T1H = T1F - T1G; cannam@167: T2b = T1F + T1G; cannam@167: T2r = FNMS(KP500000000, T2j, KP250000000 * T2k); cannam@167: T2s = T2q - T2r; cannam@167: T2B = T2q + T2r; cannam@167: T2t = FMA(KP250000000, T2h, KP500000000 * T2g); cannam@167: T2v = T2t - T2u; cannam@167: T2A = T2u + T2t; cannam@167: T1I = FNMS(KP250000000, T1B, KP500000000 * T1y); cannam@167: T1K = T1I - T1J; cannam@167: T2c = T1I + T1J; cannam@167: } cannam@167: { cannam@167: E T1M, T1X, T20, T1R; cannam@167: T1M = FNMS(KP250000000, T12, KP500000000 * T17); cannam@167: T1Q = T1M - T1P; cannam@167: T29 = T1P + T1M; cannam@167: T1X = FNMS(KP250000000, Tj, KP500000000 * Tm); cannam@167: T1Z = T1X - T1Y; cannam@167: T25 = T1Y + T1X; cannam@167: T20 = FNMS(KP250000000, TP, KP500000000 * Tw); cannam@167: T22 = T20 - T21; cannam@167: T26 = T21 + T20; cannam@167: T1R = FNMS(KP250000000, T1q, KP500000000 * T1d); cannam@167: T1V = T1R - T1U; cannam@167: T28 = T1R + T1U; cannam@167: } cannam@167: { cannam@167: E T1L, T1W, T2p, T2w; cannam@167: T1L = T1H + T1K; cannam@167: T1W = T1Q + T1V; cannam@167: Rp[WS(rs, 2)] = T1L - T1W; cannam@167: Rm[WS(rs, 3)] = T1L + T1W; cannam@167: T2p = T22 + T1Z; cannam@167: T2w = T2s - T2v; cannam@167: Ip[WS(rs, 2)] = T2p + T2w; cannam@167: Im[WS(rs, 3)] = T2w - T2p; cannam@167: } cannam@167: { cannam@167: E T23, T24, T2x, T2y; cannam@167: T23 = T1Z - T22; cannam@167: T24 = T1V - T1Q; cannam@167: Ip[WS(rs, 5)] = T23 + T24; cannam@167: Im[0] = T24 - T23; cannam@167: T2x = T1H - T1K; cannam@167: T2y = T2v + T2s; cannam@167: Rm[0] = T2x - T2y; cannam@167: Rp[WS(rs, 5)] = T2x + T2y; cannam@167: } cannam@167: { cannam@167: E T27, T2a, T2z, T2C; cannam@167: T27 = T25 - T26; cannam@167: T2a = T28 - T29; cannam@167: Ip[WS(rs, 1)] = T27 + T2a; cannam@167: Im[WS(rs, 4)] = T2a - T27; cannam@167: T2z = T2b - T2c; cannam@167: T2C = T2A - T2B; cannam@167: Rm[WS(rs, 4)] = T2z - T2C; cannam@167: Rp[WS(rs, 1)] = T2z + T2C; cannam@167: } cannam@167: { cannam@167: E T2d, T2e, T2D, T2E; cannam@167: T2d = T2b + T2c; cannam@167: T2e = T29 + T28; cannam@167: Rm[WS(rs, 1)] = T2d - T2e; cannam@167: Rp[WS(rs, 4)] = T2d + T2e; cannam@167: T2D = T26 + T25; cannam@167: T2E = T2A + T2B; cannam@167: Ip[WS(rs, 4)] = T2D + T2E; cannam@167: Im[WS(rs, 1)] = T2E - T2D; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 12}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 12, "hc2cfdft_12", twinstr, &GENUS, {112, 46, 30, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft_12) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft_12, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #endif