Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:51:30 EDT 2016 */ Chris@42: Chris@42: #include "codelet-rdft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 12 -dif -name hc2cb_12 -include hc2cb.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 118 FP additions, 68 FP multiplications, Chris@42: * (or, 72 additions, 22 multiplications, 46 fused multiply/add), Chris@42: * 64 stack variables, 2 constants, and 48 memory accesses Chris@42: */ Chris@42: #include "hc2cb.h" Chris@42: Chris@42: static void hc2cb_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: 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@42: E T1U, T1X, T1W, T1Y, T1V; Chris@42: { Chris@42: E T18, T20, T21, T1b, T2a, T1s, T29, T1p, TO, T11, To, Tb, Tg, T23, T1f; Chris@42: E Tl, Ty, Tt, T1i, T24, T1z, T2d, T1w, T2c; Chris@42: { Chris@42: E T5, Ta, TN, TI; Chris@42: { Chris@42: E T1, TE, T6, TM, T7, T1o, T4, T17, TH, T8, TJ, TK; Chris@42: T1 = Rp[0]; Chris@42: TE = Ip[0]; Chris@42: T6 = Rm[WS(rs, 5)]; Chris@42: TM = Im[WS(rs, 5)]; Chris@42: { Chris@42: E T2, T3, TF, TG; Chris@42: T2 = Rp[WS(rs, 4)]; Chris@42: T3 = Rm[WS(rs, 3)]; Chris@42: TF = Ip[WS(rs, 4)]; Chris@42: TG = Im[WS(rs, 3)]; Chris@42: T7 = Rm[WS(rs, 1)]; Chris@42: T1o = T2 - T3; Chris@42: T4 = T2 + T3; Chris@42: T17 = TF + TG; Chris@42: TH = TF - TG; Chris@42: T8 = Rp[WS(rs, 2)]; Chris@42: TJ = Ip[WS(rs, 2)]; Chris@42: TK = Im[WS(rs, 1)]; Chris@42: } Chris@42: { Chris@42: E T1r, T1a, T19, T1q, T9, TL, T16, T1n; Chris@42: T5 = T1 + T4; Chris@42: T16 = FNMS(KP500000000, T4, T1); Chris@42: T1r = T7 - T8; Chris@42: T9 = T7 + T8; Chris@42: T1a = TJ + TK; Chris@42: TL = TJ - TK; Chris@42: T18 = FNMS(KP866025403, T17, T16); Chris@42: T20 = FMA(KP866025403, T17, T16); Chris@42: T19 = FNMS(KP500000000, T9, T6); Chris@42: Ta = T6 + T9; Chris@42: TN = TL - TM; Chris@42: T1q = FMA(KP500000000, TL, TM); Chris@42: T1n = FNMS(KP500000000, TH, TE); Chris@42: TI = TE + TH; Chris@42: T21 = FNMS(KP866025403, T1a, T19); Chris@42: T1b = FMA(KP866025403, T1a, T19); Chris@42: T2a = FMA(KP866025403, T1r, T1q); Chris@42: T1s = FNMS(KP866025403, T1r, T1q); Chris@42: T29 = FNMS(KP866025403, T1o, T1n); Chris@42: T1p = FMA(KP866025403, T1o, T1n); Chris@42: } Chris@42: } Chris@42: { Chris@42: E Tc, Tp, Th, Tx, Ti, Tf, T1v, Ts, T1e, Tj, Tu, Tv; Chris@42: Tc = Rp[WS(rs, 3)]; Chris@42: TO = TI - TN; Chris@42: T11 = TI + TN; Chris@42: Tp = Ip[WS(rs, 3)]; Chris@42: To = T5 - Ta; Chris@42: Tb = T5 + Ta; Chris@42: Th = Rm[WS(rs, 2)]; Chris@42: Tx = Im[WS(rs, 2)]; Chris@42: { Chris@42: E Td, Te, Tq, Tr; Chris@42: Td = Rm[WS(rs, 4)]; Chris@42: Te = Rm[0]; Chris@42: Tq = Im[WS(rs, 4)]; Chris@42: Tr = Im[0]; Chris@42: Ti = Rp[WS(rs, 1)]; Chris@42: Tf = Td + Te; Chris@42: T1v = Td - Te; Chris@42: Ts = Tq + Tr; Chris@42: T1e = Tq - Tr; Chris@42: Tj = Rp[WS(rs, 5)]; Chris@42: Tu = Ip[WS(rs, 1)]; Chris@42: Tv = Ip[WS(rs, 5)]; Chris@42: } Chris@42: { Chris@42: E T1y, T1h, T1g, T1x, Tk, Tw, T1d, T1u; Chris@42: T1d = FNMS(KP500000000, Tf, Tc); Chris@42: Tg = Tc + Tf; Chris@42: Tk = Ti + Tj; Chris@42: T1y = Ti - Tj; Chris@42: Tw = Tu + Tv; Chris@42: T1h = Tv - Tu; Chris@42: T23 = FNMS(KP866025403, T1e, T1d); Chris@42: T1f = FMA(KP866025403, T1e, T1d); Chris@42: Tl = Th + Tk; Chris@42: T1g = FNMS(KP500000000, Tk, Th); Chris@42: T1x = FMA(KP500000000, Tw, Tx); Chris@42: Ty = Tw - Tx; Chris@42: Tt = Tp - Ts; Chris@42: T1u = FMA(KP500000000, Ts, Tp); Chris@42: T1i = FMA(KP866025403, T1h, T1g); Chris@42: T24 = FNMS(KP866025403, T1h, T1g); Chris@42: T1z = FNMS(KP866025403, T1y, T1x); Chris@42: T2d = FMA(KP866025403, T1y, T1x); Chris@42: T1w = FMA(KP866025403, T1v, T1u); Chris@42: T2c = FNMS(KP866025403, T1v, T1u); Chris@42: } Chris@42: } Chris@42: } Chris@42: { Chris@42: E TY, T13, TX, T10; Chris@42: { Chris@42: E Tn, T12, TC, Tm, TD, TS, TA, Tz; Chris@42: Tn = W[16]; Chris@42: T12 = Tt + Ty; Chris@42: Tz = Tt - Ty; Chris@42: TC = W[17]; Chris@42: Tm = Tg + Tl; Chris@42: TD = Tg - Tl; Chris@42: TS = To + Tz; Chris@42: TA = To - Tz; Chris@42: { Chris@42: E TV, TU, TW, TT; Chris@42: { Chris@42: E TQ, TR, TP, TB; Chris@42: TV = TO - TD; Chris@42: TP = TD + TO; Chris@42: Rp[0] = Tb + Tm; Chris@42: TB = Tn * TA; Chris@42: TQ = Tn * TP; Chris@42: TR = W[4]; Chris@42: Ip[WS(rs, 4)] = FNMS(TC, TP, TB); Chris@42: TU = W[5]; Chris@42: Im[WS(rs, 4)] = FMA(TC, TA, TQ); Chris@42: TW = TR * TV; Chris@42: TT = TR * TS; Chris@42: } Chris@42: Im[WS(rs, 1)] = FMA(TU, TS, TW); Chris@42: Ip[WS(rs, 1)] = FNMS(TU, TV, TT); Chris@42: TY = Tb - Tm; Chris@42: T13 = T11 - T12; Chris@42: TX = W[10]; Chris@42: T10 = W[11]; Chris@42: Rm[0] = T11 + T12; Chris@42: } Chris@42: } Chris@42: { Chris@42: E T1K, T1Q, T1P, T1L, T2o, T2u, T2t, T2p; Chris@42: { Chris@42: E T1E, T1D, T1H, T1F, T1G, T1t, T1k, T1A; Chris@42: { Chris@42: E T1c, TZ, T14, T1j; Chris@42: T1K = T18 - T1b; Chris@42: T1c = T18 + T1b; Chris@42: TZ = TX * TY; Chris@42: T14 = T10 * TY; Chris@42: T1j = T1f + T1i; Chris@42: T1Q = T1f - T1i; Chris@42: T1P = T1p + T1s; Chris@42: T1t = T1p - T1s; Chris@42: Rp[WS(rs, 3)] = FNMS(T10, T13, TZ); Chris@42: Rm[WS(rs, 3)] = FMA(TX, T13, T14); Chris@42: T1E = T1c + T1j; Chris@42: T1k = T1c - T1j; Chris@42: T1A = T1w - T1z; Chris@42: T1L = T1w + T1z; Chris@42: } Chris@42: { Chris@42: E T15, T1m, T1B, T1l, T1C; Chris@42: T15 = W[18]; Chris@42: T1m = W[19]; Chris@42: T1D = W[6]; Chris@42: T1H = T1t + T1A; Chris@42: T1B = T1t - T1A; Chris@42: T1l = T15 * T1k; Chris@42: T1C = T1m * T1k; Chris@42: T1F = T1D * T1E; Chris@42: T1G = W[7]; Chris@42: Rp[WS(rs, 5)] = FNMS(T1m, T1B, T1l); Chris@42: Rm[WS(rs, 5)] = FMA(T15, T1B, T1C); Chris@42: } Chris@42: { Chris@42: E T26, T2i, T2l, T2f, T1Z, T28; Chris@42: { Chris@42: E T22, T1I, T25, T2b, T2e; Chris@42: T22 = T20 + T21; Chris@42: T2o = T20 - T21; Chris@42: Rp[WS(rs, 2)] = FNMS(T1G, T1H, T1F); Chris@42: T1I = T1G * T1E; Chris@42: T2u = T23 - T24; Chris@42: T25 = T23 + T24; Chris@42: T2b = T29 - T2a; Chris@42: T2t = T29 + T2a; Chris@42: T2p = T2c + T2d; Chris@42: T2e = T2c - T2d; Chris@42: Rm[WS(rs, 2)] = FMA(T1D, T1H, T1I); Chris@42: T26 = T22 - T25; Chris@42: T2i = T22 + T25; Chris@42: T2l = T2b + T2e; Chris@42: T2f = T2b - T2e; Chris@42: } Chris@42: T1Z = W[2]; Chris@42: T28 = W[3]; Chris@42: { Chris@42: E T2h, T2k, T27, T2g, T2j, T2m; Chris@42: T2h = W[14]; Chris@42: T2k = W[15]; Chris@42: T27 = T1Z * T26; Chris@42: T2g = T28 * T26; Chris@42: T2j = T2h * T2i; Chris@42: T2m = T2k * T2i; Chris@42: Rp[WS(rs, 1)] = FNMS(T28, T2f, T27); Chris@42: Rm[WS(rs, 1)] = FMA(T1Z, T2f, T2g); Chris@42: Rp[WS(rs, 4)] = FNMS(T2k, T2l, T2j); Chris@42: Rm[WS(rs, 4)] = FMA(T2h, T2l, T2m); Chris@42: } Chris@42: } Chris@42: } Chris@42: { Chris@42: E T2y, T2B, T2A, T2C, T2z; Chris@42: { Chris@42: E T2n, T2q, T2v, T2s, T2r, T2x, T2w; Chris@42: T2n = W[8]; Chris@42: T2y = T2o + T2p; Chris@42: T2q = T2o - T2p; Chris@42: T2B = T2t - T2u; Chris@42: T2v = T2t + T2u; Chris@42: T2s = W[9]; Chris@42: T2r = T2n * T2q; Chris@42: T2x = W[20]; Chris@42: T2w = T2n * T2v; Chris@42: T2A = W[21]; Chris@42: Ip[WS(rs, 2)] = FNMS(T2s, T2v, T2r); Chris@42: T2C = T2x * T2B; Chris@42: T2z = T2x * T2y; Chris@42: Im[WS(rs, 2)] = FMA(T2s, T2q, T2w); Chris@42: } Chris@42: Im[WS(rs, 5)] = FMA(T2A, T2y, T2C); Chris@42: Ip[WS(rs, 5)] = FNMS(T2A, T2B, T2z); Chris@42: { Chris@42: E T1J, T1M, T1R, T1O, T1N, T1T, T1S; Chris@42: T1J = W[0]; Chris@42: T1U = T1K + T1L; Chris@42: T1M = T1K - T1L; Chris@42: T1X = T1P - T1Q; Chris@42: T1R = T1P + T1Q; Chris@42: T1O = W[1]; Chris@42: T1N = T1J * T1M; Chris@42: T1T = W[12]; Chris@42: T1S = T1J * T1R; Chris@42: T1W = W[13]; Chris@42: Ip[0] = FNMS(T1O, T1R, T1N); Chris@42: T1Y = T1T * T1X; Chris@42: T1V = T1T * T1U; Chris@42: Im[0] = FMA(T1O, T1M, T1S); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: Im[WS(rs, 3)] = FMA(T1W, T1U, T1Y); Chris@42: Ip[WS(rs, 3)] = FNMS(T1W, T1X, T1V); Chris@42: } Chris@42: } Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: {TW_FULL, 1, 12}, Chris@42: {TW_NEXT, 1, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 12, "hc2cb_12", twinstr, &GENUS, {72, 22, 46, 0} }; Chris@42: Chris@42: void X(codelet_hc2cb_12) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cb_12, &desc, HC2C_VIA_RDFT); Chris@42: } Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 12 -dif -name hc2cb_12 -include hc2cb.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 118 FP additions, 60 FP multiplications, Chris@42: * (or, 88 additions, 30 multiplications, 30 fused multiply/add), Chris@42: * 39 stack variables, 2 constants, and 48 memory accesses Chris@42: */ Chris@42: #include "hc2cb.h" Chris@42: Chris@42: static void hc2cb_12(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: { Chris@42: INT m; Chris@42: 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@42: E T5, TH, T12, T1M, T1i, T1U, Tl, Ty, T1c, T1Y, T1s, T1Q, Ta, TM, T15; Chris@42: E T1N, T1l, T1V, Tg, Tt, T19, T1X, T1p, T1P; Chris@42: { Chris@42: E T1, TD, T4, T1g, TG, T11, T10, T1h; Chris@42: T1 = Rp[0]; Chris@42: TD = Ip[0]; Chris@42: { Chris@42: E T2, T3, TE, TF; Chris@42: T2 = Rp[WS(rs, 4)]; Chris@42: T3 = Rm[WS(rs, 3)]; Chris@42: T4 = T2 + T3; Chris@42: T1g = KP866025403 * (T2 - T3); Chris@42: TE = Ip[WS(rs, 4)]; Chris@42: TF = Im[WS(rs, 3)]; Chris@42: TG = TE - TF; Chris@42: T11 = KP866025403 * (TE + TF); Chris@42: } Chris@42: T5 = T1 + T4; Chris@42: TH = TD + TG; Chris@42: T10 = FNMS(KP500000000, T4, T1); Chris@42: T12 = T10 - T11; Chris@42: T1M = T10 + T11; Chris@42: T1h = FNMS(KP500000000, TG, TD); Chris@42: T1i = T1g + T1h; Chris@42: T1U = T1h - T1g; Chris@42: } Chris@42: { Chris@42: E Th, Tx, Tk, T1a, Tw, T1r, T1b, T1q; Chris@42: Th = Rm[WS(rs, 2)]; Chris@42: Tx = Im[WS(rs, 2)]; Chris@42: { Chris@42: E Ti, Tj, Tu, Tv; Chris@42: Ti = Rp[WS(rs, 1)]; Chris@42: Tj = Rp[WS(rs, 5)]; Chris@42: Tk = Ti + Tj; Chris@42: T1a = KP866025403 * (Ti - Tj); Chris@42: Tu = Ip[WS(rs, 1)]; Chris@42: Tv = Ip[WS(rs, 5)]; Chris@42: Tw = Tu + Tv; Chris@42: T1r = KP866025403 * (Tv - Tu); Chris@42: } Chris@42: Tl = Th + Tk; Chris@42: Ty = Tw - Tx; Chris@42: T1b = FMA(KP500000000, Tw, Tx); Chris@42: T1c = T1a - T1b; Chris@42: T1Y = T1a + T1b; Chris@42: T1q = FNMS(KP500000000, Tk, Th); Chris@42: T1s = T1q + T1r; Chris@42: T1Q = T1q - T1r; Chris@42: } Chris@42: { Chris@42: E T6, TL, T9, T1j, TK, T14, T13, T1k; Chris@42: T6 = Rm[WS(rs, 5)]; Chris@42: TL = Im[WS(rs, 5)]; Chris@42: { Chris@42: E T7, T8, TI, TJ; Chris@42: T7 = Rm[WS(rs, 1)]; Chris@42: T8 = Rp[WS(rs, 2)]; Chris@42: T9 = T7 + T8; Chris@42: T1j = KP866025403 * (T7 - T8); Chris@42: TI = Ip[WS(rs, 2)]; Chris@42: TJ = Im[WS(rs, 1)]; Chris@42: TK = TI - TJ; Chris@42: T14 = KP866025403 * (TI + TJ); Chris@42: } Chris@42: Ta = T6 + T9; Chris@42: TM = TK - TL; Chris@42: T13 = FNMS(KP500000000, T9, T6); Chris@42: T15 = T13 + T14; Chris@42: T1N = T13 - T14; Chris@42: T1k = FMA(KP500000000, TK, TL); Chris@42: T1l = T1j - T1k; Chris@42: T1V = T1j + T1k; Chris@42: } Chris@42: { Chris@42: E Tc, Tp, Tf, T17, Ts, T1o, T18, T1n; Chris@42: Tc = Rp[WS(rs, 3)]; Chris@42: Tp = Ip[WS(rs, 3)]; Chris@42: { Chris@42: E Td, Te, Tq, Tr; Chris@42: Td = Rm[WS(rs, 4)]; Chris@42: Te = Rm[0]; Chris@42: Tf = Td + Te; Chris@42: T17 = KP866025403 * (Td - Te); Chris@42: Tq = Im[WS(rs, 4)]; Chris@42: Tr = Im[0]; Chris@42: Ts = Tq + Tr; Chris@42: T1o = KP866025403 * (Tq - Tr); Chris@42: } Chris@42: Tg = Tc + Tf; Chris@42: Tt = Tp - Ts; Chris@42: T18 = FMA(KP500000000, Ts, Tp); Chris@42: T19 = T17 + T18; Chris@42: T1X = T18 - T17; Chris@42: T1n = FNMS(KP500000000, Tf, Tc); Chris@42: T1p = T1n + T1o; Chris@42: T1P = T1n - T1o; Chris@42: } Chris@42: { Chris@42: E Tb, Tm, TU, TW, TX, TY, TT, TV; Chris@42: Tb = T5 + Ta; Chris@42: Tm = Tg + Tl; Chris@42: TU = Tb - Tm; Chris@42: TW = TH + TM; Chris@42: TX = Tt + Ty; Chris@42: TY = TW - TX; Chris@42: Rp[0] = Tb + Tm; Chris@42: Rm[0] = TW + TX; Chris@42: TT = W[10]; Chris@42: TV = W[11]; Chris@42: Rp[WS(rs, 3)] = FNMS(TV, TY, TT * TU); Chris@42: Rm[WS(rs, 3)] = FMA(TV, TU, TT * TY); Chris@42: } Chris@42: { Chris@42: E TA, TQ, TO, TS; Chris@42: { Chris@42: E To, Tz, TC, TN; Chris@42: To = T5 - Ta; Chris@42: Tz = Tt - Ty; Chris@42: TA = To - Tz; Chris@42: TQ = To + Tz; Chris@42: TC = Tg - Tl; Chris@42: TN = TH - TM; Chris@42: TO = TC + TN; Chris@42: TS = TN - TC; Chris@42: } Chris@42: { Chris@42: E Tn, TB, TP, TR; Chris@42: Tn = W[16]; Chris@42: TB = W[17]; Chris@42: Ip[WS(rs, 4)] = FNMS(TB, TO, Tn * TA); Chris@42: Im[WS(rs, 4)] = FMA(Tn, TO, TB * TA); Chris@42: TP = W[4]; Chris@42: TR = W[5]; Chris@42: Ip[WS(rs, 1)] = FNMS(TR, TS, TP * TQ); Chris@42: Im[WS(rs, 1)] = FMA(TP, TS, TR * TQ); Chris@42: } Chris@42: } Chris@42: { Chris@42: E T28, T2e, T2c, T2g; Chris@42: { Chris@42: E T26, T27, T2a, T2b; Chris@42: T26 = T1M - T1N; Chris@42: T27 = T1X + T1Y; Chris@42: T28 = T26 - T27; Chris@42: T2e = T26 + T27; Chris@42: T2a = T1U + T1V; Chris@42: T2b = T1P - T1Q; Chris@42: T2c = T2a + T2b; Chris@42: T2g = T2a - T2b; Chris@42: } Chris@42: { Chris@42: E T25, T29, T2d, T2f; Chris@42: T25 = W[8]; Chris@42: T29 = W[9]; Chris@42: Ip[WS(rs, 2)] = FNMS(T29, T2c, T25 * T28); Chris@42: Im[WS(rs, 2)] = FMA(T25, T2c, T29 * T28); Chris@42: T2d = W[20]; Chris@42: T2f = W[21]; Chris@42: Ip[WS(rs, 5)] = FNMS(T2f, T2g, T2d * T2e); Chris@42: Im[WS(rs, 5)] = FMA(T2d, T2g, T2f * T2e); Chris@42: } Chris@42: } Chris@42: { Chris@42: E T1S, T22, T20, T24; Chris@42: { Chris@42: E T1O, T1R, T1W, T1Z; Chris@42: T1O = T1M + T1N; Chris@42: T1R = T1P + T1Q; Chris@42: T1S = T1O - T1R; Chris@42: T22 = T1O + T1R; Chris@42: T1W = T1U - T1V; Chris@42: T1Z = T1X - T1Y; Chris@42: T20 = T1W - T1Z; Chris@42: T24 = T1W + T1Z; Chris@42: } Chris@42: { Chris@42: E T1L, T1T, T21, T23; Chris@42: T1L = W[2]; Chris@42: T1T = W[3]; Chris@42: Rp[WS(rs, 1)] = FNMS(T1T, T20, T1L * T1S); Chris@42: Rm[WS(rs, 1)] = FMA(T1T, T1S, T1L * T20); Chris@42: T21 = W[14]; Chris@42: T23 = W[15]; Chris@42: Rp[WS(rs, 4)] = FNMS(T23, T24, T21 * T22); Chris@42: Rm[WS(rs, 4)] = FMA(T23, T22, T21 * T24); Chris@42: } Chris@42: } Chris@42: { Chris@42: E T1C, T1I, T1G, T1K; Chris@42: { Chris@42: E T1A, T1B, T1E, T1F; Chris@42: T1A = T12 + T15; Chris@42: T1B = T1p + T1s; Chris@42: T1C = T1A - T1B; Chris@42: T1I = T1A + T1B; Chris@42: T1E = T1i + T1l; Chris@42: T1F = T19 + T1c; Chris@42: T1G = T1E - T1F; Chris@42: T1K = T1E + T1F; Chris@42: } Chris@42: { Chris@42: E T1z, T1D, T1H, T1J; Chris@42: T1z = W[18]; Chris@42: T1D = W[19]; Chris@42: Rp[WS(rs, 5)] = FNMS(T1D, T1G, T1z * T1C); Chris@42: Rm[WS(rs, 5)] = FMA(T1D, T1C, T1z * T1G); Chris@42: T1H = W[6]; Chris@42: T1J = W[7]; Chris@42: Rp[WS(rs, 2)] = FNMS(T1J, T1K, T1H * T1I); Chris@42: Rm[WS(rs, 2)] = FMA(T1J, T1I, T1H * T1K); Chris@42: } Chris@42: } Chris@42: { Chris@42: E T1e, T1w, T1u, T1y; Chris@42: { Chris@42: E T16, T1d, T1m, T1t; Chris@42: T16 = T12 - T15; Chris@42: T1d = T19 - T1c; Chris@42: T1e = T16 - T1d; Chris@42: T1w = T16 + T1d; Chris@42: T1m = T1i - T1l; Chris@42: T1t = T1p - T1s; Chris@42: T1u = T1m + T1t; Chris@42: T1y = T1m - T1t; Chris@42: } Chris@42: { Chris@42: E TZ, T1f, T1v, T1x; Chris@42: TZ = W[0]; Chris@42: T1f = W[1]; Chris@42: Ip[0] = FNMS(T1f, T1u, TZ * T1e); Chris@42: Im[0] = FMA(TZ, T1u, T1f * T1e); Chris@42: T1v = W[12]; Chris@42: T1x = W[13]; Chris@42: Ip[WS(rs, 3)] = FNMS(T1x, T1y, T1v * T1w); Chris@42: Im[WS(rs, 3)] = FMA(T1v, T1y, T1x * T1w); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: {TW_FULL, 1, 12}, Chris@42: {TW_NEXT, 1, 0} Chris@42: }; Chris@42: Chris@42: static const hc2c_desc desc = { 12, "hc2cb_12", twinstr, &GENUS, {88, 30, 30, 0} }; Chris@42: Chris@42: void X(codelet_hc2cb_12) (planner *p) { Chris@42: X(khc2c_register) (p, hc2cb_12, &desc, HC2C_VIA_RDFT); Chris@42: } Chris@42: #endif /* HAVE_FMA */