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 07:38:00 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 33 FP additions, 24 FP multiplications, cannam@95: * (or, 23 additions, 14 multiplications, 10 fused multiply/add), cannam@95: * 36 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "t1fu.h" cannam@95: cannam@95: static void t1fuv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: 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)) { cannam@95: V T1, T2, Th, Tj, T5, T7, Ta, Tc; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Th = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Tj = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: T5 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: T7 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Ta = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V T3, Ti, Tk, T6, T8, Tb, Td; cannam@95: T3 = BYTWJ(&(W[TWVL * 6]), T2); cannam@95: Ti = BYTWJ(&(W[TWVL * 2]), Th); cannam@95: Tk = BYTWJ(&(W[TWVL * 10]), Tj); cannam@95: T6 = BYTWJ(&(W[0]), T5); cannam@95: T8 = BYTWJ(&(W[TWVL * 8]), T7); cannam@95: Tb = BYTWJ(&(W[TWVL * 12]), Ta); cannam@95: Td = BYTWJ(&(W[TWVL * 4]), Tc); cannam@95: { cannam@95: V Tq, T4, Tr, Tl, Tt, T9, Tu, Te, Tw, Ts; cannam@95: Tq = VADD(T1, T3); cannam@95: T4 = VSUB(T1, T3); cannam@95: Tr = VADD(Ti, Tk); cannam@95: Tl = VSUB(Ti, Tk); cannam@95: Tt = VADD(T6, T8); cannam@95: T9 = VSUB(T6, T8); cannam@95: Tu = VADD(Tb, Td); cannam@95: Te = VSUB(Tb, Td); cannam@95: Tw = VSUB(Tq, Tr); cannam@95: Ts = VADD(Tq, Tr); cannam@95: { cannam@95: V Tx, Tv, Tm, Tf; cannam@95: Tx = VSUB(Tu, Tt); cannam@95: Tv = VADD(Tt, Tu); cannam@95: Tm = VSUB(Te, T9); cannam@95: Tf = VADD(T9, Te); cannam@95: { cannam@95: V Tp, Tn, To, Tg; cannam@95: ST(&(x[WS(rs, 2)]), VFMAI(Tx, Tw), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 6)]), VFNMSI(Tx, Tw), ms, &(x[0])); cannam@95: ST(&(x[0]), VADD(Ts, Tv), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 4)]), VSUB(Ts, Tv), ms, &(x[0])); cannam@95: Tp = VFMA(LDK(KP707106781), Tm, Tl); cannam@95: Tn = VFNMS(LDK(KP707106781), Tm, Tl); cannam@95: To = VFNMS(LDK(KP707106781), Tf, T4); cannam@95: Tg = VFMA(LDK(KP707106781), Tf, T4); cannam@95: ST(&(x[WS(rs, 5)]), VFNMSI(Tp, To), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 3)]), VFMAI(Tp, To), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VFMAI(Tn, Tg), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VFNMSI(Tn, Tg), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, XSIMD_STRING("t1fuv_8"), twinstr, &GENUS, {23, 14, 10, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fuv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fuv_8, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1fuv_8 -include t1fu.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 33 FP additions, 16 FP multiplications, cannam@95: * (or, 33 additions, 16 multiplications, 0 fused multiply/add), cannam@95: * 24 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "t1fu.h" cannam@95: cannam@95: static void t1fuv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: 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)) { cannam@95: V T4, Tq, Tm, Tr, T9, Tt, Te, Tu, T1, T3, T2; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: T3 = BYTWJ(&(W[TWVL * 6]), T2); cannam@95: T4 = VSUB(T1, T3); cannam@95: Tq = VADD(T1, T3); cannam@95: { cannam@95: V Tj, Tl, Ti, Tk; cannam@95: Ti = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Tj = BYTWJ(&(W[TWVL * 2]), Ti); cannam@95: Tk = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: Tl = BYTWJ(&(W[TWVL * 10]), Tk); cannam@95: Tm = VSUB(Tj, Tl); cannam@95: Tr = VADD(Tj, Tl); cannam@95: } cannam@95: { cannam@95: V T6, T8, T5, T7; cannam@95: T5 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: T6 = BYTWJ(&(W[0]), T5); cannam@95: T7 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: T8 = BYTWJ(&(W[TWVL * 8]), T7); cannam@95: T9 = VSUB(T6, T8); cannam@95: Tt = VADD(T6, T8); cannam@95: } cannam@95: { cannam@95: V Tb, Td, Ta, Tc; cannam@95: Ta = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Tb = BYTWJ(&(W[TWVL * 12]), Ta); cannam@95: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: Td = BYTWJ(&(W[TWVL * 4]), Tc); cannam@95: Te = VSUB(Tb, Td); cannam@95: Tu = VADD(Tb, Td); cannam@95: } cannam@95: { cannam@95: V Ts, Tv, Tw, Tx; cannam@95: Ts = VADD(Tq, Tr); cannam@95: Tv = VADD(Tt, Tu); cannam@95: ST(&(x[WS(rs, 4)]), VSUB(Ts, Tv), ms, &(x[0])); cannam@95: ST(&(x[0]), VADD(Ts, Tv), ms, &(x[0])); cannam@95: Tw = VSUB(Tq, Tr); cannam@95: Tx = VBYI(VSUB(Tu, Tt)); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(Tw, Tx), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 2)]), VADD(Tw, Tx), ms, &(x[0])); cannam@95: { cannam@95: V Tg, To, Tn, Tp, Tf, Th; cannam@95: Tf = VMUL(LDK(KP707106781), VADD(T9, Te)); cannam@95: Tg = VADD(T4, Tf); cannam@95: To = VSUB(T4, Tf); cannam@95: Th = VMUL(LDK(KP707106781), VSUB(Te, T9)); cannam@95: Tn = VBYI(VSUB(Th, Tm)); cannam@95: Tp = VBYI(VADD(Tm, Th)); cannam@95: ST(&(x[WS(rs, 7)]), VSUB(Tg, Tn), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 3)]), VADD(To, Tp), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VADD(Tg, Tn), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 5)]), VSUB(To, Tp), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, XSIMD_STRING("t1fuv_8"), twinstr, &GENUS, {33, 16, 0, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fuv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fuv_8, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */