/* * Copyright (c) 2003, 2007-8 Matteo Frigo * Copyright (c) 2003, 2007-8 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 Sun Jul 12 06:43:14 EDT 2009 */ #include "codelet-dft.h" #ifdef HAVE_FMA /* Generated by: ../../../genfft/gen_twiddle -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 4 -name t2sv_4 -include ts.h */ /* * This function contains 24 FP additions, 16 FP multiplications, * (or, 16 additions, 8 multiplications, 8 fused multiply/add), * 37 stack variables, 0 constants, and 16 memory accesses */ #include "ts.h" static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) { INT m; for (m = mb, W = W + (mb * 4); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 4), MAKE_VOLATILE_STRIDE(rs)) { V T2, T6, T3, T5, T1, Tx, T8, Tc, Tf, Ta, T4, Th, Tj, Tl; T2 = LDW(&(W[0])); T6 = LDW(&(W[TWVL * 3])); T3 = LDW(&(W[TWVL * 2])); T5 = LDW(&(W[TWVL * 1])); T1 = LD(&(ri[0]), ms, &(ri[0])); Tx = LD(&(ii[0]), ms, &(ii[0])); T8 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Tc = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Tf = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Ta = VMUL(T2, T6); T4 = VMUL(T2, T3); Th = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Tj = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Tl = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); { V Tg, Tb, T7, Tp, Tk, Tr, Ti; Tg = VMUL(T2, Tf); Tb = VFNMS(T5, T3, Ta); T7 = VFMA(T5, T6, T4); Tp = VMUL(T2, Th); Tk = VMUL(T3, Tj); Tr = VMUL(T3, Tl); Ti = VFMA(T5, Th, Tg); { V Tv, T9, Tq, Tm, Ts, Tw, Td; Tv = VMUL(T7, Tc); T9 = VMUL(T7, T8); Tq = VFNMS(T5, Tf, Tp); Tm = VFMA(T6, Tl, Tk); Ts = VFNMS(T6, Tj, Tr); Tw = VFNMS(Tb, T8, Tv); Td = VFMA(Tb, Tc, T9); { V Tn, TA, Tu, Tt; Tn = VADD(Ti, Tm); TA = VSUB(Ti, Tm); Tu = VADD(Tq, Ts); Tt = VSUB(Tq, Ts); { V Ty, Tz, Te, To; Ty = VADD(Tw, Tx); Tz = VSUB(Tx, Tw); Te = VADD(T1, Td); To = VSUB(T1, Td); ST(&(ii[WS(rs, 3)]), VADD(TA, Tz), ms, &(ii[WS(rs, 1)])); ST(&(ii[WS(rs, 1)]), VSUB(Tz, TA), ms, &(ii[WS(rs, 1)])); ST(&(ii[WS(rs, 2)]), VSUB(Ty, Tu), ms, &(ii[0])); ST(&(ii[0]), VADD(Tu, Ty), ms, &(ii[0])); ST(&(ri[WS(rs, 1)]), VADD(To, Tt), ms, &(ri[WS(rs, 1)])); ST(&(ri[WS(rs, 3)]), VSUB(To, Tt), ms, &(ri[WS(rs, 1)])); ST(&(ri[0]), VADD(Te, Tn), ms, &(ri[0])); ST(&(ri[WS(rs, 2)]), VSUB(Te, Tn), ms, &(ri[0])); } } } } } } static const tw_instr twinstr[] = { VTW(0, 1), VTW(0, 3), {TW_NEXT, (2 * VL), 0} }; static const ct_desc desc = { 4, "t2sv_4", twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; void X(codelet_t2sv_4) (planner *p) { X(kdft_dit_register) (p, t2sv_4, &desc); } #else /* HAVE_FMA */ /* Generated by: ../../../genfft/gen_twiddle -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 4 -name t2sv_4 -include ts.h */ /* * This function contains 24 FP additions, 16 FP multiplications, * (or, 16 additions, 8 multiplications, 8 fused multiply/add), * 21 stack variables, 0 constants, and 16 memory accesses */ #include "ts.h" static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) { INT m; for (m = mb, W = W + (mb * 4); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 4), MAKE_VOLATILE_STRIDE(rs)) { V T2, T4, T3, T5, T6, T8; T2 = LDW(&(W[0])); T4 = LDW(&(W[TWVL * 1])); T3 = LDW(&(W[TWVL * 2])); T5 = LDW(&(W[TWVL * 3])); T6 = VFMA(T2, T3, VMUL(T4, T5)); T8 = VFNMS(T4, T3, VMUL(T2, T5)); { V T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; T1 = LD(&(ri[0]), ms, &(ri[0])); Tp = LD(&(ii[0]), ms, &(ii[0])); T7 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); T9 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Ta = VFMA(T6, T7, VMUL(T8, T9)); To = VFNMS(T8, T7, VMUL(T6, T9)); { V Tc, Td, Tf, Tg; Tc = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Td = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Te = VFMA(T2, Tc, VMUL(T4, Td)); Tk = VFNMS(T4, Tc, VMUL(T2, Td)); Tf = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Tg = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Th = VFMA(T3, Tf, VMUL(T5, Tg)); Tl = VFNMS(T5, Tf, VMUL(T3, Tg)); } { V Tb, Ti, Tn, Tq; Tb = VADD(T1, Ta); Ti = VADD(Te, Th); ST(&(ri[WS(rs, 2)]), VSUB(Tb, Ti), ms, &(ri[0])); ST(&(ri[0]), VADD(Tb, Ti), ms, &(ri[0])); Tn = VADD(Tk, Tl); Tq = VADD(To, Tp); ST(&(ii[0]), VADD(Tn, Tq), ms, &(ii[0])); ST(&(ii[WS(rs, 2)]), VSUB(Tq, Tn), ms, &(ii[0])); } { V Tj, Tm, Tr, Ts; Tj = VSUB(T1, Ta); Tm = VSUB(Tk, Tl); ST(&(ri[WS(rs, 3)]), VSUB(Tj, Tm), ms, &(ri[WS(rs, 1)])); ST(&(ri[WS(rs, 1)]), VADD(Tj, Tm), ms, &(ri[WS(rs, 1)])); Tr = VSUB(Tp, To); Ts = VSUB(Te, Th); ST(&(ii[WS(rs, 1)]), VSUB(Tr, Ts), ms, &(ii[WS(rs, 1)])); ST(&(ii[WS(rs, 3)]), VADD(Ts, Tr), ms, &(ii[WS(rs, 1)])); } } } } static const tw_instr twinstr[] = { VTW(0, 1), VTW(0, 3), {TW_NEXT, (2 * VL), 0} }; static const ct_desc desc = { 4, "t2sv_4", twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; void X(codelet_t2sv_4) (planner *p) { X(kdft_dit_register) (p, t2sv_4, &desc); } #endif /* HAVE_FMA */