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 07:41:12 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_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 7 -dif -name hb_7 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 72 FP additions, 66 FP multiplications, Chris@10: * (or, 18 additions, 12 multiplications, 54 fused multiply/add), Chris@10: * 67 stack variables, 6 constants, and 28 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_7(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP974927912, +0.974927912181823607018131682993931217232785801); Chris@10: DK(KP900968867, +0.900968867902419126236102319507445051165919162); Chris@10: DK(KP801937735, +0.801937735804838252472204639014890102331838324); Chris@10: DK(KP692021471, +0.692021471630095869627814897002069140197260599); Chris@10: DK(KP356895867, +0.356895867892209443894399510021300583399127187); Chris@10: DK(KP554958132, +0.554958132087371191422194871006410481067288862); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 12); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 12, MAKE_VOLATILE_STRIDE(14, rs)) { Chris@10: E T1q, T1p, T1t, T1r, T1s, T1u; Chris@10: { Chris@10: E T1, T4, TC, T7, TB, Tt, TD, Ta, TA, T1l, TZ, T1b, Th, Tw, Td; Chris@10: E TP, Ti, Tj, Tl, Tm, T8, T9, T1a; Chris@10: T1 = cr[0]; Chris@10: { Chris@10: E T2, T3, T5, T6; Chris@10: T2 = cr[WS(rs, 1)]; Chris@10: T3 = ci[0]; Chris@10: T5 = cr[WS(rs, 2)]; Chris@10: T6 = ci[WS(rs, 1)]; Chris@10: T8 = cr[WS(rs, 3)]; Chris@10: T4 = T2 + T3; Chris@10: TC = T2 - T3; Chris@10: T7 = T5 + T6; Chris@10: TB = T5 - T6; Chris@10: T9 = ci[WS(rs, 2)]; Chris@10: } Chris@10: Tt = ci[WS(rs, 6)]; Chris@10: TD = FNMS(KP554958132, TC, TB); Chris@10: T1a = FNMS(KP356895867, T7, T4); Chris@10: Ta = T8 + T9; Chris@10: TA = T8 - T9; Chris@10: { Chris@10: E Tf, Tg, Tc, TO; Chris@10: Tf = ci[WS(rs, 3)]; Chris@10: Tg = cr[WS(rs, 4)]; Chris@10: T1l = FMA(KP554958132, TA, TC); Chris@10: TZ = FMA(KP554958132, TB, TA); Chris@10: Tc = FNMS(KP356895867, Ta, T7); Chris@10: TO = FNMS(KP356895867, T4, Ta); Chris@10: T1b = FNMS(KP692021471, T1a, Ta); Chris@10: Th = Tf + Tg; Chris@10: Tw = Tf - Tg; Chris@10: Td = FNMS(KP692021471, Tc, T4); Chris@10: TP = FNMS(KP692021471, TO, T7); Chris@10: } Chris@10: Ti = ci[WS(rs, 4)]; Chris@10: Tj = cr[WS(rs, 5)]; Chris@10: Tl = ci[WS(rs, 5)]; Chris@10: Tm = cr[WS(rs, 6)]; Chris@10: { Chris@10: E Ty, TS, TX, T1j, T1e, Tp, Tk, Tv; Chris@10: cr[0] = T1 + T4 + T7 + Ta; Chris@10: Tk = Ti + Tj; Chris@10: Tv = Ti - Tj; Chris@10: { Chris@10: E Tn, Tu, Tx, TR; Chris@10: Tn = Tl + Tm; Chris@10: Tu = Tl - Tm; Chris@10: Tx = FNMS(KP356895867, Tw, Tv); Chris@10: TR = FMA(KP554958132, Tk, Th); Chris@10: { Chris@10: E TW, T1i, T1d, To; Chris@10: TW = FNMS(KP356895867, Tu, Tw); Chris@10: T1i = FNMS(KP356895867, Tv, Tu); Chris@10: T1d = FMA(KP554958132, Th, Tn); Chris@10: To = FNMS(KP554958132, Tn, Tk); Chris@10: Ty = FNMS(KP692021471, Tx, Tu); Chris@10: TS = FNMS(KP801937735, TR, Tn); Chris@10: TX = FNMS(KP692021471, TW, Tv); Chris@10: T1j = FNMS(KP692021471, T1i, Tw); Chris@10: T1e = FMA(KP801937735, T1d, Tk); Chris@10: Tp = FNMS(KP801937735, To, Th); Chris@10: ci[0] = Tt + Tu + Tv + Tw; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TL, TH, TK, TJ, TM, Te, Tz, TE; Chris@10: Te = FNMS(KP900968867, Td, T1); Chris@10: Tz = FNMS(KP900968867, Ty, Tt); Chris@10: TE = FNMS(KP801937735, TD, TA); Chris@10: { Chris@10: E Tb, TI, Tq, TF, Ts, Tr, TG; Chris@10: Tb = W[4]; Chris@10: TI = FMA(KP974927912, Tp, Te); Chris@10: Tq = FNMS(KP974927912, Tp, Te); Chris@10: TL = FNMS(KP974927912, TE, Tz); Chris@10: TF = FMA(KP974927912, TE, Tz); Chris@10: Ts = W[5]; Chris@10: Tr = Tb * Tq; Chris@10: TH = W[6]; Chris@10: TK = W[7]; Chris@10: TG = Ts * Tq; Chris@10: cr[WS(rs, 3)] = FNMS(Ts, TF, Tr); Chris@10: TJ = TH * TI; Chris@10: TM = TK * TI; Chris@10: ci[WS(rs, 3)] = FMA(Tb, TF, TG); Chris@10: } Chris@10: { Chris@10: E T14, T13, T17, T15, T16; Chris@10: { Chris@10: E TY, TT, T10, TQ; Chris@10: TQ = FNMS(KP900968867, TP, T1); Chris@10: cr[WS(rs, 4)] = FNMS(TK, TL, TJ); Chris@10: ci[WS(rs, 4)] = FMA(TH, TL, TM); Chris@10: TY = FNMS(KP900968867, TX, Tt); Chris@10: TT = FNMS(KP974927912, TS, TQ); Chris@10: T14 = FMA(KP974927912, TS, TQ); Chris@10: T10 = FNMS(KP801937735, TZ, TC); Chris@10: { Chris@10: E TN, TV, T11, TU, T12; Chris@10: TN = W[2]; Chris@10: TV = W[3]; Chris@10: T13 = W[8]; Chris@10: T11 = FMA(KP974927912, T10, TY); Chris@10: T17 = FNMS(KP974927912, T10, TY); Chris@10: TU = TN * TT; Chris@10: T12 = TV * TT; Chris@10: T15 = T13 * T14; Chris@10: T16 = W[9]; Chris@10: cr[WS(rs, 2)] = FNMS(TV, T11, TU); Chris@10: ci[WS(rs, 2)] = FMA(TN, T11, T12); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1k, T1f, T1m, T1c, T18; Chris@10: T1c = FNMS(KP900968867, T1b, T1); Chris@10: cr[WS(rs, 5)] = FNMS(T16, T17, T15); Chris@10: T18 = T16 * T14; Chris@10: T1k = FNMS(KP900968867, T1j, Tt); Chris@10: T1f = FNMS(KP974927912, T1e, T1c); Chris@10: T1q = FMA(KP974927912, T1e, T1c); Chris@10: ci[WS(rs, 5)] = FMA(T13, T17, T18); Chris@10: T1m = FMA(KP801937735, T1l, TB); Chris@10: { Chris@10: E T19, T1h, T1n, T1g, T1o; Chris@10: T19 = W[0]; Chris@10: T1h = W[1]; Chris@10: T1p = W[10]; Chris@10: T1t = FNMS(KP974927912, T1m, T1k); Chris@10: T1n = FMA(KP974927912, T1m, T1k); Chris@10: T1g = T19 * T1f; Chris@10: T1o = T1h * T1f; Chris@10: T1r = T1p * T1q; Chris@10: T1s = W[11]; Chris@10: cr[WS(rs, 1)] = FNMS(T1h, T1n, T1g); Chris@10: ci[WS(rs, 1)] = FMA(T19, T1n, T1o); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: cr[WS(rs, 6)] = FNMS(T1s, T1t, T1r); Chris@10: T1u = T1s * T1q; Chris@10: ci[WS(rs, 6)] = FMA(T1p, T1t, T1u); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 7, "hb_7", twinstr, &GENUS, {18, 12, 54, 0} }; Chris@10: Chris@10: void X(codelet_hb_7) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_7, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 7 -dif -name hb_7 -include hb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 72 FP additions, 60 FP multiplications, Chris@10: * (or, 36 additions, 24 multiplications, 36 fused multiply/add), Chris@10: * 36 stack variables, 6 constants, and 28 memory accesses Chris@10: */ Chris@10: #include "hb.h" Chris@10: Chris@10: static void hb_7(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP222520933, +0.222520933956314404288902564496794759466355569); Chris@10: DK(KP900968867, +0.900968867902419126236102319507445051165919162); Chris@10: DK(KP623489801, +0.623489801858733530525004884004239810632274731); Chris@10: DK(KP781831482, +0.781831482468029808708444526674057750232334519); Chris@10: DK(KP974927912, +0.974927912181823607018131682993931217232785801); Chris@10: DK(KP433883739, +0.433883739117558120475768332848358754609990728); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 12); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 12, MAKE_VOLATILE_STRIDE(14, rs)) { Chris@10: E T1, T4, T7, Ta, Tx, TI, TV, TQ, TE, Tm, Tb, Te, Th, Tk, Tq; Chris@10: E TF, TR, TU, TJ, Tt; Chris@10: { Chris@10: E Tu, Tw, Tv, T2, T3; Chris@10: T1 = cr[0]; Chris@10: T2 = cr[WS(rs, 1)]; Chris@10: T3 = ci[0]; Chris@10: T4 = T2 + T3; Chris@10: Tu = T2 - T3; Chris@10: { Chris@10: E T5, T6, T8, T9; Chris@10: T5 = cr[WS(rs, 2)]; Chris@10: T6 = ci[WS(rs, 1)]; Chris@10: T7 = T5 + T6; Chris@10: Tw = T5 - T6; Chris@10: T8 = cr[WS(rs, 3)]; Chris@10: T9 = ci[WS(rs, 2)]; Chris@10: Ta = T8 + T9; Chris@10: Tv = T8 - T9; Chris@10: } Chris@10: Tx = FMA(KP433883739, Tu, KP974927912 * Tv) - (KP781831482 * Tw); Chris@10: TI = FMA(KP781831482, Tu, KP974927912 * Tw) + (KP433883739 * Tv); Chris@10: TV = FNMS(KP781831482, Tv, KP974927912 * Tu) - (KP433883739 * Tw); Chris@10: TQ = FMA(KP623489801, Ta, T1) + FNMA(KP900968867, T7, KP222520933 * T4); Chris@10: TE = FMA(KP623489801, T4, T1) + FNMA(KP900968867, Ta, KP222520933 * T7); Chris@10: Tm = FMA(KP623489801, T7, T1) + FNMA(KP222520933, Ta, KP900968867 * T4); Chris@10: } Chris@10: { Chris@10: E Tp, Tn, To, Tc, Td; Chris@10: Tb = ci[WS(rs, 6)]; Chris@10: Tc = ci[WS(rs, 5)]; Chris@10: Td = cr[WS(rs, 6)]; Chris@10: Te = Tc - Td; Chris@10: Tp = Tc + Td; Chris@10: { Chris@10: E Tf, Tg, Ti, Tj; Chris@10: Tf = ci[WS(rs, 4)]; Chris@10: Tg = cr[WS(rs, 5)]; Chris@10: Th = Tf - Tg; Chris@10: Tn = Tf + Tg; Chris@10: Ti = ci[WS(rs, 3)]; Chris@10: Tj = cr[WS(rs, 4)]; Chris@10: Tk = Ti - Tj; Chris@10: To = Ti + Tj; Chris@10: } Chris@10: Tq = FNMS(KP974927912, To, KP781831482 * Tn) - (KP433883739 * Tp); Chris@10: TF = FMA(KP781831482, Tp, KP974927912 * Tn) + (KP433883739 * To); Chris@10: TR = FMA(KP433883739, Tn, KP781831482 * To) - (KP974927912 * Tp); Chris@10: TU = FMA(KP623489801, Tk, Tb) + FNMA(KP900968867, Th, KP222520933 * Te); Chris@10: TJ = FMA(KP623489801, Te, Tb) + FNMA(KP900968867, Tk, KP222520933 * Th); Chris@10: Tt = FMA(KP623489801, Th, Tb) + FNMA(KP222520933, Tk, KP900968867 * Te); Chris@10: } Chris@10: cr[0] = T1 + T4 + T7 + Ta; Chris@10: ci[0] = Tb + Te + Th + Tk; Chris@10: { Chris@10: E Tr, Ty, Tl, Ts; Chris@10: Tr = Tm - Tq; Chris@10: Ty = Tt - Tx; Chris@10: Tl = W[6]; Chris@10: Ts = W[7]; Chris@10: cr[WS(rs, 4)] = FNMS(Ts, Ty, Tl * Tr); Chris@10: ci[WS(rs, 4)] = FMA(Tl, Ty, Ts * Tr); Chris@10: } Chris@10: { Chris@10: E TY, T10, TX, TZ; Chris@10: TY = TQ + TR; Chris@10: T10 = TV + TU; Chris@10: TX = W[2]; Chris@10: TZ = W[3]; Chris@10: cr[WS(rs, 2)] = FNMS(TZ, T10, TX * TY); Chris@10: ci[WS(rs, 2)] = FMA(TX, T10, TZ * TY); Chris@10: } Chris@10: { Chris@10: E TA, TC, Tz, TB; Chris@10: TA = Tm + Tq; Chris@10: TC = Tx + Tt; Chris@10: Tz = W[4]; Chris@10: TB = W[5]; Chris@10: cr[WS(rs, 3)] = FNMS(TB, TC, Tz * TA); Chris@10: ci[WS(rs, 3)] = FMA(Tz, TC, TB * TA); Chris@10: } Chris@10: { Chris@10: E TM, TO, TL, TN; Chris@10: TM = TE + TF; Chris@10: TO = TJ - TI; Chris@10: TL = W[10]; Chris@10: TN = W[11]; Chris@10: cr[WS(rs, 6)] = FNMS(TN, TO, TL * TM); Chris@10: ci[WS(rs, 6)] = FMA(TL, TO, TN * TM); Chris@10: } Chris@10: { Chris@10: E TS, TW, TP, TT; Chris@10: TS = TQ - TR; Chris@10: TW = TU - TV; Chris@10: TP = W[8]; Chris@10: TT = W[9]; Chris@10: cr[WS(rs, 5)] = FNMS(TT, TW, TP * TS); Chris@10: ci[WS(rs, 5)] = FMA(TP, TW, TT * TS); Chris@10: } Chris@10: { Chris@10: E TG, TK, TD, TH; Chris@10: TG = TE - TF; Chris@10: TK = TI + TJ; Chris@10: TD = W[0]; Chris@10: TH = W[1]; Chris@10: cr[WS(rs, 1)] = FNMS(TH, TK, TD * TG); Chris@10: ci[WS(rs, 1)] = FMA(TD, TK, TH * TG); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 7, "hb_7", twinstr, &GENUS, {36, 24, 36, 0} }; Chris@10: Chris@10: void X(codelet_hb_7) (planner *p) { Chris@10: X(khc2hc_register) (p, hb_7, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */