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 07:40:50 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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 90 FP additions, 66 FP multiplications, cannam@95: * (or, 60 additions, 36 multiplications, 30 fused multiply/add), cannam@95: * 68 stack variables, 2 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft2_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) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) { cannam@95: E T1G, T1F, T1C, T1D, T1N, T1B, T1R, T1L; cannam@95: { cannam@95: E T1, T2, Th, Tj, T4, T3, Ti, Tp, T5; cannam@95: T1 = W[0]; cannam@95: T2 = W[2]; cannam@95: Th = W[4]; cannam@95: Tj = W[5]; cannam@95: T4 = W[1]; cannam@95: T3 = T1 * T2; cannam@95: Ti = T1 * Th; cannam@95: Tp = T1 * Tj; cannam@95: T5 = W[3]; cannam@95: { cannam@95: E Tk, Tq, TI, T1a, T1u, TY, TF, TS, T1s, T1c, Tr, T1n, Tg, T16, Tn; cannam@95: E T13, T1f, Ts, To, T1o; cannam@95: { cannam@95: E T6, Tw, Tc, TB, TQ, TM, TC, TR, Tz, TD, TA; cannam@95: { cannam@95: E TX, TV, TT, TU; cannam@95: { cannam@95: E TG, Tb, TH, TP, TL; cannam@95: TG = Ip[0]; cannam@95: Tk = FMA(T4, Tj, Ti); cannam@95: Tq = FNMS(T4, Th, Tp); cannam@95: T6 = FMA(T4, T5, T3); cannam@95: Tw = FNMS(T4, T5, T3); cannam@95: Tb = T1 * T5; cannam@95: TH = Im[0]; cannam@95: TT = Rm[0]; cannam@95: TP = T6 * Tj; cannam@95: TL = T6 * Th; cannam@95: Tc = FNMS(T4, T2, Tb); cannam@95: TB = FMA(T4, T2, Tb); cannam@95: TX = TG + TH; cannam@95: TI = TG - TH; cannam@95: TU = Rp[0]; cannam@95: TQ = FNMS(Tc, Th, TP); cannam@95: TM = FMA(Tc, Tj, TL); cannam@95: } cannam@95: T1a = TU + TT; cannam@95: TV = TT - TU; cannam@95: { cannam@95: E Tx, Ty, T1t, TW; cannam@95: Tx = Ip[WS(rs, 2)]; cannam@95: Ty = Im[WS(rs, 2)]; cannam@95: T1t = T4 * TV; cannam@95: TW = T1 * TV; cannam@95: TC = Rp[WS(rs, 2)]; cannam@95: TR = Tx + Ty; cannam@95: Tz = Tx - Ty; cannam@95: T1u = FMA(T1, TX, T1t); cannam@95: TY = FNMS(T4, TX, TW); cannam@95: TD = Rm[WS(rs, 2)]; cannam@95: } cannam@95: TA = Tw * Tz; cannam@95: } cannam@95: { cannam@95: E Td, T9, T12, Te, Ta, T1m; cannam@95: { cannam@95: E T7, T8, TN, TE, TO, T1r, T1b; cannam@95: T7 = Ip[WS(rs, 1)]; cannam@95: T8 = Im[WS(rs, 1)]; cannam@95: TN = TD - TC; cannam@95: TE = TC + TD; cannam@95: Td = Rp[WS(rs, 1)]; cannam@95: T9 = T7 - T8; cannam@95: T12 = T7 + T8; cannam@95: TO = TM * TN; cannam@95: T1r = TQ * TN; cannam@95: T1b = Tw * TE; cannam@95: TF = FNMS(TB, TE, TA); cannam@95: TS = FNMS(TQ, TR, TO); cannam@95: T1s = FMA(TM, TR, T1r); cannam@95: T1c = FMA(TB, Tz, T1b); cannam@95: Te = Rm[WS(rs, 1)]; cannam@95: } cannam@95: Ta = T6 * T9; cannam@95: T1m = T2 * T12; cannam@95: { cannam@95: E Tl, T10, Tf, Tm, T11, T1e; cannam@95: Tl = Ip[WS(rs, 3)]; cannam@95: T10 = Td - Te; cannam@95: Tf = Td + Te; cannam@95: Tm = Im[WS(rs, 3)]; cannam@95: Tr = Rp[WS(rs, 3)]; cannam@95: T11 = T2 * T10; cannam@95: T1n = FNMS(T5, T10, T1m); cannam@95: T1e = T6 * Tf; cannam@95: Tg = FNMS(Tc, Tf, Ta); cannam@95: T16 = Tl + Tm; cannam@95: Tn = Tl - Tm; cannam@95: T13 = FMA(T5, T12, T11); cannam@95: T1f = FMA(Tc, T9, T1e); cannam@95: Ts = Rm[WS(rs, 3)]; cannam@95: } cannam@95: To = Tk * Tn; cannam@95: T1o = Th * T16; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1z, T1K, T1y, T1k, T1J, T1A, T1x, T1j; cannam@95: { cannam@95: E T1w, TK, T1l, T19, T1d, T1i; cannam@95: { cannam@95: E TJ, T14, Tt, T1v, T1h; cannam@95: T1z = TI - TF; cannam@95: TJ = TF + TI; cannam@95: T14 = Tr - Ts; cannam@95: Tt = Tr + Ts; cannam@95: T1v = T1s + T1u; cannam@95: T1G = T1u - T1s; cannam@95: { cannam@95: E TZ, T1q, Tv, T18, T15; cannam@95: T1F = TY - TS; cannam@95: TZ = TS + TY; cannam@95: T15 = Th * T14; cannam@95: { cannam@95: E T1p, T1g, Tu, T17; cannam@95: T1p = FNMS(Tj, T14, T1o); cannam@95: T1g = Tk * Tt; cannam@95: Tu = FNMS(Tq, Tt, To); cannam@95: T17 = FMA(Tj, T16, T15); cannam@95: T1C = T1p - T1n; cannam@95: T1q = T1n + T1p; cannam@95: T1h = FMA(Tq, Tn, T1g); cannam@95: T1K = Tg - Tu; cannam@95: Tv = Tg + Tu; cannam@95: T18 = T13 + T17; cannam@95: T1D = T13 - T17; cannam@95: } cannam@95: T1w = T1q - T1v; cannam@95: T1y = T1q + T1v; cannam@95: TK = Tv + TJ; cannam@95: T1l = TJ - Tv; cannam@95: T1k = T18 + TZ; cannam@95: T19 = TZ - T18; cannam@95: } cannam@95: T1J = T1a - T1c; cannam@95: T1d = T1a + T1c; cannam@95: T1i = T1f + T1h; cannam@95: T1A = T1f - T1h; cannam@95: } cannam@95: Ip[0] = KP500000000 * (TK + T19); cannam@95: Im[WS(rs, 3)] = KP500000000 * (T19 - TK); cannam@95: Im[WS(rs, 1)] = KP500000000 * (T1w - T1l); cannam@95: T1x = T1d + T1i; cannam@95: T1j = T1d - T1i; cannam@95: Ip[WS(rs, 2)] = KP500000000 * (T1l + T1w); cannam@95: } cannam@95: Rm[WS(rs, 3)] = KP500000000 * (T1x - T1y); cannam@95: Rp[0] = KP500000000 * (T1x + T1y); cannam@95: Rp[WS(rs, 2)] = KP500000000 * (T1j + T1k); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (T1j - T1k); cannam@95: T1N = T1A + T1z; cannam@95: T1B = T1z - T1A; cannam@95: T1R = T1J + T1K; cannam@95: T1L = T1J - T1K; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1E, T1O, T1H, T1P; cannam@95: T1E = T1C + T1D; cannam@95: T1O = T1C - T1D; cannam@95: T1H = T1F - T1G; cannam@95: T1P = T1F + T1G; cannam@95: { cannam@95: E T1S, T1Q, T1I, T1M; cannam@95: T1S = T1O + T1P; cannam@95: T1Q = T1O - T1P; cannam@95: T1I = T1E + T1H; cannam@95: T1M = T1H - T1E; cannam@95: Im[0] = -(KP500000000 * (FNMS(KP707106781, T1Q, T1N))); cannam@95: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1Q, T1N)); cannam@95: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1S, T1R)); cannam@95: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP707106781, T1S, T1R)); cannam@95: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1M, T1L)); cannam@95: Rm[0] = KP500000000 * (FNMS(KP707106781, T1M, T1L)); cannam@95: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP707106781, T1I, T1B))); cannam@95: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1I, T1B)); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_CEXP, 1, 7}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {60, 36, 30, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft2_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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 90 FP additions, 56 FP multiplications, cannam@95: * (or, 72 additions, 38 multiplications, 18 fused multiply/add), cannam@95: * 51 stack variables, 2 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cfdft2_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) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(32, rs)) { cannam@95: E T1, T4, T2, T5, Tu, Ty, T7, Td, Ti, Tj, Tk, TP, To, TN; cannam@95: { cannam@95: E T3, Tc, T6, Tb; cannam@95: T1 = W[0]; cannam@95: T4 = W[1]; cannam@95: T2 = W[2]; cannam@95: T5 = W[3]; cannam@95: T3 = T1 * T2; cannam@95: Tc = T4 * T2; cannam@95: T6 = T4 * T5; cannam@95: Tb = T1 * T5; cannam@95: Tu = T3 - T6; cannam@95: Ty = Tb + Tc; cannam@95: T7 = T3 + T6; cannam@95: Td = Tb - Tc; cannam@95: Ti = W[4]; cannam@95: Tj = W[5]; cannam@95: Tk = FMA(T1, Ti, T4 * Tj); cannam@95: TP = FNMS(Td, Ti, T7 * Tj); cannam@95: To = FNMS(T4, Ti, T1 * Tj); cannam@95: TN = FMA(T7, Ti, Td * Tj); cannam@95: } cannam@95: { cannam@95: E TF, T11, TC, T12, T1d, T1e, T1q, TM, TR, T1p, Th, Ts, T15, T14, T1a; cannam@95: E T1b, T1m, TV, TY, T1n; cannam@95: { cannam@95: E TD, TE, TL, TI, TJ, TK, Tx, TQ, TB, TO; cannam@95: TD = Ip[0]; cannam@95: TE = Im[0]; cannam@95: TL = TD + TE; cannam@95: TI = Rm[0]; cannam@95: TJ = Rp[0]; cannam@95: TK = TI - TJ; cannam@95: { cannam@95: E Tv, Tw, Tz, TA; cannam@95: Tv = Ip[WS(rs, 2)]; cannam@95: Tw = Im[WS(rs, 2)]; cannam@95: Tx = Tv - Tw; cannam@95: TQ = Tv + Tw; cannam@95: Tz = Rp[WS(rs, 2)]; cannam@95: TA = Rm[WS(rs, 2)]; cannam@95: TB = Tz + TA; cannam@95: TO = Tz - TA; cannam@95: } cannam@95: TF = TD - TE; cannam@95: T11 = TJ + TI; cannam@95: TC = FNMS(Ty, TB, Tu * Tx); cannam@95: T12 = FMA(Tu, TB, Ty * Tx); cannam@95: T1d = FNMS(TP, TO, TN * TQ); cannam@95: T1e = FMA(T4, TK, T1 * TL); cannam@95: T1q = T1e - T1d; cannam@95: TM = FNMS(T4, TL, T1 * TK); cannam@95: TR = FMA(TN, TO, TP * TQ); cannam@95: T1p = TR + TM; cannam@95: } cannam@95: { cannam@95: E Ta, TU, Tg, TT, Tn, TX, Tr, TW; cannam@95: { cannam@95: E T8, T9, Te, Tf; cannam@95: T8 = Ip[WS(rs, 1)]; cannam@95: T9 = Im[WS(rs, 1)]; cannam@95: Ta = T8 - T9; cannam@95: TU = T8 + T9; cannam@95: Te = Rp[WS(rs, 1)]; cannam@95: Tf = Rm[WS(rs, 1)]; cannam@95: Tg = Te + Tf; cannam@95: TT = Te - Tf; cannam@95: } cannam@95: { cannam@95: E Tl, Tm, Tp, Tq; cannam@95: Tl = Ip[WS(rs, 3)]; cannam@95: Tm = Im[WS(rs, 3)]; cannam@95: Tn = Tl - Tm; cannam@95: TX = Tl + Tm; cannam@95: Tp = Rp[WS(rs, 3)]; cannam@95: Tq = Rm[WS(rs, 3)]; cannam@95: Tr = Tp + Tq; cannam@95: TW = Tp - Tq; cannam@95: } cannam@95: Th = FNMS(Td, Tg, T7 * Ta); cannam@95: Ts = FNMS(To, Tr, Tk * Tn); cannam@95: T15 = FMA(Tk, Tr, To * Tn); cannam@95: T14 = FMA(T7, Tg, Td * Ta); cannam@95: T1a = FNMS(T5, TT, T2 * TU); cannam@95: T1b = FNMS(Tj, TW, Ti * TX); cannam@95: T1m = T1b - T1a; cannam@95: TV = FMA(T2, TT, T5 * TU); cannam@95: TY = FMA(Ti, TW, Tj * TX); cannam@95: T1n = TV - TY; cannam@95: } cannam@95: { cannam@95: E T1l, T1x, T1A, T1C, T1s, T1w, T1v, T1B; cannam@95: { cannam@95: E T1j, T1k, T1y, T1z; cannam@95: T1j = TF - TC; cannam@95: T1k = T14 - T15; cannam@95: T1l = KP500000000 * (T1j - T1k); cannam@95: T1x = KP500000000 * (T1k + T1j); cannam@95: T1y = T1m - T1n; cannam@95: T1z = T1p + T1q; cannam@95: T1A = KP353553390 * (T1y - T1z); cannam@95: T1C = KP353553390 * (T1y + T1z); cannam@95: } cannam@95: { cannam@95: E T1o, T1r, T1t, T1u; cannam@95: T1o = T1m + T1n; cannam@95: T1r = T1p - T1q; cannam@95: T1s = KP353553390 * (T1o + T1r); cannam@95: T1w = KP353553390 * (T1r - T1o); cannam@95: T1t = T11 - T12; cannam@95: T1u = Th - Ts; cannam@95: T1v = KP500000000 * (T1t - T1u); cannam@95: T1B = KP500000000 * (T1t + T1u); cannam@95: } cannam@95: Ip[WS(rs, 1)] = T1l + T1s; cannam@95: Rp[WS(rs, 1)] = T1B + T1C; cannam@95: Im[WS(rs, 2)] = T1s - T1l; cannam@95: Rm[WS(rs, 2)] = T1B - T1C; cannam@95: Rm[0] = T1v - T1w; cannam@95: Im[0] = T1A - T1x; cannam@95: Rp[WS(rs, 3)] = T1v + T1w; cannam@95: Ip[WS(rs, 3)] = T1x + T1A; cannam@95: } cannam@95: { cannam@95: E TH, T19, T1g, T1i, T10, T18, T17, T1h; cannam@95: { cannam@95: E Tt, TG, T1c, T1f; cannam@95: Tt = Th + Ts; cannam@95: TG = TC + TF; cannam@95: TH = Tt + TG; cannam@95: T19 = TG - Tt; cannam@95: T1c = T1a + T1b; cannam@95: T1f = T1d + T1e; cannam@95: T1g = T1c - T1f; cannam@95: T1i = T1c + T1f; cannam@95: } cannam@95: { cannam@95: E TS, TZ, T13, T16; cannam@95: TS = TM - TR; cannam@95: TZ = TV + TY; cannam@95: T10 = TS - TZ; cannam@95: T18 = TZ + TS; cannam@95: T13 = T11 + T12; cannam@95: T16 = T14 + T15; cannam@95: T17 = T13 - T16; cannam@95: T1h = T13 + T16; cannam@95: } cannam@95: Ip[0] = KP500000000 * (TH + T10); cannam@95: Rp[0] = KP500000000 * (T1h + T1i); cannam@95: Im[WS(rs, 3)] = KP500000000 * (T10 - TH); cannam@95: Rm[WS(rs, 3)] = KP500000000 * (T1h - T1i); cannam@95: Rm[WS(rs, 1)] = KP500000000 * (T17 - T18); cannam@95: Im[WS(rs, 1)] = KP500000000 * (T1g - T19); cannam@95: Rp[WS(rs, 2)] = KP500000000 * (T17 + T18); cannam@95: Ip[WS(rs, 2)] = KP500000000 * (T19 + T1g); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_CEXP, 1, 7}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {72, 38, 18, 0} }; cannam@95: cannam@95: void X(codelet_hc2cfdft2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cfdft2_8, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */