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 07:36:23 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.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 48 FP additions, 42 FP multiplications, cannam@95: * (or, 18 additions, 12 multiplications, 30 fused multiply/add), cannam@95: * 56 stack variables, 2 constants, and 36 memory accesses cannam@95: */ cannam@95: #include "q.h" cannam@95: cannam@95: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) { cannam@95: E Tk, Tn, Tm, To, Tl; cannam@95: { cannam@95: E T1, Td, T4, Tg, Tp, T9, Te, T6, Tf, TB, TE, Ts, TZ, Tu, Tx; cannam@95: E TC, TN, TO, TD, TV, T10, TP, Tq, Tr; cannam@95: { cannam@95: E T2, T3, T7, T8; cannam@95: T1 = rio[0]; cannam@95: T2 = rio[WS(rs, 1)]; cannam@95: T3 = rio[WS(rs, 2)]; cannam@95: Td = iio[0]; cannam@95: T7 = iio[WS(rs, 1)]; cannam@95: T8 = iio[WS(rs, 2)]; cannam@95: T4 = T2 + T3; cannam@95: Tg = T3 - T2; cannam@95: Tp = rio[WS(vs, 1)]; cannam@95: T9 = T7 - T8; cannam@95: Te = T7 + T8; cannam@95: T6 = FNMS(KP500000000, T4, T1); cannam@95: Tq = rio[WS(vs, 1) + WS(rs, 1)]; cannam@95: Tr = rio[WS(vs, 1) + WS(rs, 2)]; cannam@95: Tf = FNMS(KP500000000, Te, Td); cannam@95: } cannam@95: { cannam@95: E Tv, Tw, TT, TU; cannam@95: TB = iio[WS(vs, 1)]; cannam@95: Tv = iio[WS(vs, 1) + WS(rs, 1)]; cannam@95: TE = Tr - Tq; cannam@95: Ts = Tq + Tr; cannam@95: Tw = iio[WS(vs, 1) + WS(rs, 2)]; cannam@95: TZ = iio[WS(vs, 2)]; cannam@95: TT = iio[WS(vs, 2) + WS(rs, 1)]; cannam@95: Tu = FNMS(KP500000000, Ts, Tp); cannam@95: Tx = Tv - Tw; cannam@95: TC = Tv + Tw; cannam@95: TU = iio[WS(vs, 2) + WS(rs, 2)]; cannam@95: TN = rio[WS(vs, 2)]; cannam@95: TO = rio[WS(vs, 2) + WS(rs, 1)]; cannam@95: TD = FNMS(KP500000000, TC, TB); cannam@95: TV = TT - TU; cannam@95: T10 = TT + TU; cannam@95: TP = rio[WS(vs, 2) + WS(rs, 2)]; cannam@95: } cannam@95: { cannam@95: E T11, T12, TS, TQ; cannam@95: rio[0] = T1 + T4; cannam@95: iio[0] = Td + Te; cannam@95: T11 = FNMS(KP500000000, T10, TZ); cannam@95: T12 = TP - TO; cannam@95: TQ = TO + TP; cannam@95: rio[WS(rs, 1)] = Tp + Ts; cannam@95: iio[WS(rs, 1)] = TB + TC; cannam@95: iio[WS(rs, 2)] = TZ + T10; cannam@95: TS = FNMS(KP500000000, TQ, TN); cannam@95: rio[WS(rs, 2)] = TN + TQ; cannam@95: { cannam@95: E TW, T13, Ty, TI, TL, TF, TH, TK; cannam@95: { cannam@95: E Ta, Th, T5, Tc; cannam@95: Tk = FNMS(KP866025403, T9, T6); cannam@95: Ta = FMA(KP866025403, T9, T6); cannam@95: Th = FMA(KP866025403, Tg, Tf); cannam@95: Tn = FNMS(KP866025403, Tg, Tf); cannam@95: T5 = W[0]; cannam@95: Tc = W[1]; cannam@95: { cannam@95: E T16, T19, T18, T1a, T17, Ti, Tb, T15; cannam@95: TW = FMA(KP866025403, TV, TS); cannam@95: T16 = FNMS(KP866025403, TV, TS); cannam@95: T19 = FNMS(KP866025403, T12, T11); cannam@95: T13 = FMA(KP866025403, T12, T11); cannam@95: Ti = T5 * Th; cannam@95: Tb = T5 * Ta; cannam@95: T15 = W[2]; cannam@95: T18 = W[3]; cannam@95: iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti); cannam@95: rio[WS(vs, 1)] = FMA(Tc, Th, Tb); cannam@95: T1a = T15 * T19; cannam@95: T17 = T15 * T16; cannam@95: Ty = FMA(KP866025403, Tx, Tu); cannam@95: TI = FNMS(KP866025403, Tx, Tu); cannam@95: TL = FNMS(KP866025403, TE, TD); cannam@95: TF = FMA(KP866025403, TE, TD); cannam@95: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a); cannam@95: rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17); cannam@95: TH = W[2]; cannam@95: TK = W[3]; cannam@95: } cannam@95: } cannam@95: { cannam@95: E TA, TG, Tz, TM, TJ, Tt; cannam@95: TM = TH * TL; cannam@95: TJ = TH * TI; cannam@95: Tt = W[0]; cannam@95: TA = W[1]; cannam@95: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM); cannam@95: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ); cannam@95: TG = Tt * TF; cannam@95: Tz = Tt * Ty; cannam@95: { cannam@95: E TR, TY, T14, TX, Tj; cannam@95: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG); cannam@95: rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz); cannam@95: TR = W[0]; cannam@95: TY = W[1]; cannam@95: T14 = TR * T13; cannam@95: TX = TR * TW; cannam@95: Tj = W[2]; cannam@95: Tm = W[3]; cannam@95: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14); cannam@95: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX); cannam@95: To = Tj * Tn; cannam@95: Tl = Tj * Tk; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: iio[WS(vs, 2)] = FNMS(Tm, Tk, To); cannam@95: rio[WS(vs, 2)] = FMA(Tm, Tn, Tl); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 0, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_q1_3) (planner *p) { cannam@95: X(kdft_difsq_register) (p, q1_3, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 48 FP additions, 36 FP multiplications, cannam@95: * (or, 30 additions, 18 multiplications, 18 fused multiply/add), cannam@95: * 35 stack variables, 2 constants, and 36 memory accesses cannam@95: */ cannam@95: #include "q.h" cannam@95: cannam@95: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 4); m < me; m = m + 1, rio = rio + ms, iio = iio + ms, W = W + 4, MAKE_VOLATILE_STRIDE(6, rs), MAKE_VOLATILE_STRIDE(0, vs)) { cannam@95: E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt; cannam@95: E Tz, TR, TS, TN, TT, TF, TI, TK, TQ; cannam@95: { cannam@95: E T2, T3, Tr, Ts; cannam@95: T1 = rio[0]; cannam@95: T2 = rio[WS(rs, 1)]; cannam@95: T3 = rio[WS(rs, 2)]; cannam@95: T4 = T2 + T3; cannam@95: T6 = FNMS(KP500000000, T4, T1); cannam@95: Tc = KP866025403 * (T3 - T2); cannam@95: { cannam@95: E T7, T8, Tm, Tn; cannam@95: Td = iio[0]; cannam@95: T7 = iio[WS(rs, 1)]; cannam@95: T8 = iio[WS(rs, 2)]; cannam@95: Te = T7 + T8; cannam@95: T9 = KP866025403 * (T7 - T8); cannam@95: Tf = FNMS(KP500000000, Te, Td); cannam@95: Tl = rio[WS(vs, 1)]; cannam@95: Tm = rio[WS(vs, 1) + WS(rs, 1)]; cannam@95: Tn = rio[WS(vs, 1) + WS(rs, 2)]; cannam@95: To = Tm + Tn; cannam@95: Tq = FNMS(KP500000000, To, Tl); cannam@95: Tw = KP866025403 * (Tn - Tm); cannam@95: } cannam@95: Tx = iio[WS(vs, 1)]; cannam@95: Tr = iio[WS(vs, 1) + WS(rs, 1)]; cannam@95: Ts = iio[WS(vs, 1) + WS(rs, 2)]; cannam@95: Ty = Tr + Ts; cannam@95: Tt = KP866025403 * (Tr - Ts); cannam@95: Tz = FNMS(KP500000000, Ty, Tx); cannam@95: { cannam@95: E TL, TM, TG, TH; cannam@95: TR = iio[WS(vs, 2)]; cannam@95: TL = iio[WS(vs, 2) + WS(rs, 1)]; cannam@95: TM = iio[WS(vs, 2) + WS(rs, 2)]; cannam@95: TS = TL + TM; cannam@95: TN = KP866025403 * (TL - TM); cannam@95: TT = FNMS(KP500000000, TS, TR); cannam@95: TF = rio[WS(vs, 2)]; cannam@95: TG = rio[WS(vs, 2) + WS(rs, 1)]; cannam@95: TH = rio[WS(vs, 2) + WS(rs, 2)]; cannam@95: TI = TG + TH; cannam@95: TK = FNMS(KP500000000, TI, TF); cannam@95: TQ = KP866025403 * (TH - TG); cannam@95: } cannam@95: } cannam@95: rio[0] = T1 + T4; cannam@95: iio[0] = Td + Te; cannam@95: rio[WS(rs, 1)] = Tl + To; cannam@95: iio[WS(rs, 1)] = Tx + Ty; cannam@95: iio[WS(rs, 2)] = TR + TS; cannam@95: rio[WS(rs, 2)] = TF + TI; cannam@95: { cannam@95: E Ta, Tg, T5, Tb; cannam@95: Ta = T6 + T9; cannam@95: Tg = Tc + Tf; cannam@95: T5 = W[0]; cannam@95: Tb = W[1]; cannam@95: rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg); cannam@95: iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg); cannam@95: } cannam@95: { cannam@95: E TW, TY, TV, TX; cannam@95: TW = TK - TN; cannam@95: TY = TT - TQ; cannam@95: TV = W[2]; cannam@95: TX = W[3]; cannam@95: rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY); cannam@95: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY); cannam@95: } cannam@95: { cannam@95: E TC, TE, TB, TD; cannam@95: TC = Tq - Tt; cannam@95: TE = Tz - Tw; cannam@95: TB = W[2]; cannam@95: TD = W[3]; cannam@95: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE); cannam@95: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE); cannam@95: } cannam@95: { cannam@95: E Tu, TA, Tp, Tv; cannam@95: Tu = Tq + Tt; cannam@95: TA = Tw + Tz; cannam@95: Tp = W[0]; cannam@95: Tv = W[1]; cannam@95: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA); cannam@95: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA); cannam@95: } cannam@95: { cannam@95: E TO, TU, TJ, TP; cannam@95: TO = TK + TN; cannam@95: TU = TQ + TT; cannam@95: TJ = W[0]; cannam@95: TP = W[1]; cannam@95: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU); cannam@95: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU); cannam@95: } cannam@95: { cannam@95: E Ti, Tk, Th, Tj; cannam@95: Ti = T6 - T9; cannam@95: Tk = Tf - Tc; cannam@95: Th = W[2]; cannam@95: Tj = W[3]; cannam@95: rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk); cannam@95: iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 0, 3}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_q1_3) (planner *p) { cannam@95: X(kdft_difsq_register) (p, q1_3, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */