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 07:42:04 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 -sign 1 -n 6 -dif -name hc2cbdft_6 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 58 FP additions, 32 FP multiplications, Chris@10: * (or, 36 additions, 10 multiplications, 22 fused multiply/add), Chris@10: * 52 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_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 T18, T1b, T16, T1e, T1a, T1f, T19, T1g, T1c; Chris@10: { Chris@10: E Tw, T4, TV, Tj, TP, TH, Tr, TY, T5, T6, Ta, Ty; Chris@10: { Chris@10: E Tg, TF, Tf, TD, Tp, Th; Chris@10: { Chris@10: E Td, Te, Tn, To; Chris@10: Td = Ip[WS(rs, 1)]; Chris@10: Te = Im[WS(rs, 1)]; Chris@10: Tn = Ip[0]; Chris@10: To = Im[WS(rs, 2)]; Chris@10: Tg = Ip[WS(rs, 2)]; Chris@10: TF = Te + Td; Chris@10: Tf = Td - Te; Chris@10: TD = Tn + To; Chris@10: Tp = Tn - To; Chris@10: Th = Im[0]; Chris@10: } Chris@10: { Chris@10: E T2, T3, T8, T9; Chris@10: T2 = Rp[0]; Chris@10: T3 = Rm[WS(rs, 2)]; Chris@10: { Chris@10: E Tq, TE, Ti, TG; Chris@10: T8 = Rm[WS(rs, 1)]; Chris@10: TE = Tg + Th; Chris@10: Ti = Tg - Th; Chris@10: Tw = T2 - T3; Chris@10: T4 = T2 + T3; Chris@10: TG = TE - TF; Chris@10: TV = TF + TE; Chris@10: Tq = Tf + Ti; Chris@10: Tj = Tf - Ti; Chris@10: TP = FNMS(KP500000000, TG, TD); Chris@10: TH = TD + TG; Chris@10: T9 = Rp[WS(rs, 1)]; Chris@10: Tr = FNMS(KP500000000, Tq, Tp); Chris@10: TY = Tp + Tq; Chris@10: } Chris@10: T5 = Rp[WS(rs, 2)]; Chris@10: T6 = Rm[0]; Chris@10: Ta = T8 + T9; Chris@10: Ty = T8 - T9; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TO, TT, Ts, TA, TR, Tc, TN, TW, TS, Tx, T7; Chris@10: Tx = T5 - T6; Chris@10: T7 = T5 + T6; Chris@10: TO = W[0]; Chris@10: TT = W[1]; Chris@10: { Chris@10: E Tz, TQ, Tb, TU; Chris@10: Tz = Tx + Ty; Chris@10: TQ = Tx - Ty; Chris@10: Tb = T7 + Ta; Chris@10: Ts = T7 - Ta; Chris@10: TU = FNMS(KP500000000, Tz, Tw); Chris@10: TA = Tw + Tz; Chris@10: TR = FMA(KP866025403, TQ, TP); Chris@10: T18 = FNMS(KP866025403, TQ, TP); Chris@10: Tc = FNMS(KP500000000, Tb, T4); Chris@10: TN = T4 + Tb; Chris@10: T1b = FMA(KP866025403, TV, TU); Chris@10: TW = FNMS(KP866025403, TV, TU); Chris@10: TS = TO * TR; Chris@10: } Chris@10: { Chris@10: E T15, Tt, T12, T1, Tm, TI, TM, Tl, TJ; Chris@10: { Chris@10: E Tv, TC, TB, TL, Tk, TZ, TX, T10; Chris@10: T15 = FMA(KP866025403, Ts, Tr); Chris@10: Tt = FNMS(KP866025403, Ts, Tr); Chris@10: TZ = TO * TW; Chris@10: TX = FMA(TT, TW, TS); Chris@10: Tv = W[4]; Chris@10: TC = W[5]; Chris@10: T10 = FNMS(TT, TR, TZ); Chris@10: Rm[0] = TN + TX; Chris@10: Rp[0] = TN - TX; Chris@10: TB = Tv * TA; Chris@10: Im[0] = T10 - TY; Chris@10: Ip[0] = TY + T10; Chris@10: TL = TC * TA; Chris@10: Tk = FNMS(KP866025403, Tj, Tc); Chris@10: T12 = FMA(KP866025403, Tj, Tc); Chris@10: T1 = W[3]; Chris@10: Tm = W[2]; Chris@10: TI = FNMS(TC, TH, TB); Chris@10: TM = FMA(Tv, TH, TL); Chris@10: Tl = T1 * Tk; Chris@10: TJ = Tm * Tk; Chris@10: } Chris@10: { Chris@10: E T11, T14, T13, T1d, T17, Tu, TK; Chris@10: Tu = FMA(Tm, Tt, Tl); Chris@10: TK = FNMS(T1, Tt, TJ); Chris@10: T11 = W[6]; Chris@10: T14 = W[7]; Chris@10: Im[WS(rs, 1)] = TI - Tu; Chris@10: Ip[WS(rs, 1)] = Tu + TI; Chris@10: Rm[WS(rs, 1)] = TK + TM; Chris@10: Rp[WS(rs, 1)] = TK - TM; Chris@10: T13 = T11 * T12; Chris@10: T1d = T14 * T12; Chris@10: T17 = W[8]; Chris@10: T16 = FNMS(T14, T15, T13); Chris@10: T1e = FMA(T11, T15, T1d); Chris@10: T1a = W[9]; Chris@10: T1f = T17 * T1b; Chris@10: T19 = T17 * T18; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: T1g = FNMS(T1a, T18, T1f); Chris@10: T1c = FMA(T1a, T1b, T19); Chris@10: Im[WS(rs, 2)] = T1g - T1e; Chris@10: Ip[WS(rs, 2)] = T1e + T1g; Chris@10: Rm[WS(rs, 2)] = T16 + T1c; Chris@10: Rp[WS(rs, 2)] = T16 - T1c; 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, "hc2cbdft_6", twinstr, &GENUS, {36, 10, 22, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_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 -sign 1 -n 6 -dif -name hc2cbdft_6 -include hc2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 58 FP additions, 28 FP multiplications, Chris@10: * (or, 44 additions, 14 multiplications, 14 fused multiply/add), Chris@10: * 29 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cb.h" Chris@10: Chris@10: static void hc2cbdft_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 T4, Tv, Tr, TL, Tb, Tc, Ty, TP, To, TB, Tj, TQ, Tp, Tq, TE; Chris@10: E TM; Chris@10: { Chris@10: E Ta, Tx, T7, Tw, T2, T3; Chris@10: T2 = Rp[0]; Chris@10: T3 = Rm[WS(rs, 2)]; Chris@10: T4 = T2 + T3; Chris@10: Tv = T2 - T3; Chris@10: { Chris@10: E T8, T9, T5, T6; Chris@10: T8 = Rm[WS(rs, 1)]; Chris@10: T9 = Rp[WS(rs, 1)]; Chris@10: Ta = T8 + T9; Chris@10: Tx = T8 - T9; Chris@10: T5 = Rp[WS(rs, 2)]; Chris@10: T6 = Rm[0]; Chris@10: T7 = T5 + T6; Chris@10: Tw = T5 - T6; Chris@10: } Chris@10: Tr = KP866025403 * (T7 - Ta); Chris@10: TL = KP866025403 * (Tw - Tx); Chris@10: Tb = T7 + Ta; Chris@10: Tc = FNMS(KP500000000, Tb, T4); Chris@10: Ty = Tw + Tx; Chris@10: TP = FNMS(KP500000000, Ty, Tv); Chris@10: } Chris@10: { Chris@10: E Tf, TC, Ti, TD, Td, Te; Chris@10: Td = Ip[WS(rs, 1)]; Chris@10: Te = Im[WS(rs, 1)]; Chris@10: Tf = Td - Te; Chris@10: TC = Te + Td; Chris@10: { Chris@10: E Tm, Tn, Tg, Th; Chris@10: Tm = Ip[0]; Chris@10: Tn = Im[WS(rs, 2)]; Chris@10: To = Tm - Tn; Chris@10: TB = Tm + Tn; Chris@10: Tg = Ip[WS(rs, 2)]; Chris@10: Th = Im[0]; Chris@10: Ti = Tg - Th; Chris@10: TD = Tg + Th; Chris@10: } Chris@10: Tj = KP866025403 * (Tf - Ti); Chris@10: TQ = KP866025403 * (TC + TD); Chris@10: Tp = Tf + Ti; Chris@10: Tq = FNMS(KP500000000, Tp, To); Chris@10: TE = TC - TD; Chris@10: TM = FMA(KP500000000, TE, TB); Chris@10: } Chris@10: { Chris@10: E TJ, TT, TS, TU; Chris@10: TJ = T4 + Tb; Chris@10: TT = To + Tp; Chris@10: { Chris@10: E TN, TR, TK, TO; Chris@10: TN = TL + TM; Chris@10: TR = TP - TQ; Chris@10: TK = W[0]; Chris@10: TO = W[1]; Chris@10: TS = FMA(TK, TN, TO * TR); Chris@10: TU = FNMS(TO, TN, TK * TR); Chris@10: } Chris@10: Rp[0] = TJ - TS; Chris@10: Ip[0] = TT + TU; Chris@10: Rm[0] = TJ + TS; Chris@10: Im[0] = TU - TT; Chris@10: } Chris@10: { Chris@10: E TZ, T15, T14, T16; Chris@10: { Chris@10: E TW, TY, TV, TX; Chris@10: TW = Tc + Tj; Chris@10: TY = Tr + Tq; Chris@10: TV = W[6]; Chris@10: TX = W[7]; Chris@10: TZ = FNMS(TX, TY, TV * TW); Chris@10: T15 = FMA(TX, TW, TV * TY); Chris@10: } Chris@10: { Chris@10: E T11, T13, T10, T12; Chris@10: T11 = TM - TL; Chris@10: T13 = TP + TQ; Chris@10: T10 = W[8]; Chris@10: T12 = W[9]; Chris@10: T14 = FMA(T10, T11, T12 * T13); Chris@10: T16 = FNMS(T12, T11, T10 * T13); Chris@10: } Chris@10: Rp[WS(rs, 2)] = TZ - T14; Chris@10: Ip[WS(rs, 2)] = T15 + T16; Chris@10: Rm[WS(rs, 2)] = TZ + T14; Chris@10: Im[WS(rs, 2)] = T16 - T15; Chris@10: } Chris@10: { Chris@10: E Tt, TH, TG, TI; Chris@10: { Chris@10: E Tk, Ts, T1, Tl; Chris@10: Tk = Tc - Tj; Chris@10: Ts = Tq - Tr; Chris@10: T1 = W[3]; Chris@10: Tl = W[2]; Chris@10: Tt = FMA(T1, Tk, Tl * Ts); Chris@10: TH = FNMS(T1, Ts, Tl * Tk); Chris@10: } Chris@10: { Chris@10: E Tz, TF, Tu, TA; Chris@10: Tz = Tv + Ty; Chris@10: TF = TB - TE; Chris@10: Tu = W[4]; Chris@10: TA = W[5]; Chris@10: TG = FNMS(TA, TF, Tu * Tz); Chris@10: TI = FMA(TA, Tz, Tu * TF); Chris@10: } Chris@10: Ip[WS(rs, 1)] = Tt + TG; Chris@10: Rp[WS(rs, 1)] = TH - TI; Chris@10: Im[WS(rs, 1)] = TG - Tt; Chris@10: Rm[WS(rs, 1)] = TH + TI; 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, "hc2cbdft_6", twinstr, &GENUS, {44, 14, 14, 0} }; Chris@10: Chris@10: void X(codelet_hc2cbdft_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cbdft_6, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */