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 07:40:44 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_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cfdft_6 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 44 FP multiplications, cannam@95: * (or, 36 additions, 22 multiplications, 22 fused multiply/add), cannam@95: * 42 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft_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 TP, TT, TN, TM, TY, T13; cannam@95: { cannam@95: E T3, TQ, TJ, T12, Tu, TB, TX, T10, Tj, Tf, Ti, Td, Th, TU, TS; cannam@95: { cannam@95: E TC, TI, TF, TH, TA, Tw, TZ; cannam@95: { cannam@95: E T1, T2, TD, TE; cannam@95: T1 = Ip[0]; cannam@95: T2 = Im[0]; cannam@95: TD = Rm[0]; cannam@95: TE = Rp[0]; cannam@95: TC = W[0]; cannam@95: T3 = T1 - T2; cannam@95: TI = T1 + T2; cannam@95: TQ = TE + TD; cannam@95: TF = TD - TE; cannam@95: TH = W[1]; cannam@95: } cannam@95: { cannam@95: E Tr, To, Ts, Tl, Tq; cannam@95: { cannam@95: E Tm, Tn, TG, T11; cannam@95: Tm = Rm[WS(rs, 2)]; cannam@95: Tn = Rp[WS(rs, 2)]; cannam@95: TG = TC * TF; cannam@95: T11 = TH * TF; cannam@95: Tr = Ip[WS(rs, 2)]; cannam@95: TA = Tn + Tm; cannam@95: To = Tm - Tn; cannam@95: TJ = FNMS(TH, TI, TG); cannam@95: T12 = FMA(TC, TI, T11); cannam@95: Ts = Im[WS(rs, 2)]; cannam@95: } cannam@95: Tl = W[8]; cannam@95: Tq = W[9]; cannam@95: { cannam@95: E Tz, Ty, TW, Tx, Tt, Tp; cannam@95: Tw = W[6]; cannam@95: Tx = Tr - Ts; cannam@95: Tt = Tr + Ts; cannam@95: Tp = Tl * To; cannam@95: Tz = W[7]; cannam@95: Ty = Tw * Tx; cannam@95: TW = Tl * Tt; cannam@95: Tu = FNMS(Tq, Tt, Tp); cannam@95: TZ = Tz * Tx; cannam@95: TB = FNMS(Tz, TA, Ty); cannam@95: TX = FMA(Tq, To, TW); cannam@95: } cannam@95: } cannam@95: { cannam@95: E T5, T6, Ta, Tb; cannam@95: T5 = Ip[WS(rs, 1)]; cannam@95: T10 = FMA(Tw, TA, TZ); cannam@95: T6 = Im[WS(rs, 1)]; cannam@95: Ta = Rp[WS(rs, 1)]; cannam@95: Tb = Rm[WS(rs, 1)]; cannam@95: { cannam@95: E T4, Tg, T7, Tc, T9, T8, TR; cannam@95: T4 = W[5]; cannam@95: Tg = T5 - T6; cannam@95: T7 = T5 + T6; cannam@95: Tj = Ta + Tb; cannam@95: Tc = Ta - Tb; cannam@95: T9 = W[4]; cannam@95: T8 = T4 * T7; cannam@95: Tf = W[2]; cannam@95: Ti = W[3]; cannam@95: TR = T9 * T7; cannam@95: Td = FMA(T9, Tc, T8); cannam@95: Th = Tf * Tg; cannam@95: TU = Ti * Tg; cannam@95: TS = FNMS(T4, Tc, TR); cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E Te, T1d, TK, Tv, T1a, T1b, Tk, TV; cannam@95: TP = Td + T3; cannam@95: Te = T3 - Td; cannam@95: Tk = FNMS(Ti, Tj, Th); cannam@95: TV = FMA(Tf, Tj, TU); cannam@95: T1d = TQ + TS; cannam@95: TT = TQ - TS; cannam@95: TN = TJ - TB; cannam@95: TK = TB + TJ; cannam@95: Tv = Tk + Tu; cannam@95: TM = Tu - Tk; cannam@95: TY = TV - TX; cannam@95: T1a = TV + TX; cannam@95: T1b = T10 + T12; cannam@95: T13 = T10 - T12; cannam@95: { cannam@95: E T1g, TL, T1e, T1c, T19, T1f; cannam@95: T1g = Tv - TK; cannam@95: TL = Tv + TK; cannam@95: T1e = T1a + T1b; cannam@95: T1c = T1a - T1b; cannam@95: T19 = FNMS(KP500000000, TL, Te); cannam@95: Ip[0] = KP500000000 * (Te + TL); cannam@95: T1f = FNMS(KP500000000, T1e, T1d); cannam@95: Rp[0] = KP500000000 * (T1d + T1e); cannam@95: Im[WS(rs, 1)] = -(KP500000000 * (FNMS(KP866025403, T1c, T19))); cannam@95: Ip[WS(rs, 2)] = KP500000000 * (FMA(KP866025403, T1c, T19)); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T1g, T1f)); cannam@95: Rp[WS(rs, 2)] = KP500000000 * (FNMS(KP866025403, T1g, T1f)); cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TO, T16, T14, T18, T17, T15; cannam@95: TO = TM + TN; cannam@95: T16 = TN - TM; cannam@95: T14 = TY + T13; cannam@95: T18 = T13 - TY; cannam@95: T17 = FMA(KP500000000, TO, TP); cannam@95: Im[WS(rs, 2)] = KP500000000 * (TO - TP); cannam@95: T15 = FNMS(KP500000000, T14, TT); cannam@95: Rm[WS(rs, 2)] = KP500000000 * (TT + T14); cannam@95: Im[0] = -(KP500000000 * (FNMS(KP866025403, T18, T17))); cannam@95: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T18, T17)); cannam@95: Rm[0] = KP500000000 * (FNMS(KP866025403, T16, T15)); cannam@95: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T16, T15)); 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, "hc2cfdft_6", twinstr, &GENUS, {36, 22, 22, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft_6, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cfdft_6 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 36 FP multiplications, cannam@95: * (or, 44 additions, 22 multiplications, 14 fused multiply/add), cannam@95: * 40 stack variables, 3 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft_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(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DK(KP433012701, +0.433012701892219323381861585376468091735701313); 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, TM, Tc, TN, Ts, T10, TI, TR, TF, T11, TH, TU; cannam@95: { cannam@95: E T1, T2, TD, Tz, TA, TB, T7, Tf, Tb, Th, Tq, Tw, Tm, Tu, T4; cannam@95: E T8; cannam@95: { cannam@95: E T5, T6, T9, Ta; cannam@95: T1 = Ip[0]; cannam@95: T2 = Im[0]; cannam@95: TD = T1 + T2; cannam@95: Tz = Rm[0]; cannam@95: TA = Rp[0]; cannam@95: TB = Tz - TA; cannam@95: T5 = Ip[WS(rs, 1)]; cannam@95: T6 = Im[WS(rs, 1)]; cannam@95: T7 = T5 + T6; cannam@95: Tf = T5 - T6; cannam@95: T9 = Rp[WS(rs, 1)]; cannam@95: Ta = Rm[WS(rs, 1)]; cannam@95: Tb = T9 - Ta; cannam@95: Th = T9 + Ta; cannam@95: { cannam@95: E To, Tp, Tk, Tl; cannam@95: To = Rp[WS(rs, 2)]; cannam@95: Tp = Rm[WS(rs, 2)]; cannam@95: Tq = To - Tp; cannam@95: Tw = To + Tp; cannam@95: Tk = Ip[WS(rs, 2)]; cannam@95: Tl = Im[WS(rs, 2)]; cannam@95: Tm = Tk + Tl; cannam@95: Tu = Tk - Tl; cannam@95: } cannam@95: } cannam@95: T3 = T1 - T2; cannam@95: TM = TA + Tz; cannam@95: T4 = W[5]; cannam@95: T8 = W[4]; cannam@95: Tc = FMA(T4, T7, T8 * Tb); cannam@95: TN = FNMS(T4, Tb, T8 * T7); cannam@95: { cannam@95: E Ti, TP, Tr, TQ; cannam@95: { cannam@95: E Te, Tg, Tj, Tn; cannam@95: Te = W[2]; cannam@95: Tg = W[3]; cannam@95: Ti = FNMS(Tg, Th, Te * Tf); cannam@95: TP = FMA(Tg, Tf, Te * Th); cannam@95: Tj = W[9]; cannam@95: Tn = W[8]; cannam@95: Tr = FMA(Tj, Tm, Tn * Tq); cannam@95: TQ = FNMS(Tj, Tq, Tn * Tm); cannam@95: } cannam@95: Ts = Ti - Tr; cannam@95: T10 = TP + TQ; cannam@95: TI = Ti + Tr; cannam@95: TR = TP - TQ; cannam@95: } cannam@95: { cannam@95: E Tx, TS, TE, TT; cannam@95: { cannam@95: E Tt, Tv, Ty, TC; cannam@95: Tt = W[6]; cannam@95: Tv = W[7]; cannam@95: Tx = FNMS(Tv, Tw, Tt * Tu); cannam@95: TS = FMA(Tv, Tu, Tt * Tw); cannam@95: Ty = W[0]; cannam@95: TC = W[1]; cannam@95: TE = FNMS(TC, TD, Ty * TB); cannam@95: TT = FMA(TC, TB, Ty * TD); cannam@95: } cannam@95: TF = Tx + TE; cannam@95: T11 = TS + TT; cannam@95: TH = TE - Tx; cannam@95: TU = TS - TT; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T12, Td, TG, TZ; cannam@95: T12 = KP433012701 * (T10 - T11); cannam@95: Td = T3 - Tc; cannam@95: TG = Ts + TF; cannam@95: TZ = FNMS(KP250000000, TG, KP500000000 * Td); cannam@95: Ip[0] = KP500000000 * (Td + TG); cannam@95: Im[WS(rs, 1)] = T12 - TZ; cannam@95: Ip[WS(rs, 2)] = TZ + T12; cannam@95: } cannam@95: { cannam@95: E T16, T13, T14, T15; cannam@95: T16 = KP433012701 * (Ts - TF); cannam@95: T13 = TM + TN; cannam@95: T14 = T10 + T11; cannam@95: T15 = FNMS(KP250000000, T14, KP500000000 * T13); cannam@95: Rp[WS(rs, 2)] = T15 - T16; cannam@95: Rp[0] = KP500000000 * (T13 + T14); cannam@95: Rm[WS(rs, 1)] = T16 + T15; cannam@95: } cannam@95: { cannam@95: E TY, TJ, TK, TX; cannam@95: TY = KP433012701 * (TU - TR); cannam@95: TJ = TH - TI; cannam@95: TK = Tc + T3; cannam@95: TX = FMA(KP500000000, TK, KP250000000 * TJ); cannam@95: Im[WS(rs, 2)] = KP500000000 * (TJ - TK); cannam@95: Im[0] = TY - TX; cannam@95: Ip[WS(rs, 1)] = TX + TY; cannam@95: } cannam@95: { cannam@95: E TL, TO, TV, TW; cannam@95: TL = KP433012701 * (TI + TH); cannam@95: TO = TM - TN; cannam@95: TV = TR + TU; cannam@95: TW = FNMS(KP250000000, TV, KP500000000 * TO); cannam@95: Rp[WS(rs, 1)] = TL + TW; cannam@95: Rm[WS(rs, 2)] = KP500000000 * (TO + TV); cannam@95: Rm[0] = TW - TL; 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, "hc2cfdft_6", twinstr, &GENUS, {44, 22, 14, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft_6) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft_6, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */