Chris@82: /*
Chris@82:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82:  *
Chris@82:  * This program is free software; you can redistribute it and/or modify
Chris@82:  * it under the terms of the GNU General Public License as published by
Chris@82:  * the Free Software Foundation; either version 2 of the License, or
Chris@82:  * (at your option) any later version.
Chris@82:  *
Chris@82:  * This program is distributed in the hope that it will be useful,
Chris@82:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@82:  * GNU General Public License for more details.
Chris@82:  *
Chris@82:  * You should have received a copy of the GNU General Public License
Chris@82:  * along with this program; if not, write to the Free Software
Chris@82:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@82:  *
Chris@82:  */
Chris@82: 
Chris@82: /* This file was automatically generated --- DO NOT EDIT */
Chris@82: /* Generated on Thu May 24 08:04:30 EDT 2018 */
Chris@82: 
Chris@82: #include "dft/codelet-dft.h"
Chris@82: 
Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
Chris@82: 
Chris@82: /* Generated by: ../../../genfft/gen_twidsq.native -fma -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include dft/scalar/q.h */
Chris@82: 
Chris@82: /*
Chris@82:  * This function contains 48 FP additions, 42 FP multiplications,
Chris@82:  * (or, 18 additions, 12 multiplications, 30 fused multiply/add),
Chris@82:  * 35 stack variables, 2 constants, and 36 memory accesses
Chris@82:  */
Chris@82: #include "dft/scalar/q.h"
Chris@82: 
Chris@82: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@82: {
Chris@82:      DK(KP866025403, +0.866025403784438646763723170752936183471402627);
Chris@82:      DK(KP500000000, +0.500000000000000000000000000000000000000000000);
Chris@82:      {
Chris@82: 	  INT m;
Chris@82: 	  for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@82: 	       E T1, T4, T6, Tg, Td, Te, T9, Tf, Tp, Ts, Tu, TE, TB, TC, Tx;
Chris@82: 	       E TD, TZ, T10, TV, T11, TN, TQ, TS, T12;
Chris@82: 	       {
Chris@82: 		    E T2, T3, Tv, Tw;
Chris@82: 		    T1 = rio[0];
Chris@82: 		    T2 = rio[WS(rs, 1)];
Chris@82: 		    T3 = rio[WS(rs, 2)];
Chris@82: 		    T4 = T2 + T3;
Chris@82: 		    T6 = FNMS(KP500000000, T4, T1);
Chris@82: 		    Tg = T3 - T2;
Chris@82: 		    {
Chris@82: 			 E T7, T8, Tq, Tr;
Chris@82: 			 Td = iio[0];
Chris@82: 			 T7 = iio[WS(rs, 1)];
Chris@82: 			 T8 = iio[WS(rs, 2)];
Chris@82: 			 Te = T7 + T8;
Chris@82: 			 T9 = T7 - T8;
Chris@82: 			 Tf = FNMS(KP500000000, Te, Td);
Chris@82: 			 Tp = rio[WS(vs, 1)];
Chris@82: 			 Tq = rio[WS(vs, 1) + WS(rs, 1)];
Chris@82: 			 Tr = rio[WS(vs, 1) + WS(rs, 2)];
Chris@82: 			 Ts = Tq + Tr;
Chris@82: 			 Tu = FNMS(KP500000000, Ts, Tp);
Chris@82: 			 TE = Tr - Tq;
Chris@82: 		    }
Chris@82: 		    TB = iio[WS(vs, 1)];
Chris@82: 		    Tv = iio[WS(vs, 1) + WS(rs, 1)];
Chris@82: 		    Tw = iio[WS(vs, 1) + WS(rs, 2)];
Chris@82: 		    TC = Tv + Tw;
Chris@82: 		    Tx = Tv - Tw;
Chris@82: 		    TD = FNMS(KP500000000, TC, TB);
Chris@82: 		    {
Chris@82: 			 E TT, TU, TO, TP;
Chris@82: 			 TZ = iio[WS(vs, 2)];
Chris@82: 			 TT = iio[WS(vs, 2) + WS(rs, 1)];
Chris@82: 			 TU = iio[WS(vs, 2) + WS(rs, 2)];
Chris@82: 			 T10 = TT + TU;
Chris@82: 			 TV = TT - TU;
Chris@82: 			 T11 = FNMS(KP500000000, T10, TZ);
Chris@82: 			 TN = rio[WS(vs, 2)];
Chris@82: 			 TO = rio[WS(vs, 2) + WS(rs, 1)];
Chris@82: 			 TP = rio[WS(vs, 2) + WS(rs, 2)];
Chris@82: 			 TQ = TO + TP;
Chris@82: 			 TS = FNMS(KP500000000, TQ, TN);
Chris@82: 			 T12 = TP - TO;
Chris@82: 		    }
Chris@82: 	       }
Chris@82: 	       rio[0] = T1 + T4;
Chris@82: 	       iio[0] = Td + Te;
Chris@82: 	       rio[WS(rs, 1)] = Tp + Ts;
Chris@82: 	       iio[WS(rs, 1)] = TB + TC;
Chris@82: 	       iio[WS(rs, 2)] = TZ + T10;
Chris@82: 	       rio[WS(rs, 2)] = TN + TQ;
Chris@82: 	       {
Chris@82: 		    E Ta, Th, Tb, Ti, T5, Tc;
Chris@82: 		    Ta = FMA(KP866025403, T9, T6);
Chris@82: 		    Th = FMA(KP866025403, Tg, Tf);
Chris@82: 		    T5 = W[0];
Chris@82: 		    Tb = T5 * Ta;
Chris@82: 		    Ti = T5 * Th;
Chris@82: 		    Tc = W[1];
Chris@82: 		    rio[WS(vs, 1)] = FMA(Tc, Th, Tb);
Chris@82: 		    iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E T16, T19, T17, T1a, T15, T18;
Chris@82: 		    T16 = FNMS(KP866025403, TV, TS);
Chris@82: 		    T19 = FNMS(KP866025403, T12, T11);
Chris@82: 		    T15 = W[2];
Chris@82: 		    T17 = T15 * T16;
Chris@82: 		    T1a = T15 * T19;
Chris@82: 		    T18 = W[3];
Chris@82: 		    rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17);
Chris@82: 		    iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E TI, TL, TJ, TM, TH, TK;
Chris@82: 		    TI = FNMS(KP866025403, Tx, Tu);
Chris@82: 		    TL = FNMS(KP866025403, TE, TD);
Chris@82: 		    TH = W[2];
Chris@82: 		    TJ = TH * TI;
Chris@82: 		    TM = TH * TL;
Chris@82: 		    TK = W[3];
Chris@82: 		    rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ);
Chris@82: 		    iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E Ty, TF, Tz, TG, Tt, TA;
Chris@82: 		    Ty = FMA(KP866025403, Tx, Tu);
Chris@82: 		    TF = FMA(KP866025403, TE, TD);
Chris@82: 		    Tt = W[0];
Chris@82: 		    Tz = Tt * Ty;
Chris@82: 		    TG = Tt * TF;
Chris@82: 		    TA = W[1];
Chris@82: 		    rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz);
Chris@82: 		    iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E TW, T13, TX, T14, TR, TY;
Chris@82: 		    TW = FMA(KP866025403, TV, TS);
Chris@82: 		    T13 = FMA(KP866025403, T12, T11);
Chris@82: 		    TR = W[0];
Chris@82: 		    TX = TR * TW;
Chris@82: 		    T14 = TR * T13;
Chris@82: 		    TY = W[1];
Chris@82: 		    rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX);
Chris@82: 		    iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E Tk, Tn, Tl, To, Tj, Tm;
Chris@82: 		    Tk = FNMS(KP866025403, T9, T6);
Chris@82: 		    Tn = FNMS(KP866025403, Tg, Tf);
Chris@82: 		    Tj = W[2];
Chris@82: 		    Tl = Tj * Tk;
Chris@82: 		    To = Tj * Tn;
Chris@82: 		    Tm = W[3];
Chris@82: 		    rio[WS(vs, 2)] = FMA(Tm, Tn, Tl);
Chris@82: 		    iio[WS(vs, 2)] = FNMS(Tm, Tk, To);
Chris@82: 	       }
Chris@82: 	  }
Chris@82:      }
Chris@82: }
Chris@82: 
Chris@82: static const tw_instr twinstr[] = {
Chris@82:      {TW_FULL, 0, 3},
Chris@82:      {TW_NEXT, 1, 0}
Chris@82: };
Chris@82: 
Chris@82: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 };
Chris@82: 
Chris@82: void X(codelet_q1_3) (planner *p) {
Chris@82:      X(kdft_difsq_register) (p, q1_3, &desc);
Chris@82: }
Chris@82: #else
Chris@82: 
Chris@82: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include dft/scalar/q.h */
Chris@82: 
Chris@82: /*
Chris@82:  * This function contains 48 FP additions, 36 FP multiplications,
Chris@82:  * (or, 30 additions, 18 multiplications, 18 fused multiply/add),
Chris@82:  * 35 stack variables, 2 constants, and 36 memory accesses
Chris@82:  */
Chris@82: #include "dft/scalar/q.h"
Chris@82: 
Chris@82: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms)
Chris@82: {
Chris@82:      DK(KP866025403, +0.866025403784438646763723170752936183471402627);
Chris@82:      DK(KP500000000, +0.500000000000000000000000000000000000000000000);
Chris@82:      {
Chris@82: 	  INT m;
Chris@82: 	  for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) {
Chris@82: 	       E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt;
Chris@82: 	       E Tz, TR, TS, TN, TT, TF, TI, TK, TQ;
Chris@82: 	       {
Chris@82: 		    E T2, T3, Tr, Ts;
Chris@82: 		    T1 = rio[0];
Chris@82: 		    T2 = rio[WS(rs, 1)];
Chris@82: 		    T3 = rio[WS(rs, 2)];
Chris@82: 		    T4 = T2 + T3;
Chris@82: 		    T6 = FNMS(KP500000000, T4, T1);
Chris@82: 		    Tc = KP866025403 * (T3 - T2);
Chris@82: 		    {
Chris@82: 			 E T7, T8, Tm, Tn;
Chris@82: 			 Td = iio[0];
Chris@82: 			 T7 = iio[WS(rs, 1)];
Chris@82: 			 T8 = iio[WS(rs, 2)];
Chris@82: 			 Te = T7 + T8;
Chris@82: 			 T9 = KP866025403 * (T7 - T8);
Chris@82: 			 Tf = FNMS(KP500000000, Te, Td);
Chris@82: 			 Tl = rio[WS(vs, 1)];
Chris@82: 			 Tm = rio[WS(vs, 1) + WS(rs, 1)];
Chris@82: 			 Tn = rio[WS(vs, 1) + WS(rs, 2)];
Chris@82: 			 To = Tm + Tn;
Chris@82: 			 Tq = FNMS(KP500000000, To, Tl);
Chris@82: 			 Tw = KP866025403 * (Tn - Tm);
Chris@82: 		    }
Chris@82: 		    Tx = iio[WS(vs, 1)];
Chris@82: 		    Tr = iio[WS(vs, 1) + WS(rs, 1)];
Chris@82: 		    Ts = iio[WS(vs, 1) + WS(rs, 2)];
Chris@82: 		    Ty = Tr + Ts;
Chris@82: 		    Tt = KP866025403 * (Tr - Ts);
Chris@82: 		    Tz = FNMS(KP500000000, Ty, Tx);
Chris@82: 		    {
Chris@82: 			 E TL, TM, TG, TH;
Chris@82: 			 TR = iio[WS(vs, 2)];
Chris@82: 			 TL = iio[WS(vs, 2) + WS(rs, 1)];
Chris@82: 			 TM = iio[WS(vs, 2) + WS(rs, 2)];
Chris@82: 			 TS = TL + TM;
Chris@82: 			 TN = KP866025403 * (TL - TM);
Chris@82: 			 TT = FNMS(KP500000000, TS, TR);
Chris@82: 			 TF = rio[WS(vs, 2)];
Chris@82: 			 TG = rio[WS(vs, 2) + WS(rs, 1)];
Chris@82: 			 TH = rio[WS(vs, 2) + WS(rs, 2)];
Chris@82: 			 TI = TG + TH;
Chris@82: 			 TK = FNMS(KP500000000, TI, TF);
Chris@82: 			 TQ = KP866025403 * (TH - TG);
Chris@82: 		    }
Chris@82: 	       }
Chris@82: 	       rio[0] = T1 + T4;
Chris@82: 	       iio[0] = Td + Te;
Chris@82: 	       rio[WS(rs, 1)] = Tl + To;
Chris@82: 	       iio[WS(rs, 1)] = Tx + Ty;
Chris@82: 	       iio[WS(rs, 2)] = TR + TS;
Chris@82: 	       rio[WS(rs, 2)] = TF + TI;
Chris@82: 	       {
Chris@82: 		    E Ta, Tg, T5, Tb;
Chris@82: 		    Ta = T6 + T9;
Chris@82: 		    Tg = Tc + Tf;
Chris@82: 		    T5 = W[0];
Chris@82: 		    Tb = W[1];
Chris@82: 		    rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg);
Chris@82: 		    iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E TW, TY, TV, TX;
Chris@82: 		    TW = TK - TN;
Chris@82: 		    TY = TT - TQ;
Chris@82: 		    TV = W[2];
Chris@82: 		    TX = W[3];
Chris@82: 		    rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY);
Chris@82: 		    iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E TC, TE, TB, TD;
Chris@82: 		    TC = Tq - Tt;
Chris@82: 		    TE = Tz - Tw;
Chris@82: 		    TB = W[2];
Chris@82: 		    TD = W[3];
Chris@82: 		    rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE);
Chris@82: 		    iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E Tu, TA, Tp, Tv;
Chris@82: 		    Tu = Tq + Tt;
Chris@82: 		    TA = Tw + Tz;
Chris@82: 		    Tp = W[0];
Chris@82: 		    Tv = W[1];
Chris@82: 		    rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA);
Chris@82: 		    iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E TO, TU, TJ, TP;
Chris@82: 		    TO = TK + TN;
Chris@82: 		    TU = TQ + TT;
Chris@82: 		    TJ = W[0];
Chris@82: 		    TP = W[1];
Chris@82: 		    rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU);
Chris@82: 		    iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU);
Chris@82: 	       }
Chris@82: 	       {
Chris@82: 		    E Ti, Tk, Th, Tj;
Chris@82: 		    Ti = T6 - T9;
Chris@82: 		    Tk = Tf - Tc;
Chris@82: 		    Th = W[2];
Chris@82: 		    Tj = W[3];
Chris@82: 		    rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk);
Chris@82: 		    iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk);
Chris@82: 	       }
Chris@82: 	  }
Chris@82:      }
Chris@82: }
Chris@82: 
Chris@82: static const tw_instr twinstr[] = {
Chris@82:      {TW_FULL, 0, 3},
Chris@82:      {TW_NEXT, 1, 0}
Chris@82: };
Chris@82: 
Chris@82: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 };
Chris@82: 
Chris@82: void X(codelet_q1_3) (planner *p) {
Chris@82:      X(kdft_difsq_register) (p, q1_3, &desc);
Chris@82: }
Chris@82: #endif