Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:37:56 EDT 2016 */ Chris@42: Chris@42: #include "codelet-dft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* 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@42: Chris@42: /* Chris@42: * This function contains 48 FP additions, 42 FP multiplications, Chris@42: * (or, 18 additions, 12 multiplications, 30 fused multiply/add), Chris@42: * 56 stack variables, 2 constants, and 36 memory accesses Chris@42: */ Chris@42: #include "q.h" Chris@42: Chris@42: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: 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@42: E Tk, Tn, Tm, To, Tl; Chris@42: { Chris@42: E T1, Td, T4, Tg, Tp, T9, Te, T6, Tf, TB, TE, Ts, TZ, Tu, Tx; Chris@42: E TC, TN, TO, TD, TV, T10, TP, Tq, Tr; Chris@42: { Chris@42: E T2, T3, T7, T8; Chris@42: T1 = rio[0]; Chris@42: T2 = rio[WS(rs, 1)]; Chris@42: T3 = rio[WS(rs, 2)]; Chris@42: Td = iio[0]; Chris@42: T7 = iio[WS(rs, 1)]; Chris@42: T8 = iio[WS(rs, 2)]; Chris@42: T4 = T2 + T3; Chris@42: Tg = T3 - T2; Chris@42: Tp = rio[WS(vs, 1)]; Chris@42: T9 = T7 - T8; Chris@42: Te = T7 + T8; Chris@42: T6 = FNMS(KP500000000, T4, T1); Chris@42: Tq = rio[WS(vs, 1) + WS(rs, 1)]; Chris@42: Tr = rio[WS(vs, 1) + WS(rs, 2)]; Chris@42: Tf = FNMS(KP500000000, Te, Td); Chris@42: } Chris@42: { Chris@42: E Tv, Tw, TT, TU; Chris@42: TB = iio[WS(vs, 1)]; Chris@42: Tv = iio[WS(vs, 1) + WS(rs, 1)]; Chris@42: TE = Tr - Tq; Chris@42: Ts = Tq + Tr; Chris@42: Tw = iio[WS(vs, 1) + WS(rs, 2)]; Chris@42: TZ = iio[WS(vs, 2)]; Chris@42: TT = iio[WS(vs, 2) + WS(rs, 1)]; Chris@42: Tu = FNMS(KP500000000, Ts, Tp); Chris@42: Tx = Tv - Tw; Chris@42: TC = Tv + Tw; Chris@42: TU = iio[WS(vs, 2) + WS(rs, 2)]; Chris@42: TN = rio[WS(vs, 2)]; Chris@42: TO = rio[WS(vs, 2) + WS(rs, 1)]; Chris@42: TD = FNMS(KP500000000, TC, TB); Chris@42: TV = TT - TU; Chris@42: T10 = TT + TU; Chris@42: TP = rio[WS(vs, 2) + WS(rs, 2)]; Chris@42: } Chris@42: { Chris@42: E T11, T12, TS, TQ; Chris@42: rio[0] = T1 + T4; Chris@42: iio[0] = Td + Te; Chris@42: T11 = FNMS(KP500000000, T10, TZ); Chris@42: T12 = TP - TO; Chris@42: TQ = TO + TP; Chris@42: rio[WS(rs, 1)] = Tp + Ts; Chris@42: iio[WS(rs, 1)] = TB + TC; Chris@42: iio[WS(rs, 2)] = TZ + T10; Chris@42: TS = FNMS(KP500000000, TQ, TN); Chris@42: rio[WS(rs, 2)] = TN + TQ; Chris@42: { Chris@42: E TW, T13, Ty, TI, TL, TF, TH, TK; Chris@42: { Chris@42: E Ta, Th, T5, Tc; Chris@42: Tk = FNMS(KP866025403, T9, T6); Chris@42: Ta = FMA(KP866025403, T9, T6); Chris@42: Th = FMA(KP866025403, Tg, Tf); Chris@42: Tn = FNMS(KP866025403, Tg, Tf); Chris@42: T5 = W[0]; Chris@42: Tc = W[1]; Chris@42: { Chris@42: E T16, T19, T18, T1a, T17, Ti, Tb, T15; Chris@42: TW = FMA(KP866025403, TV, TS); Chris@42: T16 = FNMS(KP866025403, TV, TS); Chris@42: T19 = FNMS(KP866025403, T12, T11); Chris@42: T13 = FMA(KP866025403, T12, T11); Chris@42: Ti = T5 * Th; Chris@42: Tb = T5 * Ta; Chris@42: T15 = W[2]; Chris@42: T18 = W[3]; Chris@42: iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti); Chris@42: rio[WS(vs, 1)] = FMA(Tc, Th, Tb); Chris@42: T1a = T15 * T19; Chris@42: T17 = T15 * T16; Chris@42: Ty = FMA(KP866025403, Tx, Tu); Chris@42: TI = FNMS(KP866025403, Tx, Tu); Chris@42: TL = FNMS(KP866025403, TE, TD); Chris@42: TF = FMA(KP866025403, TE, TD); Chris@42: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a); Chris@42: rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17); Chris@42: TH = W[2]; Chris@42: TK = W[3]; Chris@42: } Chris@42: } Chris@42: { Chris@42: E TA, TG, Tz, TM, TJ, Tt; Chris@42: TM = TH * TL; Chris@42: TJ = TH * TI; Chris@42: Tt = W[0]; Chris@42: TA = W[1]; Chris@42: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM); Chris@42: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ); Chris@42: TG = Tt * TF; Chris@42: Tz = Tt * Ty; Chris@42: { Chris@42: E TR, TY, T14, TX, Tj; Chris@42: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG); Chris@42: rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz); Chris@42: TR = W[0]; Chris@42: TY = W[1]; Chris@42: T14 = TR * T13; Chris@42: TX = TR * TW; Chris@42: Tj = W[2]; Chris@42: Tm = W[3]; Chris@42: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14); Chris@42: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX); Chris@42: To = Tj * Tn; Chris@42: Tl = Tj * Tk; Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: iio[WS(vs, 2)] = FNMS(Tm, Tk, To); Chris@42: rio[WS(vs, 2)] = FMA(Tm, Tn, Tl); Chris@42: } Chris@42: } Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: {TW_FULL, 0, 3}, Chris@42: {TW_NEXT, 1, 0} Chris@42: }; Chris@42: Chris@42: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 }; Chris@42: Chris@42: void X(codelet_q1_3) (planner *p) { Chris@42: X(kdft_difsq_register) (p, q1_3, &desc); Chris@42: } Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include q.h */ Chris@42: Chris@42: /* Chris@42: * This function contains 48 FP additions, 36 FP multiplications, Chris@42: * (or, 30 additions, 18 multiplications, 18 fused multiply/add), Chris@42: * 35 stack variables, 2 constants, and 36 memory accesses Chris@42: */ Chris@42: #include "q.h" Chris@42: Chris@42: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: 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@42: E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt; Chris@42: E Tz, TR, TS, TN, TT, TF, TI, TK, TQ; Chris@42: { Chris@42: E T2, T3, Tr, Ts; Chris@42: T1 = rio[0]; Chris@42: T2 = rio[WS(rs, 1)]; Chris@42: T3 = rio[WS(rs, 2)]; Chris@42: T4 = T2 + T3; Chris@42: T6 = FNMS(KP500000000, T4, T1); Chris@42: Tc = KP866025403 * (T3 - T2); Chris@42: { Chris@42: E T7, T8, Tm, Tn; Chris@42: Td = iio[0]; Chris@42: T7 = iio[WS(rs, 1)]; Chris@42: T8 = iio[WS(rs, 2)]; Chris@42: Te = T7 + T8; Chris@42: T9 = KP866025403 * (T7 - T8); Chris@42: Tf = FNMS(KP500000000, Te, Td); Chris@42: Tl = rio[WS(vs, 1)]; Chris@42: Tm = rio[WS(vs, 1) + WS(rs, 1)]; Chris@42: Tn = rio[WS(vs, 1) + WS(rs, 2)]; Chris@42: To = Tm + Tn; Chris@42: Tq = FNMS(KP500000000, To, Tl); Chris@42: Tw = KP866025403 * (Tn - Tm); Chris@42: } Chris@42: Tx = iio[WS(vs, 1)]; Chris@42: Tr = iio[WS(vs, 1) + WS(rs, 1)]; Chris@42: Ts = iio[WS(vs, 1) + WS(rs, 2)]; Chris@42: Ty = Tr + Ts; Chris@42: Tt = KP866025403 * (Tr - Ts); Chris@42: Tz = FNMS(KP500000000, Ty, Tx); Chris@42: { Chris@42: E TL, TM, TG, TH; Chris@42: TR = iio[WS(vs, 2)]; Chris@42: TL = iio[WS(vs, 2) + WS(rs, 1)]; Chris@42: TM = iio[WS(vs, 2) + WS(rs, 2)]; Chris@42: TS = TL + TM; Chris@42: TN = KP866025403 * (TL - TM); Chris@42: TT = FNMS(KP500000000, TS, TR); Chris@42: TF = rio[WS(vs, 2)]; Chris@42: TG = rio[WS(vs, 2) + WS(rs, 1)]; Chris@42: TH = rio[WS(vs, 2) + WS(rs, 2)]; Chris@42: TI = TG + TH; Chris@42: TK = FNMS(KP500000000, TI, TF); Chris@42: TQ = KP866025403 * (TH - TG); Chris@42: } Chris@42: } Chris@42: rio[0] = T1 + T4; Chris@42: iio[0] = Td + Te; Chris@42: rio[WS(rs, 1)] = Tl + To; Chris@42: iio[WS(rs, 1)] = Tx + Ty; Chris@42: iio[WS(rs, 2)] = TR + TS; Chris@42: rio[WS(rs, 2)] = TF + TI; Chris@42: { Chris@42: E Ta, Tg, T5, Tb; Chris@42: Ta = T6 + T9; Chris@42: Tg = Tc + Tf; Chris@42: T5 = W[0]; Chris@42: Tb = W[1]; Chris@42: rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg); Chris@42: iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg); Chris@42: } Chris@42: { Chris@42: E TW, TY, TV, TX; Chris@42: TW = TK - TN; Chris@42: TY = TT - TQ; Chris@42: TV = W[2]; Chris@42: TX = W[3]; Chris@42: rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY); Chris@42: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY); Chris@42: } Chris@42: { Chris@42: E TC, TE, TB, TD; Chris@42: TC = Tq - Tt; Chris@42: TE = Tz - Tw; Chris@42: TB = W[2]; Chris@42: TD = W[3]; Chris@42: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE); Chris@42: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE); Chris@42: } Chris@42: { Chris@42: E Tu, TA, Tp, Tv; Chris@42: Tu = Tq + Tt; Chris@42: TA = Tw + Tz; Chris@42: Tp = W[0]; Chris@42: Tv = W[1]; Chris@42: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA); Chris@42: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA); Chris@42: } Chris@42: { Chris@42: E TO, TU, TJ, TP; Chris@42: TO = TK + TN; Chris@42: TU = TQ + TT; Chris@42: TJ = W[0]; Chris@42: TP = W[1]; Chris@42: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU); Chris@42: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU); Chris@42: } Chris@42: { Chris@42: E Ti, Tk, Th, Tj; Chris@42: Ti = T6 - T9; Chris@42: Tk = Tf - Tc; Chris@42: Th = W[2]; Chris@42: Tj = W[3]; Chris@42: rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk); Chris@42: iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: {TW_FULL, 0, 3}, Chris@42: {TW_NEXT, 1, 0} Chris@42: }; Chris@42: Chris@42: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 }; Chris@42: Chris@42: void X(codelet_q1_3) (planner *p) { Chris@42: X(kdft_difsq_register) (p, q1_3, &desc); Chris@42: } Chris@42: #endif /* HAVE_FMA */