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 08:07:57 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_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cbdft_6 -include rdft/scalar/hc2cb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 58 FP additions, 32 FP multiplications, Chris@82: * (or, 36 additions, 10 multiplications, 22 fused multiply/add), Chris@82: * 34 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hc2cb.h" Chris@82: Chris@82: static void hc2cbdft_6(R *Rp, R *Ip, R *Rm, R *Im, 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@82: E Tp, TD, Tj, TV, Tq, Tr, TG, TP, T4, Ts, TQ, Tb, Tc, TA, TU; Chris@82: { Chris@82: E Tf, TF, Ti, TE, Td, Te; Chris@82: Td = Ip[WS(rs, 1)]; Chris@82: Te = Im[WS(rs, 1)]; Chris@82: Tf = Td - Te; Chris@82: TF = Te + Td; Chris@82: { Chris@82: E Tn, To, Tg, Th; Chris@82: Tn = Ip[0]; Chris@82: To = Im[WS(rs, 2)]; Chris@82: Tp = Tn - To; Chris@82: TD = Tn + To; Chris@82: Tg = Ip[WS(rs, 2)]; Chris@82: Th = Im[0]; Chris@82: Ti = Tg - Th; Chris@82: TE = Tg + Th; Chris@82: } Chris@82: Tj = Tf - Ti; Chris@82: TV = TF + TE; Chris@82: Tq = Tf + Ti; Chris@82: Tr = FNMS(KP500000000, Tq, Tp); Chris@82: TG = TE - TF; Chris@82: TP = FNMS(KP500000000, TG, TD); Chris@82: } Chris@82: { Chris@82: E Tw, Ta, Ty, T7, Tx, T2, T3, Tz; Chris@82: T2 = Rp[0]; Chris@82: T3 = Rm[WS(rs, 2)]; Chris@82: T4 = T2 + T3; Chris@82: Tw = T2 - T3; Chris@82: { Chris@82: E T8, T9, T5, T6; Chris@82: T8 = Rm[WS(rs, 1)]; Chris@82: T9 = Rp[WS(rs, 1)]; Chris@82: Ta = T8 + T9; Chris@82: Ty = T8 - T9; Chris@82: T5 = Rp[WS(rs, 2)]; Chris@82: T6 = Rm[0]; Chris@82: T7 = T5 + T6; Chris@82: Tx = T5 - T6; Chris@82: } Chris@82: Ts = T7 - Ta; Chris@82: TQ = Tx - Ty; Chris@82: Tb = T7 + Ta; Chris@82: Tc = FNMS(KP500000000, Tb, T4); Chris@82: Tz = Tx + Ty; Chris@82: TA = Tw + Tz; Chris@82: TU = FNMS(KP500000000, Tz, Tw); Chris@82: } Chris@82: { Chris@82: E TN, TY, TR, TW, TS, TZ, TO, TX, T10, TT; Chris@82: TN = T4 + Tb; Chris@82: TY = Tp + Tq; Chris@82: TR = FMA(KP866025403, TQ, TP); Chris@82: TW = FNMS(KP866025403, TV, TU); Chris@82: TO = W[0]; Chris@82: TS = TO * TR; Chris@82: TZ = TO * TW; Chris@82: TT = W[1]; Chris@82: TX = FMA(TT, TW, TS); Chris@82: T10 = FNMS(TT, TR, TZ); Chris@82: Rp[0] = TN - TX; Chris@82: Ip[0] = TY + T10; Chris@82: Rm[0] = TN + TX; Chris@82: Im[0] = T10 - TY; Chris@82: } Chris@82: { Chris@82: E Tt, TH, Tv, TB, TC, TL, T1, Tl, Tm, TJ, Tk; Chris@82: Tt = FNMS(KP866025403, Ts, Tr); Chris@82: TH = TD + TG; Chris@82: Tv = W[4]; Chris@82: TB = Tv * TA; Chris@82: TC = W[5]; Chris@82: TL = TC * TA; Chris@82: Tk = FNMS(KP866025403, Tj, Tc); Chris@82: T1 = W[3]; Chris@82: Tl = T1 * Tk; Chris@82: Tm = W[2]; Chris@82: TJ = Tm * Tk; Chris@82: { Chris@82: E Tu, TI, TK, TM; Chris@82: Tu = FMA(Tm, Tt, Tl); Chris@82: TI = FNMS(TC, TH, TB); Chris@82: Ip[WS(rs, 1)] = Tu + TI; Chris@82: Im[WS(rs, 1)] = TI - Tu; Chris@82: TK = FNMS(T1, Tt, TJ); Chris@82: TM = FMA(Tv, TH, TL); Chris@82: Rp[WS(rs, 1)] = TK - TM; Chris@82: Rm[WS(rs, 1)] = TK + TM; Chris@82: } Chris@82: } Chris@82: { Chris@82: E T15, T11, T13, T14, T1d, T18, T1b, T19, T1f, T12, T17; Chris@82: T15 = FMA(KP866025403, Ts, Tr); Chris@82: T12 = FMA(KP866025403, Tj, Tc); Chris@82: T11 = W[6]; Chris@82: T13 = T11 * T12; Chris@82: T14 = W[7]; Chris@82: T1d = T14 * T12; Chris@82: T18 = FNMS(KP866025403, TQ, TP); Chris@82: T1b = FMA(KP866025403, TV, TU); Chris@82: T17 = W[8]; Chris@82: T19 = T17 * T18; Chris@82: T1f = T17 * T1b; Chris@82: { Chris@82: E T16, T1e, T1c, T1g, T1a; Chris@82: T16 = FNMS(T14, T15, T13); Chris@82: T1e = FMA(T11, T15, T1d); Chris@82: T1a = W[9]; Chris@82: T1c = FMA(T1a, T1b, T19); Chris@82: T1g = FNMS(T1a, T18, T1f); Chris@82: Rp[WS(rs, 2)] = T16 - T1c; Chris@82: Ip[WS(rs, 2)] = T1e + T1g; Chris@82: Rm[WS(rs, 2)] = T16 + T1c; Chris@82: Im[WS(rs, 2)] = T1g - T1e; 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 hc2c_desc desc = { 6, "hc2cbdft_6", twinstr, &GENUS, {36, 10, 22, 0} }; Chris@82: Chris@82: void X(codelet_hc2cbdft_6) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cbdft_6, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 6 -dif -name hc2cbdft_6 -include rdft/scalar/hc2cb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 58 FP additions, 28 FP multiplications, Chris@82: * (or, 44 additions, 14 multiplications, 14 fused multiply/add), Chris@82: * 29 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hc2cb.h" Chris@82: Chris@82: static void hc2cbdft_6(R *Rp, R *Ip, R *Rm, R *Im, 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@82: E T4, Tv, Tr, TL, Tb, Tc, Ty, TP, To, TB, Tj, TQ, Tp, Tq, TE; Chris@82: E TM; Chris@82: { Chris@82: E Ta, Tx, T7, Tw, T2, T3; Chris@82: T2 = Rp[0]; Chris@82: T3 = Rm[WS(rs, 2)]; Chris@82: T4 = T2 + T3; Chris@82: Tv = T2 - T3; Chris@82: { Chris@82: E T8, T9, T5, T6; Chris@82: T8 = Rm[WS(rs, 1)]; Chris@82: T9 = Rp[WS(rs, 1)]; Chris@82: Ta = T8 + T9; Chris@82: Tx = T8 - T9; Chris@82: T5 = Rp[WS(rs, 2)]; Chris@82: T6 = Rm[0]; Chris@82: T7 = T5 + T6; Chris@82: Tw = T5 - T6; Chris@82: } Chris@82: Tr = KP866025403 * (T7 - Ta); Chris@82: TL = KP866025403 * (Tw - Tx); Chris@82: Tb = T7 + Ta; Chris@82: Tc = FNMS(KP500000000, Tb, T4); Chris@82: Ty = Tw + Tx; Chris@82: TP = FNMS(KP500000000, Ty, Tv); Chris@82: } Chris@82: { Chris@82: E Tf, TC, Ti, TD, Td, Te; Chris@82: Td = Ip[WS(rs, 1)]; Chris@82: Te = Im[WS(rs, 1)]; Chris@82: Tf = Td - Te; Chris@82: TC = Te + Td; Chris@82: { Chris@82: E Tm, Tn, Tg, Th; Chris@82: Tm = Ip[0]; Chris@82: Tn = Im[WS(rs, 2)]; Chris@82: To = Tm - Tn; Chris@82: TB = Tm + Tn; Chris@82: Tg = Ip[WS(rs, 2)]; Chris@82: Th = Im[0]; Chris@82: Ti = Tg - Th; Chris@82: TD = Tg + Th; Chris@82: } Chris@82: Tj = KP866025403 * (Tf - Ti); Chris@82: TQ = KP866025403 * (TC + TD); Chris@82: Tp = Tf + Ti; Chris@82: Tq = FNMS(KP500000000, Tp, To); Chris@82: TE = TC - TD; Chris@82: TM = FMA(KP500000000, TE, TB); Chris@82: } Chris@82: { Chris@82: E TJ, TT, TS, TU; Chris@82: TJ = T4 + Tb; Chris@82: TT = To + Tp; Chris@82: { Chris@82: E TN, TR, TK, TO; Chris@82: TN = TL + TM; Chris@82: TR = TP - TQ; Chris@82: TK = W[0]; Chris@82: TO = W[1]; Chris@82: TS = FMA(TK, TN, TO * TR); Chris@82: TU = FNMS(TO, TN, TK * TR); Chris@82: } Chris@82: Rp[0] = TJ - TS; Chris@82: Ip[0] = TT + TU; Chris@82: Rm[0] = TJ + TS; Chris@82: Im[0] = TU - TT; Chris@82: } Chris@82: { Chris@82: E TZ, T15, T14, T16; Chris@82: { Chris@82: E TW, TY, TV, TX; Chris@82: TW = Tc + Tj; Chris@82: TY = Tr + Tq; Chris@82: TV = W[6]; Chris@82: TX = W[7]; Chris@82: TZ = FNMS(TX, TY, TV * TW); Chris@82: T15 = FMA(TX, TW, TV * TY); Chris@82: } Chris@82: { Chris@82: E T11, T13, T10, T12; Chris@82: T11 = TM - TL; Chris@82: T13 = TP + TQ; Chris@82: T10 = W[8]; Chris@82: T12 = W[9]; Chris@82: T14 = FMA(T10, T11, T12 * T13); Chris@82: T16 = FNMS(T12, T11, T10 * T13); Chris@82: } Chris@82: Rp[WS(rs, 2)] = TZ - T14; Chris@82: Ip[WS(rs, 2)] = T15 + T16; Chris@82: Rm[WS(rs, 2)] = TZ + T14; Chris@82: Im[WS(rs, 2)] = T16 - T15; Chris@82: } Chris@82: { Chris@82: E Tt, TH, TG, TI; Chris@82: { Chris@82: E Tk, Ts, T1, Tl; Chris@82: Tk = Tc - Tj; Chris@82: Ts = Tq - Tr; Chris@82: T1 = W[3]; Chris@82: Tl = W[2]; Chris@82: Tt = FMA(T1, Tk, Tl * Ts); Chris@82: TH = FNMS(T1, Ts, Tl * Tk); Chris@82: } Chris@82: { Chris@82: E Tz, TF, Tu, TA; Chris@82: Tz = Tv + Ty; Chris@82: TF = TB - TE; Chris@82: Tu = W[4]; Chris@82: TA = W[5]; Chris@82: TG = FNMS(TA, TF, Tu * Tz); Chris@82: TI = FMA(TA, Tz, Tu * TF); Chris@82: } Chris@82: Ip[WS(rs, 1)] = Tt + TG; Chris@82: Rp[WS(rs, 1)] = TH - TI; Chris@82: Im[WS(rs, 1)] = TG - Tt; Chris@82: Rm[WS(rs, 1)] = TH + TI; 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 hc2c_desc desc = { 6, "hc2cbdft_6", twinstr, &GENUS, {44, 14, 14, 0} }; Chris@82: Chris@82: void X(codelet_hc2cbdft_6) (planner *p) { Chris@82: X(khc2c_register) (p, hc2cbdft_6, &desc, HC2C_VIA_DFT); Chris@82: } Chris@82: #endif