Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:07:31 EDT 2018 */ Chris@82: Chris@82: #include "rdft/codelet-rdft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hb_6 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 32 FP multiplications, Chris@82: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), Chris@82: * 31 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@82: E Td, Tn, TO, TJ, TN, Tk, Tr, T3, TC, Ts, TQ, Ta, Tm, TF, TG; Chris@82: { Chris@82: E Tb, Tc, Tg, TH, Tj, TI; Chris@82: Tb = ci[WS(rs, 5)]; Chris@82: Tc = cr[WS(rs, 3)]; Chris@82: Td = Tb - Tc; Chris@82: { Chris@82: E Te, Tf, Th, Ti; Chris@82: Te = ci[WS(rs, 3)]; Chris@82: Tf = cr[WS(rs, 5)]; Chris@82: Tg = Te - Tf; Chris@82: TH = Te + Tf; Chris@82: Th = ci[WS(rs, 4)]; Chris@82: Ti = cr[WS(rs, 4)]; Chris@82: Tj = Th - Ti; Chris@82: TI = Th + Ti; Chris@82: } Chris@82: Tn = Tj - Tg; Chris@82: TO = TH - TI; Chris@82: TJ = TH + TI; Chris@82: TN = Tb + Tc; Chris@82: Tk = Tg + Tj; Chris@82: Tr = FNMS(KP500000000, Tk, Td); Chris@82: } Chris@82: { Chris@82: E T6, TD, T9, TE, T1, T2; Chris@82: T1 = cr[0]; Chris@82: T2 = ci[WS(rs, 2)]; Chris@82: T3 = T1 + T2; Chris@82: TC = T1 - T2; Chris@82: { Chris@82: E T4, T5, T7, T8; Chris@82: T4 = cr[WS(rs, 2)]; Chris@82: T5 = ci[0]; Chris@82: T6 = T4 + T5; Chris@82: TD = T4 - T5; Chris@82: T7 = ci[WS(rs, 1)]; Chris@82: T8 = cr[WS(rs, 1)]; Chris@82: T9 = T7 + T8; Chris@82: TE = T7 - T8; Chris@82: } Chris@82: Ts = T6 - T9; Chris@82: TQ = TD - TE; Chris@82: Ta = T6 + T9; Chris@82: Tm = FNMS(KP500000000, Ta, T3); Chris@82: TF = TD + TE; Chris@82: TG = FNMS(KP500000000, TF, TC); Chris@82: } Chris@82: cr[0] = T3 + Ta; Chris@82: ci[0] = Td + Tk; Chris@82: { Chris@82: E To, Tt, Tp, Tu, Tl, Tq; Chris@82: To = FNMS(KP866025403, Tn, Tm); Chris@82: Tt = FNMS(KP866025403, Ts, Tr); Chris@82: Tl = W[2]; Chris@82: Tp = Tl * To; Chris@82: Tu = Tl * Tt; Chris@82: Tq = W[3]; Chris@82: cr[WS(rs, 2)] = FNMS(Tq, Tt, Tp); Chris@82: ci[WS(rs, 2)] = FMA(Tq, To, Tu); Chris@82: } Chris@82: { Chris@82: E T13, TZ, T11, T12, T14, T10; Chris@82: T13 = TN + TO; Chris@82: T10 = TC + TF; Chris@82: TZ = W[4]; Chris@82: T11 = TZ * T10; Chris@82: T12 = W[5]; Chris@82: T14 = T12 * T10; Chris@82: cr[WS(rs, 3)] = FNMS(T12, T13, T11); Chris@82: ci[WS(rs, 3)] = FMA(TZ, T13, T14); Chris@82: } Chris@82: { Chris@82: E Tw, Tz, Tx, TA, Tv, Ty; Chris@82: Tw = FMA(KP866025403, Tn, Tm); Chris@82: Tz = FMA(KP866025403, Ts, Tr); Chris@82: Tv = W[6]; Chris@82: Tx = Tv * Tw; Chris@82: TA = Tv * Tz; Chris@82: Ty = W[7]; Chris@82: cr[WS(rs, 4)] = FNMS(Ty, Tz, Tx); Chris@82: ci[WS(rs, 4)] = FMA(Ty, Tw, TA); Chris@82: } Chris@82: { Chris@82: E TR, TX, TT, TV, TW, TY, TB, TL, TM, TS, TP, TU, TK; Chris@82: TP = FNMS(KP500000000, TO, TN); Chris@82: TR = FMA(KP866025403, TQ, TP); Chris@82: TX = FNMS(KP866025403, TQ, TP); Chris@82: TU = FMA(KP866025403, TJ, TG); Chris@82: TT = W[8]; Chris@82: TV = TT * TU; Chris@82: TW = W[9]; Chris@82: TY = TW * TU; Chris@82: TK = FNMS(KP866025403, TJ, TG); Chris@82: TB = W[0]; Chris@82: TL = TB * TK; Chris@82: TM = W[1]; Chris@82: TS = TM * TK; Chris@82: cr[WS(rs, 1)] = FNMS(TM, TR, TL); Chris@82: ci[WS(rs, 1)] = FMA(TB, TR, TS); Chris@82: cr[WS(rs, 5)] = FNMS(TW, TX, TV); Chris@82: ci[WS(rs, 5)] = FMA(TT, TX, TY); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 6}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 6, "hb_6", twinstr, &GENUS, {24, 10, 22, 0} }; Chris@82: Chris@82: void X(codelet_hb_6) (planner *p) { Chris@82: X(khc2hc_register) (p, hb_6, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hb_6 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 28 FP multiplications, Chris@82: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), Chris@82: * 27 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT m; Chris@82: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@82: E T3, Ty, Ta, TO, Tr, TB, Td, TE, Tk, TL, Tn, TH; Chris@82: { Chris@82: E T1, T2, Tb, Tc; Chris@82: T1 = cr[0]; Chris@82: T2 = ci[WS(rs, 2)]; Chris@82: T3 = T1 + T2; Chris@82: Ty = T1 - T2; Chris@82: { Chris@82: E T6, Tz, T9, TA; Chris@82: { Chris@82: E T4, T5, T7, T8; Chris@82: T4 = cr[WS(rs, 2)]; Chris@82: T5 = ci[0]; Chris@82: T6 = T4 + T5; Chris@82: Tz = T4 - T5; Chris@82: T7 = ci[WS(rs, 1)]; Chris@82: T8 = cr[WS(rs, 1)]; Chris@82: T9 = T7 + T8; Chris@82: TA = T7 - T8; Chris@82: } Chris@82: Ta = T6 + T9; Chris@82: TO = KP866025403 * (Tz - TA); Chris@82: Tr = KP866025403 * (T6 - T9); Chris@82: TB = Tz + TA; Chris@82: } Chris@82: Tb = ci[WS(rs, 5)]; Chris@82: Tc = cr[WS(rs, 3)]; Chris@82: Td = Tb - Tc; Chris@82: TE = Tb + Tc; Chris@82: { Chris@82: E Tg, TG, Tj, TF; Chris@82: { Chris@82: E Te, Tf, Th, Ti; Chris@82: Te = ci[WS(rs, 3)]; Chris@82: Tf = cr[WS(rs, 5)]; Chris@82: Tg = Te - Tf; Chris@82: TG = Te + Tf; Chris@82: Th = ci[WS(rs, 4)]; Chris@82: Ti = cr[WS(rs, 4)]; Chris@82: Tj = Th - Ti; Chris@82: TF = Th + Ti; Chris@82: } Chris@82: Tk = Tg + Tj; Chris@82: TL = KP866025403 * (TG + TF); Chris@82: Tn = KP866025403 * (Tj - Tg); Chris@82: TH = TF - TG; Chris@82: } Chris@82: } Chris@82: cr[0] = T3 + Ta; Chris@82: ci[0] = Td + Tk; Chris@82: { Chris@82: E TC, TI, Tx, TD; Chris@82: TC = Ty + TB; Chris@82: TI = TE - TH; Chris@82: Tx = W[4]; Chris@82: TD = W[5]; Chris@82: cr[WS(rs, 3)] = FNMS(TD, TI, Tx * TC); Chris@82: ci[WS(rs, 3)] = FMA(TD, TC, Tx * TI); Chris@82: } Chris@82: { Chris@82: E To, Tu, Ts, Tw, Tm, Tq; Chris@82: Tm = FNMS(KP500000000, Ta, T3); Chris@82: To = Tm - Tn; Chris@82: Tu = Tm + Tn; Chris@82: Tq = FNMS(KP500000000, Tk, Td); Chris@82: Ts = Tq - Tr; Chris@82: Tw = Tr + Tq; Chris@82: { Chris@82: E Tl, Tp, Tt, Tv; Chris@82: Tl = W[2]; Chris@82: Tp = W[3]; Chris@82: cr[WS(rs, 2)] = FNMS(Tp, Ts, Tl * To); Chris@82: ci[WS(rs, 2)] = FMA(Tl, Ts, Tp * To); Chris@82: Tt = W[6]; Chris@82: Tv = W[7]; Chris@82: cr[WS(rs, 4)] = FNMS(Tv, Tw, Tt * Tu); Chris@82: ci[WS(rs, 4)] = FMA(Tt, Tw, Tv * Tu); Chris@82: } Chris@82: } Chris@82: { Chris@82: E TM, TS, TQ, TU, TK, TP; Chris@82: TK = FNMS(KP500000000, TB, Ty); Chris@82: TM = TK - TL; Chris@82: TS = TK + TL; Chris@82: TP = FMA(KP500000000, TH, TE); Chris@82: TQ = TO + TP; Chris@82: TU = TP - TO; Chris@82: { Chris@82: E TJ, TN, TR, TT; Chris@82: TJ = W[0]; Chris@82: TN = W[1]; Chris@82: cr[WS(rs, 1)] = FNMS(TN, TQ, TJ * TM); Chris@82: ci[WS(rs, 1)] = FMA(TN, TM, TJ * TQ); Chris@82: TR = W[8]; Chris@82: TT = W[9]; Chris@82: cr[WS(rs, 5)] = FNMS(TT, TU, TR * TS); Chris@82: ci[WS(rs, 5)] = FMA(TT, TS, TR * TU); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 1, 6}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 6, "hb_6", twinstr, &GENUS, {32, 14, 14, 0} }; Chris@82: Chris@82: void X(codelet_hb_6) (planner *p) { Chris@82: X(khc2hc_register) (p, hb_6, &desc); Chris@82: } Chris@82: #endif