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