cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:35:49 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 8 -name t1_8 -include t.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 36 FP multiplications, cannam@95: * (or, 44 additions, 14 multiplications, 22 fused multiply/add), cannam@95: * 61 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "t.h" cannam@95: cannam@95: static void t1_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 14); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 14, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: E T1g, T1f, T1e, Tm, T1q, T1o, T1p, TN, T1h, T1i; cannam@95: { cannam@95: E T1, T1m, T1l, T7, TS, Tk, TQ, Te, To, Tr, T17, TM, T12, Tu, TW; cannam@95: E Tp, Tx, Tt, Tq, Tw; cannam@95: { cannam@95: E T3, T6, T2, T5; cannam@95: T1 = ri[0]; cannam@95: T1m = ii[0]; cannam@95: T3 = ri[WS(rs, 4)]; cannam@95: T6 = ii[WS(rs, 4)]; cannam@95: T2 = W[6]; cannam@95: T5 = W[7]; cannam@95: { cannam@95: E Ta, Td, T9, Tc; cannam@95: { cannam@95: E Tg, Tj, Ti, TR, Th, T1k, T4, Tf; cannam@95: Tg = ri[WS(rs, 6)]; cannam@95: Tj = ii[WS(rs, 6)]; cannam@95: T1k = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: Tf = W[10]; cannam@95: Ti = W[11]; cannam@95: T1l = FNMS(T5, T3, T1k); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: TR = Tf * Tj; cannam@95: Th = Tf * Tg; cannam@95: Ta = ri[WS(rs, 2)]; cannam@95: Td = ii[WS(rs, 2)]; cannam@95: TS = FNMS(Ti, Tg, TR); cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: T9 = W[2]; cannam@95: Tc = W[3]; cannam@95: } cannam@95: { cannam@95: E TB, TE, TH, T13, TC, TK, TG, TD, TJ, TP, Tb, TA, Tn; cannam@95: TB = ri[WS(rs, 7)]; cannam@95: TE = ii[WS(rs, 7)]; cannam@95: TP = T9 * Td; cannam@95: Tb = T9 * Ta; cannam@95: TA = W[12]; cannam@95: TH = ri[WS(rs, 3)]; cannam@95: TQ = FNMS(Tc, Ta, TP); cannam@95: Te = FMA(Tc, Td, Tb); cannam@95: T13 = TA * TE; cannam@95: TC = TA * TB; cannam@95: TK = ii[WS(rs, 3)]; cannam@95: TG = W[4]; cannam@95: TD = W[13]; cannam@95: TJ = W[5]; cannam@95: { cannam@95: E T14, TF, T16, TL, T15, TI; cannam@95: To = ri[WS(rs, 1)]; cannam@95: T15 = TG * TK; cannam@95: TI = TG * TH; cannam@95: T14 = FNMS(TD, TB, T13); cannam@95: TF = FMA(TD, TE, TC); cannam@95: T16 = FNMS(TJ, TH, T15); cannam@95: TL = FMA(TJ, TK, TI); cannam@95: Tr = ii[WS(rs, 1)]; cannam@95: Tn = W[0]; cannam@95: T17 = T14 - T16; cannam@95: T1g = T14 + T16; cannam@95: TM = TF + TL; cannam@95: T12 = TF - TL; cannam@95: } cannam@95: Tu = ri[WS(rs, 5)]; cannam@95: TW = Tn * Tr; cannam@95: Tp = Tn * To; cannam@95: Tx = ii[WS(rs, 5)]; cannam@95: Tt = W[8]; cannam@95: Tq = W[1]; cannam@95: Tw = W[9]; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E T8, T1j, T1n, Tz, T1a, TU, Tl, T1b, T1c, T1v, T1t, T1w, T19, T1u, T1d; cannam@95: { cannam@95: E T1r, T10, TV, T1s, T11, T18; cannam@95: { cannam@95: E TO, TX, Ts, TZ, Ty, TT, TY, Tv; cannam@95: T8 = T1 + T7; cannam@95: TO = T1 - T7; cannam@95: TY = Tt * Tx; cannam@95: Tv = Tt * Tu; cannam@95: TX = FNMS(Tq, To, TW); cannam@95: Ts = FMA(Tq, Tr, Tp); cannam@95: TZ = FNMS(Tw, Tu, TY); cannam@95: Ty = FMA(Tw, Tx, Tv); cannam@95: TT = TQ - TS; cannam@95: T1j = TQ + TS; cannam@95: T1n = T1l + T1m; cannam@95: T1r = T1m - T1l; cannam@95: T10 = TX - TZ; cannam@95: T1f = TX + TZ; cannam@95: Tz = Ts + Ty; cannam@95: TV = Ts - Ty; cannam@95: T1a = TO - TT; cannam@95: TU = TO + TT; cannam@95: T1s = Te - Tk; cannam@95: Tl = Te + Tk; cannam@95: } cannam@95: T1b = T10 - TV; cannam@95: T11 = TV + T10; cannam@95: T18 = T12 - T17; cannam@95: T1c = T12 + T17; cannam@95: T1v = T1s + T1r; cannam@95: T1t = T1r - T1s; cannam@95: T1w = T18 - T11; cannam@95: T19 = T11 + T18; cannam@95: } cannam@95: ii[WS(rs, 3)] = FMA(KP707106781, T1w, T1v); cannam@95: ii[WS(rs, 7)] = FNMS(KP707106781, T1w, T1v); cannam@95: ri[WS(rs, 1)] = FMA(KP707106781, T19, TU); cannam@95: ri[WS(rs, 5)] = FNMS(KP707106781, T19, TU); cannam@95: T1u = T1b + T1c; cannam@95: T1d = T1b - T1c; cannam@95: ii[WS(rs, 1)] = FMA(KP707106781, T1u, T1t); cannam@95: ii[WS(rs, 5)] = FNMS(KP707106781, T1u, T1t); cannam@95: ri[WS(rs, 3)] = FMA(KP707106781, T1d, T1a); cannam@95: ri[WS(rs, 7)] = FNMS(KP707106781, T1d, T1a); cannam@95: T1e = T8 - Tl; cannam@95: Tm = T8 + Tl; cannam@95: T1q = T1n - T1j; cannam@95: T1o = T1j + T1n; cannam@95: T1p = TM - Tz; cannam@95: TN = Tz + TM; cannam@95: } cannam@95: } cannam@95: ii[WS(rs, 2)] = T1p + T1q; cannam@95: ii[WS(rs, 6)] = T1q - T1p; cannam@95: ri[0] = Tm + TN; cannam@95: ri[WS(rs, 4)] = Tm - TN; cannam@95: T1h = T1f - T1g; cannam@95: T1i = T1f + T1g; cannam@95: ii[0] = T1i + T1o; cannam@95: ii[WS(rs, 4)] = T1o - T1i; cannam@95: ri[WS(rs, 2)] = T1e + T1h; cannam@95: ri[WS(rs, 6)] = T1e - T1h; cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 0, 8}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, "t1_8", twinstr, &GENUS, {44, 14, 22, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_t1_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1_8, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -n 8 -name t1_8 -include t.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 32 FP multiplications, cannam@95: * (or, 52 additions, 18 multiplications, 14 fused multiply/add), cannam@95: * 28 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "t.h" cannam@95: cannam@95: static void t1_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 14); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 14, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: E T7, T1e, TH, T19, TF, T13, TR, TU, Ti, T1f, TK, T16, Tu, T12, TM; cannam@95: E TP; cannam@95: { cannam@95: E T1, T18, T6, T17; cannam@95: T1 = ri[0]; cannam@95: T18 = ii[0]; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = ri[WS(rs, 4)]; cannam@95: T5 = ii[WS(rs, 4)]; cannam@95: T2 = W[6]; cannam@95: T4 = W[7]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: T17 = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: T7 = T1 + T6; cannam@95: T1e = T18 - T17; cannam@95: TH = T1 - T6; cannam@95: T19 = T17 + T18; cannam@95: } cannam@95: { cannam@95: E Tz, TS, TE, TT; cannam@95: { cannam@95: E Tw, Ty, Tv, Tx; cannam@95: Tw = ri[WS(rs, 7)]; cannam@95: Ty = ii[WS(rs, 7)]; cannam@95: Tv = W[12]; cannam@95: Tx = W[13]; cannam@95: Tz = FMA(Tv, Tw, Tx * Ty); cannam@95: TS = FNMS(Tx, Tw, Tv * Ty); cannam@95: } cannam@95: { cannam@95: E TB, TD, TA, TC; cannam@95: TB = ri[WS(rs, 3)]; cannam@95: TD = ii[WS(rs, 3)]; cannam@95: TA = W[4]; cannam@95: TC = W[5]; cannam@95: TE = FMA(TA, TB, TC * TD); cannam@95: TT = FNMS(TC, TB, TA * TD); cannam@95: } cannam@95: TF = Tz + TE; cannam@95: T13 = TS + TT; cannam@95: TR = Tz - TE; cannam@95: TU = TS - TT; cannam@95: } cannam@95: { cannam@95: E Tc, TI, Th, TJ; cannam@95: { cannam@95: E T9, Tb, T8, Ta; cannam@95: T9 = ri[WS(rs, 2)]; cannam@95: Tb = ii[WS(rs, 2)]; cannam@95: T8 = W[2]; cannam@95: Ta = W[3]; cannam@95: Tc = FMA(T8, T9, Ta * Tb); cannam@95: TI = FNMS(Ta, T9, T8 * Tb); cannam@95: } cannam@95: { cannam@95: E Te, Tg, Td, Tf; cannam@95: Te = ri[WS(rs, 6)]; cannam@95: Tg = ii[WS(rs, 6)]; cannam@95: Td = W[10]; cannam@95: Tf = W[11]; cannam@95: Th = FMA(Td, Te, Tf * Tg); cannam@95: TJ = FNMS(Tf, Te, Td * Tg); cannam@95: } cannam@95: Ti = Tc + Th; cannam@95: T1f = Tc - Th; cannam@95: TK = TI - TJ; cannam@95: T16 = TI + TJ; cannam@95: } cannam@95: { cannam@95: E To, TN, Tt, TO; cannam@95: { cannam@95: E Tl, Tn, Tk, Tm; cannam@95: Tl = ri[WS(rs, 1)]; cannam@95: Tn = ii[WS(rs, 1)]; cannam@95: Tk = W[0]; cannam@95: Tm = W[1]; cannam@95: To = FMA(Tk, Tl, Tm * Tn); cannam@95: TN = FNMS(Tm, Tl, Tk * Tn); cannam@95: } cannam@95: { cannam@95: E Tq, Ts, Tp, Tr; cannam@95: Tq = ri[WS(rs, 5)]; cannam@95: Ts = ii[WS(rs, 5)]; cannam@95: Tp = W[8]; cannam@95: Tr = W[9]; cannam@95: Tt = FMA(Tp, Tq, Tr * Ts); cannam@95: TO = FNMS(Tr, Tq, Tp * Ts); cannam@95: } cannam@95: Tu = To + Tt; cannam@95: T12 = TN + TO; cannam@95: TM = To - Tt; cannam@95: TP = TN - TO; cannam@95: } cannam@95: { cannam@95: E Tj, TG, T1b, T1c; cannam@95: Tj = T7 + Ti; cannam@95: TG = Tu + TF; cannam@95: ri[WS(rs, 4)] = Tj - TG; cannam@95: ri[0] = Tj + TG; cannam@95: { cannam@95: E T15, T1a, T11, T14; cannam@95: T15 = T12 + T13; cannam@95: T1a = T16 + T19; cannam@95: ii[0] = T15 + T1a; cannam@95: ii[WS(rs, 4)] = T1a - T15; cannam@95: T11 = T7 - Ti; cannam@95: T14 = T12 - T13; cannam@95: ri[WS(rs, 6)] = T11 - T14; cannam@95: ri[WS(rs, 2)] = T11 + T14; cannam@95: } cannam@95: T1b = TF - Tu; cannam@95: T1c = T19 - T16; cannam@95: ii[WS(rs, 2)] = T1b + T1c; cannam@95: ii[WS(rs, 6)] = T1c - T1b; cannam@95: { cannam@95: E TX, T1g, T10, T1d, TY, TZ; cannam@95: TX = TH - TK; cannam@95: T1g = T1e - T1f; cannam@95: TY = TP - TM; cannam@95: TZ = TR + TU; cannam@95: T10 = KP707106781 * (TY - TZ); cannam@95: T1d = KP707106781 * (TY + TZ); cannam@95: ri[WS(rs, 7)] = TX - T10; cannam@95: ii[WS(rs, 5)] = T1g - T1d; cannam@95: ri[WS(rs, 3)] = TX + T10; cannam@95: ii[WS(rs, 1)] = T1d + T1g; cannam@95: } cannam@95: { cannam@95: E TL, T1i, TW, T1h, TQ, TV; cannam@95: TL = TH + TK; cannam@95: T1i = T1f + T1e; cannam@95: TQ = TM + TP; cannam@95: TV = TR - TU; cannam@95: TW = KP707106781 * (TQ + TV); cannam@95: T1h = KP707106781 * (TV - TQ); cannam@95: ri[WS(rs, 5)] = TL - TW; cannam@95: ii[WS(rs, 7)] = T1i - T1h; cannam@95: ri[WS(rs, 1)] = TL + TW; cannam@95: ii[WS(rs, 3)] = T1h + T1i; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 0, 8}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, "t1_8", twinstr, &GENUS, {52, 18, 14, 0}, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_t1_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1_8, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */