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 07:38:00 EST 2012 */
Chris@10: 
Chris@10: #include "codelet-dft.h"
Chris@10: 
Chris@10: #ifdef HAVE_FMA
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1fuv_8 -include t1fu.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 33 FP additions, 24 FP multiplications,
Chris@10:  * (or, 23 additions, 14 multiplications, 10 fused multiply/add),
Chris@10:  * 36 stack variables, 1 constants, and 16 memory accesses
Chris@10:  */
Chris@10: #include "t1fu.h"
Chris@10: 
Chris@10: static void t1fuv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  R *x;
Chris@10: 	  x = ri;
Chris@10: 	  for (m = mb, W = W + (mb * ((TWVL / VL) * 14)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(8, rs)) {
Chris@10: 	       V T1, T2, Th, Tj, T5, T7, Ta, Tc;
Chris@10: 	       T1 = LD(&(x[0]), ms, &(x[0]));
Chris@10: 	       T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
Chris@10: 	       Th = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
Chris@10: 	       Tj = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
Chris@10: 	       T5 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
Chris@10: 	       T7 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
Chris@10: 	       Ta = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
Chris@10: 	       Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
Chris@10: 	       {
Chris@10: 		    V T3, Ti, Tk, T6, T8, Tb, Td;
Chris@10: 		    T3 = BYTWJ(&(W[TWVL * 6]), T2);
Chris@10: 		    Ti = BYTWJ(&(W[TWVL * 2]), Th);
Chris@10: 		    Tk = BYTWJ(&(W[TWVL * 10]), Tj);
Chris@10: 		    T6 = BYTWJ(&(W[0]), T5);
Chris@10: 		    T8 = BYTWJ(&(W[TWVL * 8]), T7);
Chris@10: 		    Tb = BYTWJ(&(W[TWVL * 12]), Ta);
Chris@10: 		    Td = BYTWJ(&(W[TWVL * 4]), Tc);
Chris@10: 		    {
Chris@10: 			 V Tq, T4, Tr, Tl, Tt, T9, Tu, Te, Tw, Ts;
Chris@10: 			 Tq = VADD(T1, T3);
Chris@10: 			 T4 = VSUB(T1, T3);
Chris@10: 			 Tr = VADD(Ti, Tk);
Chris@10: 			 Tl = VSUB(Ti, Tk);
Chris@10: 			 Tt = VADD(T6, T8);
Chris@10: 			 T9 = VSUB(T6, T8);
Chris@10: 			 Tu = VADD(Tb, Td);
Chris@10: 			 Te = VSUB(Tb, Td);
Chris@10: 			 Tw = VSUB(Tq, Tr);
Chris@10: 			 Ts = VADD(Tq, Tr);
Chris@10: 			 {
Chris@10: 			      V Tx, Tv, Tm, Tf;
Chris@10: 			      Tx = VSUB(Tu, Tt);
Chris@10: 			      Tv = VADD(Tt, Tu);
Chris@10: 			      Tm = VSUB(Te, T9);
Chris@10: 			      Tf = VADD(T9, Te);
Chris@10: 			      {
Chris@10: 				   V Tp, Tn, To, Tg;
Chris@10: 				   ST(&(x[WS(rs, 2)]), VFMAI(Tx, Tw), ms, &(x[0]));
Chris@10: 				   ST(&(x[WS(rs, 6)]), VFNMSI(Tx, Tw), ms, &(x[0]));
Chris@10: 				   ST(&(x[0]), VADD(Ts, Tv), ms, &(x[0]));
Chris@10: 				   ST(&(x[WS(rs, 4)]), VSUB(Ts, Tv), ms, &(x[0]));
Chris@10: 				   Tp = VFMA(LDK(KP707106781), Tm, Tl);
Chris@10: 				   Tn = VFNMS(LDK(KP707106781), Tm, Tl);
Chris@10: 				   To = VFNMS(LDK(KP707106781), Tf, T4);
Chris@10: 				   Tg = VFMA(LDK(KP707106781), Tf, T4);
Chris@10: 				   ST(&(x[WS(rs, 5)]), VFNMSI(Tp, To), ms, &(x[WS(rs, 1)]));
Chris@10: 				   ST(&(x[WS(rs, 3)]), VFMAI(Tp, To), ms, &(x[WS(rs, 1)]));
Chris@10: 				   ST(&(x[WS(rs, 7)]), VFMAI(Tn, Tg), ms, &(x[WS(rs, 1)]));
Chris@10: 				   ST(&(x[WS(rs, 1)]), VFNMSI(Tn, Tg), ms, &(x[WS(rs, 1)]));
Chris@10: 			      }
Chris@10: 			 }
Chris@10: 		    }
Chris@10: 	       }
Chris@10: 	  }
Chris@10:      }
Chris@10:      VLEAVE();
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      VTW(0, 1),
Chris@10:      VTW(0, 2),
Chris@10:      VTW(0, 3),
Chris@10:      VTW(0, 4),
Chris@10:      VTW(0, 5),
Chris@10:      VTW(0, 6),
Chris@10:      VTW(0, 7),
Chris@10:      {TW_NEXT, VL, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 8, XSIMD_STRING("t1fuv_8"), twinstr, &GENUS, {23, 14, 10, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void XSIMD(codelet_t1fuv_8) (planner *p) {
Chris@10:      X(kdft_dit_register) (p, t1fuv_8, &desc);
Chris@10: }
Chris@10: #else				/* HAVE_FMA */
Chris@10: 
Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1fuv_8 -include t1fu.h */
Chris@10: 
Chris@10: /*
Chris@10:  * This function contains 33 FP additions, 16 FP multiplications,
Chris@10:  * (or, 33 additions, 16 multiplications, 0 fused multiply/add),
Chris@10:  * 24 stack variables, 1 constants, and 16 memory accesses
Chris@10:  */
Chris@10: #include "t1fu.h"
Chris@10: 
Chris@10: static void t1fuv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms)
Chris@10: {
Chris@10:      DVK(KP707106781, +0.707106781186547524400844362104849039284835938);
Chris@10:      {
Chris@10: 	  INT m;
Chris@10: 	  R *x;
Chris@10: 	  x = ri;
Chris@10: 	  for (m = mb, W = W + (mb * ((TWVL / VL) * 14)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(8, rs)) {
Chris@10: 	       V T4, Tq, Tm, Tr, T9, Tt, Te, Tu, T1, T3, T2;
Chris@10: 	       T1 = LD(&(x[0]), ms, &(x[0]));
Chris@10: 	       T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0]));
Chris@10: 	       T3 = BYTWJ(&(W[TWVL * 6]), T2);
Chris@10: 	       T4 = VSUB(T1, T3);
Chris@10: 	       Tq = VADD(T1, T3);
Chris@10: 	       {
Chris@10: 		    V Tj, Tl, Ti, Tk;
Chris@10: 		    Ti = LD(&(x[WS(rs, 2)]), ms, &(x[0]));
Chris@10: 		    Tj = BYTWJ(&(W[TWVL * 2]), Ti);
Chris@10: 		    Tk = LD(&(x[WS(rs, 6)]), ms, &(x[0]));
Chris@10: 		    Tl = BYTWJ(&(W[TWVL * 10]), Tk);
Chris@10: 		    Tm = VSUB(Tj, Tl);
Chris@10: 		    Tr = VADD(Tj, Tl);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    V T6, T8, T5, T7;
Chris@10: 		    T5 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)]));
Chris@10: 		    T6 = BYTWJ(&(W[0]), T5);
Chris@10: 		    T7 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)]));
Chris@10: 		    T8 = BYTWJ(&(W[TWVL * 8]), T7);
Chris@10: 		    T9 = VSUB(T6, T8);
Chris@10: 		    Tt = VADD(T6, T8);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    V Tb, Td, Ta, Tc;
Chris@10: 		    Ta = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)]));
Chris@10: 		    Tb = BYTWJ(&(W[TWVL * 12]), Ta);
Chris@10: 		    Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)]));
Chris@10: 		    Td = BYTWJ(&(W[TWVL * 4]), Tc);
Chris@10: 		    Te = VSUB(Tb, Td);
Chris@10: 		    Tu = VADD(Tb, Td);
Chris@10: 	       }
Chris@10: 	       {
Chris@10: 		    V Ts, Tv, Tw, Tx;
Chris@10: 		    Ts = VADD(Tq, Tr);
Chris@10: 		    Tv = VADD(Tt, Tu);
Chris@10: 		    ST(&(x[WS(rs, 4)]), VSUB(Ts, Tv), ms, &(x[0]));
Chris@10: 		    ST(&(x[0]), VADD(Ts, Tv), ms, &(x[0]));
Chris@10: 		    Tw = VSUB(Tq, Tr);
Chris@10: 		    Tx = VBYI(VSUB(Tu, Tt));
Chris@10: 		    ST(&(x[WS(rs, 6)]), VSUB(Tw, Tx), ms, &(x[0]));
Chris@10: 		    ST(&(x[WS(rs, 2)]), VADD(Tw, Tx), ms, &(x[0]));
Chris@10: 		    {
Chris@10: 			 V Tg, To, Tn, Tp, Tf, Th;
Chris@10: 			 Tf = VMUL(LDK(KP707106781), VADD(T9, Te));
Chris@10: 			 Tg = VADD(T4, Tf);
Chris@10: 			 To = VSUB(T4, Tf);
Chris@10: 			 Th = VMUL(LDK(KP707106781), VSUB(Te, T9));
Chris@10: 			 Tn = VBYI(VSUB(Th, Tm));
Chris@10: 			 Tp = VBYI(VADD(Tm, Th));
Chris@10: 			 ST(&(x[WS(rs, 7)]), VSUB(Tg, Tn), ms, &(x[WS(rs, 1)]));
Chris@10: 			 ST(&(x[WS(rs, 3)]), VADD(To, Tp), ms, &(x[WS(rs, 1)]));
Chris@10: 			 ST(&(x[WS(rs, 1)]), VADD(Tg, Tn), ms, &(x[WS(rs, 1)]));
Chris@10: 			 ST(&(x[WS(rs, 5)]), VSUB(To, Tp), ms, &(x[WS(rs, 1)]));
Chris@10: 		    }
Chris@10: 	       }
Chris@10: 	  }
Chris@10:      }
Chris@10:      VLEAVE();
Chris@10: }
Chris@10: 
Chris@10: static const tw_instr twinstr[] = {
Chris@10:      VTW(0, 1),
Chris@10:      VTW(0, 2),
Chris@10:      VTW(0, 3),
Chris@10:      VTW(0, 4),
Chris@10:      VTW(0, 5),
Chris@10:      VTW(0, 6),
Chris@10:      VTW(0, 7),
Chris@10:      {TW_NEXT, VL, 0}
Chris@10: };
Chris@10: 
Chris@10: static const ct_desc desc = { 8, XSIMD_STRING("t1fuv_8"), twinstr, &GENUS, {33, 16, 0, 0}, 0, 0, 0 };
Chris@10: 
Chris@10: void XSIMD(codelet_t1fuv_8) (planner *p) {
Chris@10:      X(kdft_dit_register) (p, t1fuv_8, &desc);
Chris@10: }
Chris@10: #endif				/* HAVE_FMA */