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 08:07:51 EDT 2018 */ cannam@167: cannam@167: #include "rdft/codelet-rdft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2c.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cb_6 -include rdft/scalar/hc2cb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 46 FP additions, 32 FP multiplications, cannam@167: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), cannam@167: * 31 stack variables, 2 constants, and 24 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cb.h" cannam@167: cannam@167: static void hc2cb_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, 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 - 1) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { cannam@167: E Td, Tn, TO, TJ, TN, Tk, Tr, T3, TC, Ts, TQ, Ta, Tm, TF, TG; cannam@167: { cannam@167: E Tb, Tc, Tj, TI, Tg, TH; cannam@167: Tb = Ip[0]; cannam@167: Tc = Im[WS(rs, 2)]; cannam@167: Td = Tb - Tc; cannam@167: { cannam@167: E Th, Ti, Te, Tf; cannam@167: Th = Ip[WS(rs, 1)]; cannam@167: Ti = Im[WS(rs, 1)]; cannam@167: Tj = Th - Ti; cannam@167: TI = Th + Ti; cannam@167: Te = Ip[WS(rs, 2)]; cannam@167: Tf = Im[0]; cannam@167: Tg = Te - Tf; cannam@167: TH = Te + Tf; cannam@167: } cannam@167: Tn = Tj - Tg; cannam@167: TO = TH - TI; cannam@167: TJ = TH + TI; cannam@167: TN = Tb + Tc; cannam@167: Tk = Tg + Tj; cannam@167: Tr = FNMS(KP500000000, Tk, Td); cannam@167: } cannam@167: { cannam@167: E T9, TE, T6, TD, T1, T2; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 2)]; cannam@167: T3 = T1 + T2; cannam@167: TC = T1 - T2; cannam@167: { cannam@167: E T7, T8, T4, T5; cannam@167: T7 = Rm[WS(rs, 1)]; cannam@167: T8 = Rp[WS(rs, 1)]; cannam@167: T9 = T7 + T8; cannam@167: TE = T7 - T8; cannam@167: T4 = Rp[WS(rs, 2)]; cannam@167: T5 = Rm[0]; cannam@167: T6 = T4 + T5; cannam@167: TD = T4 - T5; cannam@167: } cannam@167: Ts = T6 - T9; cannam@167: TQ = TD - TE; cannam@167: Ta = T6 + T9; cannam@167: Tm = FNMS(KP500000000, Ta, T3); cannam@167: TF = TD + TE; cannam@167: TG = FNMS(KP500000000, TF, TC); cannam@167: } cannam@167: Rp[0] = T3 + Ta; cannam@167: Rm[0] = Td + Tk; cannam@167: { cannam@167: E To, Tt, Tp, Tu, Tl, Tq; cannam@167: To = FNMS(KP866025403, Tn, Tm); cannam@167: Tt = FNMS(KP866025403, Ts, Tr); cannam@167: Tl = W[2]; cannam@167: Tp = Tl * To; cannam@167: Tu = Tl * Tt; cannam@167: Tq = W[3]; cannam@167: Rp[WS(rs, 1)] = FNMS(Tq, Tt, Tp); cannam@167: Rm[WS(rs, 1)] = FMA(Tq, To, Tu); cannam@167: } cannam@167: { cannam@167: E T13, TZ, T11, T12, T14, T10; cannam@167: T13 = TN + TO; cannam@167: T10 = TC + TF; cannam@167: TZ = W[4]; cannam@167: T11 = TZ * T10; cannam@167: T12 = W[5]; cannam@167: T14 = T12 * T10; cannam@167: Ip[WS(rs, 1)] = FNMS(T12, T13, T11); cannam@167: Im[WS(rs, 1)] = FMA(TZ, T13, T14); cannam@167: } cannam@167: { cannam@167: E Tw, Tz, Tx, TA, Tv, Ty; cannam@167: Tw = FMA(KP866025403, Tn, Tm); cannam@167: Tz = FMA(KP866025403, Ts, Tr); cannam@167: Tv = W[6]; cannam@167: Tx = Tv * Tw; cannam@167: TA = Tv * Tz; cannam@167: Ty = W[7]; cannam@167: Rp[WS(rs, 2)] = FNMS(Ty, Tz, Tx); cannam@167: Rm[WS(rs, 2)] = FMA(Ty, Tw, TA); cannam@167: } cannam@167: { cannam@167: E TR, TX, TT, TV, TW, TY, TB, TL, TM, TS, TP, TU, TK; cannam@167: TP = FNMS(KP500000000, TO, TN); cannam@167: TR = FMA(KP866025403, TQ, TP); cannam@167: TX = FNMS(KP866025403, TQ, TP); cannam@167: TU = FMA(KP866025403, TJ, TG); cannam@167: TT = W[8]; cannam@167: TV = TT * TU; cannam@167: TW = W[9]; cannam@167: TY = TW * TU; cannam@167: TK = FNMS(KP866025403, TJ, TG); cannam@167: TB = W[0]; cannam@167: TL = TB * TK; cannam@167: TM = W[1]; cannam@167: TS = TM * TK; cannam@167: Ip[0] = FNMS(TM, TR, TL); cannam@167: Im[0] = FMA(TB, TR, TS); cannam@167: Ip[WS(rs, 2)] = FNMS(TW, TX, TV); cannam@167: Im[WS(rs, 2)] = FMA(TT, TX, TY); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 6}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {24, 10, 22, 0} }; cannam@167: cannam@167: void X(codelet_hc2cb_6) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cb_6 -include rdft/scalar/hc2cb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 46 FP additions, 28 FP multiplications, cannam@167: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), cannam@167: * 25 stack variables, 2 constants, and 24 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cb.h" cannam@167: cannam@167: static void hc2cb_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { cannam@167: E T3, Ty, Td, TE, Ta, TO, Tr, TB, Tk, TL, Tn, TH; cannam@167: { cannam@167: E T1, T2, Tb, Tc; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 2)]; cannam@167: T3 = T1 + T2; cannam@167: Ty = T1 - T2; cannam@167: Tb = Ip[0]; cannam@167: Tc = Im[WS(rs, 2)]; cannam@167: Td = Tb - Tc; cannam@167: TE = Tb + Tc; cannam@167: } cannam@167: { cannam@167: E T6, Tz, T9, TA; cannam@167: { cannam@167: E T4, T5, T7, T8; cannam@167: T4 = Rp[WS(rs, 2)]; cannam@167: T5 = Rm[0]; cannam@167: T6 = T4 + T5; cannam@167: Tz = T4 - T5; cannam@167: T7 = Rm[WS(rs, 1)]; cannam@167: T8 = Rp[WS(rs, 1)]; cannam@167: T9 = T7 + T8; cannam@167: TA = T7 - T8; cannam@167: } cannam@167: Ta = T6 + T9; cannam@167: TO = KP866025403 * (Tz - TA); cannam@167: Tr = KP866025403 * (T6 - T9); cannam@167: TB = Tz + TA; cannam@167: } cannam@167: { cannam@167: E Tg, TG, Tj, TF; cannam@167: { cannam@167: E Te, Tf, Th, Ti; cannam@167: Te = Ip[WS(rs, 2)]; cannam@167: Tf = Im[0]; cannam@167: Tg = Te - Tf; cannam@167: TG = Te + Tf; cannam@167: Th = Ip[WS(rs, 1)]; cannam@167: Ti = Im[WS(rs, 1)]; cannam@167: Tj = Th - Ti; cannam@167: TF = Th + Ti; cannam@167: } cannam@167: Tk = Tg + Tj; cannam@167: TL = KP866025403 * (TG + TF); cannam@167: Tn = KP866025403 * (Tj - Tg); cannam@167: TH = TF - TG; cannam@167: } cannam@167: Rp[0] = T3 + Ta; cannam@167: Rm[0] = Td + Tk; cannam@167: { cannam@167: E TC, TI, Tx, TD; cannam@167: TC = Ty + TB; cannam@167: TI = TE - TH; cannam@167: Tx = W[4]; cannam@167: TD = W[5]; cannam@167: Ip[WS(rs, 1)] = FNMS(TD, TI, Tx * TC); cannam@167: Im[WS(rs, 1)] = FMA(TD, TC, Tx * TI); cannam@167: } cannam@167: { cannam@167: E To, Tu, Ts, Tw, Tm, Tq; cannam@167: Tm = FNMS(KP500000000, Ta, T3); cannam@167: To = Tm - Tn; cannam@167: Tu = Tm + Tn; cannam@167: Tq = FNMS(KP500000000, Tk, Td); cannam@167: Ts = Tq - Tr; cannam@167: Tw = Tr + Tq; cannam@167: { cannam@167: E Tl, Tp, Tt, Tv; cannam@167: Tl = W[2]; cannam@167: Tp = W[3]; cannam@167: Rp[WS(rs, 1)] = FNMS(Tp, Ts, Tl * To); cannam@167: Rm[WS(rs, 1)] = FMA(Tl, Ts, Tp * To); cannam@167: Tt = W[6]; cannam@167: Tv = W[7]; cannam@167: Rp[WS(rs, 2)] = FNMS(Tv, Tw, Tt * Tu); cannam@167: Rm[WS(rs, 2)] = FMA(Tt, Tw, Tv * Tu); cannam@167: } cannam@167: } cannam@167: { cannam@167: E TM, TS, TQ, TU, TK, TP; cannam@167: TK = FNMS(KP500000000, TB, Ty); cannam@167: TM = TK - TL; cannam@167: TS = TK + TL; cannam@167: TP = FMA(KP500000000, TH, TE); cannam@167: TQ = TO + TP; cannam@167: TU = TP - TO; cannam@167: { cannam@167: E TJ, TN, TR, TT; cannam@167: TJ = W[0]; cannam@167: TN = W[1]; cannam@167: Ip[0] = FNMS(TN, TQ, TJ * TM); cannam@167: Im[0] = FMA(TN, TM, TJ * TQ); cannam@167: TR = W[8]; cannam@167: TT = W[9]; cannam@167: Ip[WS(rs, 2)] = FNMS(TT, TU, TR * TS); cannam@167: Im[WS(rs, 2)] = FMA(TT, TS, TR * TU); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 6}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 6, "hc2cb_6", twinstr, &GENUS, {32, 14, 14, 0} }; cannam@167: cannam@167: void X(codelet_hc2cb_6) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cb_6, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #endif