cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:51:55 EDT 2016 */ cannam@127: cannam@127: #include "codelet-rdft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cbdft_6 -include hc2cb.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 58 FP additions, 32 FP multiplications, cannam@127: * (or, 36 additions, 10 multiplications, 22 fused multiply/add), cannam@127: * 52 stack variables, 2 constants, and 24 memory accesses cannam@127: */ cannam@127: #include "hc2cb.h" cannam@127: cannam@127: static void hc2cbdft_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T18, T1b, T16, T1e, T1a, T1f, T19, T1g, T1c; cannam@127: { cannam@127: E Tw, T4, TV, Tj, TP, TH, Tr, TY, T5, T6, Ta, Ty; cannam@127: { cannam@127: E Tg, TF, Tf, TD, Tp, Th; cannam@127: { cannam@127: E Td, Te, Tn, To; cannam@127: Td = Ip[WS(rs, 1)]; cannam@127: Te = Im[WS(rs, 1)]; cannam@127: Tn = Ip[0]; cannam@127: To = Im[WS(rs, 2)]; cannam@127: Tg = Ip[WS(rs, 2)]; cannam@127: TF = Te + Td; cannam@127: Tf = Td - Te; cannam@127: TD = Tn + To; cannam@127: Tp = Tn - To; cannam@127: Th = Im[0]; cannam@127: } cannam@127: { cannam@127: E T2, T3, T8, T9; cannam@127: T2 = Rp[0]; cannam@127: T3 = Rm[WS(rs, 2)]; cannam@127: { cannam@127: E Tq, TE, Ti, TG; cannam@127: T8 = Rm[WS(rs, 1)]; cannam@127: TE = Tg + Th; cannam@127: Ti = Tg - Th; cannam@127: Tw = T2 - T3; cannam@127: T4 = T2 + T3; cannam@127: TG = TE - TF; cannam@127: TV = TF + TE; cannam@127: Tq = Tf + Ti; cannam@127: Tj = Tf - Ti; cannam@127: TP = FNMS(KP500000000, TG, TD); cannam@127: TH = TD + TG; cannam@127: T9 = Rp[WS(rs, 1)]; cannam@127: Tr = FNMS(KP500000000, Tq, Tp); cannam@127: TY = Tp + Tq; cannam@127: } cannam@127: T5 = Rp[WS(rs, 2)]; cannam@127: T6 = Rm[0]; cannam@127: Ta = T8 + T9; cannam@127: Ty = T8 - T9; cannam@127: } cannam@127: } cannam@127: { cannam@127: E TO, TT, Ts, TA, TR, Tc, TN, TW, TS, Tx, T7; cannam@127: Tx = T5 - T6; cannam@127: T7 = T5 + T6; cannam@127: TO = W[0]; cannam@127: TT = W[1]; cannam@127: { cannam@127: E Tz, TQ, Tb, TU; cannam@127: Tz = Tx + Ty; cannam@127: TQ = Tx - Ty; cannam@127: Tb = T7 + Ta; cannam@127: Ts = T7 - Ta; cannam@127: TU = FNMS(KP500000000, Tz, Tw); cannam@127: TA = Tw + Tz; cannam@127: TR = FMA(KP866025403, TQ, TP); cannam@127: T18 = FNMS(KP866025403, TQ, TP); cannam@127: Tc = FNMS(KP500000000, Tb, T4); cannam@127: TN = T4 + Tb; cannam@127: T1b = FMA(KP866025403, TV, TU); cannam@127: TW = FNMS(KP866025403, TV, TU); cannam@127: TS = TO * TR; cannam@127: } cannam@127: { cannam@127: E T15, Tt, T12, T1, Tm, TI, TM, Tl, TJ; cannam@127: { cannam@127: E Tv, TC, TB, TL, Tk, TZ, TX, T10; cannam@127: T15 = FMA(KP866025403, Ts, Tr); cannam@127: Tt = FNMS(KP866025403, Ts, Tr); cannam@127: TZ = TO * TW; cannam@127: TX = FMA(TT, TW, TS); cannam@127: Tv = W[4]; cannam@127: TC = W[5]; cannam@127: T10 = FNMS(TT, TR, TZ); cannam@127: Rm[0] = TN + TX; cannam@127: Rp[0] = TN - TX; cannam@127: TB = Tv * TA; cannam@127: Im[0] = T10 - TY; cannam@127: Ip[0] = TY + T10; cannam@127: TL = TC * TA; cannam@127: Tk = FNMS(KP866025403, Tj, Tc); cannam@127: T12 = FMA(KP866025403, Tj, Tc); cannam@127: T1 = W[3]; cannam@127: Tm = W[2]; cannam@127: TI = FNMS(TC, TH, TB); cannam@127: TM = FMA(Tv, TH, TL); cannam@127: Tl = T1 * Tk; cannam@127: TJ = Tm * Tk; cannam@127: } cannam@127: { cannam@127: E T11, T14, T13, T1d, T17, Tu, TK; cannam@127: Tu = FMA(Tm, Tt, Tl); cannam@127: TK = FNMS(T1, Tt, TJ); cannam@127: T11 = W[6]; cannam@127: T14 = W[7]; cannam@127: Im[WS(rs, 1)] = TI - Tu; cannam@127: Ip[WS(rs, 1)] = Tu + TI; cannam@127: Rm[WS(rs, 1)] = TK + TM; cannam@127: Rp[WS(rs, 1)] = TK - TM; cannam@127: T13 = T11 * T12; cannam@127: T1d = T14 * T12; cannam@127: T17 = W[8]; cannam@127: T16 = FNMS(T14, T15, T13); cannam@127: T1e = FMA(T11, T15, T1d); cannam@127: T1a = W[9]; cannam@127: T1f = T17 * T1b; cannam@127: T19 = T17 * T18; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: T1g = FNMS(T1a, T18, T1f); cannam@127: T1c = FMA(T1a, T1b, T19); cannam@127: Im[WS(rs, 2)] = T1g - T1e; cannam@127: Ip[WS(rs, 2)] = T1e + T1g; cannam@127: Rm[WS(rs, 2)] = T16 + T1c; cannam@127: Rp[WS(rs, 2)] = T16 - T1c; cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 6}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2c_desc desc = { 6, "hc2cbdft_6", twinstr, &GENUS, {36, 10, 22, 0} }; cannam@127: cannam@127: void X(codelet_hc2cbdft_6) (planner *p) { cannam@127: X(khc2c_register) (p, hc2cbdft_6, &desc, HC2C_VIA_DFT); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cbdft_6 -include hc2cb.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 58 FP additions, 28 FP multiplications, cannam@127: * (or, 44 additions, 14 multiplications, 14 fused multiply/add), cannam@127: * 29 stack variables, 2 constants, and 24 memory accesses cannam@127: */ cannam@127: #include "hc2cb.h" cannam@127: cannam@127: static void hc2cbdft_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@127: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T4, Tv, Tr, TL, Tb, Tc, Ty, TP, To, TB, Tj, TQ, Tp, Tq, TE; cannam@127: E TM; cannam@127: { cannam@127: E Ta, Tx, T7, Tw, T2, T3; cannam@127: T2 = Rp[0]; cannam@127: T3 = Rm[WS(rs, 2)]; cannam@127: T4 = T2 + T3; cannam@127: Tv = T2 - T3; cannam@127: { cannam@127: E T8, T9, T5, T6; cannam@127: T8 = Rm[WS(rs, 1)]; cannam@127: T9 = Rp[WS(rs, 1)]; cannam@127: Ta = T8 + T9; cannam@127: Tx = T8 - T9; cannam@127: T5 = Rp[WS(rs, 2)]; cannam@127: T6 = Rm[0]; cannam@127: T7 = T5 + T6; cannam@127: Tw = T5 - T6; cannam@127: } cannam@127: Tr = KP866025403 * (T7 - Ta); cannam@127: TL = KP866025403 * (Tw - Tx); cannam@127: Tb = T7 + Ta; cannam@127: Tc = FNMS(KP500000000, Tb, T4); cannam@127: Ty = Tw + Tx; cannam@127: TP = FNMS(KP500000000, Ty, Tv); cannam@127: } cannam@127: { cannam@127: E Tf, TC, Ti, TD, Td, Te; cannam@127: Td = Ip[WS(rs, 1)]; cannam@127: Te = Im[WS(rs, 1)]; cannam@127: Tf = Td - Te; cannam@127: TC = Te + Td; cannam@127: { cannam@127: E Tm, Tn, Tg, Th; cannam@127: Tm = Ip[0]; cannam@127: Tn = Im[WS(rs, 2)]; cannam@127: To = Tm - Tn; cannam@127: TB = Tm + Tn; cannam@127: Tg = Ip[WS(rs, 2)]; cannam@127: Th = Im[0]; cannam@127: Ti = Tg - Th; cannam@127: TD = Tg + Th; cannam@127: } cannam@127: Tj = KP866025403 * (Tf - Ti); cannam@127: TQ = KP866025403 * (TC + TD); cannam@127: Tp = Tf + Ti; cannam@127: Tq = FNMS(KP500000000, Tp, To); cannam@127: TE = TC - TD; cannam@127: TM = FMA(KP500000000, TE, TB); cannam@127: } cannam@127: { cannam@127: E TJ, TT, TS, TU; cannam@127: TJ = T4 + Tb; cannam@127: TT = To + Tp; cannam@127: { cannam@127: E TN, TR, TK, TO; cannam@127: TN = TL + TM; cannam@127: TR = TP - TQ; cannam@127: TK = W[0]; cannam@127: TO = W[1]; cannam@127: TS = FMA(TK, TN, TO * TR); cannam@127: TU = FNMS(TO, TN, TK * TR); cannam@127: } cannam@127: Rp[0] = TJ - TS; cannam@127: Ip[0] = TT + TU; cannam@127: Rm[0] = TJ + TS; cannam@127: Im[0] = TU - TT; cannam@127: } cannam@127: { cannam@127: E TZ, T15, T14, T16; cannam@127: { cannam@127: E TW, TY, TV, TX; cannam@127: TW = Tc + Tj; cannam@127: TY = Tr + Tq; cannam@127: TV = W[6]; cannam@127: TX = W[7]; cannam@127: TZ = FNMS(TX, TY, TV * TW); cannam@127: T15 = FMA(TX, TW, TV * TY); cannam@127: } cannam@127: { cannam@127: E T11, T13, T10, T12; cannam@127: T11 = TM - TL; cannam@127: T13 = TP + TQ; cannam@127: T10 = W[8]; cannam@127: T12 = W[9]; cannam@127: T14 = FMA(T10, T11, T12 * T13); cannam@127: T16 = FNMS(T12, T11, T10 * T13); cannam@127: } cannam@127: Rp[WS(rs, 2)] = TZ - T14; cannam@127: Ip[WS(rs, 2)] = T15 + T16; cannam@127: Rm[WS(rs, 2)] = TZ + T14; cannam@127: Im[WS(rs, 2)] = T16 - T15; cannam@127: } cannam@127: { cannam@127: E Tt, TH, TG, TI; cannam@127: { cannam@127: E Tk, Ts, T1, Tl; cannam@127: Tk = Tc - Tj; cannam@127: Ts = Tq - Tr; cannam@127: T1 = W[3]; cannam@127: Tl = W[2]; cannam@127: Tt = FMA(T1, Tk, Tl * Ts); cannam@127: TH = FNMS(T1, Ts, Tl * Tk); cannam@127: } cannam@127: { cannam@127: E Tz, TF, Tu, TA; cannam@127: Tz = Tv + Ty; cannam@127: TF = TB - TE; cannam@127: Tu = W[4]; cannam@127: TA = W[5]; cannam@127: TG = FNMS(TA, TF, Tu * Tz); cannam@127: TI = FMA(TA, Tz, Tu * TF); cannam@127: } cannam@127: Ip[WS(rs, 1)] = Tt + TG; cannam@127: Rp[WS(rs, 1)] = TH - TI; cannam@127: Im[WS(rs, 1)] = TG - Tt; cannam@127: Rm[WS(rs, 1)] = TH + TI; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_FULL, 1, 6}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2c_desc desc = { 6, "hc2cbdft_6", twinstr, &GENUS, {44, 14, 14, 0} }; cannam@127: cannam@127: void X(codelet_hc2cbdft_6) (planner *p) { cannam@127: X(khc2c_register) (p, hc2cbdft_6, &desc, HC2C_VIA_DFT); cannam@127: } cannam@127: #endif /* HAVE_FMA */