Mercurial > hg > sv-dependency-builds
diff src/fftw-3.3.3/rdft/scalar/r2cb/hc2cbdft_8.c @ 95:89f5e221ed7b
Add FFTW3
| author | Chris Cannam <cannam@all-day-breakfast.com> |
|---|---|
| date | Wed, 20 Mar 2013 15:35:50 +0000 |
| parents | |
| children |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/src/fftw-3.3.3/rdft/scalar/r2cb/hc2cbdft_8.c Wed Mar 20 15:35:50 2013 +0000 @@ -0,0 +1,427 @@ +/* + * Copyright (c) 2003, 2007-11 Matteo Frigo + * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + */ + +/* This file was automatically generated --- DO NOT EDIT */ +/* Generated on Sun Nov 25 07:42:04 EST 2012 */ + +#include "codelet-rdft.h" + +#ifdef HAVE_FMA + +/* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cbdft_8 -include hc2cb.h */ + +/* + * This function contains 82 FP additions, 36 FP multiplications, + * (or, 60 additions, 14 multiplications, 22 fused multiply/add), + * 55 stack variables, 1 constants, and 32 memory accesses + */ +#include "hc2cb.h" + +static void hc2cbdft_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) +{ + DK(KP707106781, +0.707106781186547524400844362104849039284835938); + { + INT m; + for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) { + E T1m, T1r, T1i, T1u, T1o, T1v, T1n, T1w, T1s; + { + E T1k, Tl, T1p, TE, TP, T1g, TM, T1b, T1f, T1a, TU, Tf, T1l, TH, Tw; + E T1q; + { + E TA, T3, TN, Tk, Th, T6, TO, TD, Tb, Tm, Ta, TK, Tp, Tc, Ts; + E Tt; + { + E T4, T5, TB, TC; + { + E T1, T2, Ti, Tj; + T1 = Rp[0]; + T2 = Rm[WS(rs, 3)]; + Ti = Ip[0]; + Tj = Im[WS(rs, 3)]; + T4 = Rp[WS(rs, 2)]; + TA = T1 - T2; + T3 = T1 + T2; + TN = Ti - Tj; + Tk = Ti + Tj; + T5 = Rm[WS(rs, 1)]; + TB = Ip[WS(rs, 2)]; + TC = Im[WS(rs, 1)]; + } + { + E T8, T9, Tn, To; + T8 = Rp[WS(rs, 1)]; + Th = T4 - T5; + T6 = T4 + T5; + TO = TB - TC; + TD = TB + TC; + T9 = Rm[WS(rs, 2)]; + Tn = Ip[WS(rs, 1)]; + To = Im[WS(rs, 2)]; + Tb = Rm[0]; + Tm = T8 - T9; + Ta = T8 + T9; + TK = Tn - To; + Tp = Tn + To; + Tc = Rp[WS(rs, 3)]; + Ts = Im[0]; + Tt = Ip[WS(rs, 3)]; + } + } + { + E Tr, Td, Tu, TL, Te, T7; + T1k = Tk - Th; + Tl = Th + Tk; + Tr = Tb - Tc; + Td = Tb + Tc; + TL = Tt - Ts; + Tu = Ts + Tt; + T1p = TA + TD; + TE = TA - TD; + TP = TN + TO; + T1g = TN - TO; + TM = TK + TL; + T1b = TL - TK; + T1f = Ta - Td; + Te = Ta + Td; + T1a = T3 - T6; + T7 = T3 + T6; + { + E Tq, TF, TG, Tv; + Tq = Tm + Tp; + TF = Tm - Tp; + TG = Tr - Tu; + Tv = Tr + Tu; + TU = T7 - Te; + Tf = T7 + Te; + T1l = TF - TG; + TH = TF + TG; + Tw = Tq - Tv; + T1q = Tq + Tv; + } + } + } + { + E TX, T10, T1c, T13, T1h, T1E, T1H, T1C, T1K, T1G, T1L, T1F; + { + E TQ, Tx, T1y, TI, Tg, Tz; + TX = TP - TM; + TQ = TM + TP; + Tx = FMA(KP707106781, Tw, Tl); + T10 = FNMS(KP707106781, Tw, Tl); + T1c = T1a + T1b; + T1y = T1a - T1b; + T13 = FNMS(KP707106781, TH, TE); + TI = FMA(KP707106781, TH, TE); + Tg = W[0]; + Tz = W[1]; + { + E T1B, T1A, T1x, T1J, T1z, T1D; + { + E TR, Ty, TS, TJ; + T1B = T1g - T1f; + T1h = T1f + T1g; + T1A = W[11]; + TR = Tg * TI; + Ty = Tg * Tx; + T1x = W[10]; + T1J = T1A * T1y; + TS = FNMS(Tz, Tx, TR); + TJ = FMA(Tz, TI, Ty); + T1z = T1x * T1y; + T1m = FMA(KP707106781, T1l, T1k); + T1E = FNMS(KP707106781, T1l, T1k); + Im[0] = TS - TQ; + Ip[0] = TQ + TS; + Rm[0] = Tf + TJ; + Rp[0] = Tf - TJ; + T1H = FMA(KP707106781, T1q, T1p); + T1r = FNMS(KP707106781, T1q, T1p); + T1D = W[12]; + } + T1C = FNMS(T1A, T1B, T1z); + T1K = FMA(T1x, T1B, T1J); + T1G = W[13]; + T1L = T1D * T1H; + T1F = T1D * T1E; + } + } + { + E TY, T16, T12, T17, T11; + { + E TW, TT, T15, TV, TZ, T1M, T1I; + TW = W[7]; + T1M = FNMS(T1G, T1E, T1L); + T1I = FMA(T1G, T1H, T1F); + TT = W[6]; + T15 = TW * TU; + Im[WS(rs, 3)] = T1M - T1K; + Ip[WS(rs, 3)] = T1K + T1M; + Rm[WS(rs, 3)] = T1C + T1I; + Rp[WS(rs, 3)] = T1C - T1I; + TV = TT * TU; + TZ = W[8]; + TY = FNMS(TW, TX, TV); + T16 = FMA(TT, TX, T15); + T12 = W[9]; + T17 = TZ * T13; + T11 = TZ * T10; + } + { + E T1e, T19, T1t, T1d, T1j, T18, T14; + T1e = W[3]; + T18 = FNMS(T12, T10, T17); + T14 = FMA(T12, T13, T11); + T19 = W[2]; + T1t = T1e * T1c; + Im[WS(rs, 2)] = T18 - T16; + Ip[WS(rs, 2)] = T16 + T18; + Rm[WS(rs, 2)] = TY + T14; + Rp[WS(rs, 2)] = TY - T14; + T1d = T19 * T1c; + T1j = W[4]; + T1i = FNMS(T1e, T1h, T1d); + T1u = FMA(T19, T1h, T1t); + T1o = W[5]; + T1v = T1j * T1r; + T1n = T1j * T1m; + } + } + } + } + T1w = FNMS(T1o, T1m, T1v); + T1s = FMA(T1o, T1r, T1n); + Im[WS(rs, 1)] = T1w - T1u; + Ip[WS(rs, 1)] = T1u + T1w; + Rm[WS(rs, 1)] = T1i + T1s; + Rp[WS(rs, 1)] = T1i - T1s; + } + } +} + +static const tw_instr twinstr[] = { + {TW_FULL, 1, 8}, + {TW_NEXT, 1, 0} +}; + +static const hc2c_desc desc = { 8, "hc2cbdft_8", twinstr, &GENUS, {60, 14, 22, 0} }; + +void X(codelet_hc2cbdft_8) (planner *p) { + X(khc2c_register) (p, hc2cbdft_8, &desc, HC2C_VIA_DFT); +} +#else /* HAVE_FMA */ + +/* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cbdft_8 -include hc2cb.h */ + +/* + * This function contains 82 FP additions, 32 FP multiplications, + * (or, 68 additions, 18 multiplications, 14 fused multiply/add), + * 30 stack variables, 1 constants, and 32 memory accesses + */ +#include "hc2cb.h" + +static void hc2cbdft_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) +{ + DK(KP707106781, +0.707106781186547524400844362104849039284835938); + { + INT m; + for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) { + E T7, T1d, T1h, Tl, TG, T14, T19, TO, Te, TL, T18, T15, TB, T1e, Tw; + E T1i; + { + E T3, TC, Tk, TM, T6, Th, TF, TN; + { + E T1, T2, Ti, Tj; + T1 = Rp[0]; + T2 = Rm[WS(rs, 3)]; + T3 = T1 + T2; + TC = T1 - T2; + Ti = Ip[0]; + Tj = Im[WS(rs, 3)]; + Tk = Ti + Tj; + TM = Ti - Tj; + } + { + E T4, T5, TD, TE; + T4 = Rp[WS(rs, 2)]; + T5 = Rm[WS(rs, 1)]; + T6 = T4 + T5; + Th = T4 - T5; + TD = Ip[WS(rs, 2)]; + TE = Im[WS(rs, 1)]; + TF = TD + TE; + TN = TD - TE; + } + T7 = T3 + T6; + T1d = Tk - Th; + T1h = TC + TF; + Tl = Th + Tk; + TG = TC - TF; + T14 = T3 - T6; + T19 = TM - TN; + TO = TM + TN; + } + { + E Ta, Tm, Tp, TJ, Td, Tr, Tu, TK; + { + E T8, T9, Tn, To; + T8 = Rp[WS(rs, 1)]; + T9 = Rm[WS(rs, 2)]; + Ta = T8 + T9; + Tm = T8 - T9; + Tn = Ip[WS(rs, 1)]; + To = Im[WS(rs, 2)]; + Tp = Tn + To; + TJ = Tn - To; + } + { + E Tb, Tc, Ts, Tt; + Tb = Rm[0]; + Tc = Rp[WS(rs, 3)]; + Td = Tb + Tc; + Tr = Tb - Tc; + Ts = Im[0]; + Tt = Ip[WS(rs, 3)]; + Tu = Ts + Tt; + TK = Tt - Ts; + } + Te = Ta + Td; + TL = TJ + TK; + T18 = Ta - Td; + T15 = TK - TJ; + { + E Tz, TA, Tq, Tv; + Tz = Tm - Tp; + TA = Tr - Tu; + TB = KP707106781 * (Tz + TA); + T1e = KP707106781 * (Tz - TA); + Tq = Tm + Tp; + Tv = Tr + Tu; + Tw = KP707106781 * (Tq - Tv); + T1i = KP707106781 * (Tq + Tv); + } + } + { + E Tf, TP, TI, TQ; + Tf = T7 + Te; + TP = TL + TO; + { + E Tx, TH, Tg, Ty; + Tx = Tl + Tw; + TH = TB + TG; + Tg = W[0]; + Ty = W[1]; + TI = FMA(Tg, Tx, Ty * TH); + TQ = FNMS(Ty, Tx, Tg * TH); + } + Rp[0] = Tf - TI; + Ip[0] = TP + TQ; + Rm[0] = Tf + TI; + Im[0] = TQ - TP; + } + { + E T1r, T1x, T1w, T1y; + { + E T1o, T1q, T1n, T1p; + T1o = T14 - T15; + T1q = T19 - T18; + T1n = W[10]; + T1p = W[11]; + T1r = FNMS(T1p, T1q, T1n * T1o); + T1x = FMA(T1p, T1o, T1n * T1q); + } + { + E T1t, T1v, T1s, T1u; + T1t = T1d - T1e; + T1v = T1i + T1h; + T1s = W[12]; + T1u = W[13]; + T1w = FMA(T1s, T1t, T1u * T1v); + T1y = FNMS(T1u, T1t, T1s * T1v); + } + Rp[WS(rs, 3)] = T1r - T1w; + Ip[WS(rs, 3)] = T1x + T1y; + Rm[WS(rs, 3)] = T1r + T1w; + Im[WS(rs, 3)] = T1y - T1x; + } + { + E TV, T11, T10, T12; + { + E TS, TU, TR, TT; + TS = T7 - Te; + TU = TO - TL; + TR = W[6]; + TT = W[7]; + TV = FNMS(TT, TU, TR * TS); + T11 = FMA(TT, TS, TR * TU); + } + { + E TX, TZ, TW, TY; + TX = Tl - Tw; + TZ = TG - TB; + TW = W[8]; + TY = W[9]; + T10 = FMA(TW, TX, TY * TZ); + T12 = FNMS(TY, TX, TW * TZ); + } + Rp[WS(rs, 2)] = TV - T10; + Ip[WS(rs, 2)] = T11 + T12; + Rm[WS(rs, 2)] = TV + T10; + Im[WS(rs, 2)] = T12 - T11; + } + { + E T1b, T1l, T1k, T1m; + { + E T16, T1a, T13, T17; + T16 = T14 + T15; + T1a = T18 + T19; + T13 = W[2]; + T17 = W[3]; + T1b = FNMS(T17, T1a, T13 * T16); + T1l = FMA(T17, T16, T13 * T1a); + } + { + E T1f, T1j, T1c, T1g; + T1f = T1d + T1e; + T1j = T1h - T1i; + T1c = W[4]; + T1g = W[5]; + T1k = FMA(T1c, T1f, T1g * T1j); + T1m = FNMS(T1g, T1f, T1c * T1j); + } + Rp[WS(rs, 1)] = T1b - T1k; + Ip[WS(rs, 1)] = T1l + T1m; + Rm[WS(rs, 1)] = T1b + T1k; + Im[WS(rs, 1)] = T1m - T1l; + } + } + } +} + +static const tw_instr twinstr[] = { + {TW_FULL, 1, 8}, + {TW_NEXT, 1, 0} +}; + +static const hc2c_desc desc = { 8, "hc2cbdft_8", twinstr, &GENUS, {68, 18, 14, 0} }; + +void X(codelet_hc2cbdft_8) (planner *p) { + X(khc2c_register) (p, hc2cbdft_8, &desc, HC2C_VIA_DFT); +} +#endif /* HAVE_FMA */