cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:44:15 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 7 -name t1bv_7 -include t1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 36 FP additions, 36 FP multiplications, cannam@127: * (or, 15 additions, 15 multiplications, 21 fused multiply/add), cannam@127: * 42 stack variables, 6 constants, and 14 memory accesses cannam@127: */ cannam@127: #include "t1b.h" cannam@127: cannam@127: static void t1bv_7(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP900968867, +0.900968867902419126236102319507445051165919162); cannam@127: DVK(KP801937735, +0.801937735804838252472204639014890102331838324); cannam@127: DVK(KP974927912, +0.974927912181823607018131682993931217232785801); cannam@127: DVK(KP692021471, +0.692021471630095869627814897002069140197260599); cannam@127: DVK(KP554958132, +0.554958132087371191422194871006410481067288862); cannam@127: DVK(KP356895867, +0.356895867892209443894399510021300583399127187); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 12)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 12), MAKE_VOLATILE_STRIDE(7, rs)) { cannam@127: V T1, T2, T4, Te, Tc, T9, T7; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T2 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T4 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@127: Te = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T9 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@127: T7 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: { cannam@127: V T3, T5, Tf, Td, Ta, T8; cannam@127: T3 = BYTW(&(W[0]), T2); cannam@127: T5 = BYTW(&(W[TWVL * 10]), T4); cannam@127: Tf = BYTW(&(W[TWVL * 6]), Te); cannam@127: Td = BYTW(&(W[TWVL * 4]), Tc); cannam@127: Ta = BYTW(&(W[TWVL * 8]), T9); cannam@127: T8 = BYTW(&(W[TWVL * 2]), T7); cannam@127: { cannam@127: V T6, Tm, Tg, Tk, Tb, Tl; cannam@127: T6 = VADD(T3, T5); cannam@127: Tm = VSUB(T3, T5); cannam@127: Tg = VADD(Td, Tf); cannam@127: Tk = VSUB(Td, Tf); cannam@127: Tb = VADD(T8, Ta); cannam@127: Tl = VSUB(T8, Ta); cannam@127: { cannam@127: V Tp, Tx, Tu, Th, Ts, Tn, Tq, Ty; cannam@127: Tp = VFNMS(LDK(KP356895867), T6, Tg); cannam@127: Tx = VFMA(LDK(KP554958132), Tk, Tm); cannam@127: ST(&(x[0]), VADD(T1, VADD(T6, VADD(Tb, Tg))), ms, &(x[0])); cannam@127: Tu = VFNMS(LDK(KP356895867), Tb, T6); cannam@127: Th = VFNMS(LDK(KP356895867), Tg, Tb); cannam@127: Ts = VFMA(LDK(KP554958132), Tl, Tk); cannam@127: Tn = VFNMS(LDK(KP554958132), Tm, Tl); cannam@127: Tq = VFNMS(LDK(KP692021471), Tp, Tb); cannam@127: Ty = VMUL(LDK(KP974927912), VFMA(LDK(KP801937735), Tx, Tl)); cannam@127: { cannam@127: V Tv, Ti, Tt, To, Tr, Tw, Tj; cannam@127: Tv = VFNMS(LDK(KP692021471), Tu, Tg); cannam@127: Ti = VFNMS(LDK(KP692021471), Th, T6); cannam@127: Tt = VMUL(LDK(KP974927912), VFNMS(LDK(KP801937735), Ts, Tm)); cannam@127: To = VMUL(LDK(KP974927912), VFNMS(LDK(KP801937735), Tn, Tk)); cannam@127: Tr = VFNMS(LDK(KP900968867), Tq, T1); cannam@127: Tw = VFNMS(LDK(KP900968867), Tv, T1); cannam@127: Tj = VFNMS(LDK(KP900968867), Ti, T1); cannam@127: ST(&(x[WS(rs, 5)]), VFNMSI(Tt, Tr), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 2)]), VFMAI(Tt, Tr), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 6)]), VFNMSI(Ty, Tw), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 1)]), VFMAI(Ty, Tw), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 4)]), VFNMSI(To, Tj), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 3)]), VFMAI(To, Tj), ms, &(x[WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: VTW(0, 5), cannam@127: VTW(0, 6), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 7, XSIMD_STRING("t1bv_7"), twinstr, &GENUS, {15, 15, 21, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1bv_7) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1bv_7, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 7 -name t1bv_7 -include t1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 36 FP additions, 30 FP multiplications, cannam@127: * (or, 24 additions, 18 multiplications, 12 fused multiply/add), cannam@127: * 21 stack variables, 6 constants, and 14 memory accesses cannam@127: */ cannam@127: #include "t1b.h" cannam@127: cannam@127: static void t1bv_7(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP222520933, +0.222520933956314404288902564496794759466355569); cannam@127: DVK(KP900968867, +0.900968867902419126236102319507445051165919162); cannam@127: DVK(KP623489801, +0.623489801858733530525004884004239810632274731); cannam@127: DVK(KP433883739, +0.433883739117558120475768332848358754609990728); cannam@127: DVK(KP781831482, +0.781831482468029808708444526674057750232334519); cannam@127: DVK(KP974927912, +0.974927912181823607018131682993931217232785801); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 12)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 12), MAKE_VOLATILE_STRIDE(7, rs)) { cannam@127: V Th, Tf, Ti, T5, Tk, Ta, Tj, To, Tp; cannam@127: Th = LD(&(x[0]), ms, &(x[0])); cannam@127: { cannam@127: V Tc, Te, Tb, Td; cannam@127: Tb = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: Tc = BYTW(&(W[TWVL * 2]), Tb); cannam@127: Td = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@127: Te = BYTW(&(W[TWVL * 8]), Td); cannam@127: Tf = VSUB(Tc, Te); cannam@127: Ti = VADD(Tc, Te); cannam@127: } cannam@127: { cannam@127: V T2, T4, T1, T3; cannam@127: T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T2 = BYTW(&(W[0]), T1); cannam@127: T3 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@127: T4 = BYTW(&(W[TWVL * 10]), T3); cannam@127: T5 = VSUB(T2, T4); cannam@127: Tk = VADD(T2, T4); cannam@127: } cannam@127: { cannam@127: V T7, T9, T6, T8; cannam@127: T6 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T7 = BYTW(&(W[TWVL * 4]), T6); cannam@127: T8 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: T9 = BYTW(&(W[TWVL * 6]), T8); cannam@127: Ta = VSUB(T7, T9); cannam@127: Tj = VADD(T7, T9); cannam@127: } cannam@127: ST(&(x[0]), VADD(Th, VADD(Tk, VADD(Ti, Tj))), ms, &(x[0])); cannam@127: To = VBYI(VFNMS(LDK(KP781831482), Ta, VFNMS(LDK(KP433883739), Tf, VMUL(LDK(KP974927912), T5)))); cannam@127: Tp = VFMA(LDK(KP623489801), Tj, VFNMS(LDK(KP900968867), Ti, VFNMS(LDK(KP222520933), Tk, Th))); cannam@127: ST(&(x[WS(rs, 2)]), VADD(To, Tp), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 5)]), VSUB(Tp, To), ms, &(x[WS(rs, 1)])); cannam@127: { cannam@127: V Tg, Tl, Tm, Tn; cannam@127: Tg = VBYI(VFMA(LDK(KP433883739), T5, VFNMS(LDK(KP781831482), Tf, VMUL(LDK(KP974927912), Ta)))); cannam@127: Tl = VFMA(LDK(KP623489801), Ti, VFNMS(LDK(KP222520933), Tj, VFNMS(LDK(KP900968867), Tk, Th))); cannam@127: ST(&(x[WS(rs, 3)]), VADD(Tg, Tl), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 4)]), VSUB(Tl, Tg), ms, &(x[0])); cannam@127: Tm = VBYI(VFMA(LDK(KP781831482), T5, VFMA(LDK(KP974927912), Tf, VMUL(LDK(KP433883739), Ta)))); cannam@127: Tn = VFMA(LDK(KP623489801), Tk, VFNMS(LDK(KP900968867), Tj, VFNMS(LDK(KP222520933), Ti, Th))); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tm, Tn), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 6)]), VSUB(Tn, Tm), ms, &(x[0])); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: VTW(0, 5), cannam@127: VTW(0, 6), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 7, XSIMD_STRING("t1bv_7"), twinstr, &GENUS, {24, 18, 12, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1bv_7) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1bv_7, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */