/* * Copyright (c) 2003, 2006 Matteo Frigo * Copyright (c) 2003, 2006 Massachusetts Institute of Technology * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * */ /* This file was automatically generated --- DO NOT EDIT */ /* Generated on Sat Jul 1 22:36:34 EDT 2006 */ #include "codelet-dft.h" #ifdef HAVE_FMA /* Generated by: ../../../genfft/gen_twidsq_c -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include q1f.h */ /* * This function contains 44 FP additions, 32 FP multiplications, * (or, 36 additions, 24 multiplications, 8 fused multiply/add), * 38 stack variables, and 32 memory accesses */ /* * Generator Id's : * $Id: algsimp.ml,v 1.9 2006-02-12 23:34:12 athena Exp $ * $Id: fft.ml,v 1.4 2006-01-05 03:04:27 stevenj Exp $ * $Id: gen_twidsq_c.ml,v 1.8 2006-02-12 23:34:12 athena Exp $ */ #include "q1f.h" static const R *q1fv_4(R *ri, R *ii, const R *W, stride is, stride vs, INT m, INT dist) { INT i; R *x; x = ri; for (i = 0; i < m; i = i + VL, x = x + (VL * dist), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(is), MAKE_VOLATILE_STRIDE(vs)) { V Tb, Tm, Tx, TI; { V Tc, T9, T3, TG, TA, TH, TD, Ta, T6, Td, Tn, To, Tq, Tr, Tf; V Tg; { V T1, T2, Ty, Tz, TB, TC, T4, T5; T1 = LD(&(x[0]), dist, &(x[0])); T2 = LD(&(x[WS(is, 2)]), dist, &(x[0])); Ty = LD(&(x[WS(vs, 3)]), dist, &(x[WS(vs, 3)])); Tz = LD(&(x[WS(vs, 3) + WS(is, 2)]), dist, &(x[WS(vs, 3)])); TB = LD(&(x[WS(vs, 3) + WS(is, 1)]), dist, &(x[WS(vs, 3) + WS(is, 1)])); TC = LD(&(x[WS(vs, 3) + WS(is, 3)]), dist, &(x[WS(vs, 3) + WS(is, 1)])); T4 = LD(&(x[WS(is, 1)]), dist, &(x[WS(is, 1)])); T5 = LD(&(x[WS(is, 3)]), dist, &(x[WS(is, 1)])); Tc = LD(&(x[WS(vs, 1)]), dist, &(x[WS(vs, 1)])); T9 = VADD(T1, T2); T3 = VSUB(T1, T2); TG = VADD(Ty, Tz); TA = VSUB(Ty, Tz); TH = VADD(TB, TC); TD = VSUB(TB, TC); Ta = VADD(T4, T5); T6 = VSUB(T4, T5); Td = LD(&(x[WS(vs, 1) + WS(is, 2)]), dist, &(x[WS(vs, 1)])); Tn = LD(&(x[WS(vs, 2)]), dist, &(x[WS(vs, 2)])); To = LD(&(x[WS(vs, 2) + WS(is, 2)]), dist, &(x[WS(vs, 2)])); Tq = LD(&(x[WS(vs, 2) + WS(is, 1)]), dist, &(x[WS(vs, 2) + WS(is, 1)])); Tr = LD(&(x[WS(vs, 2) + WS(is, 3)]), dist, &(x[WS(vs, 2) + WS(is, 1)])); Tf = LD(&(x[WS(vs, 1) + WS(is, 1)]), dist, &(x[WS(vs, 1) + WS(is, 1)])); Tg = LD(&(x[WS(vs, 1) + WS(is, 3)]), dist, &(x[WS(vs, 1) + WS(is, 1)])); } { V Tk, Te, Tv, Tp, Tw, Ts, Tl, Th, T7, TE, Tu, TF; ST(&(x[0]), VADD(T9, Ta), dist, &(x[0])); Tk = VADD(Tc, Td); Te = VSUB(Tc, Td); Tv = VADD(Tn, To); Tp = VSUB(Tn, To); Tw = VADD(Tq, Tr); Ts = VSUB(Tq, Tr); Tl = VADD(Tf, Tg); Th = VSUB(Tf, Tg); ST(&(x[WS(is, 3)]), VADD(TG, TH), dist, &(x[WS(is, 1)])); T7 = BYTWJ(&(W[0]), VFNMSI(T6, T3)); TE = BYTWJ(&(W[0]), VFNMSI(TD, TA)); { V Tt, Ti, Tj, T8; T8 = BYTWJ(&(W[TWVL * 4]), VFMAI(T6, T3)); ST(&(x[WS(is, 2)]), VADD(Tv, Tw), dist, &(x[0])); Tt = BYTWJ(&(W[0]), VFNMSI(Ts, Tp)); ST(&(x[WS(is, 1)]), VADD(Tk, Tl), dist, &(x[WS(is, 1)])); Ti = BYTWJ(&(W[0]), VFNMSI(Th, Te)); Tj = BYTWJ(&(W[TWVL * 4]), VFMAI(Th, Te)); ST(&(x[WS(vs, 1)]), T7, dist, &(x[WS(vs, 1)])); ST(&(x[WS(vs, 1) + WS(is, 3)]), TE, dist, &(x[WS(vs, 1) + WS(is, 1)])); ST(&(x[WS(vs, 3)]), T8, dist, &(x[WS(vs, 3)])); Tu = BYTWJ(&(W[TWVL * 4]), VFMAI(Ts, Tp)); ST(&(x[WS(vs, 1) + WS(is, 2)]), Tt, dist, &(x[WS(vs, 1)])); TF = BYTWJ(&(W[TWVL * 4]), VFMAI(TD, TA)); ST(&(x[WS(vs, 1) + WS(is, 1)]), Ti, dist, &(x[WS(vs, 1) + WS(is, 1)])); ST(&(x[WS(vs, 3) + WS(is, 1)]), Tj, dist, &(x[WS(vs, 3) + WS(is, 1)])); } Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); ST(&(x[WS(vs, 3) + WS(is, 2)]), Tu, dist, &(x[WS(vs, 3)])); TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); ST(&(x[WS(vs, 3) + WS(is, 3)]), TF, dist, &(x[WS(vs, 3) + WS(is, 1)])); } } ST(&(x[WS(vs, 2)]), Tb, dist, &(x[WS(vs, 2)])); ST(&(x[WS(vs, 2) + WS(is, 1)]), Tm, dist, &(x[WS(vs, 2) + WS(is, 1)])); ST(&(x[WS(vs, 2) + WS(is, 2)]), Tx, dist, &(x[WS(vs, 2)])); ST(&(x[WS(vs, 2) + WS(is, 3)]), TI, dist, &(x[WS(vs, 2) + WS(is, 1)])); } return W; } static const tw_instr twinstr[] = { VTW(1), VTW(2), VTW(3), {TW_NEXT, VL, 0} }; static const ct_desc desc = { 4, "q1fv_4", twinstr, &GENUS, {36, 24, 8, 0}, 0, 0, 0 }; void X(codelet_q1fv_4) (planner *p) { X(kdft_difsq_register) (p, q1fv_4, &desc); } #else /* HAVE_FMA */ /* Generated by: ../../../genfft/gen_twidsq_c -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include q1f.h */ /* * This function contains 44 FP additions, 24 FP multiplications, * (or, 44 additions, 24 multiplications, 0 fused multiply/add), * 22 stack variables, and 32 memory accesses */ /* * Generator Id's : * $Id: algsimp.ml,v 1.9 2006-02-12 23:34:12 athena Exp $ * $Id: fft.ml,v 1.4 2006-01-05 03:04:27 stevenj Exp $ * $Id: gen_twidsq_c.ml,v 1.8 2006-02-12 23:34:12 athena Exp $ */ #include "q1f.h" static const R *q1fv_4(R *ri, R *ii, const R *W, stride is, stride vs, INT m, INT dist) { INT i; R *x; x = ri; for (i = 0; i < m; i = i + VL, x = x + (VL * dist), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(is), MAKE_VOLATILE_STRIDE(vs)) { V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; V Tl; { V T1, T2, Ty, Tz; T1 = LD(&(x[0]), dist, &(x[0])); T2 = LD(&(x[WS(is, 2)]), dist, &(x[0])); T3 = VSUB(T1, T2); T9 = VADD(T1, T2); Ty = LD(&(x[WS(vs, 3)]), dist, &(x[WS(vs, 3)])); Tz = LD(&(x[WS(vs, 3) + WS(is, 2)]), dist, &(x[WS(vs, 3)])); TA = VSUB(Ty, Tz); TG = VADD(Ty, Tz); } { V TB, TC, T4, T5; TB = LD(&(x[WS(vs, 3) + WS(is, 1)]), dist, &(x[WS(vs, 3) + WS(is, 1)])); TC = LD(&(x[WS(vs, 3) + WS(is, 3)]), dist, &(x[WS(vs, 3) + WS(is, 1)])); TD = VBYI(VSUB(TB, TC)); TH = VADD(TB, TC); T4 = LD(&(x[WS(is, 1)]), dist, &(x[WS(is, 1)])); T5 = LD(&(x[WS(is, 3)]), dist, &(x[WS(is, 1)])); T6 = VBYI(VSUB(T4, T5)); Ta = VADD(T4, T5); } { V Tc, Td, Tn, To; Tc = LD(&(x[WS(vs, 1)]), dist, &(x[WS(vs, 1)])); Td = LD(&(x[WS(vs, 1) + WS(is, 2)]), dist, &(x[WS(vs, 1)])); Te = VSUB(Tc, Td); Tk = VADD(Tc, Td); Tn = LD(&(x[WS(vs, 2)]), dist, &(x[WS(vs, 2)])); To = LD(&(x[WS(vs, 2) + WS(is, 2)]), dist, &(x[WS(vs, 2)])); Tp = VSUB(Tn, To); Tv = VADD(Tn, To); } { V Tq, Tr, Tf, Tg; Tq = LD(&(x[WS(vs, 2) + WS(is, 1)]), dist, &(x[WS(vs, 2) + WS(is, 1)])); Tr = LD(&(x[WS(vs, 2) + WS(is, 3)]), dist, &(x[WS(vs, 2) + WS(is, 1)])); Ts = VBYI(VSUB(Tq, Tr)); Tw = VADD(Tq, Tr); Tf = LD(&(x[WS(vs, 1) + WS(is, 1)]), dist, &(x[WS(vs, 1) + WS(is, 1)])); Tg = LD(&(x[WS(vs, 1) + WS(is, 3)]), dist, &(x[WS(vs, 1) + WS(is, 1)])); Th = VBYI(VSUB(Tf, Tg)); Tl = VADD(Tf, Tg); } ST(&(x[0]), VADD(T9, Ta), dist, &(x[0])); ST(&(x[WS(is, 1)]), VADD(Tk, Tl), dist, &(x[WS(is, 1)])); ST(&(x[WS(is, 2)]), VADD(Tv, Tw), dist, &(x[0])); ST(&(x[WS(is, 3)]), VADD(TG, TH), dist, &(x[WS(is, 1)])); { V T7, Ti, Tt, TE; T7 = BYTWJ(&(W[0]), VSUB(T3, T6)); ST(&(x[WS(vs, 1)]), T7, dist, &(x[WS(vs, 1)])); Ti = BYTWJ(&(W[0]), VSUB(Te, Th)); ST(&(x[WS(vs, 1) + WS(is, 1)]), Ti, dist, &(x[WS(vs, 1) + WS(is, 1)])); Tt = BYTWJ(&(W[0]), VSUB(Tp, Ts)); ST(&(x[WS(vs, 1) + WS(is, 2)]), Tt, dist, &(x[WS(vs, 1)])); TE = BYTWJ(&(W[0]), VSUB(TA, TD)); ST(&(x[WS(vs, 1) + WS(is, 3)]), TE, dist, &(x[WS(vs, 1) + WS(is, 1)])); } { V T8, Tj, Tu, TF; T8 = BYTWJ(&(W[TWVL * 4]), VADD(T3, T6)); ST(&(x[WS(vs, 3)]), T8, dist, &(x[WS(vs, 3)])); Tj = BYTWJ(&(W[TWVL * 4]), VADD(Te, Th)); ST(&(x[WS(vs, 3) + WS(is, 1)]), Tj, dist, &(x[WS(vs, 3) + WS(is, 1)])); Tu = BYTWJ(&(W[TWVL * 4]), VADD(Tp, Ts)); ST(&(x[WS(vs, 3) + WS(is, 2)]), Tu, dist, &(x[WS(vs, 3)])); TF = BYTWJ(&(W[TWVL * 4]), VADD(TA, TD)); ST(&(x[WS(vs, 3) + WS(is, 3)]), TF, dist, &(x[WS(vs, 3) + WS(is, 1)])); } { V Tb, Tm, Tx, TI; Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); ST(&(x[WS(vs, 2)]), Tb, dist, &(x[WS(vs, 2)])); Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); ST(&(x[WS(vs, 2) + WS(is, 1)]), Tm, dist, &(x[WS(vs, 2) + WS(is, 1)])); Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); ST(&(x[WS(vs, 2) + WS(is, 2)]), Tx, dist, &(x[WS(vs, 2)])); TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); ST(&(x[WS(vs, 2) + WS(is, 3)]), TI, dist, &(x[WS(vs, 2) + WS(is, 1)])); } } return W; } static const tw_instr twinstr[] = { VTW(1), VTW(2), VTW(3), {TW_NEXT, VL, 0} }; static const ct_desc desc = { 4, "q1fv_4", twinstr, &GENUS, {44, 24, 0, 0}, 0, 0, 0 }; void X(codelet_q1fv_4) (planner *p) { X(kdft_difsq_register) (p, q1fv_4, &desc); } #endif /* HAVE_FMA */

Generated by Doxygen 1.6.0 Back to index