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 07:40:44 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cfdft_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 82 FP additions, 52 FP multiplications, cannam@95: * (or, 60 additions, 30 multiplications, 22 fused multiply/add), cannam@95: * 55 stack variables, 2 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T1A, T1w, T1z, T1x, T1H, T1v, T1L, T1F; cannam@95: { cannam@95: E Ty, T14, TO, T1o, Tv, TG, T16, T1m, Ta, T19, T1h, TV, T10, TX, TZ; cannam@95: E Tk, T1i, TY, T1b, TF, TB, T1l; cannam@95: { cannam@95: E TH, TN, TK, TM; cannam@95: { cannam@95: E Tw, Tx, TI, TJ; cannam@95: Tw = Ip[0]; cannam@95: Tx = Im[0]; cannam@95: TI = Rm[0]; cannam@95: TJ = Rp[0]; cannam@95: TH = W[0]; cannam@95: Ty = Tw - Tx; cannam@95: TN = Tw + Tx; cannam@95: T14 = TJ + TI; cannam@95: TK = TI - TJ; cannam@95: TM = W[1]; cannam@95: } cannam@95: { cannam@95: E Ts, Tp, Tt, Tm, Tr; cannam@95: { cannam@95: E Tn, To, TL, T1n; cannam@95: Tn = Ip[WS(rs, 2)]; cannam@95: To = Im[WS(rs, 2)]; cannam@95: TL = TH * TK; cannam@95: T1n = TM * TK; cannam@95: Ts = Rp[WS(rs, 2)]; cannam@95: TF = Tn + To; cannam@95: Tp = Tn - To; cannam@95: TO = FNMS(TM, TN, TL); cannam@95: T1o = FMA(TH, TN, T1n); cannam@95: Tt = Rm[WS(rs, 2)]; cannam@95: } cannam@95: Tm = W[6]; cannam@95: Tr = W[7]; cannam@95: { cannam@95: E TE, TD, T15, TC, Tu, Tq; cannam@95: TB = W[8]; cannam@95: TC = Tt - Ts; cannam@95: Tu = Ts + Tt; cannam@95: Tq = Tm * Tp; cannam@95: TE = W[9]; cannam@95: TD = TB * TC; cannam@95: T15 = Tm * Tu; cannam@95: Tv = FNMS(Tr, Tu, Tq); cannam@95: T1l = TE * TC; cannam@95: TG = FNMS(TE, TF, TD); cannam@95: T16 = FMA(Tr, Tp, T15); cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TU, TR, TT, T1g, TS; cannam@95: { cannam@95: E T2, T3, T7, T8; cannam@95: T2 = Ip[WS(rs, 1)]; cannam@95: T1m = FMA(TB, TF, T1l); cannam@95: T3 = Im[WS(rs, 1)]; cannam@95: T7 = Rp[WS(rs, 1)]; cannam@95: T8 = Rm[WS(rs, 1)]; cannam@95: { cannam@95: E T1, T4, T9, T6, T5, TQ, T18; cannam@95: T1 = W[2]; cannam@95: TU = T2 + T3; cannam@95: T4 = T2 - T3; cannam@95: TR = T7 - T8; cannam@95: T9 = T7 + T8; cannam@95: T6 = W[3]; cannam@95: T5 = T1 * T4; cannam@95: TQ = W[4]; cannam@95: T18 = T1 * T9; cannam@95: TT = W[5]; cannam@95: Ta = FNMS(T6, T9, T5); cannam@95: T1g = TQ * TU; cannam@95: TS = TQ * TR; cannam@95: T19 = FMA(T6, T4, T18); cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tc, Td, Th, Ti; cannam@95: Tc = Ip[WS(rs, 3)]; cannam@95: T1h = FNMS(TT, TR, T1g); cannam@95: TV = FMA(TT, TU, TS); cannam@95: Td = Im[WS(rs, 3)]; cannam@95: Th = Rp[WS(rs, 3)]; cannam@95: Ti = Rm[WS(rs, 3)]; cannam@95: { cannam@95: E Tb, Te, Tj, Tg, Tf, TW, T1a; cannam@95: Tb = W[10]; cannam@95: T10 = Tc + Td; cannam@95: Te = Tc - Td; cannam@95: TX = Th - Ti; cannam@95: Tj = Th + Ti; cannam@95: Tg = W[11]; cannam@95: Tf = Tb * Te; cannam@95: TW = W[12]; cannam@95: T1a = Tb * Tj; cannam@95: TZ = W[13]; cannam@95: Tk = FNMS(Tg, Tj, Tf); cannam@95: T1i = TW * T10; cannam@95: TY = TW * TX; cannam@95: T1b = FMA(Tg, Te, T1a); cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1E, T1t, TA, T1s, T1D, T1u, T1e, T13, T1r, T1d; cannam@95: { cannam@95: E TP, T1f, T1q, T12, T17, T1c; cannam@95: { cannam@95: E Tl, T11, Tz, T1p, T1k, T1j; cannam@95: T1E = Ta - Tk; cannam@95: Tl = Ta + Tk; cannam@95: T1j = FNMS(TZ, TX, T1i); cannam@95: T11 = FMA(TZ, T10, TY); cannam@95: Tz = Tv + Ty; cannam@95: T1t = Ty - Tv; cannam@95: T1A = T1o - T1m; cannam@95: T1p = T1m + T1o; cannam@95: T1k = T1h + T1j; cannam@95: T1w = T1j - T1h; cannam@95: T1z = TO - TG; cannam@95: TP = TG + TO; cannam@95: T1f = Tz - Tl; cannam@95: TA = Tl + Tz; cannam@95: T1s = T1k + T1p; cannam@95: T1q = T1k - T1p; cannam@95: T12 = TV + T11; cannam@95: T1x = TV - T11; cannam@95: T1D = T14 - T16; cannam@95: T17 = T14 + T16; cannam@95: T1c = T19 + T1b; cannam@95: T1u = T19 - T1b; cannam@95: } cannam@95: Im[WS(rs, 1)] = KP500000000 * (T1q - T1f); cannam@95: T1e = T12 + TP; cannam@95: T13 = TP - T12; cannam@95: T1r = T17 + T1c; cannam@95: T1d = T17 - T1c; cannam@95: Ip[WS(rs, 2)] = KP500000000 * (T1f + T1q); cannam@95: } cannam@95: Im[WS(rs, 3)] = KP500000000 * (T13 - TA); cannam@95: Ip[0] = KP500000000 * (TA + T13); cannam@95: Rm[WS(rs, 3)] = KP500000000 * (T1r - T1s); cannam@95: Rp[0] = KP500000000 * (T1r + T1s); cannam@95: Rp[WS(rs, 2)] = KP500000000 * (T1d + T1e); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (T1d - T1e); cannam@95: T1H = T1u + T1t; cannam@95: T1v = T1t - T1u; cannam@95: T1L = T1D + T1E; cannam@95: T1F = T1D - T1E; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1y, T1I, T1B, T1J; cannam@95: T1y = T1w + T1x; cannam@95: T1I = T1w - T1x; cannam@95: T1B = T1z - T1A; cannam@95: T1J = T1z + T1A; cannam@95: { cannam@95: E T1M, T1K, T1C, T1G; cannam@95: T1M = T1I + T1J; cannam@95: T1K = T1I - T1J; cannam@95: T1C = T1y + T1B; cannam@95: T1G = T1B - T1y; cannam@95: Im[0] = -(KP500000000 * (FNMS(KP707106781, T1K, T1H))); cannam@95: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1K, T1H)); cannam@95: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1M, T1L)); cannam@95: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP707106781, T1M, T1L)); cannam@95: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1G, T1F)); cannam@95: Rm[0] = KP500000000 * (FNMS(KP707106781, T1G, T1F)); cannam@95: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP707106781, T1C, T1v))); cannam@95: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1C, T1v)); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 8}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cfdft_8", twinstr, &GENUS, {60, 30, 22, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft_8, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cfdft_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 82 FP additions, 44 FP multiplications, cannam@95: * (or, 68 additions, 30 multiplications, 14 fused multiply/add), cannam@95: * 39 stack variables, 2 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP353553390, +0.353553390593273762200422181052424519642417969); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E Tv, TX, Ts, TY, TE, T1a, TJ, T19, T1l, T1m, T9, T10, Ti, T11, TP; cannam@95: E T16, TU, T17, T1i, T1j; cannam@95: { cannam@95: E Tt, Tu, TD, Tz, TA, TB, Tn, TI, Tr, TG, Tk, To; cannam@95: Tt = Ip[0]; cannam@95: Tu = Im[0]; cannam@95: TD = Tt + Tu; cannam@95: Tz = Rm[0]; cannam@95: TA = Rp[0]; cannam@95: TB = Tz - TA; cannam@95: { cannam@95: E Tl, Tm, Tp, Tq; cannam@95: Tl = Ip[WS(rs, 2)]; cannam@95: Tm = Im[WS(rs, 2)]; cannam@95: Tn = Tl - Tm; cannam@95: TI = Tl + Tm; cannam@95: Tp = Rp[WS(rs, 2)]; cannam@95: Tq = Rm[WS(rs, 2)]; cannam@95: Tr = Tp + Tq; cannam@95: TG = Tp - Tq; cannam@95: } cannam@95: Tv = Tt - Tu; cannam@95: TX = TA + Tz; cannam@95: Tk = W[6]; cannam@95: To = W[7]; cannam@95: Ts = FNMS(To, Tr, Tk * Tn); cannam@95: TY = FMA(Tk, Tr, To * Tn); cannam@95: { cannam@95: E Ty, TC, TF, TH; cannam@95: Ty = W[0]; cannam@95: TC = W[1]; cannam@95: TE = FNMS(TC, TD, Ty * TB); cannam@95: T1a = FMA(TC, TB, Ty * TD); cannam@95: TF = W[8]; cannam@95: TH = W[9]; cannam@95: TJ = FMA(TF, TG, TH * TI); cannam@95: T19 = FNMS(TH, TG, TF * TI); cannam@95: } cannam@95: T1l = TJ + TE; cannam@95: T1m = T1a - T19; cannam@95: } cannam@95: { cannam@95: E T4, TO, T8, TM, Td, TT, Th, TR; cannam@95: { cannam@95: E T2, T3, T6, T7; cannam@95: T2 = Ip[WS(rs, 1)]; cannam@95: T3 = Im[WS(rs, 1)]; cannam@95: T4 = T2 - T3; cannam@95: TO = T2 + T3; cannam@95: T6 = Rp[WS(rs, 1)]; cannam@95: T7 = Rm[WS(rs, 1)]; cannam@95: T8 = T6 + T7; cannam@95: TM = T6 - T7; cannam@95: } cannam@95: { cannam@95: E Tb, Tc, Tf, Tg; cannam@95: Tb = Ip[WS(rs, 3)]; cannam@95: Tc = Im[WS(rs, 3)]; cannam@95: Td = Tb - Tc; cannam@95: TT = Tb + Tc; cannam@95: Tf = Rp[WS(rs, 3)]; cannam@95: Tg = Rm[WS(rs, 3)]; cannam@95: Th = Tf + Tg; cannam@95: TR = Tf - Tg; cannam@95: } cannam@95: { cannam@95: E T1, T5, Ta, Te; cannam@95: T1 = W[2]; cannam@95: T5 = W[3]; cannam@95: T9 = FNMS(T5, T8, T1 * T4); cannam@95: T10 = FMA(T1, T8, T5 * T4); cannam@95: Ta = W[10]; cannam@95: Te = W[11]; cannam@95: Ti = FNMS(Te, Th, Ta * Td); cannam@95: T11 = FMA(Ta, Th, Te * Td); cannam@95: { cannam@95: E TL, TN, TQ, TS; cannam@95: TL = W[4]; cannam@95: TN = W[5]; cannam@95: TP = FMA(TL, TM, TN * TO); cannam@95: T16 = FNMS(TN, TM, TL * TO); cannam@95: TQ = W[12]; cannam@95: TS = W[13]; cannam@95: TU = FMA(TQ, TR, TS * TT); cannam@95: T17 = FNMS(TS, TR, TQ * TT); cannam@95: } cannam@95: T1i = T17 - T16; cannam@95: T1j = TP - TU; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1h, T1t, T1w, T1y, T1o, T1s, T1r, T1x; cannam@95: { cannam@95: E T1f, T1g, T1u, T1v; cannam@95: T1f = Tv - Ts; cannam@95: T1g = T10 - T11; cannam@95: T1h = KP500000000 * (T1f - T1g); cannam@95: T1t = KP500000000 * (T1g + T1f); cannam@95: T1u = T1i - T1j; cannam@95: T1v = T1l + T1m; cannam@95: T1w = KP353553390 * (T1u - T1v); cannam@95: T1y = KP353553390 * (T1u + T1v); cannam@95: } cannam@95: { cannam@95: E T1k, T1n, T1p, T1q; cannam@95: T1k = T1i + T1j; cannam@95: T1n = T1l - T1m; cannam@95: T1o = KP353553390 * (T1k + T1n); cannam@95: T1s = KP353553390 * (T1n - T1k); cannam@95: T1p = TX - TY; cannam@95: T1q = T9 - Ti; cannam@95: T1r = KP500000000 * (T1p - T1q); cannam@95: T1x = KP500000000 * (T1p + T1q); cannam@95: } cannam@95: Ip[WS(rs, 1)] = T1h + T1o; cannam@95: Rp[WS(rs, 1)] = T1x + T1y; cannam@95: Im[WS(rs, 2)] = T1o - T1h; cannam@95: Rm[WS(rs, 2)] = T1x - T1y; cannam@95: Rm[0] = T1r - T1s; cannam@95: Im[0] = T1w - T1t; cannam@95: Rp[WS(rs, 3)] = T1r + T1s; cannam@95: Ip[WS(rs, 3)] = T1t + T1w; cannam@95: } cannam@95: { cannam@95: E Tx, T15, T1c, T1e, TW, T14, T13, T1d; cannam@95: { cannam@95: E Tj, Tw, T18, T1b; cannam@95: Tj = T9 + Ti; cannam@95: Tw = Ts + Tv; cannam@95: Tx = Tj + Tw; cannam@95: T15 = Tw - Tj; cannam@95: T18 = T16 + T17; cannam@95: T1b = T19 + T1a; cannam@95: T1c = T18 - T1b; cannam@95: T1e = T18 + T1b; cannam@95: } cannam@95: { cannam@95: E TK, TV, TZ, T12; cannam@95: TK = TE - TJ; cannam@95: TV = TP + TU; cannam@95: TW = TK - TV; cannam@95: T14 = TV + TK; cannam@95: TZ = TX + TY; cannam@95: T12 = T10 + T11; cannam@95: T13 = TZ - T12; cannam@95: T1d = TZ + T12; cannam@95: } cannam@95: Ip[0] = KP500000000 * (Tx + TW); cannam@95: Rp[0] = KP500000000 * (T1d + T1e); cannam@95: Im[WS(rs, 3)] = KP500000000 * (TW - Tx); cannam@95: Rm[WS(rs, 3)] = KP500000000 * (T1d - T1e); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (T13 - T14); cannam@95: Im[WS(rs, 1)] = KP500000000 * (T1c - T15); cannam@95: Rp[WS(rs, 2)] = KP500000000 * (T13 + T14); cannam@95: Ip[WS(rs, 2)] = KP500000000 * (T15 + T1c); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 8}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cfdft_8", twinstr, &GENUS, {68, 30, 14, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft_8, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */