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 07:42:07 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 -sign 1 -n 8 -dif -name hc2cbdft2_8 -include hc2cb.h */
cannam@95: 
cannam@95: /*
cannam@95:  * This function contains 82 FP additions, 36 FP multiplications,
cannam@95:  * (or, 60 additions, 14 multiplications, 22 fused multiply/add),
cannam@95:  * 55 stack variables, 1 constants, and 32 memory accesses
cannam@95:  */
cannam@95: #include "hc2cb.h"
cannam@95: 
cannam@95: static void hc2cbdft2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
cannam@95: {
cannam@95:      DK(KP707106781, +0.707106781186547524400844362104849039284835938);
cannam@95:      {
cannam@95: 	  INT m;
cannam@95: 	  for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) {
cannam@95: 	       E T1m, T1r, T1i, T1u, T1o, T1v, T1n, T1w, T1s;
cannam@95: 	       {
cannam@95: 		    E T1k, Tl, T1p, TE, TP, T1g, TM, T1b, T1f, T1a, TU, Tf, T1l, TH, Tw;
cannam@95: 		    E T1q;
cannam@95: 		    {
cannam@95: 			 E TA, T3, TN, Tk, Th, T6, TO, TD, Tb, Tm, Ta, TK, Tp, Tc, Ts;
cannam@95: 			 E Tt;
cannam@95: 			 {
cannam@95: 			      E T4, T5, TB, TC;
cannam@95: 			      {
cannam@95: 				   E T1, T2, Ti, Tj;
cannam@95: 				   T1 = Rp[0];
cannam@95: 				   T2 = Rm[WS(rs, 3)];
cannam@95: 				   Ti = Ip[0];
cannam@95: 				   Tj = Im[WS(rs, 3)];
cannam@95: 				   T4 = Rp[WS(rs, 2)];
cannam@95: 				   TA = T1 - T2;
cannam@95: 				   T3 = T1 + T2;
cannam@95: 				   TN = Ti - Tj;
cannam@95: 				   Tk = Ti + Tj;
cannam@95: 				   T5 = Rm[WS(rs, 1)];
cannam@95: 				   TB = Ip[WS(rs, 2)];
cannam@95: 				   TC = Im[WS(rs, 1)];
cannam@95: 			      }
cannam@95: 			      {
cannam@95: 				   E T8, T9, Tn, To;
cannam@95: 				   T8 = Rp[WS(rs, 1)];
cannam@95: 				   Th = T4 - T5;
cannam@95: 				   T6 = T4 + T5;
cannam@95: 				   TO = TB - TC;
cannam@95: 				   TD = TB + TC;
cannam@95: 				   T9 = Rm[WS(rs, 2)];
cannam@95: 				   Tn = Ip[WS(rs, 1)];
cannam@95: 				   To = Im[WS(rs, 2)];
cannam@95: 				   Tb = Rm[0];
cannam@95: 				   Tm = T8 - T9;
cannam@95: 				   Ta = T8 + T9;
cannam@95: 				   TK = Tn - To;
cannam@95: 				   Tp = Tn + To;
cannam@95: 				   Tc = Rp[WS(rs, 3)];
cannam@95: 				   Ts = Im[0];
cannam@95: 				   Tt = Ip[WS(rs, 3)];
cannam@95: 			      }
cannam@95: 			 }
cannam@95: 			 {
cannam@95: 			      E Tr, Td, Tu, TL, Te, T7;
cannam@95: 			      T1k = Tk - Th;
cannam@95: 			      Tl = Th + Tk;
cannam@95: 			      Tr = Tb - Tc;
cannam@95: 			      Td = Tb + Tc;
cannam@95: 			      TL = Tt - Ts;
cannam@95: 			      Tu = Ts + Tt;
cannam@95: 			      T1p = TA + TD;
cannam@95: 			      TE = TA - TD;
cannam@95: 			      TP = TN + TO;
cannam@95: 			      T1g = TN - TO;
cannam@95: 			      TM = TK + TL;
cannam@95: 			      T1b = TL - TK;
cannam@95: 			      T1f = Ta - Td;
cannam@95: 			      Te = Ta + Td;
cannam@95: 			      T1a = T3 - T6;
cannam@95: 			      T7 = T3 + T6;
cannam@95: 			      {
cannam@95: 				   E Tq, TF, TG, Tv;
cannam@95: 				   Tq = Tm + Tp;
cannam@95: 				   TF = Tm - Tp;
cannam@95: 				   TG = Tr - Tu;
cannam@95: 				   Tv = Tr + Tu;
cannam@95: 				   TU = T7 - Te;
cannam@95: 				   Tf = T7 + Te;
cannam@95: 				   T1l = TF - TG;
cannam@95: 				   TH = TF + TG;
cannam@95: 				   Tw = Tq - Tv;
cannam@95: 				   T1q = Tq + Tv;
cannam@95: 			      }
cannam@95: 			 }
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E TX, T10, T1c, T13, T1h, T1E, T1H, T1C, T1K, T1G, T1L, T1F;
cannam@95: 			 {
cannam@95: 			      E TQ, Tx, T1y, TI, Tg, Tz;
cannam@95: 			      TX = TP - TM;
cannam@95: 			      TQ = TM + TP;
cannam@95: 			      Tx = FMA(KP707106781, Tw, Tl);
cannam@95: 			      T10 = FNMS(KP707106781, Tw, Tl);
cannam@95: 			      T1c = T1a + T1b;
cannam@95: 			      T1y = T1a - T1b;
cannam@95: 			      T13 = FNMS(KP707106781, TH, TE);
cannam@95: 			      TI = FMA(KP707106781, TH, TE);
cannam@95: 			      Tg = W[0];
cannam@95: 			      Tz = W[1];
cannam@95: 			      {
cannam@95: 				   E T1B, T1A, T1x, T1J, T1z, T1D;
cannam@95: 				   {
cannam@95: 					E TR, Ty, TS, TJ;
cannam@95: 					T1B = T1g - T1f;
cannam@95: 					T1h = T1f + T1g;
cannam@95: 					T1A = W[11];
cannam@95: 					TR = Tg * TI;
cannam@95: 					Ty = Tg * Tx;
cannam@95: 					T1x = W[10];
cannam@95: 					T1J = T1A * T1y;
cannam@95: 					TS = FNMS(Tz, Tx, TR);
cannam@95: 					TJ = FMA(Tz, TI, Ty);
cannam@95: 					T1z = T1x * T1y;
cannam@95: 					T1m = FMA(KP707106781, T1l, T1k);
cannam@95: 					T1E = FNMS(KP707106781, T1l, T1k);
cannam@95: 					Im[0] = TS - TQ;
cannam@95: 					Ip[0] = TQ + TS;
cannam@95: 					Rm[0] = Tf + TJ;
cannam@95: 					Rp[0] = Tf - TJ;
cannam@95: 					T1H = FMA(KP707106781, T1q, T1p);
cannam@95: 					T1r = FNMS(KP707106781, T1q, T1p);
cannam@95: 					T1D = W[12];
cannam@95: 				   }
cannam@95: 				   T1C = FNMS(T1A, T1B, T1z);
cannam@95: 				   T1K = FMA(T1x, T1B, T1J);
cannam@95: 				   T1G = W[13];
cannam@95: 				   T1L = T1D * T1H;
cannam@95: 				   T1F = T1D * T1E;
cannam@95: 			      }
cannam@95: 			 }
cannam@95: 			 {
cannam@95: 			      E TY, T16, T12, T17, T11;
cannam@95: 			      {
cannam@95: 				   E TW, TT, T15, TV, TZ, T1M, T1I;
cannam@95: 				   TW = W[7];
cannam@95: 				   T1M = FNMS(T1G, T1E, T1L);
cannam@95: 				   T1I = FMA(T1G, T1H, T1F);
cannam@95: 				   TT = W[6];
cannam@95: 				   T15 = TW * TU;
cannam@95: 				   Im[WS(rs, 3)] = T1M - T1K;
cannam@95: 				   Ip[WS(rs, 3)] = T1K + T1M;
cannam@95: 				   Rm[WS(rs, 3)] = T1C + T1I;
cannam@95: 				   Rp[WS(rs, 3)] = T1C - T1I;
cannam@95: 				   TV = TT * TU;
cannam@95: 				   TZ = W[8];
cannam@95: 				   TY = FNMS(TW, TX, TV);
cannam@95: 				   T16 = FMA(TT, TX, T15);
cannam@95: 				   T12 = W[9];
cannam@95: 				   T17 = TZ * T13;
cannam@95: 				   T11 = TZ * T10;
cannam@95: 			      }
cannam@95: 			      {
cannam@95: 				   E T1e, T19, T1t, T1d, T1j, T18, T14;
cannam@95: 				   T1e = W[3];
cannam@95: 				   T18 = FNMS(T12, T10, T17);
cannam@95: 				   T14 = FMA(T12, T13, T11);
cannam@95: 				   T19 = W[2];
cannam@95: 				   T1t = T1e * T1c;
cannam@95: 				   Im[WS(rs, 2)] = T18 - T16;
cannam@95: 				   Ip[WS(rs, 2)] = T16 + T18;
cannam@95: 				   Rm[WS(rs, 2)] = TY + T14;
cannam@95: 				   Rp[WS(rs, 2)] = TY - T14;
cannam@95: 				   T1d = T19 * T1c;
cannam@95: 				   T1j = W[4];
cannam@95: 				   T1i = FNMS(T1e, T1h, T1d);
cannam@95: 				   T1u = FMA(T19, T1h, T1t);
cannam@95: 				   T1o = W[5];
cannam@95: 				   T1v = T1j * T1r;
cannam@95: 				   T1n = T1j * T1m;
cannam@95: 			      }
cannam@95: 			 }
cannam@95: 		    }
cannam@95: 	       }
cannam@95: 	       T1w = FNMS(T1o, T1m, T1v);
cannam@95: 	       T1s = FMA(T1o, T1r, T1n);
cannam@95: 	       Im[WS(rs, 1)] = T1w - T1u;
cannam@95: 	       Ip[WS(rs, 1)] = T1u + T1w;
cannam@95: 	       Rm[WS(rs, 1)] = T1i + T1s;
cannam@95: 	       Rp[WS(rs, 1)] = T1i - T1s;
cannam@95: 	  }
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: static const tw_instr twinstr[] = {
cannam@95:      {TW_FULL, 1, 8},
cannam@95:      {TW_NEXT, 1, 0}
cannam@95: };
cannam@95: 
cannam@95: static const hc2c_desc desc = { 8, "hc2cbdft2_8", twinstr, &GENUS, {60, 14, 22, 0} };
cannam@95: 
cannam@95: void X(codelet_hc2cbdft2_8) (planner *p) {
cannam@95:      X(khc2c_register) (p, hc2cbdft2_8, &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 -sign 1 -n 8 -dif -name hc2cbdft2_8 -include hc2cb.h */
cannam@95: 
cannam@95: /*
cannam@95:  * This function contains 82 FP additions, 32 FP multiplications,
cannam@95:  * (or, 68 additions, 18 multiplications, 14 fused multiply/add),
cannam@95:  * 30 stack variables, 1 constants, and 32 memory accesses
cannam@95:  */
cannam@95: #include "hc2cb.h"
cannam@95: 
cannam@95: static void hc2cbdft2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms)
cannam@95: {
cannam@95:      DK(KP707106781, +0.707106781186547524400844362104849039284835938);
cannam@95:      {
cannam@95: 	  INT m;
cannam@95: 	  for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) {
cannam@95: 	       E T7, T1d, T1h, Tl, TG, T14, T19, TO, Te, TL, T18, T15, TB, T1e, Tw;
cannam@95: 	       E T1i;
cannam@95: 	       {
cannam@95: 		    E T3, TC, Tk, TM, T6, Th, TF, TN;
cannam@95: 		    {
cannam@95: 			 E T1, T2, Ti, Tj;
cannam@95: 			 T1 = Rp[0];
cannam@95: 			 T2 = Rm[WS(rs, 3)];
cannam@95: 			 T3 = T1 + T2;
cannam@95: 			 TC = T1 - T2;
cannam@95: 			 Ti = Ip[0];
cannam@95: 			 Tj = Im[WS(rs, 3)];
cannam@95: 			 Tk = Ti + Tj;
cannam@95: 			 TM = Ti - Tj;
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E T4, T5, TD, TE;
cannam@95: 			 T4 = Rp[WS(rs, 2)];
cannam@95: 			 T5 = Rm[WS(rs, 1)];
cannam@95: 			 T6 = T4 + T5;
cannam@95: 			 Th = T4 - T5;
cannam@95: 			 TD = Ip[WS(rs, 2)];
cannam@95: 			 TE = Im[WS(rs, 1)];
cannam@95: 			 TF = TD + TE;
cannam@95: 			 TN = TD - TE;
cannam@95: 		    }
cannam@95: 		    T7 = T3 + T6;
cannam@95: 		    T1d = Tk - Th;
cannam@95: 		    T1h = TC + TF;
cannam@95: 		    Tl = Th + Tk;
cannam@95: 		    TG = TC - TF;
cannam@95: 		    T14 = T3 - T6;
cannam@95: 		    T19 = TM - TN;
cannam@95: 		    TO = TM + TN;
cannam@95: 	       }
cannam@95: 	       {
cannam@95: 		    E Ta, Tm, Tp, TJ, Td, Tr, Tu, TK;
cannam@95: 		    {
cannam@95: 			 E T8, T9, Tn, To;
cannam@95: 			 T8 = Rp[WS(rs, 1)];
cannam@95: 			 T9 = Rm[WS(rs, 2)];
cannam@95: 			 Ta = T8 + T9;
cannam@95: 			 Tm = T8 - T9;
cannam@95: 			 Tn = Ip[WS(rs, 1)];
cannam@95: 			 To = Im[WS(rs, 2)];
cannam@95: 			 Tp = Tn + To;
cannam@95: 			 TJ = Tn - To;
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E Tb, Tc, Ts, Tt;
cannam@95: 			 Tb = Rm[0];
cannam@95: 			 Tc = Rp[WS(rs, 3)];
cannam@95: 			 Td = Tb + Tc;
cannam@95: 			 Tr = Tb - Tc;
cannam@95: 			 Ts = Im[0];
cannam@95: 			 Tt = Ip[WS(rs, 3)];
cannam@95: 			 Tu = Ts + Tt;
cannam@95: 			 TK = Tt - Ts;
cannam@95: 		    }
cannam@95: 		    Te = Ta + Td;
cannam@95: 		    TL = TJ + TK;
cannam@95: 		    T18 = Ta - Td;
cannam@95: 		    T15 = TK - TJ;
cannam@95: 		    {
cannam@95: 			 E Tz, TA, Tq, Tv;
cannam@95: 			 Tz = Tm - Tp;
cannam@95: 			 TA = Tr - Tu;
cannam@95: 			 TB = KP707106781 * (Tz + TA);
cannam@95: 			 T1e = KP707106781 * (Tz - TA);
cannam@95: 			 Tq = Tm + Tp;
cannam@95: 			 Tv = Tr + Tu;
cannam@95: 			 Tw = KP707106781 * (Tq - Tv);
cannam@95: 			 T1i = KP707106781 * (Tq + Tv);
cannam@95: 		    }
cannam@95: 	       }
cannam@95: 	       {
cannam@95: 		    E Tf, TP, TI, TQ;
cannam@95: 		    Tf = T7 + Te;
cannam@95: 		    TP = TL + TO;
cannam@95: 		    {
cannam@95: 			 E Tx, TH, Tg, Ty;
cannam@95: 			 Tx = Tl + Tw;
cannam@95: 			 TH = TB + TG;
cannam@95: 			 Tg = W[0];
cannam@95: 			 Ty = W[1];
cannam@95: 			 TI = FMA(Tg, Tx, Ty * TH);
cannam@95: 			 TQ = FNMS(Ty, Tx, Tg * TH);
cannam@95: 		    }
cannam@95: 		    Rp[0] = Tf - TI;
cannam@95: 		    Ip[0] = TP + TQ;
cannam@95: 		    Rm[0] = Tf + TI;
cannam@95: 		    Im[0] = TQ - TP;
cannam@95: 	       }
cannam@95: 	       {
cannam@95: 		    E T1r, T1x, T1w, T1y;
cannam@95: 		    {
cannam@95: 			 E T1o, T1q, T1n, T1p;
cannam@95: 			 T1o = T14 - T15;
cannam@95: 			 T1q = T19 - T18;
cannam@95: 			 T1n = W[10];
cannam@95: 			 T1p = W[11];
cannam@95: 			 T1r = FNMS(T1p, T1q, T1n * T1o);
cannam@95: 			 T1x = FMA(T1p, T1o, T1n * T1q);
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E T1t, T1v, T1s, T1u;
cannam@95: 			 T1t = T1d - T1e;
cannam@95: 			 T1v = T1i + T1h;
cannam@95: 			 T1s = W[12];
cannam@95: 			 T1u = W[13];
cannam@95: 			 T1w = FMA(T1s, T1t, T1u * T1v);
cannam@95: 			 T1y = FNMS(T1u, T1t, T1s * T1v);
cannam@95: 		    }
cannam@95: 		    Rp[WS(rs, 3)] = T1r - T1w;
cannam@95: 		    Ip[WS(rs, 3)] = T1x + T1y;
cannam@95: 		    Rm[WS(rs, 3)] = T1r + T1w;
cannam@95: 		    Im[WS(rs, 3)] = T1y - T1x;
cannam@95: 	       }
cannam@95: 	       {
cannam@95: 		    E TV, T11, T10, T12;
cannam@95: 		    {
cannam@95: 			 E TS, TU, TR, TT;
cannam@95: 			 TS = T7 - Te;
cannam@95: 			 TU = TO - TL;
cannam@95: 			 TR = W[6];
cannam@95: 			 TT = W[7];
cannam@95: 			 TV = FNMS(TT, TU, TR * TS);
cannam@95: 			 T11 = FMA(TT, TS, TR * TU);
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E TX, TZ, TW, TY;
cannam@95: 			 TX = Tl - Tw;
cannam@95: 			 TZ = TG - TB;
cannam@95: 			 TW = W[8];
cannam@95: 			 TY = W[9];
cannam@95: 			 T10 = FMA(TW, TX, TY * TZ);
cannam@95: 			 T12 = FNMS(TY, TX, TW * TZ);
cannam@95: 		    }
cannam@95: 		    Rp[WS(rs, 2)] = TV - T10;
cannam@95: 		    Ip[WS(rs, 2)] = T11 + T12;
cannam@95: 		    Rm[WS(rs, 2)] = TV + T10;
cannam@95: 		    Im[WS(rs, 2)] = T12 - T11;
cannam@95: 	       }
cannam@95: 	       {
cannam@95: 		    E T1b, T1l, T1k, T1m;
cannam@95: 		    {
cannam@95: 			 E T16, T1a, T13, T17;
cannam@95: 			 T16 = T14 + T15;
cannam@95: 			 T1a = T18 + T19;
cannam@95: 			 T13 = W[2];
cannam@95: 			 T17 = W[3];
cannam@95: 			 T1b = FNMS(T17, T1a, T13 * T16);
cannam@95: 			 T1l = FMA(T17, T16, T13 * T1a);
cannam@95: 		    }
cannam@95: 		    {
cannam@95: 			 E T1f, T1j, T1c, T1g;
cannam@95: 			 T1f = T1d + T1e;
cannam@95: 			 T1j = T1h - T1i;
cannam@95: 			 T1c = W[4];
cannam@95: 			 T1g = W[5];
cannam@95: 			 T1k = FMA(T1c, T1f, T1g * T1j);
cannam@95: 			 T1m = FNMS(T1g, T1f, T1c * T1j);
cannam@95: 		    }
cannam@95: 		    Rp[WS(rs, 1)] = T1b - T1k;
cannam@95: 		    Ip[WS(rs, 1)] = T1l + T1m;
cannam@95: 		    Rm[WS(rs, 1)] = T1b + T1k;
cannam@95: 		    Im[WS(rs, 1)] = T1m - T1l;
cannam@95: 	       }
cannam@95: 	  }
cannam@95:      }
cannam@95: }
cannam@95: 
cannam@95: static const tw_instr twinstr[] = {
cannam@95:      {TW_FULL, 1, 8},
cannam@95:      {TW_NEXT, 1, 0}
cannam@95: };
cannam@95: 
cannam@95: static const hc2c_desc desc = { 8, "hc2cbdft2_8", twinstr, &GENUS, {68, 18, 14, 0} };
cannam@95: 
cannam@95: void X(codelet_hc2cbdft2_8) (planner *p) {
cannam@95:      X(khc2c_register) (p, hc2cbdft2_8, &desc, HC2C_VIA_DFT);
cannam@95: }
cannam@95: #endif				/* HAVE_FMA */