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 07:40:44 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_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cfdft_6 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 58 FP additions, 44 FP multiplications, Chris@10: * (or, 36 additions, 22 multiplications, 22 fused multiply/add), Chris@10: * 42 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_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 TP, TT, TN, TM, TY, T13; Chris@10: { Chris@10: E T3, TQ, TJ, T12, Tu, TB, TX, T10, Tj, Tf, Ti, Td, Th, TU, TS; Chris@10: { Chris@10: E TC, TI, TF, TH, TA, Tw, TZ; Chris@10: { Chris@10: E T1, T2, TD, TE; Chris@10: T1 = Ip[0]; Chris@10: T2 = Im[0]; Chris@10: TD = Rm[0]; Chris@10: TE = Rp[0]; Chris@10: TC = W[0]; Chris@10: T3 = T1 - T2; Chris@10: TI = T1 + T2; Chris@10: TQ = TE + TD; Chris@10: TF = TD - TE; Chris@10: TH = W[1]; Chris@10: } Chris@10: { Chris@10: E Tr, To, Ts, Tl, Tq; Chris@10: { Chris@10: E Tm, Tn, TG, T11; Chris@10: Tm = Rm[WS(rs, 2)]; Chris@10: Tn = Rp[WS(rs, 2)]; Chris@10: TG = TC * TF; Chris@10: T11 = TH * TF; Chris@10: Tr = Ip[WS(rs, 2)]; Chris@10: TA = Tn + Tm; Chris@10: To = Tm - Tn; Chris@10: TJ = FNMS(TH, TI, TG); Chris@10: T12 = FMA(TC, TI, T11); Chris@10: Ts = Im[WS(rs, 2)]; Chris@10: } Chris@10: Tl = W[8]; Chris@10: Tq = W[9]; Chris@10: { Chris@10: E Tz, Ty, TW, Tx, Tt, Tp; Chris@10: Tw = W[6]; Chris@10: Tx = Tr - Ts; Chris@10: Tt = Tr + Ts; Chris@10: Tp = Tl * To; Chris@10: Tz = W[7]; Chris@10: Ty = Tw * Tx; Chris@10: TW = Tl * Tt; Chris@10: Tu = FNMS(Tq, Tt, Tp); Chris@10: TZ = Tz * Tx; Chris@10: TB = FNMS(Tz, TA, Ty); Chris@10: TX = FMA(Tq, To, TW); Chris@10: } Chris@10: } Chris@10: { Chris@10: E T5, T6, Ta, Tb; Chris@10: T5 = Ip[WS(rs, 1)]; Chris@10: T10 = FMA(Tw, TA, TZ); Chris@10: T6 = Im[WS(rs, 1)]; Chris@10: Ta = Rp[WS(rs, 1)]; Chris@10: Tb = Rm[WS(rs, 1)]; Chris@10: { Chris@10: E T4, Tg, T7, Tc, T9, T8, TR; Chris@10: T4 = W[5]; Chris@10: Tg = T5 - T6; Chris@10: T7 = T5 + T6; Chris@10: Tj = Ta + Tb; Chris@10: Tc = Ta - Tb; Chris@10: T9 = W[4]; Chris@10: T8 = T4 * T7; Chris@10: Tf = W[2]; Chris@10: Ti = W[3]; Chris@10: TR = T9 * T7; Chris@10: Td = FMA(T9, Tc, T8); Chris@10: Th = Tf * Tg; Chris@10: TU = Ti * Tg; Chris@10: TS = FNMS(T4, Tc, TR); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Te, T1d, TK, Tv, T1a, T1b, Tk, TV; Chris@10: TP = Td + T3; Chris@10: Te = T3 - Td; Chris@10: Tk = FNMS(Ti, Tj, Th); Chris@10: TV = FMA(Tf, Tj, TU); Chris@10: T1d = TQ + TS; Chris@10: TT = TQ - TS; Chris@10: TN = TJ - TB; Chris@10: TK = TB + TJ; Chris@10: Tv = Tk + Tu; Chris@10: TM = Tu - Tk; Chris@10: TY = TV - TX; Chris@10: T1a = TV + TX; Chris@10: T1b = T10 + T12; Chris@10: T13 = T10 - T12; Chris@10: { Chris@10: E T1g, TL, T1e, T1c, T19, T1f; Chris@10: T1g = Tv - TK; Chris@10: TL = Tv + TK; Chris@10: T1e = T1a + T1b; Chris@10: T1c = T1a - T1b; Chris@10: T19 = FNMS(KP500000000, TL, Te); Chris@10: Ip[0] = KP500000000 * (Te + TL); Chris@10: T1f = FNMS(KP500000000, T1e, T1d); Chris@10: Rp[0] = KP500000000 * (T1d + T1e); Chris@10: Im[WS(rs, 1)] = -(KP500000000 * (FNMS(KP866025403, T1c, T19))); Chris@10: Ip[WS(rs, 2)] = KP500000000 * (FMA(KP866025403, T1c, T19)); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T1g, T1f)); Chris@10: Rp[WS(rs, 2)] = KP500000000 * (FNMS(KP866025403, T1g, T1f)); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E TO, T16, T14, T18, T17, T15; Chris@10: TO = TM + TN; Chris@10: T16 = TN - TM; Chris@10: T14 = TY + T13; Chris@10: T18 = T13 - TY; Chris@10: T17 = FMA(KP500000000, TO, TP); Chris@10: Im[WS(rs, 2)] = KP500000000 * (TO - TP); Chris@10: T15 = FNMS(KP500000000, T14, TT); Chris@10: Rm[WS(rs, 2)] = KP500000000 * (TT + T14); Chris@10: Im[0] = -(KP500000000 * (FNMS(KP866025403, T18, T17))); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T18, T17)); Chris@10: Rm[0] = KP500000000 * (FNMS(KP866025403, T16, T15)); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP866025403, T16, T15)); 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, "hc2cfdft_6", twinstr, &GENUS, {36, 22, 22, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_6, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cfdft_6 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 58 FP additions, 36 FP multiplications, Chris@10: * (or, 44 additions, 22 multiplications, 14 fused multiply/add), Chris@10: * 40 stack variables, 3 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft_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(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP433012701, +0.433012701892219323381861585376468091735701313); 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, TM, Tc, TN, Ts, T10, TI, TR, TF, T11, TH, TU; Chris@10: { Chris@10: E T1, T2, TD, Tz, TA, TB, T7, Tf, Tb, Th, Tq, Tw, Tm, Tu, T4; Chris@10: E T8; Chris@10: { Chris@10: E T5, T6, T9, Ta; Chris@10: T1 = Ip[0]; Chris@10: T2 = Im[0]; Chris@10: TD = T1 + T2; Chris@10: Tz = Rm[0]; Chris@10: TA = Rp[0]; Chris@10: TB = Tz - TA; Chris@10: T5 = Ip[WS(rs, 1)]; Chris@10: T6 = Im[WS(rs, 1)]; Chris@10: T7 = T5 + T6; Chris@10: Tf = T5 - T6; Chris@10: T9 = Rp[WS(rs, 1)]; Chris@10: Ta = Rm[WS(rs, 1)]; Chris@10: Tb = T9 - Ta; Chris@10: Th = T9 + Ta; Chris@10: { Chris@10: E To, Tp, Tk, Tl; Chris@10: To = Rp[WS(rs, 2)]; Chris@10: Tp = Rm[WS(rs, 2)]; Chris@10: Tq = To - Tp; Chris@10: Tw = To + Tp; Chris@10: Tk = Ip[WS(rs, 2)]; Chris@10: Tl = Im[WS(rs, 2)]; Chris@10: Tm = Tk + Tl; Chris@10: Tu = Tk - Tl; Chris@10: } Chris@10: } Chris@10: T3 = T1 - T2; Chris@10: TM = TA + Tz; Chris@10: T4 = W[5]; Chris@10: T8 = W[4]; Chris@10: Tc = FMA(T4, T7, T8 * Tb); Chris@10: TN = FNMS(T4, Tb, T8 * T7); Chris@10: { Chris@10: E Ti, TP, Tr, TQ; Chris@10: { Chris@10: E Te, Tg, Tj, Tn; Chris@10: Te = W[2]; Chris@10: Tg = W[3]; Chris@10: Ti = FNMS(Tg, Th, Te * Tf); Chris@10: TP = FMA(Tg, Tf, Te * Th); Chris@10: Tj = W[9]; Chris@10: Tn = W[8]; Chris@10: Tr = FMA(Tj, Tm, Tn * Tq); Chris@10: TQ = FNMS(Tj, Tq, Tn * Tm); Chris@10: } Chris@10: Ts = Ti - Tr; Chris@10: T10 = TP + TQ; Chris@10: TI = Ti + Tr; Chris@10: TR = TP - TQ; Chris@10: } Chris@10: { Chris@10: E Tx, TS, TE, TT; Chris@10: { Chris@10: E Tt, Tv, Ty, TC; Chris@10: Tt = W[6]; Chris@10: Tv = W[7]; Chris@10: Tx = FNMS(Tv, Tw, Tt * Tu); Chris@10: TS = FMA(Tv, Tu, Tt * Tw); Chris@10: Ty = W[0]; Chris@10: TC = W[1]; Chris@10: TE = FNMS(TC, TD, Ty * TB); Chris@10: TT = FMA(TC, TB, Ty * TD); Chris@10: } Chris@10: TF = Tx + TE; Chris@10: T11 = TS + TT; Chris@10: TH = TE - Tx; Chris@10: TU = TS - TT; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T12, Td, TG, TZ; Chris@10: T12 = KP433012701 * (T10 - T11); Chris@10: Td = T3 - Tc; Chris@10: TG = Ts + TF; Chris@10: TZ = FNMS(KP250000000, TG, KP500000000 * Td); Chris@10: Ip[0] = KP500000000 * (Td + TG); Chris@10: Im[WS(rs, 1)] = T12 - TZ; Chris@10: Ip[WS(rs, 2)] = TZ + T12; Chris@10: } Chris@10: { Chris@10: E T16, T13, T14, T15; Chris@10: T16 = KP433012701 * (Ts - TF); Chris@10: T13 = TM + TN; Chris@10: T14 = T10 + T11; Chris@10: T15 = FNMS(KP250000000, T14, KP500000000 * T13); Chris@10: Rp[WS(rs, 2)] = T15 - T16; Chris@10: Rp[0] = KP500000000 * (T13 + T14); Chris@10: Rm[WS(rs, 1)] = T16 + T15; Chris@10: } Chris@10: { Chris@10: E TY, TJ, TK, TX; Chris@10: TY = KP433012701 * (TU - TR); Chris@10: TJ = TH - TI; Chris@10: TK = Tc + T3; Chris@10: TX = FMA(KP500000000, TK, KP250000000 * TJ); Chris@10: Im[WS(rs, 2)] = KP500000000 * (TJ - TK); Chris@10: Im[0] = TY - TX; Chris@10: Ip[WS(rs, 1)] = TX + TY; Chris@10: } Chris@10: { Chris@10: E TL, TO, TV, TW; Chris@10: TL = KP433012701 * (TI + TH); Chris@10: TO = TM - TN; Chris@10: TV = TR + TU; Chris@10: TW = FNMS(KP250000000, TV, KP500000000 * TO); Chris@10: Rp[WS(rs, 1)] = TL + TW; Chris@10: Rm[WS(rs, 2)] = KP500000000 * (TO + TV); Chris@10: Rm[0] = TW - TL; 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, "hc2cfdft_6", twinstr, &GENUS, {44, 22, 14, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft_6, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */