d@0: /* d@0: * Copyright (c) 2003, 2007-8 Matteo Frigo d@0: * Copyright (c) 2003, 2007-8 Massachusetts Institute of Technology d@0: * d@0: * This program is free software; you can redistribute it and/or modify d@0: * it under the terms of the GNU General Public License as published by d@0: * the Free Software Foundation; either version 2 of the License, or d@0: * (at your option) any later version. d@0: * d@0: * This program is distributed in the hope that it will be useful, d@0: * but WITHOUT ANY WARRANTY; without even the implied warranty of d@0: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the d@0: * GNU General Public License for more details. d@0: * d@0: * You should have received a copy of the GNU General Public License d@0: * along with this program; if not, write to the Free Software d@0: * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA d@0: * d@0: */ d@0: d@0: /* This file was automatically generated --- DO NOT EDIT */ d@0: /* Generated on Mon Feb 9 19:50:58 EST 2009 */ d@0: d@0: #include "codelet-dft.h" d@0: d@0: #ifdef HAVE_FMA d@0: d@0: /* Generated by: ../../../genfft/gen_twiddle -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 10 -name t1_10 -include t.h */ d@0: d@0: /* d@0: * This function contains 102 FP additions, 72 FP multiplications, d@0: * (or, 48 additions, 18 multiplications, 54 fused multiply/add), d@0: * 70 stack variables, 4 constants, and 40 memory accesses d@0: */ d@0: #include "t.h" d@0: d@0: static void t1_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) d@0: { d@0: DK(KP951056516, +0.951056516295153572116439333379382143405698634); d@0: DK(KP559016994, +0.559016994374947424102293417182819058860154590); d@0: DK(KP250000000, +0.250000000000000000000000000000000000000000000); d@0: DK(KP618033988, +0.618033988749894848204586834365638117720309180); d@0: INT m; d@0: for (m = mb, W = W + (mb * 18); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 18, MAKE_VOLATILE_STRIDE(rs)) { d@0: E T1X, T21, T20, T22; d@0: { d@0: E T23, T1U, T8, T12, T1y, T25, T1P, T1H, T1Y, T18, T10, T2b, T1K, T1O, T15; d@0: E T1Z, T2a, Tz, T24, T1n; d@0: { d@0: E T1, T1T, T3, T6, T2, T5; d@0: T1 = ri[0]; d@0: T1T = ii[0]; d@0: T3 = ri[WS(rs, 5)]; d@0: T6 = ii[WS(rs, 5)]; d@0: T2 = W[8]; d@0: T5 = W[9]; d@0: { d@0: E T1w, TY, T1s, T1F, TM, T16, T1u, TS; d@0: { d@0: E TF, T1p, TO, TR, T1r, TL, TN, TQ, T1t, TP; d@0: { d@0: E TU, TX, TT, TW; d@0: { d@0: E TB, TE, T1R, T4, TA, TD; d@0: TB = ri[WS(rs, 4)]; d@0: TE = ii[WS(rs, 4)]; d@0: T1R = T2 * T6; d@0: T4 = T2 * T3; d@0: TA = W[6]; d@0: TD = W[7]; d@0: { d@0: E T1S, T7, T1o, TC; d@0: T1S = FNMS(T5, T3, T1R); d@0: T7 = FMA(T5, T6, T4); d@0: T1o = TA * TE; d@0: TC = TA * TB; d@0: T23 = T1T - T1S; d@0: T1U = T1S + T1T; d@0: T8 = T1 - T7; d@0: T12 = T1 + T7; d@0: TF = FMA(TD, TE, TC); d@0: T1p = FNMS(TD, TB, T1o); d@0: } d@0: } d@0: TU = ri[WS(rs, 1)]; d@0: TX = ii[WS(rs, 1)]; d@0: TT = W[0]; d@0: TW = W[1]; d@0: { d@0: E TH, TK, TJ, T1q, TI, T1v, TV, TG; d@0: TH = ri[WS(rs, 9)]; d@0: TK = ii[WS(rs, 9)]; d@0: T1v = TT * TX; d@0: TV = TT * TU; d@0: TG = W[16]; d@0: TJ = W[17]; d@0: T1w = FNMS(TW, TU, T1v); d@0: TY = FMA(TW, TX, TV); d@0: T1q = TG * TK; d@0: TI = TG * TH; d@0: TO = ri[WS(rs, 6)]; d@0: TR = ii[WS(rs, 6)]; d@0: T1r = FNMS(TJ, TH, T1q); d@0: TL = FMA(TJ, TK, TI); d@0: TN = W[10]; d@0: TQ = W[11]; d@0: } d@0: } d@0: T1s = T1p - T1r; d@0: T1F = T1p + T1r; d@0: TM = TF - TL; d@0: T16 = TF + TL; d@0: T1t = TN * TR; d@0: TP = TN * TO; d@0: T1u = FNMS(TQ, TO, T1t); d@0: TS = FMA(TQ, TR, TP); d@0: } d@0: { d@0: E T1e, Te, T1l, Tx, Tn, Tq, Tp, T1g, Tk, T1i, To; d@0: { d@0: E Tt, Tw, Tv, T1k, Tu; d@0: { d@0: E Ta, Td, T9, Tc, T1d, Tb, Ts; d@0: Ta = ri[WS(rs, 2)]; d@0: Td = ii[WS(rs, 2)]; d@0: { d@0: E T1G, T1x, TZ, T17; d@0: T1G = T1u + T1w; d@0: T1x = T1u - T1w; d@0: TZ = TS - TY; d@0: T17 = TS + TY; d@0: T1y = T1s - T1x; d@0: T25 = T1s + T1x; d@0: T1P = T1F + T1G; d@0: T1H = T1F - T1G; d@0: T1Y = T16 - T17; d@0: T18 = T16 + T17; d@0: T10 = TM + TZ; d@0: T2b = TM - TZ; d@0: T9 = W[2]; d@0: } d@0: Tc = W[3]; d@0: Tt = ri[WS(rs, 3)]; d@0: Tw = ii[WS(rs, 3)]; d@0: T1d = T9 * Td; d@0: Tb = T9 * Ta; d@0: Ts = W[4]; d@0: Tv = W[5]; d@0: T1e = FNMS(Tc, Ta, T1d); d@0: Te = FMA(Tc, Td, Tb); d@0: T1k = Ts * Tw; d@0: Tu = Ts * Tt; d@0: } d@0: { d@0: E Tg, Tj, Tf, Ti, T1f, Th, Tm; d@0: Tg = ri[WS(rs, 7)]; d@0: Tj = ii[WS(rs, 7)]; d@0: T1l = FNMS(Tv, Tt, T1k); d@0: Tx = FMA(Tv, Tw, Tu); d@0: Tf = W[12]; d@0: Ti = W[13]; d@0: Tn = ri[WS(rs, 8)]; d@0: Tq = ii[WS(rs, 8)]; d@0: T1f = Tf * Tj; d@0: Th = Tf * Tg; d@0: Tm = W[14]; d@0: Tp = W[15]; d@0: T1g = FNMS(Ti, Tg, T1f); d@0: Tk = FMA(Ti, Tj, Th); d@0: T1i = Tm * Tq; d@0: To = Tm * Tn; d@0: } d@0: } d@0: { d@0: E T1h, T1I, Tl, T13, T1j, Tr; d@0: T1h = T1e - T1g; d@0: T1I = T1e + T1g; d@0: Tl = Te - Tk; d@0: T13 = Te + Tk; d@0: T1j = FNMS(Tp, Tn, T1i); d@0: Tr = FMA(Tp, Tq, To); d@0: { d@0: E T1m, T1J, T14, Ty; d@0: T1m = T1j - T1l; d@0: T1J = T1j + T1l; d@0: T14 = Tr + Tx; d@0: Ty = Tr - Tx; d@0: T1K = T1I - T1J; d@0: T1O = T1I + T1J; d@0: T15 = T13 + T14; d@0: T1Z = T13 - T14; d@0: T2a = Tl - Ty; d@0: Tz = Tl + Ty; d@0: T24 = T1h + T1m; d@0: T1n = T1h - T1m; d@0: } d@0: } d@0: } d@0: } d@0: } d@0: { d@0: E T2c, T2e, T29, T2d; d@0: { d@0: E T1b, T11, T26, T28, T27; d@0: T1b = Tz - T10; d@0: T11 = Tz + T10; d@0: T26 = T24 + T25; d@0: T28 = T24 - T25; d@0: { d@0: E T1B, T1z, T1a, T1A, T1c; d@0: T1B = FNMS(KP618033988, T1n, T1y); d@0: T1z = FMA(KP618033988, T1y, T1n); d@0: ri[WS(rs, 5)] = T8 + T11; d@0: T1a = FNMS(KP250000000, T11, T8); d@0: T1A = FNMS(KP559016994, T1b, T1a); d@0: T1c = FMA(KP559016994, T1b, T1a); d@0: T27 = FNMS(KP250000000, T26, T23); d@0: T2c = FMA(KP618033988, T2b, T2a); d@0: T2e = FNMS(KP618033988, T2a, T2b); d@0: ri[WS(rs, 1)] = FMA(KP951056516, T1z, T1c); d@0: ri[WS(rs, 9)] = FNMS(KP951056516, T1z, T1c); d@0: ri[WS(rs, 3)] = FMA(KP951056516, T1B, T1A); d@0: ri[WS(rs, 7)] = FNMS(KP951056516, T1B, T1A); d@0: } d@0: ii[WS(rs, 5)] = T26 + T23; d@0: T29 = FMA(KP559016994, T28, T27); d@0: T2d = FNMS(KP559016994, T28, T27); d@0: } d@0: { d@0: E T1E, T1M, T1L, T1N, T19, T1D, T1C, T1Q, T1W, T1V; d@0: T19 = T15 + T18; d@0: T1D = T15 - T18; d@0: ii[WS(rs, 7)] = FMA(KP951056516, T2e, T2d); d@0: ii[WS(rs, 3)] = FNMS(KP951056516, T2e, T2d); d@0: ii[WS(rs, 9)] = FMA(KP951056516, T2c, T29); d@0: ii[WS(rs, 1)] = FNMS(KP951056516, T2c, T29); d@0: T1C = FNMS(KP250000000, T19, T12); d@0: ri[0] = T12 + T19; d@0: T1E = FNMS(KP559016994, T1D, T1C); d@0: T1M = FMA(KP559016994, T1D, T1C); d@0: T1L = FNMS(KP618033988, T1K, T1H); d@0: T1N = FMA(KP618033988, T1H, T1K); d@0: T1Q = T1O + T1P; d@0: T1W = T1O - T1P; d@0: ri[WS(rs, 6)] = FMA(KP951056516, T1N, T1M); d@0: ri[WS(rs, 4)] = FNMS(KP951056516, T1N, T1M); d@0: ri[WS(rs, 8)] = FMA(KP951056516, T1L, T1E); d@0: ri[WS(rs, 2)] = FNMS(KP951056516, T1L, T1E); d@0: T1V = FNMS(KP250000000, T1Q, T1U); d@0: ii[0] = T1Q + T1U; d@0: T1X = FNMS(KP559016994, T1W, T1V); d@0: T21 = FMA(KP559016994, T1W, T1V); d@0: T20 = FNMS(KP618033988, T1Z, T1Y); d@0: T22 = FMA(KP618033988, T1Y, T1Z); d@0: } d@0: } d@0: } d@0: ii[WS(rs, 6)] = FNMS(KP951056516, T22, T21); d@0: ii[WS(rs, 4)] = FMA(KP951056516, T22, T21); d@0: ii[WS(rs, 8)] = FNMS(KP951056516, T20, T1X); d@0: ii[WS(rs, 2)] = FMA(KP951056516, T20, T1X); d@0: } d@0: } d@0: d@0: static const tw_instr twinstr[] = { d@0: {TW_FULL, 0, 10}, d@0: {TW_NEXT, 1, 0} d@0: }; d@0: d@0: static const ct_desc desc = { 10, "t1_10", twinstr, &GENUS, {48, 18, 54, 0}, 0, 0, 0 }; d@0: d@0: void X(codelet_t1_10) (planner *p) { d@0: X(kdft_dit_register) (p, t1_10, &desc); d@0: } d@0: #else /* HAVE_FMA */ d@0: d@0: /* Generated by: ../../../genfft/gen_twiddle -compact -variables 4 -pipeline-latency 4 -n 10 -name t1_10 -include t.h */ d@0: d@0: /* d@0: * This function contains 102 FP additions, 60 FP multiplications, d@0: * (or, 72 additions, 30 multiplications, 30 fused multiply/add), d@0: * 45 stack variables, 4 constants, and 40 memory accesses d@0: */ d@0: #include "t.h" d@0: d@0: static void t1_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) d@0: { d@0: DK(KP587785252, +0.587785252292473129168705954639072768597652438); d@0: DK(KP951056516, +0.951056516295153572116439333379382143405698634); d@0: DK(KP250000000, +0.250000000000000000000000000000000000000000000); d@0: DK(KP559016994, +0.559016994374947424102293417182819058860154590); d@0: INT m; d@0: for (m = mb, W = W + (mb * 18); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 18, MAKE_VOLATILE_STRIDE(rs)) { d@0: E T7, T1O, TT, T1C, TF, TQ, TR, T1o, T1p, T1y, TX, TY, TZ, T1d, T1g; d@0: E T1M, Ti, Tt, Tu, T1r, T1s, T1x, TU, TV, TW, T16, T19, T1L; d@0: { d@0: E T1, T1B, T6, T1A; d@0: T1 = ri[0]; d@0: T1B = ii[0]; d@0: { d@0: E T3, T5, T2, T4; d@0: T3 = ri[WS(rs, 5)]; d@0: T5 = ii[WS(rs, 5)]; d@0: T2 = W[8]; d@0: T4 = W[9]; d@0: T6 = FMA(T2, T3, T4 * T5); d@0: T1A = FNMS(T4, T3, T2 * T5); d@0: } d@0: T7 = T1 - T6; d@0: T1O = T1B - T1A; d@0: TT = T1 + T6; d@0: T1C = T1A + T1B; d@0: } d@0: { d@0: E Tz, T1b, TP, T1f, TE, T1c, TK, T1e; d@0: { d@0: E Tw, Ty, Tv, Tx; d@0: Tw = ri[WS(rs, 4)]; d@0: Ty = ii[WS(rs, 4)]; d@0: Tv = W[6]; d@0: Tx = W[7]; d@0: Tz = FMA(Tv, Tw, Tx * Ty); d@0: T1b = FNMS(Tx, Tw, Tv * Ty); d@0: } d@0: { d@0: E TM, TO, TL, TN; d@0: TM = ri[WS(rs, 1)]; d@0: TO = ii[WS(rs, 1)]; d@0: TL = W[0]; d@0: TN = W[1]; d@0: TP = FMA(TL, TM, TN * TO); d@0: T1f = FNMS(TN, TM, TL * TO); d@0: } d@0: { d@0: E TB, TD, TA, TC; d@0: TB = ri[WS(rs, 9)]; d@0: TD = ii[WS(rs, 9)]; d@0: TA = W[16]; d@0: TC = W[17]; d@0: TE = FMA(TA, TB, TC * TD); d@0: T1c = FNMS(TC, TB, TA * TD); d@0: } d@0: { d@0: E TH, TJ, TG, TI; d@0: TH = ri[WS(rs, 6)]; d@0: TJ = ii[WS(rs, 6)]; d@0: TG = W[10]; d@0: TI = W[11]; d@0: TK = FMA(TG, TH, TI * TJ); d@0: T1e = FNMS(TI, TH, TG * TJ); d@0: } d@0: TF = Tz - TE; d@0: TQ = TK - TP; d@0: TR = TF + TQ; d@0: T1o = T1b + T1c; d@0: T1p = T1e + T1f; d@0: T1y = T1o + T1p; d@0: TX = Tz + TE; d@0: TY = TK + TP; d@0: TZ = TX + TY; d@0: T1d = T1b - T1c; d@0: T1g = T1e - T1f; d@0: T1M = T1d + T1g; d@0: } d@0: { d@0: E Tc, T14, Ts, T18, Th, T15, Tn, T17; d@0: { d@0: E T9, Tb, T8, Ta; d@0: T9 = ri[WS(rs, 2)]; d@0: Tb = ii[WS(rs, 2)]; d@0: T8 = W[2]; d@0: Ta = W[3]; d@0: Tc = FMA(T8, T9, Ta * Tb); d@0: T14 = FNMS(Ta, T9, T8 * Tb); d@0: } d@0: { d@0: E Tp, Tr, To, Tq; d@0: Tp = ri[WS(rs, 3)]; d@0: Tr = ii[WS(rs, 3)]; d@0: To = W[4]; d@0: Tq = W[5]; d@0: Ts = FMA(To, Tp, Tq * Tr); d@0: T18 = FNMS(Tq, Tp, To * Tr); d@0: } d@0: { d@0: E Te, Tg, Td, Tf; d@0: Te = ri[WS(rs, 7)]; d@0: Tg = ii[WS(rs, 7)]; d@0: Td = W[12]; d@0: Tf = W[13]; d@0: Th = FMA(Td, Te, Tf * Tg); d@0: T15 = FNMS(Tf, Te, Td * Tg); d@0: } d@0: { d@0: E Tk, Tm, Tj, Tl; d@0: Tk = ri[WS(rs, 8)]; d@0: Tm = ii[WS(rs, 8)]; d@0: Tj = W[14]; d@0: Tl = W[15]; d@0: Tn = FMA(Tj, Tk, Tl * Tm); d@0: T17 = FNMS(Tl, Tk, Tj * Tm); d@0: } d@0: Ti = Tc - Th; d@0: Tt = Tn - Ts; d@0: Tu = Ti + Tt; d@0: T1r = T14 + T15; d@0: T1s = T17 + T18; d@0: T1x = T1r + T1s; d@0: TU = Tc + Th; d@0: TV = Tn + Ts; d@0: TW = TU + TV; d@0: T16 = T14 - T15; d@0: T19 = T17 - T18; d@0: T1L = T16 + T19; d@0: } d@0: { d@0: E T11, TS, T12, T1i, T1k, T1a, T1h, T1j, T13; d@0: T11 = KP559016994 * (Tu - TR); d@0: TS = Tu + TR; d@0: T12 = FNMS(KP250000000, TS, T7); d@0: T1a = T16 - T19; d@0: T1h = T1d - T1g; d@0: T1i = FMA(KP951056516, T1a, KP587785252 * T1h); d@0: T1k = FNMS(KP587785252, T1a, KP951056516 * T1h); d@0: ri[WS(rs, 5)] = T7 + TS; d@0: T1j = T12 - T11; d@0: ri[WS(rs, 7)] = T1j - T1k; d@0: ri[WS(rs, 3)] = T1j + T1k; d@0: T13 = T11 + T12; d@0: ri[WS(rs, 9)] = T13 - T1i; d@0: ri[WS(rs, 1)] = T13 + T1i; d@0: } d@0: { d@0: E T1N, T1P, T1Q, T1U, T1W, T1S, T1T, T1V, T1R; d@0: T1N = KP559016994 * (T1L - T1M); d@0: T1P = T1L + T1M; d@0: T1Q = FNMS(KP250000000, T1P, T1O); d@0: T1S = Ti - Tt; d@0: T1T = TF - TQ; d@0: T1U = FMA(KP951056516, T1S, KP587785252 * T1T); d@0: T1W = FNMS(KP587785252, T1S, KP951056516 * T1T); d@0: ii[WS(rs, 5)] = T1P + T1O; d@0: T1V = T1Q - T1N; d@0: ii[WS(rs, 3)] = T1V - T1W; d@0: ii[WS(rs, 7)] = T1W + T1V; d@0: T1R = T1N + T1Q; d@0: ii[WS(rs, 1)] = T1R - T1U; d@0: ii[WS(rs, 9)] = T1U + T1R; d@0: } d@0: { d@0: E T1m, T10, T1l, T1u, T1w, T1q, T1t, T1v, T1n; d@0: T1m = KP559016994 * (TW - TZ); d@0: T10 = TW + TZ; d@0: T1l = FNMS(KP250000000, T10, TT); d@0: T1q = T1o - T1p; d@0: T1t = T1r - T1s; d@0: T1u = FNMS(KP587785252, T1t, KP951056516 * T1q); d@0: T1w = FMA(KP951056516, T1t, KP587785252 * T1q); d@0: ri[0] = TT + T10; d@0: T1v = T1m + T1l; d@0: ri[WS(rs, 4)] = T1v - T1w; d@0: ri[WS(rs, 6)] = T1v + T1w; d@0: T1n = T1l - T1m; d@0: ri[WS(rs, 2)] = T1n - T1u; d@0: ri[WS(rs, 8)] = T1n + T1u; d@0: } d@0: { d@0: E T1H, T1z, T1G, T1F, T1J, T1D, T1E, T1K, T1I; d@0: T1H = KP559016994 * (T1x - T1y); d@0: T1z = T1x + T1y; d@0: T1G = FNMS(KP250000000, T1z, T1C); d@0: T1D = TX - TY; d@0: T1E = TU - TV; d@0: T1F = FNMS(KP587785252, T1E, KP951056516 * T1D); d@0: T1J = FMA(KP951056516, T1E, KP587785252 * T1D); d@0: ii[0] = T1z + T1C; d@0: T1K = T1H + T1G; d@0: ii[WS(rs, 4)] = T1J + T1K; d@0: ii[WS(rs, 6)] = T1K - T1J; d@0: T1I = T1G - T1H; d@0: ii[WS(rs, 2)] = T1F + T1I; d@0: ii[WS(rs, 8)] = T1I - T1F; d@0: } d@0: } d@0: } d@0: d@0: static const tw_instr twinstr[] = { d@0: {TW_FULL, 0, 10}, d@0: {TW_NEXT, 1, 0} d@0: }; d@0: d@0: static const ct_desc desc = { 10, "t1_10", twinstr, &GENUS, {72, 30, 30, 0}, 0, 0, 0 }; d@0: d@0: void X(codelet_t1_10) (planner *p) { d@0: X(kdft_dit_register) (p, t1_10, &desc); d@0: } d@0: #endif /* HAVE_FMA */