cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:41:52 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 46 FP additions, 32 FP multiplications, cannam@95: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), cannam@95: * 45 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E TK, TR, TB, TM, TL, TS; cannam@95: { cannam@95: E Td, TN, TO, TJ, Tn, Tk, TC, T3, Tr, T7, T8, T4, T5; cannam@95: { cannam@95: E TI, Tj, Tg, TH, Te, Tf, T1, T2; cannam@95: { cannam@95: E Tb, Tc, Th, Ti; cannam@95: Tb = Ip[0]; cannam@95: Tc = Im[WS(rs, 2)]; cannam@95: Th = Ip[WS(rs, 1)]; cannam@95: Ti = Im[WS(rs, 1)]; cannam@95: Te = Ip[WS(rs, 2)]; cannam@95: Td = Tb - Tc; cannam@95: TN = Tb + Tc; cannam@95: Tf = Im[0]; cannam@95: TI = Th + Ti; cannam@95: Tj = Th - Ti; cannam@95: } cannam@95: Tg = Te - Tf; cannam@95: TH = Te + Tf; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 2)]; cannam@95: TO = TH - TI; cannam@95: TJ = TH + TI; cannam@95: Tn = Tj - Tg; cannam@95: Tk = Tg + Tj; cannam@95: TC = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: Tr = FNMS(KP500000000, Tk, Td); cannam@95: T7 = Rm[WS(rs, 1)]; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[0]; cannam@95: } cannam@95: { cannam@95: E Tl, Tq, TQ, Ts, Ta, T10, TG; cannam@95: Rm[0] = Td + Tk; cannam@95: { cannam@95: E T9, TE, T6, TD, TF; cannam@95: T9 = T7 + T8; cannam@95: TE = T7 - T8; cannam@95: T6 = T4 + T5; cannam@95: TD = T4 - T5; cannam@95: Tl = W[2]; cannam@95: Tq = W[3]; cannam@95: TQ = TD - TE; cannam@95: TF = TD + TE; cannam@95: Ts = T6 - T9; cannam@95: Ta = T6 + T9; cannam@95: T10 = TC + TF; cannam@95: TG = FNMS(KP500000000, TF, TC); cannam@95: } cannam@95: { cannam@95: E T13, TP, Tz, TZ, Tw, T14, Tv, Ty; cannam@95: { cannam@95: E Tt, T12, T11, Tp, Tm, To, Tu; cannam@95: T13 = TN + TO; cannam@95: TP = FNMS(KP500000000, TO, TN); cannam@95: Rp[0] = T3 + Ta; cannam@95: Tm = FNMS(KP500000000, Ta, T3); cannam@95: Tz = FMA(KP866025403, Ts, Tr); cannam@95: Tt = FNMS(KP866025403, Ts, Tr); cannam@95: TZ = W[4]; cannam@95: To = FNMS(KP866025403, Tn, Tm); cannam@95: Tw = FMA(KP866025403, Tn, Tm); cannam@95: Tu = Tl * Tt; cannam@95: T12 = W[5]; cannam@95: T11 = TZ * T10; cannam@95: Tp = Tl * To; cannam@95: Rm[WS(rs, 1)] = FMA(Tq, To, Tu); cannam@95: T14 = T12 * T10; cannam@95: Ip[WS(rs, 1)] = FNMS(T12, T13, T11); cannam@95: Rp[WS(rs, 1)] = FNMS(Tq, Tt, Tp); cannam@95: } cannam@95: Im[WS(rs, 1)] = FMA(TZ, T13, T14); cannam@95: Tv = W[6]; cannam@95: Ty = W[7]; cannam@95: { cannam@95: E TX, TT, TW, TV, TY, TU, TA, Tx; cannam@95: TK = FNMS(KP866025403, TJ, TG); cannam@95: TU = FMA(KP866025403, TJ, TG); cannam@95: TA = Tv * Tz; cannam@95: Tx = Tv * Tw; cannam@95: TX = FNMS(KP866025403, TQ, TP); cannam@95: TR = FMA(KP866025403, TQ, TP); cannam@95: Rm[WS(rs, 2)] = FMA(Ty, Tw, TA); cannam@95: Rp[WS(rs, 2)] = FNMS(Ty, Tz, Tx); cannam@95: TT = W[8]; cannam@95: TW = W[9]; cannam@95: TB = W[0]; cannam@95: TV = TT * TU; cannam@95: TY = TW * TU; cannam@95: TM = W[1]; cannam@95: TL = TB * TK; cannam@95: Ip[WS(rs, 2)] = FNMS(TW, TX, TV); cannam@95: Im[WS(rs, 2)] = FMA(TT, TX, TY); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: Ip[0] = FNMS(TM, TR, TL); cannam@95: TS = TM * TK; cannam@95: Im[0] = FMA(TB, TR, TS); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 6}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {24, 10, 22, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cb_6 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 46 FP additions, 28 FP multiplications, cannam@95: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), cannam@95: * 25 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T3, Ty, Td, TE, Ta, TO, Tr, TB, Tk, TL, Tn, TH; cannam@95: { cannam@95: E T1, T2, Tb, Tc; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 2)]; cannam@95: T3 = T1 + T2; cannam@95: Ty = T1 - T2; cannam@95: Tb = Ip[0]; cannam@95: Tc = Im[WS(rs, 2)]; cannam@95: Td = Tb - Tc; cannam@95: TE = Tb + Tc; cannam@95: } cannam@95: { cannam@95: E T6, Tz, T9, TA; cannam@95: { cannam@95: E T4, T5, T7, T8; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[0]; cannam@95: T6 = T4 + T5; cannam@95: Tz = T4 - T5; cannam@95: T7 = Rm[WS(rs, 1)]; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: T9 = T7 + T8; cannam@95: TA = T7 - T8; cannam@95: } cannam@95: Ta = T6 + T9; cannam@95: TO = KP866025403 * (Tz - TA); cannam@95: Tr = KP866025403 * (T6 - T9); cannam@95: TB = Tz + TA; cannam@95: } cannam@95: { cannam@95: E Tg, TG, Tj, TF; cannam@95: { cannam@95: E Te, Tf, Th, Ti; cannam@95: Te = Ip[WS(rs, 2)]; cannam@95: Tf = Im[0]; cannam@95: Tg = Te - Tf; cannam@95: TG = Te + Tf; cannam@95: Th = Ip[WS(rs, 1)]; cannam@95: Ti = Im[WS(rs, 1)]; cannam@95: Tj = Th - Ti; cannam@95: TF = Th + Ti; cannam@95: } cannam@95: Tk = Tg + Tj; cannam@95: TL = KP866025403 * (TG + TF); cannam@95: Tn = KP866025403 * (Tj - Tg); cannam@95: TH = TF - TG; cannam@95: } cannam@95: Rp[0] = T3 + Ta; cannam@95: Rm[0] = Td + Tk; cannam@95: { cannam@95: E TC, TI, Tx, TD; cannam@95: TC = Ty + TB; cannam@95: TI = TE - TH; cannam@95: Tx = W[4]; cannam@95: TD = W[5]; cannam@95: Ip[WS(rs, 1)] = FNMS(TD, TI, Tx * TC); cannam@95: Im[WS(rs, 1)] = FMA(TD, TC, Tx * TI); cannam@95: } cannam@95: { cannam@95: E To, Tu, Ts, Tw, Tm, Tq; cannam@95: Tm = FNMS(KP500000000, Ta, T3); cannam@95: To = Tm - Tn; cannam@95: Tu = Tm + Tn; cannam@95: Tq = FNMS(KP500000000, Tk, Td); cannam@95: Ts = Tq - Tr; cannam@95: Tw = Tr + Tq; cannam@95: { cannam@95: E Tl, Tp, Tt, Tv; cannam@95: Tl = W[2]; cannam@95: Tp = W[3]; cannam@95: Rp[WS(rs, 1)] = FNMS(Tp, Ts, Tl * To); cannam@95: Rm[WS(rs, 1)] = FMA(Tl, Ts, Tp * To); cannam@95: Tt = W[6]; cannam@95: Tv = W[7]; cannam@95: Rp[WS(rs, 2)] = FNMS(Tv, Tw, Tt * Tu); cannam@95: Rm[WS(rs, 2)] = FMA(Tt, Tw, Tv * Tu); cannam@95: } cannam@95: } cannam@95: { cannam@95: E TM, TS, TQ, TU, TK, TP; cannam@95: TK = FNMS(KP500000000, TB, Ty); cannam@95: TM = TK - TL; cannam@95: TS = TK + TL; cannam@95: TP = FMA(KP500000000, TH, TE); cannam@95: TQ = TO + TP; cannam@95: TU = TP - TO; cannam@95: { cannam@95: E TJ, TN, TR, TT; cannam@95: TJ = W[0]; cannam@95: TN = W[1]; cannam@95: Ip[0] = FNMS(TN, TQ, TJ * TM); cannam@95: Im[0] = FMA(TN, TM, TJ * TQ); cannam@95: TR = W[8]; cannam@95: TT = W[9]; cannam@95: Ip[WS(rs, 2)] = FNMS(TT, TU, TR * TS); cannam@95: Im[WS(rs, 2)] = FMA(TT, TS, TR * TU); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 6}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {32, 14, 14, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */