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 07:39:31 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_twidsq_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include q1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 44 FP additions, 32 FP multiplications, cannam@95: * (or, 36 additions, 24 multiplications, 8 fused multiply/add), cannam@95: * 38 stack variables, 0 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "q1f.h" cannam@95: cannam@95: static void q1fv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(8, vs)) { cannam@95: V Tb, Tm, Tx, TI; cannam@95: { cannam@95: V Tc, T9, T3, TG, TA, TH, TD, Ta, T6, Td, Tn, To, Tq, Tr, Tf; cannam@95: V Tg; cannam@95: { cannam@95: V T1, T2, Ty, Tz, TB, TC, T4, T5; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@95: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@95: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@95: T9 = VADD(T1, T2); cannam@95: T3 = VSUB(T1, T2); cannam@95: TG = VADD(Ty, Tz); cannam@95: TA = VSUB(Ty, Tz); cannam@95: TH = VADD(TB, TC); cannam@95: TD = VSUB(TB, TC); cannam@95: Ta = VADD(T4, T5); cannam@95: T6 = VSUB(T4, T5); cannam@95: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@95: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@95: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@95: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V Tk, Te, Tv, Tp, Tw, Ts, Tl, Th, T7, TE, Tu, TF; cannam@95: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@95: Tk = VADD(Tc, Td); cannam@95: Te = VSUB(Tc, Td); cannam@95: Tv = VADD(Tn, To); cannam@95: Tp = VSUB(Tn, To); cannam@95: Tw = VADD(Tq, Tr); cannam@95: Ts = VSUB(Tq, Tr); cannam@95: Tl = VADD(Tf, Tg); cannam@95: Th = VSUB(Tf, Tg); cannam@95: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@95: T7 = BYTWJ(&(W[0]), VFNMSI(T6, T3)); cannam@95: TE = BYTWJ(&(W[0]), VFNMSI(TD, TA)); cannam@95: { cannam@95: V Tt, Ti, Tj, T8; cannam@95: T8 = BYTWJ(&(W[TWVL * 4]), VFMAI(T6, T3)); cannam@95: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@95: Tt = BYTWJ(&(W[0]), VFNMSI(Ts, Tp)); cannam@95: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@95: Ti = BYTWJ(&(W[0]), VFNMSI(Th, Te)); cannam@95: Tj = BYTWJ(&(W[TWVL * 4]), VFMAI(Th, Te)); cannam@95: ST(&(x[WS(vs, 1)]), T7, ms, &(x[WS(vs, 1)])); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: ST(&(x[WS(vs, 3)]), T8, ms, &(x[WS(vs, 3)])); cannam@95: Tu = BYTWJ(&(W[TWVL * 4]), VFMAI(Ts, Tp)); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 1)])); cannam@95: TF = BYTWJ(&(W[TWVL * 4]), VFMAI(TD, TA)); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: } cannam@95: Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@95: Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@95: Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 3)])); cannam@95: TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); 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: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 4, XSIMD_STRING("q1fv_4"), twinstr, &GENUS, {36, 24, 8, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_q1fv_4) (planner *p) { cannam@95: X(kdft_difsq_register) (p, q1fv_4, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twidsq_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 4 -dif -name q1fv_4 -include q1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 44 FP additions, 24 FP multiplications, cannam@95: * (or, 44 additions, 24 multiplications, 0 fused multiply/add), cannam@95: * 22 stack variables, 0 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "q1f.h" cannam@95: cannam@95: static void q1fv_4(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: for (m = mb, W = W + (mb * ((TWVL / VL) * 6)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 6), MAKE_VOLATILE_STRIDE(8, rs), MAKE_VOLATILE_STRIDE(8, vs)) { cannam@95: V T3, T9, TA, TG, TD, TH, T6, Ta, Te, Tk, Tp, Tv, Ts, Tw, Th; cannam@95: V Tl; cannam@95: { cannam@95: V T1, T2, Ty, Tz; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: T3 = VSUB(T1, T2); cannam@95: T9 = VADD(T1, T2); cannam@95: Ty = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@95: Tz = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@95: TA = VSUB(Ty, Tz); cannam@95: TG = VADD(Ty, Tz); cannam@95: } cannam@95: { cannam@95: V TB, TC, T4, T5; cannam@95: TB = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: TC = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: TD = VBYI(VSUB(TB, TC)); cannam@95: TH = VADD(TB, TC); cannam@95: T4 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: T6 = VBYI(VSUB(T4, T5)); cannam@95: Ta = VADD(T4, T5); cannam@95: } cannam@95: { cannam@95: V Tc, Td, Tn, To; cannam@95: Tc = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@95: Td = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@95: Te = VSUB(Tc, Td); cannam@95: Tk = VADD(Tc, Td); cannam@95: Tn = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@95: To = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@95: Tp = VSUB(Tn, To); cannam@95: Tv = VADD(Tn, To); cannam@95: } cannam@95: { cannam@95: V Tq, Tr, Tf, Tg; cannam@95: Tq = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: Tr = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: Ts = VBYI(VSUB(Tq, Tr)); cannam@95: Tw = VADD(Tq, Tr); cannam@95: Tf = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: Tg = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: Th = VBYI(VSUB(Tf, Tg)); cannam@95: Tl = VADD(Tf, Tg); cannam@95: } cannam@95: ST(&(x[0]), VADD(T9, Ta), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 1)]), VADD(Tk, Tl), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 2)]), VADD(Tv, Tw), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 3)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V T7, Ti, Tt, TE; cannam@95: T7 = BYTWJ(&(W[0]), VSUB(T3, T6)); cannam@95: ST(&(x[WS(vs, 1)]), T7, ms, &(x[WS(vs, 1)])); cannam@95: Ti = BYTWJ(&(W[0]), VSUB(Te, Th)); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Ti, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: Tt = BYTWJ(&(W[0]), VSUB(Tp, Ts)); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 2)]), Tt, ms, &(x[WS(vs, 1)])); cannam@95: TE = BYTWJ(&(W[0]), VSUB(TA, TD)); cannam@95: ST(&(x[WS(vs, 1) + WS(rs, 3)]), TE, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V T8, Tj, Tu, TF; cannam@95: T8 = BYTWJ(&(W[TWVL * 4]), VADD(T3, T6)); cannam@95: ST(&(x[WS(vs, 3)]), T8, ms, &(x[WS(vs, 3)])); cannam@95: Tj = BYTWJ(&(W[TWVL * 4]), VADD(Te, Th)); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 1)]), Tj, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: Tu = BYTWJ(&(W[TWVL * 4]), VADD(Tp, Ts)); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 2)]), Tu, ms, &(x[WS(vs, 3)])); cannam@95: TF = BYTWJ(&(W[TWVL * 4]), VADD(TA, TD)); cannam@95: ST(&(x[WS(vs, 3) + WS(rs, 3)]), TF, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V Tb, Tm, Tx, TI; cannam@95: Tb = BYTWJ(&(W[TWVL * 2]), VSUB(T9, Ta)); cannam@95: ST(&(x[WS(vs, 2)]), Tb, ms, &(x[WS(vs, 2)])); cannam@95: Tm = BYTWJ(&(W[TWVL * 2]), VSUB(Tk, Tl)); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 1)]), Tm, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@95: Tx = BYTWJ(&(W[TWVL * 2]), VSUB(Tv, Tw)); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 2)]), Tx, ms, &(x[WS(vs, 2)])); cannam@95: TI = BYTWJ(&(W[TWVL * 2]), VSUB(TG, TH)); cannam@95: ST(&(x[WS(vs, 2) + WS(rs, 3)]), TI, ms, &(x[WS(vs, 2) + WS(rs, 1)])); 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: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 4, XSIMD_STRING("q1fv_4"), twinstr, &GENUS, {44, 24, 0, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_q1fv_4) (planner *p) { cannam@95: X(kdft_difsq_register) (p, q1fv_4, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */