Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:35:59 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -name t2_8 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 74 FP additions, 50 FP multiplications, Chris@10: * (or, 44 additions, 20 multiplications, 30 fused multiply/add), Chris@10: * 64 stack variables, 1 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t2_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + (mb * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E TS, T1m, TJ, T1l, T1k, Tw, T1w, T1u; Chris@10: { Chris@10: E T2, T3, Tl, Tn, T5, T4, Tm, Tr, T6; Chris@10: T2 = W[0]; Chris@10: T3 = W[2]; Chris@10: Tl = W[4]; Chris@10: Tn = W[5]; Chris@10: T5 = W[1]; Chris@10: T4 = T2 * T3; Chris@10: Tm = T2 * Tl; Chris@10: Tr = T2 * Tn; Chris@10: T6 = W[3]; Chris@10: { Chris@10: E T1, T1s, TG, Td, T1r, Tu, TY, Tk, TW, T18, T1d, TD, TH, TA, T13; Chris@10: E TE, T14; Chris@10: { Chris@10: E To, Ts, Tf, T7, T8, Ti, Tb, T9, Tc, TC, Ta, TF, TB, Tg, Th; Chris@10: E Tj; Chris@10: T1 = ri[0]; Chris@10: To = FMA(T5, Tn, Tm); Chris@10: Ts = FNMS(T5, Tl, Tr); Chris@10: Tf = FMA(T5, T6, T4); Chris@10: T7 = FNMS(T5, T6, T4); Chris@10: Ta = T2 * T6; Chris@10: T1s = ii[0]; Chris@10: T8 = ri[WS(rs, 4)]; Chris@10: TF = Tf * Tn; Chris@10: TB = Tf * Tl; Chris@10: Ti = FNMS(T5, T3, Ta); Chris@10: Tb = FMA(T5, T3, Ta); Chris@10: T9 = T7 * T8; Chris@10: Tc = ii[WS(rs, 4)]; Chris@10: TG = FNMS(Ti, Tl, TF); Chris@10: TC = FMA(Ti, Tn, TB); Chris@10: { Chris@10: E Tp, T1q, Tt, Tq, TX; Chris@10: Tp = ri[WS(rs, 6)]; Chris@10: Td = FMA(Tb, Tc, T9); Chris@10: T1q = T7 * Tc; Chris@10: Tt = ii[WS(rs, 6)]; Chris@10: Tq = To * Tp; Chris@10: Tg = ri[WS(rs, 2)]; Chris@10: T1r = FNMS(Tb, T8, T1q); Chris@10: TX = To * Tt; Chris@10: Tu = FMA(Ts, Tt, Tq); Chris@10: Th = Tf * Tg; Chris@10: Tj = ii[WS(rs, 2)]; Chris@10: TY = FNMS(Ts, Tp, TX); Chris@10: } Chris@10: { Chris@10: E TO, TQ, TN, TP, T1a, T1b; Chris@10: { Chris@10: E TK, TM, TL, T19, TV; Chris@10: TK = ri[WS(rs, 7)]; Chris@10: TM = ii[WS(rs, 7)]; Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: TV = Tf * Tj; Chris@10: TL = Tl * TK; Chris@10: T19 = Tl * TM; Chris@10: TO = ri[WS(rs, 3)]; Chris@10: TW = FNMS(Ti, Tg, TV); Chris@10: TQ = ii[WS(rs, 3)]; Chris@10: TN = FMA(Tn, TM, TL); Chris@10: TP = T3 * TO; Chris@10: T1a = FNMS(Tn, TK, T19); Chris@10: T1b = T3 * TQ; Chris@10: } Chris@10: { Chris@10: E Tx, Tz, Ty, T12, T1c, TR; Chris@10: Tx = ri[WS(rs, 1)]; Chris@10: TR = FMA(T6, TQ, TP); Chris@10: Tz = ii[WS(rs, 1)]; Chris@10: T1c = FNMS(T6, TO, T1b); Chris@10: Ty = T2 * Tx; Chris@10: T18 = TN - TR; Chris@10: TS = TN + TR; Chris@10: T12 = T2 * Tz; Chris@10: T1d = T1a - T1c; Chris@10: T1m = T1a + T1c; Chris@10: TD = ri[WS(rs, 5)]; Chris@10: TH = ii[WS(rs, 5)]; Chris@10: TA = FMA(T5, Tz, Ty); Chris@10: T13 = FNMS(T5, Tx, T12); Chris@10: TE = TC * TD; Chris@10: T14 = TC * TH; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Te, T1p, T1t, Tv; Chris@10: { Chris@10: E T1g, T10, T1z, T1B, T1A, T1j, T1C, T1f; Chris@10: { Chris@10: E T1x, T11, T16, T1y; Chris@10: { Chris@10: E TU, TZ, TI, T15; Chris@10: Te = T1 + Td; Chris@10: TU = T1 - Td; Chris@10: TZ = TW - TY; Chris@10: T1p = TW + TY; Chris@10: TI = FMA(TG, TH, TE); Chris@10: T15 = FNMS(TG, TD, T14); Chris@10: T1t = T1r + T1s; Chris@10: T1x = T1s - T1r; Chris@10: T1g = TU - TZ; Chris@10: T10 = TU + TZ; Chris@10: T11 = TA - TI; Chris@10: TJ = TA + TI; Chris@10: T1l = T13 + T15; Chris@10: T16 = T13 - T15; Chris@10: T1y = Tk - Tu; Chris@10: Tv = Tk + Tu; Chris@10: } Chris@10: { Chris@10: E T1i, T1e, T17, T1h; Chris@10: T1i = T18 + T1d; Chris@10: T1e = T18 - T1d; Chris@10: T17 = T11 + T16; Chris@10: T1h = T16 - T11; Chris@10: T1z = T1x - T1y; Chris@10: T1B = T1y + T1x; Chris@10: T1A = T1h + T1i; Chris@10: T1j = T1h - T1i; Chris@10: T1C = T1e - T17; Chris@10: T1f = T17 + T1e; Chris@10: } Chris@10: } Chris@10: ri[WS(rs, 7)] = FNMS(KP707106781, T1j, T1g); Chris@10: ii[WS(rs, 7)] = FNMS(KP707106781, T1C, T1B); Chris@10: ri[WS(rs, 1)] = FMA(KP707106781, T1f, T10); Chris@10: ri[WS(rs, 5)] = FNMS(KP707106781, T1f, T10); Chris@10: ii[WS(rs, 1)] = FMA(KP707106781, T1A, T1z); Chris@10: ii[WS(rs, 5)] = FNMS(KP707106781, T1A, T1z); Chris@10: ri[WS(rs, 3)] = FMA(KP707106781, T1j, T1g); Chris@10: ii[WS(rs, 3)] = FMA(KP707106781, T1C, T1B); Chris@10: } Chris@10: T1k = Te - Tv; Chris@10: Tw = Te + Tv; Chris@10: T1w = T1t - T1p; Chris@10: T1u = T1p + T1t; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E TT, T1v, T1n, T1o; Chris@10: TT = TJ + TS; Chris@10: T1v = TS - TJ; Chris@10: T1n = T1l - T1m; Chris@10: T1o = T1l + T1m; Chris@10: ii[WS(rs, 2)] = T1v + T1w; Chris@10: ii[WS(rs, 6)] = T1w - T1v; Chris@10: ri[0] = Tw + TT; Chris@10: ri[WS(rs, 4)] = Tw - TT; Chris@10: ii[0] = T1o + T1u; Chris@10: ii[WS(rs, 4)] = T1u - T1o; Chris@10: ri[WS(rs, 2)] = T1k + T1n; Chris@10: ri[WS(rs, 6)] = T1k - T1n; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 0, 1}, Chris@10: {TW_CEXP, 0, 3}, Chris@10: {TW_CEXP, 0, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 8, "t2_8", twinstr, &GENUS, {44, 20, 30, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t2_8) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2_8, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 8 -name t2_8 -include t.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 74 FP additions, 44 FP multiplications, Chris@10: * (or, 56 additions, 26 multiplications, 18 fused multiply/add), Chris@10: * 42 stack variables, 1 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "t.h" Chris@10: Chris@10: static void t2_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + (mb * 6); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { Chris@10: E T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx; Chris@10: { Chris@10: E T4, Tb, T7, Ta; Chris@10: T2 = W[0]; Chris@10: T5 = W[1]; Chris@10: T3 = W[2]; Chris@10: T6 = W[3]; Chris@10: T4 = T2 * T3; Chris@10: Tb = T5 * T3; Chris@10: T7 = T5 * T6; Chris@10: Ta = T2 * T6; Chris@10: T8 = T4 - T7; Chris@10: Tc = Ta + Tb; Chris@10: Tg = T4 + T7; Chris@10: Ti = Ta - Tb; Chris@10: Tl = W[4]; Chris@10: Tm = W[5]; Chris@10: Tn = FMA(T2, Tl, T5 * Tm); Chris@10: Tz = FNMS(Ti, Tl, Tg * Tm); Chris@10: Tp = FNMS(T5, Tl, T2 * Tm); Chris@10: Tx = FMA(Tg, Tl, Ti * Tm); Chris@10: } Chris@10: { Chris@10: E Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ; Chris@10: E TT; Chris@10: { Chris@10: E T1, T1c, Te, T1b, T9, Td; Chris@10: T1 = ri[0]; Chris@10: T1c = ii[0]; Chris@10: T9 = ri[WS(rs, 4)]; Chris@10: Td = ii[WS(rs, 4)]; Chris@10: Te = FMA(T8, T9, Tc * Td); Chris@10: T1b = FNMS(Tc, T9, T8 * Td); Chris@10: Tf = T1 + Te; Chris@10: T1i = T1c - T1b; Chris@10: TL = T1 - Te; Chris@10: T1d = T1b + T1c; Chris@10: } Chris@10: { Chris@10: E TF, TW, TI, TX; Chris@10: { Chris@10: E TD, TE, TG, TH; Chris@10: TD = ri[WS(rs, 7)]; Chris@10: TE = ii[WS(rs, 7)]; Chris@10: TF = FMA(Tl, TD, Tm * TE); Chris@10: TW = FNMS(Tm, TD, Tl * TE); Chris@10: TG = ri[WS(rs, 3)]; Chris@10: TH = ii[WS(rs, 3)]; Chris@10: TI = FMA(T3, TG, T6 * TH); Chris@10: TX = FNMS(T6, TG, T3 * TH); Chris@10: } Chris@10: TJ = TF + TI; Chris@10: T17 = TW + TX; Chris@10: TV = TF - TI; Chris@10: TY = TW - TX; Chris@10: } Chris@10: { Chris@10: E Tk, TM, Tr, TN; Chris@10: { Chris@10: E Th, Tj, To, Tq; Chris@10: Th = ri[WS(rs, 2)]; Chris@10: Tj = ii[WS(rs, 2)]; Chris@10: Tk = FMA(Tg, Th, Ti * Tj); Chris@10: TM = FNMS(Ti, Th, Tg * Tj); Chris@10: To = ri[WS(rs, 6)]; Chris@10: Tq = ii[WS(rs, 6)]; Chris@10: Tr = FMA(Tn, To, Tp * Tq); Chris@10: TN = FNMS(Tp, To, Tn * Tq); Chris@10: } Chris@10: Ts = Tk + Tr; Chris@10: T1j = Tk - Tr; Chris@10: TO = TM - TN; Chris@10: T1a = TM + TN; Chris@10: } Chris@10: { Chris@10: E Tw, TR, TB, TS; Chris@10: { Chris@10: E Tu, Tv, Ty, TA; Chris@10: Tu = ri[WS(rs, 1)]; Chris@10: Tv = ii[WS(rs, 1)]; Chris@10: Tw = FMA(T2, Tu, T5 * Tv); Chris@10: TR = FNMS(T5, Tu, T2 * Tv); Chris@10: Ty = ri[WS(rs, 5)]; Chris@10: TA = ii[WS(rs, 5)]; Chris@10: TB = FMA(Tx, Ty, Tz * TA); Chris@10: TS = FNMS(Tz, Ty, Tx * TA); Chris@10: } Chris@10: TC = Tw + TB; Chris@10: T16 = TR + TS; Chris@10: TQ = Tw - TB; Chris@10: TT = TR - TS; Chris@10: } Chris@10: { Chris@10: E Tt, TK, T1f, T1g; Chris@10: Tt = Tf + Ts; Chris@10: TK = TC + TJ; Chris@10: ri[WS(rs, 4)] = Tt - TK; Chris@10: ri[0] = Tt + TK; Chris@10: { Chris@10: E T19, T1e, T15, T18; Chris@10: T19 = T16 + T17; Chris@10: T1e = T1a + T1d; Chris@10: ii[0] = T19 + T1e; Chris@10: ii[WS(rs, 4)] = T1e - T19; Chris@10: T15 = Tf - Ts; Chris@10: T18 = T16 - T17; Chris@10: ri[WS(rs, 6)] = T15 - T18; Chris@10: ri[WS(rs, 2)] = T15 + T18; Chris@10: } Chris@10: T1f = TJ - TC; Chris@10: T1g = T1d - T1a; Chris@10: ii[WS(rs, 2)] = T1f + T1g; Chris@10: ii[WS(rs, 6)] = T1g - T1f; Chris@10: { Chris@10: E T11, T1k, T14, T1h, T12, T13; Chris@10: T11 = TL - TO; Chris@10: T1k = T1i - T1j; Chris@10: T12 = TT - TQ; Chris@10: T13 = TV + TY; Chris@10: T14 = KP707106781 * (T12 - T13); Chris@10: T1h = KP707106781 * (T12 + T13); Chris@10: ri[WS(rs, 7)] = T11 - T14; Chris@10: ii[WS(rs, 5)] = T1k - T1h; Chris@10: ri[WS(rs, 3)] = T11 + T14; Chris@10: ii[WS(rs, 1)] = T1h + T1k; Chris@10: } Chris@10: { Chris@10: E TP, T1m, T10, T1l, TU, TZ; Chris@10: TP = TL + TO; Chris@10: T1m = T1j + T1i; Chris@10: TU = TQ + TT; Chris@10: TZ = TV - TY; Chris@10: T10 = KP707106781 * (TU + TZ); Chris@10: T1l = KP707106781 * (TZ - TU); Chris@10: ri[WS(rs, 5)] = TP - T10; Chris@10: ii[WS(rs, 7)] = T1m - T1l; Chris@10: ri[WS(rs, 1)] = TP + T10; Chris@10: ii[WS(rs, 3)] = T1l + T1m; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 0, 1}, Chris@10: {TW_CEXP, 0, 3}, Chris@10: {TW_CEXP, 0, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 8, "t2_8", twinstr, &GENUS, {56, 26, 18, 0}, 0, 0, 0 }; Chris@10: Chris@10: void X(codelet_t2_8) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2_8, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */