Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:36:23 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* 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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 48 FP additions, 42 FP multiplications, Chris@10: * (or, 18 additions, 12 multiplications, 30 fused multiply/add), Chris@10: * 56 stack variables, 2 constants, and 36 memory accesses Chris@10: */ Chris@10: #include "q.h" Chris@10: Chris@10: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: E Tk, Tn, Tm, To, Tl; Chris@10: { Chris@10: E T1, Td, T4, Tg, Tp, T9, Te, T6, Tf, TB, TE, Ts, TZ, Tu, Tx; Chris@10: E TC, TN, TO, TD, TV, T10, TP, Tq, Tr; Chris@10: { Chris@10: E T2, T3, T7, T8; Chris@10: T1 = rio[0]; Chris@10: T2 = rio[WS(rs, 1)]; Chris@10: T3 = rio[WS(rs, 2)]; Chris@10: Td = iio[0]; Chris@10: T7 = iio[WS(rs, 1)]; Chris@10: T8 = iio[WS(rs, 2)]; Chris@10: T4 = T2 + T3; Chris@10: Tg = T3 - T2; Chris@10: Tp = rio[WS(vs, 1)]; Chris@10: T9 = T7 - T8; Chris@10: Te = T7 + T8; Chris@10: T6 = FNMS(KP500000000, T4, T1); Chris@10: Tq = rio[WS(vs, 1) + WS(rs, 1)]; Chris@10: Tr = rio[WS(vs, 1) + WS(rs, 2)]; Chris@10: Tf = FNMS(KP500000000, Te, Td); Chris@10: } Chris@10: { Chris@10: E Tv, Tw, TT, TU; Chris@10: TB = iio[WS(vs, 1)]; Chris@10: Tv = iio[WS(vs, 1) + WS(rs, 1)]; Chris@10: TE = Tr - Tq; Chris@10: Ts = Tq + Tr; Chris@10: Tw = iio[WS(vs, 1) + WS(rs, 2)]; Chris@10: TZ = iio[WS(vs, 2)]; Chris@10: TT = iio[WS(vs, 2) + WS(rs, 1)]; Chris@10: Tu = FNMS(KP500000000, Ts, Tp); Chris@10: Tx = Tv - Tw; Chris@10: TC = Tv + Tw; Chris@10: TU = iio[WS(vs, 2) + WS(rs, 2)]; Chris@10: TN = rio[WS(vs, 2)]; Chris@10: TO = rio[WS(vs, 2) + WS(rs, 1)]; Chris@10: TD = FNMS(KP500000000, TC, TB); Chris@10: TV = TT - TU; Chris@10: T10 = TT + TU; Chris@10: TP = rio[WS(vs, 2) + WS(rs, 2)]; Chris@10: } Chris@10: { Chris@10: E T11, T12, TS, TQ; Chris@10: rio[0] = T1 + T4; Chris@10: iio[0] = Td + Te; Chris@10: T11 = FNMS(KP500000000, T10, TZ); Chris@10: T12 = TP - TO; Chris@10: TQ = TO + TP; Chris@10: rio[WS(rs, 1)] = Tp + Ts; Chris@10: iio[WS(rs, 1)] = TB + TC; Chris@10: iio[WS(rs, 2)] = TZ + T10; Chris@10: TS = FNMS(KP500000000, TQ, TN); Chris@10: rio[WS(rs, 2)] = TN + TQ; Chris@10: { Chris@10: E TW, T13, Ty, TI, TL, TF, TH, TK; Chris@10: { Chris@10: E Ta, Th, T5, Tc; Chris@10: Tk = FNMS(KP866025403, T9, T6); Chris@10: Ta = FMA(KP866025403, T9, T6); Chris@10: Th = FMA(KP866025403, Tg, Tf); Chris@10: Tn = FNMS(KP866025403, Tg, Tf); Chris@10: T5 = W[0]; Chris@10: Tc = W[1]; Chris@10: { Chris@10: E T16, T19, T18, T1a, T17, Ti, Tb, T15; Chris@10: TW = FMA(KP866025403, TV, TS); Chris@10: T16 = FNMS(KP866025403, TV, TS); Chris@10: T19 = FNMS(KP866025403, T12, T11); Chris@10: T13 = FMA(KP866025403, T12, T11); Chris@10: Ti = T5 * Th; Chris@10: Tb = T5 * Ta; Chris@10: T15 = W[2]; Chris@10: T18 = W[3]; Chris@10: iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti); Chris@10: rio[WS(vs, 1)] = FMA(Tc, Th, Tb); Chris@10: T1a = T15 * T19; Chris@10: T17 = T15 * T16; Chris@10: Ty = FMA(KP866025403, Tx, Tu); Chris@10: TI = FNMS(KP866025403, Tx, Tu); Chris@10: TL = FNMS(KP866025403, TE, TD); Chris@10: TF = FMA(KP866025403, TE, TD); Chris@10: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a); Chris@10: rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17); Chris@10: TH = W[2]; Chris@10: TK = W[3]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TA, TG, Tz, TM, TJ, Tt; Chris@10: TM = TH * TL; Chris@10: TJ = TH * TI; Chris@10: Tt = W[0]; Chris@10: TA = W[1]; Chris@10: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM); Chris@10: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ); Chris@10: TG = Tt * TF; Chris@10: Tz = Tt * Ty; Chris@10: { Chris@10: E TR, TY, T14, TX, Tj; Chris@10: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG); Chris@10: rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz); Chris@10: TR = W[0]; Chris@10: TY = W[1]; Chris@10: T14 = TR * T13; Chris@10: TX = TR * TW; Chris@10: Tj = W[2]; Chris@10: Tm = W[3]; Chris@10: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14); Chris@10: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX); Chris@10: To = Tj * Tn; Chris@10: Tl = Tj * Tk; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: iio[WS(vs, 2)] = FNMS(Tm, Tk, To); Chris@10: rio[WS(vs, 2)] = FMA(Tm, Tn, Tl); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 0, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_q1_3) (planner *p) { Chris@10: X(kdft_difsq_register) (p, q1_3, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 48 FP additions, 36 FP multiplications, Chris@10: * (or, 30 additions, 18 multiplications, 18 fused multiply/add), Chris@10: * 35 stack variables, 2 constants, and 36 memory accesses Chris@10: */ Chris@10: #include "q.h" Chris@10: Chris@10: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt; Chris@10: E Tz, TR, TS, TN, TT, TF, TI, TK, TQ; Chris@10: { Chris@10: E T2, T3, Tr, Ts; Chris@10: T1 = rio[0]; Chris@10: T2 = rio[WS(rs, 1)]; Chris@10: T3 = rio[WS(rs, 2)]; Chris@10: T4 = T2 + T3; Chris@10: T6 = FNMS(KP500000000, T4, T1); Chris@10: Tc = KP866025403 * (T3 - T2); Chris@10: { Chris@10: E T7, T8, Tm, Tn; Chris@10: Td = iio[0]; Chris@10: T7 = iio[WS(rs, 1)]; Chris@10: T8 = iio[WS(rs, 2)]; Chris@10: Te = T7 + T8; Chris@10: T9 = KP866025403 * (T7 - T8); Chris@10: Tf = FNMS(KP500000000, Te, Td); Chris@10: Tl = rio[WS(vs, 1)]; Chris@10: Tm = rio[WS(vs, 1) + WS(rs, 1)]; Chris@10: Tn = rio[WS(vs, 1) + WS(rs, 2)]; Chris@10: To = Tm + Tn; Chris@10: Tq = FNMS(KP500000000, To, Tl); Chris@10: Tw = KP866025403 * (Tn - Tm); Chris@10: } Chris@10: Tx = iio[WS(vs, 1)]; Chris@10: Tr = iio[WS(vs, 1) + WS(rs, 1)]; Chris@10: Ts = iio[WS(vs, 1) + WS(rs, 2)]; Chris@10: Ty = Tr + Ts; Chris@10: Tt = KP866025403 * (Tr - Ts); Chris@10: Tz = FNMS(KP500000000, Ty, Tx); Chris@10: { Chris@10: E TL, TM, TG, TH; Chris@10: TR = iio[WS(vs, 2)]; Chris@10: TL = iio[WS(vs, 2) + WS(rs, 1)]; Chris@10: TM = iio[WS(vs, 2) + WS(rs, 2)]; Chris@10: TS = TL + TM; Chris@10: TN = KP866025403 * (TL - TM); Chris@10: TT = FNMS(KP500000000, TS, TR); Chris@10: TF = rio[WS(vs, 2)]; Chris@10: TG = rio[WS(vs, 2) + WS(rs, 1)]; Chris@10: TH = rio[WS(vs, 2) + WS(rs, 2)]; Chris@10: TI = TG + TH; Chris@10: TK = FNMS(KP500000000, TI, TF); Chris@10: TQ = KP866025403 * (TH - TG); Chris@10: } Chris@10: } Chris@10: rio[0] = T1 + T4; Chris@10: iio[0] = Td + Te; Chris@10: rio[WS(rs, 1)] = Tl + To; Chris@10: iio[WS(rs, 1)] = Tx + Ty; Chris@10: iio[WS(rs, 2)] = TR + TS; Chris@10: rio[WS(rs, 2)] = TF + TI; Chris@10: { Chris@10: E Ta, Tg, T5, Tb; Chris@10: Ta = T6 + T9; Chris@10: Tg = Tc + Tf; Chris@10: T5 = W[0]; Chris@10: Tb = W[1]; Chris@10: rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg); Chris@10: iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg); Chris@10: } Chris@10: { Chris@10: E TW, TY, TV, TX; Chris@10: TW = TK - TN; Chris@10: TY = TT - TQ; Chris@10: TV = W[2]; Chris@10: TX = W[3]; Chris@10: rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY); Chris@10: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY); Chris@10: } Chris@10: { Chris@10: E TC, TE, TB, TD; Chris@10: TC = Tq - Tt; Chris@10: TE = Tz - Tw; Chris@10: TB = W[2]; Chris@10: TD = W[3]; Chris@10: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE); Chris@10: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE); Chris@10: } Chris@10: { Chris@10: E Tu, TA, Tp, Tv; Chris@10: Tu = Tq + Tt; Chris@10: TA = Tw + Tz; Chris@10: Tp = W[0]; Chris@10: Tv = W[1]; Chris@10: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA); Chris@10: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA); Chris@10: } Chris@10: { Chris@10: E TO, TU, TJ, TP; Chris@10: TO = TK + TN; Chris@10: TU = TQ + TT; Chris@10: TJ = W[0]; Chris@10: TP = W[1]; Chris@10: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU); Chris@10: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU); Chris@10: } Chris@10: { Chris@10: E Ti, Tk, Th, Tj; Chris@10: Ti = T6 - T9; Chris@10: Tk = Tf - Tc; Chris@10: Th = W[2]; Chris@10: Tj = W[3]; Chris@10: rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk); Chris@10: iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 0, 3}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_q1_3) (planner *p) { Chris@10: X(kdft_difsq_register) (p, q1_3, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */