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 07:41:52 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_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cb_6 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 32 FP multiplications, Chris@10: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), Chris@10: * 45 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cb_6(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) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: E TK, TR, TB, TM, TL, TS; Chris@10: { Chris@10: E Td, TN, TO, TJ, Tn, Tk, TC, T3, Tr, T7, T8, T4, T5; Chris@10: { Chris@10: E TI, Tj, Tg, TH, Te, Tf, T1, T2; Chris@10: { Chris@10: E Tb, Tc, Th, Ti; Chris@10: Tb = Ip[0]; Chris@10: Tc = Im[WS(rs, 2)]; Chris@10: Th = Ip[WS(rs, 1)]; Chris@10: Ti = Im[WS(rs, 1)]; Chris@10: Te = Ip[WS(rs, 2)]; Chris@10: Td = Tb - Tc; Chris@10: TN = Tb + Tc; Chris@10: Tf = Im[0]; Chris@10: TI = Th + Ti; Chris@10: Tj = Th - Ti; Chris@10: } Chris@10: Tg = Te - Tf; Chris@10: TH = Te + Tf; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 2)]; Chris@10: TO = TH - TI; Chris@10: TJ = TH + TI; Chris@10: Tn = Tj - Tg; Chris@10: Tk = Tg + Tj; Chris@10: TC = T1 - T2; Chris@10: T3 = T1 + T2; Chris@10: Tr = FNMS(KP500000000, Tk, Td); Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = Rp[WS(rs, 1)]; Chris@10: T4 = Rp[WS(rs, 2)]; Chris@10: T5 = Rm[0]; Chris@10: } Chris@10: { Chris@10: E Tl, Tq, TQ, Ts, Ta, T10, TG; Chris@10: Rm[0] = Td + Tk; Chris@10: { Chris@10: E T9, TE, T6, TD, TF; Chris@10: T9 = T7 + T8; Chris@10: TE = T7 - T8; Chris@10: T6 = T4 + T5; Chris@10: TD = T4 - T5; Chris@10: Tl = W[2]; Chris@10: Tq = W[3]; Chris@10: TQ = TD - TE; Chris@10: TF = TD + TE; Chris@10: Ts = T6 - T9; Chris@10: Ta = T6 + T9; Chris@10: T10 = TC + TF; Chris@10: TG = FNMS(KP500000000, TF, TC); Chris@10: } Chris@10: { Chris@10: E T13, TP, Tz, TZ, Tw, T14, Tv, Ty; Chris@10: { Chris@10: E Tt, T12, T11, Tp, Tm, To, Tu; Chris@10: T13 = TN + TO; Chris@10: TP = FNMS(KP500000000, TO, TN); Chris@10: Rp[0] = T3 + Ta; Chris@10: Tm = FNMS(KP500000000, Ta, T3); Chris@10: Tz = FMA(KP866025403, Ts, Tr); Chris@10: Tt = FNMS(KP866025403, Ts, Tr); Chris@10: TZ = W[4]; Chris@10: To = FNMS(KP866025403, Tn, Tm); Chris@10: Tw = FMA(KP866025403, Tn, Tm); Chris@10: Tu = Tl * Tt; Chris@10: T12 = W[5]; Chris@10: T11 = TZ * T10; Chris@10: Tp = Tl * To; Chris@10: Rm[WS(rs, 1)] = FMA(Tq, To, Tu); Chris@10: T14 = T12 * T10; Chris@10: Ip[WS(rs, 1)] = FNMS(T12, T13, T11); Chris@10: Rp[WS(rs, 1)] = FNMS(Tq, Tt, Tp); Chris@10: } Chris@10: Im[WS(rs, 1)] = FMA(TZ, T13, T14); Chris@10: Tv = W[6]; Chris@10: Ty = W[7]; Chris@10: { Chris@10: E TX, TT, TW, TV, TY, TU, TA, Tx; Chris@10: TK = FNMS(KP866025403, TJ, TG); Chris@10: TU = FMA(KP866025403, TJ, TG); Chris@10: TA = Tv * Tz; Chris@10: Tx = Tv * Tw; Chris@10: TX = FNMS(KP866025403, TQ, TP); Chris@10: TR = FMA(KP866025403, TQ, TP); Chris@10: Rm[WS(rs, 2)] = FMA(Ty, Tw, TA); Chris@10: Rp[WS(rs, 2)] = FNMS(Ty, Tz, Tx); Chris@10: TT = W[8]; Chris@10: TW = W[9]; Chris@10: TB = W[0]; Chris@10: TV = TT * TU; Chris@10: TY = TW * TU; Chris@10: TM = W[1]; Chris@10: TL = TB * TK; Chris@10: Ip[WS(rs, 2)] = FNMS(TW, TX, TV); Chris@10: Im[WS(rs, 2)] = FMA(TT, TX, TY); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Ip[0] = FNMS(TM, TR, TL); Chris@10: TS = TM * TK; Chris@10: Im[0] = FMA(TB, TR, TS); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {24, 10, 22, 0} }; Chris@10: Chris@10: void X(codelet_hc2cb_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cb_6 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 28 FP multiplications, Chris@10: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), Chris@10: * 25 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cb_6(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) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: E T3, Ty, Td, TE, Ta, TO, Tr, TB, Tk, TL, Tn, TH; Chris@10: { Chris@10: E T1, T2, Tb, Tc; Chris@10: T1 = Rp[0]; Chris@10: T2 = Rm[WS(rs, 2)]; Chris@10: T3 = T1 + T2; Chris@10: Ty = T1 - T2; Chris@10: Tb = Ip[0]; Chris@10: Tc = Im[WS(rs, 2)]; Chris@10: Td = Tb - Tc; Chris@10: TE = Tb + Tc; Chris@10: } Chris@10: { Chris@10: E T6, Tz, T9, TA; Chris@10: { Chris@10: E T4, T5, T7, T8; Chris@10: T4 = Rp[WS(rs, 2)]; Chris@10: T5 = Rm[0]; Chris@10: T6 = T4 + T5; Chris@10: Tz = T4 - T5; Chris@10: T7 = Rm[WS(rs, 1)]; Chris@10: T8 = Rp[WS(rs, 1)]; Chris@10: T9 = T7 + T8; Chris@10: TA = T7 - T8; Chris@10: } Chris@10: Ta = T6 + T9; Chris@10: TO = KP866025403 * (Tz - TA); Chris@10: Tr = KP866025403 * (T6 - T9); Chris@10: TB = Tz + TA; Chris@10: } Chris@10: { Chris@10: E Tg, TG, Tj, TF; Chris@10: { Chris@10: E Te, Tf, Th, Ti; Chris@10: Te = Ip[WS(rs, 2)]; Chris@10: Tf = Im[0]; Chris@10: Tg = Te - Tf; Chris@10: TG = Te + Tf; Chris@10: Th = Ip[WS(rs, 1)]; Chris@10: Ti = Im[WS(rs, 1)]; Chris@10: Tj = Th - Ti; Chris@10: TF = Th + Ti; Chris@10: } Chris@10: Tk = Tg + Tj; Chris@10: TL = KP866025403 * (TG + TF); Chris@10: Tn = KP866025403 * (Tj - Tg); Chris@10: TH = TF - TG; Chris@10: } Chris@10: Rp[0] = T3 + Ta; Chris@10: Rm[0] = Td + Tk; Chris@10: { Chris@10: E TC, TI, Tx, TD; Chris@10: TC = Ty + TB; Chris@10: TI = TE - TH; Chris@10: Tx = W[4]; Chris@10: TD = W[5]; Chris@10: Ip[WS(rs, 1)] = FNMS(TD, TI, Tx * TC); Chris@10: Im[WS(rs, 1)] = FMA(TD, TC, Tx * TI); Chris@10: } Chris@10: { Chris@10: E To, Tu, Ts, Tw, Tm, Tq; Chris@10: Tm = FNMS(KP500000000, Ta, T3); Chris@10: To = Tm - Tn; Chris@10: Tu = Tm + Tn; Chris@10: Tq = FNMS(KP500000000, Tk, Td); Chris@10: Ts = Tq - Tr; Chris@10: Tw = Tr + Tq; Chris@10: { Chris@10: E Tl, Tp, Tt, Tv; Chris@10: Tl = W[2]; Chris@10: Tp = W[3]; Chris@10: Rp[WS(rs, 1)] = FNMS(Tp, Ts, Tl * To); Chris@10: Rm[WS(rs, 1)] = FMA(Tl, Ts, Tp * To); Chris@10: Tt = W[6]; Chris@10: Tv = W[7]; Chris@10: Rp[WS(rs, 2)] = FNMS(Tv, Tw, Tt * Tu); Chris@10: Rm[WS(rs, 2)] = FMA(Tt, Tw, Tv * Tu); Chris@10: } Chris@10: } Chris@10: { Chris@10: E TM, TS, TQ, TU, TK, TP; Chris@10: TK = FNMS(KP500000000, TB, Ty); Chris@10: TM = TK - TL; Chris@10: TS = TK + TL; Chris@10: TP = FMA(KP500000000, TH, TE); Chris@10: TQ = TO + TP; Chris@10: TU = TP - TO; Chris@10: { Chris@10: E TJ, TN, TR, TT; Chris@10: TJ = W[0]; Chris@10: TN = W[1]; Chris@10: Ip[0] = FNMS(TN, TQ, TJ * TM); Chris@10: Im[0] = FMA(TN, TM, TJ * TQ); Chris@10: TR = W[8]; Chris@10: TT = W[9]; Chris@10: Ip[WS(rs, 2)] = FNMS(TT, TU, TR * TS); Chris@10: Im[WS(rs, 2)] = FMA(TT, TS, TR * TU); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {32, 14, 14, 0} }; Chris@10: Chris@10: void X(codelet_hc2cb_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */