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 07:42:04 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 12 -dif -name hc2cbdft_12 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 142 FP additions, 68 FP multiplications, Chris@10: * (or, 96 additions, 22 multiplications, 46 fused multiply/add), Chris@10: * 81 stack variables, 2 constants, and 48 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 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)) { Chris@10: E T2S, T2V, T2w, T2Z, T2T, T2I, T2Q, T2Y, T2U, T2K, T2G, T30, T2W; Chris@10: { Chris@10: E Tb, T1Z, T2D, T1E, T1N, T2y, TD, T2t, T1U, T1e, T2o, TY, T1f, TI, T1g; Chris@10: E TN, Tm, T1V, T2z, T1H, T1Q, T2E, T19, T2u; Chris@10: { Chris@10: E T1c, TU, T1d, TX; Chris@10: { Chris@10: E Tu, T6, TT, TS, T5, Tt, Tw, Tx, TB, T9, Ty; Chris@10: { Chris@10: E T1, Tp, Tq, Tr, T4, T2, T3, T7, T8, Ts; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rp[WS(rs, 4)]; Chris@10: T3 = Rm[WS(rs, 3)]; Chris@10: Tp = Ip[0]; Chris@10: Tq = Ip[WS(rs, 4)]; Chris@10: Tr = Im[WS(rs, 3)]; Chris@10: T4 = T2 + T3; Chris@10: Tu = T2 - T3; Chris@10: T6 = Rm[WS(rs, 5)]; Chris@10: TT = Tr + Tq; Chris@10: Ts = Tq - Tr; Chris@10: TS = FNMS(KP500000000, T4, T1); Chris@10: T5 = T1 + T4; Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = Rp[WS(rs, 2)]; Chris@10: T1c = Tp + Ts; Chris@10: Tt = FNMS(KP500000000, Ts, Tp); Chris@10: Tw = Im[WS(rs, 5)]; Chris@10: Tx = Im[WS(rs, 1)]; Chris@10: TB = T7 - T8; Chris@10: T9 = T7 + T8; Chris@10: Ty = Ip[WS(rs, 2)]; Chris@10: } Chris@10: { Chris@10: E T1L, Tv, Ta, TV, TW, Tz; Chris@10: T1L = FNMS(KP866025403, Tu, Tt); Chris@10: Tv = FMA(KP866025403, Tu, Tt); Chris@10: Ta = T6 + T9; Chris@10: TV = FNMS(KP500000000, T9, T6); Chris@10: TW = Tx + Ty; Chris@10: Tz = Tx - Ty; Chris@10: { Chris@10: E TC, T1M, T1C, TA, T1D; Chris@10: T1C = FMA(KP866025403, TT, TS); Chris@10: TU = FNMS(KP866025403, TT, TS); Chris@10: T1d = Tw + Tz; Chris@10: TA = FNMS(KP500000000, Tz, Tw); Chris@10: T1D = FNMS(KP866025403, TW, TV); Chris@10: TX = FMA(KP866025403, TW, TV); Chris@10: Tb = T5 + Ta; Chris@10: T1Z = T5 - Ta; Chris@10: TC = FNMS(KP866025403, TB, TA); Chris@10: T1M = FMA(KP866025403, TB, TA); Chris@10: T2D = T1C - T1D; Chris@10: T1E = T1C + T1D; Chris@10: T1N = T1L - T1M; Chris@10: T2y = T1L + T1M; Chris@10: TD = Tv + TC; Chris@10: T2t = Tv - TC; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T12, Th, TH, TE, Tg, T11, T14, TK, T17, Tk, TL; Chris@10: { Chris@10: E Tc, TZ, TF, TG, Tf, Td, Te, Ti, Tj, T10; Chris@10: Tc = Rp[WS(rs, 3)]; Chris@10: T1U = T1c + T1d; Chris@10: T1e = T1c - T1d; Chris@10: T2o = TU + TX; Chris@10: TY = TU - TX; Chris@10: Td = Rm[WS(rs, 4)]; Chris@10: Te = Rm[0]; Chris@10: TZ = Ip[WS(rs, 3)]; Chris@10: TF = Im[WS(rs, 4)]; Chris@10: TG = Im[0]; Chris@10: Tf = Td + Te; Chris@10: T12 = Td - Te; Chris@10: Th = Rm[WS(rs, 2)]; Chris@10: TH = TF - TG; Chris@10: T10 = TF + TG; Chris@10: TE = FNMS(KP500000000, Tf, Tc); Chris@10: Tg = Tc + Tf; Chris@10: Ti = Rp[WS(rs, 1)]; Chris@10: Tj = Rp[WS(rs, 5)]; Chris@10: T1f = TZ - T10; Chris@10: T11 = FMA(KP500000000, T10, TZ); Chris@10: T14 = Im[WS(rs, 2)]; Chris@10: TK = Ip[WS(rs, 5)]; Chris@10: T17 = Ti - Tj; Chris@10: Tk = Ti + Tj; Chris@10: TL = Ip[WS(rs, 1)]; Chris@10: } Chris@10: { Chris@10: E T1O, T13, Tl, TJ, TM, T15; Chris@10: T1O = FNMS(KP866025403, T12, T11); Chris@10: T13 = FMA(KP866025403, T12, T11); Chris@10: Tl = Th + Tk; Chris@10: TJ = FNMS(KP500000000, Tk, Th); Chris@10: TM = TK - TL; Chris@10: T15 = TK + TL; Chris@10: { Chris@10: E T18, T1P, T1F, T16, T1G; Chris@10: T1F = FNMS(KP866025403, TH, TE); Chris@10: TI = FMA(KP866025403, TH, TE); Chris@10: T1g = T15 - T14; Chris@10: T16 = FMA(KP500000000, T15, T14); Chris@10: T1G = FNMS(KP866025403, TM, TJ); Chris@10: TN = FMA(KP866025403, TM, TJ); Chris@10: Tm = Tg + Tl; Chris@10: T1V = Tg - Tl; Chris@10: T18 = FNMS(KP866025403, T17, T16); Chris@10: T1P = FMA(KP866025403, T17, T16); Chris@10: T2z = T1F - T1G; Chris@10: T1H = T1F + T1G; Chris@10: T1Q = T1O - T1P; Chris@10: T2E = T1O + T1P; Chris@10: T19 = T13 + T18; Chris@10: T2u = T13 - T18; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T20, T2p, T1v, T1s, T1q, T1y, T1u, T1z, T1t; Chris@10: { Chris@10: E T1m, Tn, T1a, T1p, T1i, To, TP, TR, T1h, TO; Chris@10: T1m = Tb - Tm; Chris@10: Tn = Tb + Tm; Chris@10: T20 = T1f - T1g; Chris@10: T1h = T1f + T1g; Chris@10: T2p = TI + TN; Chris@10: TO = TI - TN; Chris@10: T1a = TY - T19; Chris@10: T1v = TY + T19; Chris@10: T1p = T1e - T1h; Chris@10: T1i = T1e + T1h; Chris@10: To = W[0]; Chris@10: T1s = TD - TO; Chris@10: TP = TD + TO; Chris@10: TR = W[1]; Chris@10: { Chris@10: E T1l, T1o, T1n, T1x, T1r; Chris@10: { Chris@10: E T1j, TQ, T1k, T1b; Chris@10: T1j = To * T1a; Chris@10: TQ = To * TP; Chris@10: T1l = W[10]; Chris@10: T1k = FNMS(TR, TP, T1j); Chris@10: T1b = FMA(TR, T1a, TQ); Chris@10: T1o = W[11]; Chris@10: T1n = T1l * T1m; Chris@10: Im[0] = T1k - T1i; Chris@10: Ip[0] = T1i + T1k; Chris@10: Rm[0] = Tn + T1b; Chris@10: Rp[0] = Tn - T1b; Chris@10: T1x = T1o * T1m; Chris@10: T1r = W[12]; Chris@10: } Chris@10: T1q = FNMS(T1o, T1p, T1n); Chris@10: T1y = FMA(T1l, T1p, T1x); Chris@10: T1u = W[13]; Chris@10: T1z = T1r * T1v; Chris@10: T1t = T1r * T1s; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2e, T2h, T1S, T2j, T2f, T26, T2c, T2m, T2g, T24, T22; Chris@10: { Chris@10: E T2b, T1R, T27, T2a, T1B, T29, T2l, T1K, T1J, T1W, T21, T25, T2d, T23, T1X; Chris@10: E T1Y; Chris@10: { Chris@10: E T1I, T28, T1A, T1w, T1T; Chris@10: T1A = FNMS(T1u, T1s, T1z); Chris@10: T1w = FMA(T1u, T1v, T1t); Chris@10: T1I = T1E - T1H; Chris@10: T28 = T1E + T1H; Chris@10: T2b = T1N + T1Q; Chris@10: T1R = T1N - T1Q; Chris@10: Im[WS(rs, 3)] = T1A - T1y; Chris@10: Ip[WS(rs, 3)] = T1y + T1A; Chris@10: Rm[WS(rs, 3)] = T1q + T1w; Chris@10: Rp[WS(rs, 3)] = T1q - T1w; Chris@10: T27 = W[14]; Chris@10: T2a = W[15]; Chris@10: T1B = W[2]; Chris@10: T29 = T27 * T28; Chris@10: T2l = T2a * T28; Chris@10: T1K = W[3]; Chris@10: T1J = T1B * T1I; Chris@10: T1W = T1U - T1V; Chris@10: T2e = T1V + T1U; Chris@10: T2h = T1Z - T20; Chris@10: T21 = T1Z + T20; Chris@10: T25 = T1K * T1I; Chris@10: T1T = W[4]; Chris@10: T2d = W[16]; Chris@10: T23 = T1T * T21; Chris@10: T1X = T1T * T1W; Chris@10: } Chris@10: T1S = FNMS(T1K, T1R, T1J); Chris@10: T2j = T2d * T2h; Chris@10: T2f = T2d * T2e; Chris@10: T26 = FMA(T1B, T1R, T25); Chris@10: T1Y = W[5]; Chris@10: T2c = FNMS(T2a, T2b, T29); Chris@10: T2m = FMA(T27, T2b, T2l); Chris@10: T2g = W[17]; Chris@10: T24 = FNMS(T1Y, T1W, T23); Chris@10: T22 = FMA(T1Y, T21, T1X); Chris@10: } Chris@10: { Chris@10: E T2L, T2O, T2P, T2v, T2N, T2X, T2n, T2s, T2A, T2F, T2r, T2H, T2R, T2J, T2B; Chris@10: E T2C; Chris@10: { Chris@10: E T2q, T2k, T2i, T2M, T2x; Chris@10: T2k = FNMS(T2g, T2e, T2j); Chris@10: T2i = FMA(T2g, T2h, T2f); Chris@10: Im[WS(rs, 1)] = T24 - T26; Chris@10: Ip[WS(rs, 1)] = T24 + T26; Chris@10: Rm[WS(rs, 1)] = T22 + T1S; Chris@10: Rp[WS(rs, 1)] = T1S - T22; Chris@10: Im[WS(rs, 4)] = T2k - T2m; Chris@10: Ip[WS(rs, 4)] = T2k + T2m; Chris@10: Rm[WS(rs, 4)] = T2i + T2c; Chris@10: Rp[WS(rs, 4)] = T2c - T2i; Chris@10: T2q = T2o + T2p; Chris@10: T2M = T2o - T2p; Chris@10: T2L = W[18]; Chris@10: T2O = W[19]; Chris@10: T2P = T2t - T2u; Chris@10: T2v = T2t + T2u; Chris@10: T2N = T2L * T2M; Chris@10: T2X = T2O * T2M; Chris@10: T2n = W[6]; Chris@10: T2s = W[7]; Chris@10: T2S = T2y - T2z; Chris@10: T2A = T2y + T2z; Chris@10: T2F = T2D - T2E; Chris@10: T2V = T2D + T2E; Chris@10: T2r = T2n * T2q; Chris@10: T2H = T2s * T2q; Chris@10: T2x = W[8]; Chris@10: T2R = W[20]; Chris@10: T2J = T2x * T2F; Chris@10: T2B = T2x * T2A; Chris@10: } Chris@10: T2w = FNMS(T2s, T2v, T2r); Chris@10: T2Z = T2R * T2V; Chris@10: T2T = T2R * T2S; Chris@10: T2I = FMA(T2n, T2v, T2H); Chris@10: T2C = W[9]; Chris@10: T2Q = FNMS(T2O, T2P, T2N); Chris@10: T2Y = FMA(T2L, T2P, T2X); Chris@10: T2U = W[21]; Chris@10: T2K = FNMS(T2C, T2A, T2J); Chris@10: T2G = FMA(T2C, T2F, T2B); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: T30 = FNMS(T2U, T2S, T2Z); Chris@10: T2W = FMA(T2U, T2V, T2T); Chris@10: Im[WS(rs, 2)] = T2K - T2I; Chris@10: Ip[WS(rs, 2)] = T2I + T2K; Chris@10: Rm[WS(rs, 2)] = T2w + T2G; Chris@10: Rp[WS(rs, 2)] = T2w - T2G; Chris@10: Im[WS(rs, 5)] = T30 - T2Y; Chris@10: Ip[WS(rs, 5)] = T2Y + T30; Chris@10: Rm[WS(rs, 5)] = T2Q + T2W; Chris@10: Rp[WS(rs, 5)] = T2Q - T2W; Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 12}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, "hc2cbdft_12", twinstr, &GENUS, {96, 22, 46, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_12, &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 12 -dif -name hc2cbdft_12 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 142 FP additions, 60 FP multiplications, Chris@10: * (or, 112 additions, 30 multiplications, 30 fused multiply/add), Chris@10: * 47 stack variables, 2 constants, and 48 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 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)) { Chris@10: E Tv, T1E, TC, T1F, TW, T1x, TT, T1w, T1d, T1N, Tb, T1R, TI, T1z, TN; Chris@10: E T1A, T17, T1I, T12, T1H, T1g, T1S, Tm, T1O; Chris@10: { Chris@10: E T1, Tq, T6, TA, T4, Tp, Tt, TS, T9, Tw, Tz, TV; Chris@10: T1 = Rp[0]; Chris@10: Tq = Ip[0]; Chris@10: T6 = Rm[WS(rs, 5)]; Chris@10: TA = Im[WS(rs, 5)]; Chris@10: { Chris@10: E T2, T3, Tr, Ts; Chris@10: T2 = Rp[WS(rs, 4)]; Chris@10: T3 = Rm[WS(rs, 3)]; Chris@10: T4 = T2 + T3; Chris@10: Tp = KP866025403 * (T2 - T3); Chris@10: Tr = Im[WS(rs, 3)]; Chris@10: Ts = Ip[WS(rs, 4)]; Chris@10: Tt = Tr - Ts; Chris@10: TS = KP866025403 * (Tr + Ts); Chris@10: } Chris@10: { Chris@10: E T7, T8, Tx, Ty; Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = Rp[WS(rs, 2)]; Chris@10: T9 = T7 + T8; Chris@10: Tw = KP866025403 * (T7 - T8); Chris@10: Tx = Im[WS(rs, 1)]; Chris@10: Ty = Ip[WS(rs, 2)]; Chris@10: Tz = Tx - Ty; Chris@10: TV = KP866025403 * (Tx + Ty); Chris@10: } Chris@10: { Chris@10: E Tu, TB, TU, TR; Chris@10: Tu = FMA(KP500000000, Tt, Tq); Chris@10: Tv = Tp + Tu; Chris@10: T1E = Tu - Tp; Chris@10: TB = FMS(KP500000000, Tz, TA); Chris@10: TC = Tw + TB; Chris@10: T1F = TB - Tw; Chris@10: TU = FNMS(KP500000000, T9, T6); Chris@10: TW = TU + TV; Chris@10: T1x = TU - TV; Chris@10: TR = FNMS(KP500000000, T4, T1); Chris@10: TT = TR - TS; Chris@10: T1w = TR + TS; Chris@10: { Chris@10: E T1b, T1c, T5, Ta; Chris@10: T1b = Tq - Tt; Chris@10: T1c = Tz + TA; Chris@10: T1d = T1b - T1c; Chris@10: T1N = T1b + T1c; Chris@10: T5 = T1 + T4; Chris@10: Ta = T6 + T9; Chris@10: Tb = T5 + Ta; Chris@10: T1R = T5 - Ta; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tc, T10, Th, T15, Tf, TY, TH, TZ, Tk, T13, TM, T14; Chris@10: Tc = Rp[WS(rs, 3)]; Chris@10: T10 = Ip[WS(rs, 3)]; Chris@10: Th = Rm[WS(rs, 2)]; Chris@10: T15 = Im[WS(rs, 2)]; Chris@10: { Chris@10: E Td, Te, TF, TG; Chris@10: Td = Rm[WS(rs, 4)]; Chris@10: Te = Rm[0]; Chris@10: Tf = Td + Te; Chris@10: TY = KP866025403 * (Td - Te); Chris@10: TF = Im[WS(rs, 4)]; Chris@10: TG = Im[0]; Chris@10: TH = KP866025403 * (TF - TG); Chris@10: TZ = TF + TG; Chris@10: } Chris@10: { Chris@10: E Ti, Tj, TK, TL; Chris@10: Ti = Rp[WS(rs, 1)]; Chris@10: Tj = Rp[WS(rs, 5)]; Chris@10: Tk = Ti + Tj; Chris@10: T13 = KP866025403 * (Ti - Tj); Chris@10: TK = Ip[WS(rs, 5)]; Chris@10: TL = Ip[WS(rs, 1)]; Chris@10: TM = KP866025403 * (TK - TL); Chris@10: T14 = TK + TL; Chris@10: } Chris@10: { Chris@10: E TE, TJ, T16, T11; Chris@10: TE = FNMS(KP500000000, Tf, Tc); Chris@10: TI = TE + TH; Chris@10: T1z = TE - TH; Chris@10: TJ = FNMS(KP500000000, Tk, Th); Chris@10: TN = TJ + TM; Chris@10: T1A = TJ - TM; Chris@10: T16 = FMA(KP500000000, T14, T15); Chris@10: T17 = T13 - T16; Chris@10: T1I = T13 + T16; Chris@10: T11 = FMA(KP500000000, TZ, T10); Chris@10: T12 = TY + T11; Chris@10: T1H = T11 - TY; Chris@10: { Chris@10: E T1e, T1f, Tg, Tl; Chris@10: T1e = T10 - TZ; Chris@10: T1f = T14 - T15; Chris@10: T1g = T1e + T1f; Chris@10: T1S = T1e - T1f; Chris@10: Tg = Tc + Tf; Chris@10: Tl = Th + Tk; Chris@10: Tm = Tg + Tl; Chris@10: T1O = Tg - Tl; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tn, T1h, TP, T1p, T19, T1r, T1n, T1t; Chris@10: Tn = Tb + Tm; Chris@10: T1h = T1d + T1g; Chris@10: { Chris@10: E TD, TO, TX, T18; Chris@10: TD = Tv - TC; Chris@10: TO = TI - TN; Chris@10: TP = TD + TO; Chris@10: T1p = TD - TO; Chris@10: TX = TT - TW; Chris@10: T18 = T12 - T17; Chris@10: T19 = TX - T18; Chris@10: T1r = TX + T18; Chris@10: { Chris@10: E T1k, T1m, T1j, T1l; Chris@10: T1k = Tb - Tm; Chris@10: T1m = T1d - T1g; Chris@10: T1j = W[10]; Chris@10: T1l = W[11]; Chris@10: T1n = FNMS(T1l, T1m, T1j * T1k); Chris@10: T1t = FMA(T1l, T1k, T1j * T1m); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1a, T1i, To, TQ; Chris@10: To = W[0]; Chris@10: TQ = W[1]; Chris@10: T1a = FMA(To, TP, TQ * T19); Chris@10: T1i = FNMS(TQ, TP, To * T19); Chris@10: Rp[0] = Tn - T1a; Chris@10: Ip[0] = T1h + T1i; Chris@10: Rm[0] = Tn + T1a; Chris@10: Im[0] = T1i - T1h; Chris@10: } Chris@10: { Chris@10: E T1s, T1u, T1o, T1q; Chris@10: T1o = W[12]; Chris@10: T1q = W[13]; Chris@10: T1s = FMA(T1o, T1p, T1q * T1r); Chris@10: T1u = FNMS(T1q, T1p, T1o * T1r); Chris@10: Rp[WS(rs, 3)] = T1n - T1s; Chris@10: Ip[WS(rs, 3)] = T1t + T1u; Chris@10: Rm[WS(rs, 3)] = T1n + T1s; Chris@10: Im[WS(rs, 3)] = T1u - T1t; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1C, T1Y, T1K, T20, T1U, T1V, T26, T27; Chris@10: { Chris@10: E T1y, T1B, T1G, T1J; Chris@10: T1y = T1w + T1x; Chris@10: T1B = T1z + T1A; Chris@10: T1C = T1y - T1B; Chris@10: T1Y = T1y + T1B; Chris@10: T1G = T1E + T1F; Chris@10: T1J = T1H - T1I; Chris@10: T1K = T1G - T1J; Chris@10: T20 = T1G + T1J; Chris@10: } Chris@10: { Chris@10: E T1P, T1T, T1M, T1Q; Chris@10: T1P = T1N - T1O; Chris@10: T1T = T1R + T1S; Chris@10: T1M = W[4]; Chris@10: T1Q = W[5]; Chris@10: T1U = FMA(T1M, T1P, T1Q * T1T); Chris@10: T1V = FNMS(T1Q, T1P, T1M * T1T); Chris@10: } Chris@10: { Chris@10: E T23, T25, T22, T24; Chris@10: T23 = T1O + T1N; Chris@10: T25 = T1R - T1S; Chris@10: T22 = W[16]; Chris@10: T24 = W[17]; Chris@10: T26 = FMA(T22, T23, T24 * T25); Chris@10: T27 = FNMS(T24, T23, T22 * T25); Chris@10: } Chris@10: { Chris@10: E T1L, T1W, T1v, T1D; Chris@10: T1v = W[2]; Chris@10: T1D = W[3]; Chris@10: T1L = FNMS(T1D, T1K, T1v * T1C); Chris@10: T1W = FMA(T1D, T1C, T1v * T1K); Chris@10: Rp[WS(rs, 1)] = T1L - T1U; Chris@10: Ip[WS(rs, 1)] = T1V + T1W; Chris@10: Rm[WS(rs, 1)] = T1U + T1L; Chris@10: Im[WS(rs, 1)] = T1V - T1W; Chris@10: } Chris@10: { Chris@10: E T21, T28, T1X, T1Z; Chris@10: T1X = W[14]; Chris@10: T1Z = W[15]; Chris@10: T21 = FNMS(T1Z, T20, T1X * T1Y); Chris@10: T28 = FMA(T1Z, T1Y, T1X * T20); Chris@10: Rp[WS(rs, 4)] = T21 - T26; Chris@10: Ip[WS(rs, 4)] = T27 + T28; Chris@10: Rm[WS(rs, 4)] = T26 + T21; Chris@10: Im[WS(rs, 4)] = T27 - T28; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T2c, T2u, T2p, T2B, T2g, T2w, T2l, T2z; Chris@10: { Chris@10: E T2a, T2b, T2n, T2o; Chris@10: T2a = TT + TW; Chris@10: T2b = TI + TN; Chris@10: T2c = T2a + T2b; Chris@10: T2u = T2a - T2b; Chris@10: T2n = T1w - T1x; Chris@10: T2o = T1H + T1I; Chris@10: T2p = T2n - T2o; Chris@10: T2B = T2n + T2o; Chris@10: } Chris@10: { Chris@10: E T2e, T2f, T2j, T2k; Chris@10: T2e = Tv + TC; Chris@10: T2f = T12 + T17; Chris@10: T2g = T2e + T2f; Chris@10: T2w = T2e - T2f; Chris@10: T2j = T1E - T1F; Chris@10: T2k = T1z - T1A; Chris@10: T2l = T2j + T2k; Chris@10: T2z = T2j - T2k; Chris@10: } Chris@10: { Chris@10: E T2h, T2r, T2q, T2s; Chris@10: { Chris@10: E T29, T2d, T2i, T2m; Chris@10: T29 = W[6]; Chris@10: T2d = W[7]; Chris@10: T2h = FNMS(T2d, T2g, T29 * T2c); Chris@10: T2r = FMA(T2d, T2c, T29 * T2g); Chris@10: T2i = W[8]; Chris@10: T2m = W[9]; Chris@10: T2q = FMA(T2i, T2l, T2m * T2p); Chris@10: T2s = FNMS(T2m, T2l, T2i * T2p); Chris@10: } Chris@10: Rp[WS(rs, 2)] = T2h - T2q; Chris@10: Ip[WS(rs, 2)] = T2r + T2s; Chris@10: Rm[WS(rs, 2)] = T2h + T2q; Chris@10: Im[WS(rs, 2)] = T2s - T2r; Chris@10: } Chris@10: { Chris@10: E T2x, T2D, T2C, T2E; Chris@10: { Chris@10: E T2t, T2v, T2y, T2A; Chris@10: T2t = W[18]; Chris@10: T2v = W[19]; Chris@10: T2x = FNMS(T2v, T2w, T2t * T2u); Chris@10: T2D = FMA(T2v, T2u, T2t * T2w); Chris@10: T2y = W[20]; Chris@10: T2A = W[21]; Chris@10: T2C = FMA(T2y, T2z, T2A * T2B); Chris@10: T2E = FNMS(T2A, T2z, T2y * T2B); Chris@10: } Chris@10: Rp[WS(rs, 5)] = T2x - T2C; Chris@10: Ip[WS(rs, 5)] = T2D + T2E; Chris@10: Rm[WS(rs, 5)] = T2x + T2C; Chris@10: Im[WS(rs, 5)] = T2E - T2D; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 12}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 12, "hc2cbdft_12", twinstr, &GENUS, {112, 30, 30, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_12) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_12, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */