cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:04:30 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq.native -fma -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include dft/scalar/q.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 48 FP additions, 42 FP multiplications, cannam@167: * (or, 18 additions, 12 multiplications, 30 fused multiply/add), cannam@167: * 35 stack variables, 2 constants, and 36 memory accesses cannam@167: */ cannam@167: #include "dft/scalar/q.h" cannam@167: cannam@167: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: 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@167: E T1, T4, T6, Tg, Td, Te, T9, Tf, Tp, Ts, Tu, TE, TB, TC, Tx; cannam@167: E TD, TZ, T10, TV, T11, TN, TQ, TS, T12; cannam@167: { cannam@167: E T2, T3, Tv, Tw; cannam@167: T1 = rio[0]; cannam@167: T2 = rio[WS(rs, 1)]; cannam@167: T3 = rio[WS(rs, 2)]; cannam@167: T4 = T2 + T3; cannam@167: T6 = FNMS(KP500000000, T4, T1); cannam@167: Tg = T3 - T2; cannam@167: { cannam@167: E T7, T8, Tq, Tr; cannam@167: Td = iio[0]; cannam@167: T7 = iio[WS(rs, 1)]; cannam@167: T8 = iio[WS(rs, 2)]; cannam@167: Te = T7 + T8; cannam@167: T9 = T7 - T8; cannam@167: Tf = FNMS(KP500000000, Te, Td); cannam@167: Tp = rio[WS(vs, 1)]; cannam@167: Tq = rio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tr = rio[WS(vs, 1) + WS(rs, 2)]; cannam@167: Ts = Tq + Tr; cannam@167: Tu = FNMS(KP500000000, Ts, Tp); cannam@167: TE = Tr - Tq; cannam@167: } cannam@167: TB = iio[WS(vs, 1)]; cannam@167: Tv = iio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tw = iio[WS(vs, 1) + WS(rs, 2)]; cannam@167: TC = Tv + Tw; cannam@167: Tx = Tv - Tw; cannam@167: TD = FNMS(KP500000000, TC, TB); cannam@167: { cannam@167: E TT, TU, TO, TP; cannam@167: TZ = iio[WS(vs, 2)]; cannam@167: TT = iio[WS(vs, 2) + WS(rs, 1)]; cannam@167: TU = iio[WS(vs, 2) + WS(rs, 2)]; cannam@167: T10 = TT + TU; cannam@167: TV = TT - TU; cannam@167: T11 = FNMS(KP500000000, T10, TZ); cannam@167: TN = rio[WS(vs, 2)]; cannam@167: TO = rio[WS(vs, 2) + WS(rs, 1)]; cannam@167: TP = rio[WS(vs, 2) + WS(rs, 2)]; cannam@167: TQ = TO + TP; cannam@167: TS = FNMS(KP500000000, TQ, TN); cannam@167: T12 = TP - TO; cannam@167: } cannam@167: } cannam@167: rio[0] = T1 + T4; cannam@167: iio[0] = Td + Te; cannam@167: rio[WS(rs, 1)] = Tp + Ts; cannam@167: iio[WS(rs, 1)] = TB + TC; cannam@167: iio[WS(rs, 2)] = TZ + T10; cannam@167: rio[WS(rs, 2)] = TN + TQ; cannam@167: { cannam@167: E Ta, Th, Tb, Ti, T5, Tc; cannam@167: Ta = FMA(KP866025403, T9, T6); cannam@167: Th = FMA(KP866025403, Tg, Tf); cannam@167: T5 = W[0]; cannam@167: Tb = T5 * Ta; cannam@167: Ti = T5 * Th; cannam@167: Tc = W[1]; cannam@167: rio[WS(vs, 1)] = FMA(Tc, Th, Tb); cannam@167: iio[WS(vs, 1)] = FNMS(Tc, Ta, Ti); cannam@167: } cannam@167: { cannam@167: E T16, T19, T17, T1a, T15, T18; cannam@167: T16 = FNMS(KP866025403, TV, TS); cannam@167: T19 = FNMS(KP866025403, T12, T11); cannam@167: T15 = W[2]; cannam@167: T17 = T15 * T16; cannam@167: T1a = T15 * T19; cannam@167: T18 = W[3]; cannam@167: rio[WS(vs, 2) + WS(rs, 2)] = FMA(T18, T19, T17); cannam@167: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(T18, T16, T1a); cannam@167: } cannam@167: { cannam@167: E TI, TL, TJ, TM, TH, TK; cannam@167: TI = FNMS(KP866025403, Tx, Tu); cannam@167: TL = FNMS(KP866025403, TE, TD); cannam@167: TH = W[2]; cannam@167: TJ = TH * TI; cannam@167: TM = TH * TL; cannam@167: TK = W[3]; cannam@167: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TK, TL, TJ); cannam@167: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TK, TI, TM); cannam@167: } cannam@167: { cannam@167: E Ty, TF, Tz, TG, Tt, TA; cannam@167: Ty = FMA(KP866025403, Tx, Tu); cannam@167: TF = FMA(KP866025403, TE, TD); cannam@167: Tt = W[0]; cannam@167: Tz = Tt * Ty; cannam@167: TG = Tt * TF; cannam@167: TA = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 1)] = FMA(TA, TF, Tz); cannam@167: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(TA, Ty, TG); cannam@167: } cannam@167: { cannam@167: E TW, T13, TX, T14, TR, TY; cannam@167: TW = FMA(KP866025403, TV, TS); cannam@167: T13 = FMA(KP866025403, T12, T11); cannam@167: TR = W[0]; cannam@167: TX = TR * TW; cannam@167: T14 = TR * T13; cannam@167: TY = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TY, T13, TX); cannam@167: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TY, TW, T14); cannam@167: } cannam@167: { cannam@167: E Tk, Tn, Tl, To, Tj, Tm; cannam@167: Tk = FNMS(KP866025403, T9, T6); cannam@167: Tn = FNMS(KP866025403, Tg, Tf); cannam@167: Tj = W[2]; cannam@167: Tl = Tj * Tk; cannam@167: To = Tj * Tn; cannam@167: Tm = W[3]; cannam@167: rio[WS(vs, 2)] = FMA(Tm, Tn, Tl); cannam@167: iio[WS(vs, 2)] = FNMS(Tm, Tk, To); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 0, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {18, 12, 30, 0}, 0, 0, 0 }; cannam@167: cannam@167: void X(codelet_q1_3) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1_3, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq.native -compact -variables 4 -pipeline-latency 4 -reload-twiddle -dif -n 3 -name q1_3 -include dft/scalar/q.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 48 FP additions, 36 FP multiplications, cannam@167: * (or, 30 additions, 18 multiplications, 18 fused multiply/add), cannam@167: * 35 stack variables, 2 constants, and 36 memory accesses cannam@167: */ cannam@167: #include "dft/scalar/q.h" cannam@167: cannam@167: static void q1_3(R *rio, R *iio, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: 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@167: E T1, T4, T6, Tc, Td, Te, T9, Tf, Tl, To, Tq, Tw, Tx, Ty, Tt; cannam@167: E Tz, TR, TS, TN, TT, TF, TI, TK, TQ; cannam@167: { cannam@167: E T2, T3, Tr, Ts; cannam@167: T1 = rio[0]; cannam@167: T2 = rio[WS(rs, 1)]; cannam@167: T3 = rio[WS(rs, 2)]; cannam@167: T4 = T2 + T3; cannam@167: T6 = FNMS(KP500000000, T4, T1); cannam@167: Tc = KP866025403 * (T3 - T2); cannam@167: { cannam@167: E T7, T8, Tm, Tn; cannam@167: Td = iio[0]; cannam@167: T7 = iio[WS(rs, 1)]; cannam@167: T8 = iio[WS(rs, 2)]; cannam@167: Te = T7 + T8; cannam@167: T9 = KP866025403 * (T7 - T8); cannam@167: Tf = FNMS(KP500000000, Te, Td); cannam@167: Tl = rio[WS(vs, 1)]; cannam@167: Tm = rio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Tn = rio[WS(vs, 1) + WS(rs, 2)]; cannam@167: To = Tm + Tn; cannam@167: Tq = FNMS(KP500000000, To, Tl); cannam@167: Tw = KP866025403 * (Tn - Tm); cannam@167: } cannam@167: Tx = iio[WS(vs, 1)]; cannam@167: Tr = iio[WS(vs, 1) + WS(rs, 1)]; cannam@167: Ts = iio[WS(vs, 1) + WS(rs, 2)]; cannam@167: Ty = Tr + Ts; cannam@167: Tt = KP866025403 * (Tr - Ts); cannam@167: Tz = FNMS(KP500000000, Ty, Tx); cannam@167: { cannam@167: E TL, TM, TG, TH; cannam@167: TR = iio[WS(vs, 2)]; cannam@167: TL = iio[WS(vs, 2) + WS(rs, 1)]; cannam@167: TM = iio[WS(vs, 2) + WS(rs, 2)]; cannam@167: TS = TL + TM; cannam@167: TN = KP866025403 * (TL - TM); cannam@167: TT = FNMS(KP500000000, TS, TR); cannam@167: TF = rio[WS(vs, 2)]; cannam@167: TG = rio[WS(vs, 2) + WS(rs, 1)]; cannam@167: TH = rio[WS(vs, 2) + WS(rs, 2)]; cannam@167: TI = TG + TH; cannam@167: TK = FNMS(KP500000000, TI, TF); cannam@167: TQ = KP866025403 * (TH - TG); cannam@167: } cannam@167: } cannam@167: rio[0] = T1 + T4; cannam@167: iio[0] = Td + Te; cannam@167: rio[WS(rs, 1)] = Tl + To; cannam@167: iio[WS(rs, 1)] = Tx + Ty; cannam@167: iio[WS(rs, 2)] = TR + TS; cannam@167: rio[WS(rs, 2)] = TF + TI; cannam@167: { cannam@167: E Ta, Tg, T5, Tb; cannam@167: Ta = T6 + T9; cannam@167: Tg = Tc + Tf; cannam@167: T5 = W[0]; cannam@167: Tb = W[1]; cannam@167: rio[WS(vs, 1)] = FMA(T5, Ta, Tb * Tg); cannam@167: iio[WS(vs, 1)] = FNMS(Tb, Ta, T5 * Tg); cannam@167: } cannam@167: { cannam@167: E TW, TY, TV, TX; cannam@167: TW = TK - TN; cannam@167: TY = TT - TQ; cannam@167: TV = W[2]; cannam@167: TX = W[3]; cannam@167: rio[WS(vs, 2) + WS(rs, 2)] = FMA(TV, TW, TX * TY); cannam@167: iio[WS(vs, 2) + WS(rs, 2)] = FNMS(TX, TW, TV * TY); cannam@167: } cannam@167: { cannam@167: E TC, TE, TB, TD; cannam@167: TC = Tq - Tt; cannam@167: TE = Tz - Tw; cannam@167: TB = W[2]; cannam@167: TD = W[3]; cannam@167: rio[WS(vs, 2) + WS(rs, 1)] = FMA(TB, TC, TD * TE); cannam@167: iio[WS(vs, 2) + WS(rs, 1)] = FNMS(TD, TC, TB * TE); cannam@167: } cannam@167: { cannam@167: E Tu, TA, Tp, Tv; cannam@167: Tu = Tq + Tt; cannam@167: TA = Tw + Tz; cannam@167: Tp = W[0]; cannam@167: Tv = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 1)] = FMA(Tp, Tu, Tv * TA); cannam@167: iio[WS(vs, 1) + WS(rs, 1)] = FNMS(Tv, Tu, Tp * TA); cannam@167: } cannam@167: { cannam@167: E TO, TU, TJ, TP; cannam@167: TO = TK + TN; cannam@167: TU = TQ + TT; cannam@167: TJ = W[0]; cannam@167: TP = W[1]; cannam@167: rio[WS(vs, 1) + WS(rs, 2)] = FMA(TJ, TO, TP * TU); cannam@167: iio[WS(vs, 1) + WS(rs, 2)] = FNMS(TP, TO, TJ * TU); cannam@167: } cannam@167: { cannam@167: E Ti, Tk, Th, Tj; cannam@167: Ti = T6 - T9; cannam@167: Tk = Tf - Tc; cannam@167: Th = W[2]; cannam@167: Tj = W[3]; cannam@167: rio[WS(vs, 2)] = FMA(Th, Ti, Tj * Tk); cannam@167: iio[WS(vs, 2)] = FNMS(Tj, Ti, Th * Tk); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 0, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 3, "q1_3", twinstr, &GENUS, {30, 18, 18, 0}, 0, 0, 0 }; cannam@167: cannam@167: void X(codelet_q1_3) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1_3, &desc); cannam@167: } cannam@167: #endif