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 07:40:50 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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 90 FP additions, 66 FP multiplications, Chris@10: * (or, 60 additions, 36 multiplications, 30 fused multiply/add), Chris@10: * 68 stack variables, 2 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft2_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) * 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)) { Chris@10: E T1G, T1F, T1C, T1D, T1N, T1B, T1R, T1L; Chris@10: { Chris@10: E T1, T2, Th, Tj, T4, T3, Ti, Tp, T5; Chris@10: T1 = W[0]; Chris@10: T2 = W[2]; Chris@10: Th = W[4]; Chris@10: Tj = W[5]; Chris@10: T4 = W[1]; Chris@10: T3 = T1 * T2; Chris@10: Ti = T1 * Th; Chris@10: Tp = T1 * Tj; Chris@10: T5 = W[3]; Chris@10: { Chris@10: E Tk, Tq, TI, T1a, T1u, TY, TF, TS, T1s, T1c, Tr, T1n, Tg, T16, Tn; Chris@10: E T13, T1f, Ts, To, T1o; Chris@10: { Chris@10: E T6, Tw, Tc, TB, TQ, TM, TC, TR, Tz, TD, TA; Chris@10: { Chris@10: E TX, TV, TT, TU; Chris@10: { Chris@10: E TG, Tb, TH, TP, TL; Chris@10: TG = Ip[0]; Chris@10: Tk = FMA(T4, Tj, Ti); Chris@10: Tq = FNMS(T4, Th, Tp); Chris@10: T6 = FMA(T4, T5, T3); Chris@10: Tw = FNMS(T4, T5, T3); Chris@10: Tb = T1 * T5; Chris@10: TH = Im[0]; Chris@10: TT = Rm[0]; Chris@10: TP = T6 * Tj; Chris@10: TL = T6 * Th; Chris@10: Tc = FNMS(T4, T2, Tb); Chris@10: TB = FMA(T4, T2, Tb); Chris@10: TX = TG + TH; Chris@10: TI = TG - TH; Chris@10: TU = Rp[0]; Chris@10: TQ = FNMS(Tc, Th, TP); Chris@10: TM = FMA(Tc, Tj, TL); Chris@10: } Chris@10: T1a = TU + TT; Chris@10: TV = TT - TU; Chris@10: { Chris@10: E Tx, Ty, T1t, TW; Chris@10: Tx = Ip[WS(rs, 2)]; Chris@10: Ty = Im[WS(rs, 2)]; Chris@10: T1t = T4 * TV; Chris@10: TW = T1 * TV; Chris@10: TC = Rp[WS(rs, 2)]; Chris@10: TR = Tx + Ty; Chris@10: Tz = Tx - Ty; Chris@10: T1u = FMA(T1, TX, T1t); Chris@10: TY = FNMS(T4, TX, TW); Chris@10: TD = Rm[WS(rs, 2)]; Chris@10: } Chris@10: TA = Tw * Tz; Chris@10: } Chris@10: { Chris@10: E Td, T9, T12, Te, Ta, T1m; Chris@10: { Chris@10: E T7, T8, TN, TE, TO, T1r, T1b; Chris@10: T7 = Ip[WS(rs, 1)]; Chris@10: T8 = Im[WS(rs, 1)]; Chris@10: TN = TD - TC; Chris@10: TE = TC + TD; Chris@10: Td = Rp[WS(rs, 1)]; Chris@10: T9 = T7 - T8; Chris@10: T12 = T7 + T8; Chris@10: TO = TM * TN; Chris@10: T1r = TQ * TN; Chris@10: T1b = Tw * TE; Chris@10: TF = FNMS(TB, TE, TA); Chris@10: TS = FNMS(TQ, TR, TO); Chris@10: T1s = FMA(TM, TR, T1r); Chris@10: T1c = FMA(TB, Tz, T1b); Chris@10: Te = Rm[WS(rs, 1)]; Chris@10: } Chris@10: Ta = T6 * T9; Chris@10: T1m = T2 * T12; Chris@10: { Chris@10: E Tl, T10, Tf, Tm, T11, T1e; Chris@10: Tl = Ip[WS(rs, 3)]; Chris@10: T10 = Td - Te; Chris@10: Tf = Td + Te; Chris@10: Tm = Im[WS(rs, 3)]; Chris@10: Tr = Rp[WS(rs, 3)]; Chris@10: T11 = T2 * T10; Chris@10: T1n = FNMS(T5, T10, T1m); Chris@10: T1e = T6 * Tf; Chris@10: Tg = FNMS(Tc, Tf, Ta); Chris@10: T16 = Tl + Tm; Chris@10: Tn = Tl - Tm; Chris@10: T13 = FMA(T5, T12, T11); Chris@10: T1f = FMA(Tc, T9, T1e); Chris@10: Ts = Rm[WS(rs, 3)]; Chris@10: } Chris@10: To = Tk * Tn; Chris@10: T1o = Th * T16; Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1z, T1K, T1y, T1k, T1J, T1A, T1x, T1j; Chris@10: { Chris@10: E T1w, TK, T1l, T19, T1d, T1i; Chris@10: { Chris@10: E TJ, T14, Tt, T1v, T1h; Chris@10: T1z = TI - TF; Chris@10: TJ = TF + TI; Chris@10: T14 = Tr - Ts; Chris@10: Tt = Tr + Ts; Chris@10: T1v = T1s + T1u; Chris@10: T1G = T1u - T1s; Chris@10: { Chris@10: E TZ, T1q, Tv, T18, T15; Chris@10: T1F = TY - TS; Chris@10: TZ = TS + TY; Chris@10: T15 = Th * T14; Chris@10: { Chris@10: E T1p, T1g, Tu, T17; Chris@10: T1p = FNMS(Tj, T14, T1o); Chris@10: T1g = Tk * Tt; Chris@10: Tu = FNMS(Tq, Tt, To); Chris@10: T17 = FMA(Tj, T16, T15); Chris@10: T1C = T1p - T1n; Chris@10: T1q = T1n + T1p; Chris@10: T1h = FMA(Tq, Tn, T1g); Chris@10: T1K = Tg - Tu; Chris@10: Tv = Tg + Tu; Chris@10: T18 = T13 + T17; Chris@10: T1D = T13 - T17; Chris@10: } Chris@10: T1w = T1q - T1v; Chris@10: T1y = T1q + T1v; Chris@10: TK = Tv + TJ; Chris@10: T1l = TJ - Tv; Chris@10: T1k = T18 + TZ; Chris@10: T19 = TZ - T18; Chris@10: } Chris@10: T1J = T1a - T1c; Chris@10: T1d = T1a + T1c; Chris@10: T1i = T1f + T1h; Chris@10: T1A = T1f - T1h; Chris@10: } Chris@10: Ip[0] = KP500000000 * (TK + T19); Chris@10: Im[WS(rs, 3)] = KP500000000 * (T19 - TK); Chris@10: Im[WS(rs, 1)] = KP500000000 * (T1w - T1l); Chris@10: T1x = T1d + T1i; Chris@10: T1j = T1d - T1i; Chris@10: Ip[WS(rs, 2)] = KP500000000 * (T1l + T1w); Chris@10: } Chris@10: Rm[WS(rs, 3)] = KP500000000 * (T1x - T1y); Chris@10: Rp[0] = KP500000000 * (T1x + T1y); Chris@10: Rp[WS(rs, 2)] = KP500000000 * (T1j + T1k); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (T1j - T1k); Chris@10: T1N = T1A + T1z; Chris@10: T1B = T1z - T1A; Chris@10: T1R = T1J + T1K; Chris@10: T1L = T1J - T1K; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E T1E, T1O, T1H, T1P; Chris@10: T1E = T1C + T1D; Chris@10: T1O = T1C - T1D; Chris@10: T1H = T1F - T1G; Chris@10: T1P = T1F + T1G; Chris@10: { Chris@10: E T1S, T1Q, T1I, T1M; Chris@10: T1S = T1O + T1P; Chris@10: T1Q = T1O - T1P; Chris@10: T1I = T1E + T1H; Chris@10: T1M = T1H - T1E; Chris@10: Im[0] = -(KP500000000 * (FNMS(KP707106781, T1Q, T1N))); Chris@10: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1Q, T1N)); Chris@10: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1S, T1R)); Chris@10: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP707106781, T1S, T1R)); Chris@10: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1M, T1L)); Chris@10: Rm[0] = KP500000000 * (FNMS(KP707106781, T1M, T1L)); Chris@10: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP707106781, T1I, T1B))); Chris@10: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1I, T1B)); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_CEXP, 1, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {60, 36, 30, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft2_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft2_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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cfdft2_8 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 90 FP additions, 56 FP multiplications, Chris@10: * (or, 72 additions, 38 multiplications, 18 fused multiply/add), Chris@10: * 51 stack variables, 2 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cfdft2_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) * 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)) { Chris@10: E T1, T4, T2, T5, Tu, Ty, T7, Td, Ti, Tj, Tk, TP, To, TN; Chris@10: { Chris@10: E T3, Tc, T6, Tb; Chris@10: T1 = W[0]; Chris@10: T4 = W[1]; Chris@10: T2 = W[2]; Chris@10: T5 = W[3]; Chris@10: T3 = T1 * T2; Chris@10: Tc = T4 * T2; Chris@10: T6 = T4 * T5; Chris@10: Tb = T1 * T5; Chris@10: Tu = T3 - T6; Chris@10: Ty = Tb + Tc; Chris@10: T7 = T3 + T6; Chris@10: Td = Tb - Tc; Chris@10: Ti = W[4]; Chris@10: Tj = W[5]; Chris@10: Tk = FMA(T1, Ti, T4 * Tj); Chris@10: TP = FNMS(Td, Ti, T7 * Tj); Chris@10: To = FNMS(T4, Ti, T1 * Tj); Chris@10: TN = FMA(T7, Ti, Td * Tj); Chris@10: } Chris@10: { Chris@10: E TF, T11, TC, T12, T1d, T1e, T1q, TM, TR, T1p, Th, Ts, T15, T14, T1a; Chris@10: E T1b, T1m, TV, TY, T1n; Chris@10: { Chris@10: E TD, TE, TL, TI, TJ, TK, Tx, TQ, TB, TO; Chris@10: TD = Ip[0]; Chris@10: TE = Im[0]; Chris@10: TL = TD + TE; Chris@10: TI = Rm[0]; Chris@10: TJ = Rp[0]; Chris@10: TK = TI - TJ; Chris@10: { Chris@10: E Tv, Tw, Tz, TA; Chris@10: Tv = Ip[WS(rs, 2)]; Chris@10: Tw = Im[WS(rs, 2)]; Chris@10: Tx = Tv - Tw; Chris@10: TQ = Tv + Tw; Chris@10: Tz = Rp[WS(rs, 2)]; Chris@10: TA = Rm[WS(rs, 2)]; Chris@10: TB = Tz + TA; Chris@10: TO = Tz - TA; Chris@10: } Chris@10: TF = TD - TE; Chris@10: T11 = TJ + TI; Chris@10: TC = FNMS(Ty, TB, Tu * Tx); Chris@10: T12 = FMA(Tu, TB, Ty * Tx); Chris@10: T1d = FNMS(TP, TO, TN * TQ); Chris@10: T1e = FMA(T4, TK, T1 * TL); Chris@10: T1q = T1e - T1d; Chris@10: TM = FNMS(T4, TL, T1 * TK); Chris@10: TR = FMA(TN, TO, TP * TQ); Chris@10: T1p = TR + TM; Chris@10: } Chris@10: { Chris@10: E Ta, TU, Tg, TT, Tn, TX, Tr, TW; Chris@10: { Chris@10: E T8, T9, Te, Tf; Chris@10: T8 = Ip[WS(rs, 1)]; Chris@10: T9 = Im[WS(rs, 1)]; Chris@10: Ta = T8 - T9; Chris@10: TU = T8 + T9; Chris@10: Te = Rp[WS(rs, 1)]; Chris@10: Tf = Rm[WS(rs, 1)]; Chris@10: Tg = Te + Tf; Chris@10: TT = Te - Tf; Chris@10: } Chris@10: { Chris@10: E Tl, Tm, Tp, Tq; Chris@10: Tl = Ip[WS(rs, 3)]; Chris@10: Tm = Im[WS(rs, 3)]; Chris@10: Tn = Tl - Tm; Chris@10: TX = Tl + Tm; Chris@10: Tp = Rp[WS(rs, 3)]; Chris@10: Tq = Rm[WS(rs, 3)]; Chris@10: Tr = Tp + Tq; Chris@10: TW = Tp - Tq; Chris@10: } Chris@10: Th = FNMS(Td, Tg, T7 * Ta); Chris@10: Ts = FNMS(To, Tr, Tk * Tn); Chris@10: T15 = FMA(Tk, Tr, To * Tn); Chris@10: T14 = FMA(T7, Tg, Td * Ta); Chris@10: T1a = FNMS(T5, TT, T2 * TU); Chris@10: T1b = FNMS(Tj, TW, Ti * TX); Chris@10: T1m = T1b - T1a; Chris@10: TV = FMA(T2, TT, T5 * TU); Chris@10: TY = FMA(Ti, TW, Tj * TX); Chris@10: T1n = TV - TY; Chris@10: } Chris@10: { Chris@10: E T1l, T1x, T1A, T1C, T1s, T1w, T1v, T1B; Chris@10: { Chris@10: E T1j, T1k, T1y, T1z; Chris@10: T1j = TF - TC; Chris@10: T1k = T14 - T15; Chris@10: T1l = KP500000000 * (T1j - T1k); Chris@10: T1x = KP500000000 * (T1k + T1j); Chris@10: T1y = T1m - T1n; Chris@10: T1z = T1p + T1q; Chris@10: T1A = KP353553390 * (T1y - T1z); Chris@10: T1C = KP353553390 * (T1y + T1z); Chris@10: } Chris@10: { Chris@10: E T1o, T1r, T1t, T1u; Chris@10: T1o = T1m + T1n; Chris@10: T1r = T1p - T1q; Chris@10: T1s = KP353553390 * (T1o + T1r); Chris@10: T1w = KP353553390 * (T1r - T1o); Chris@10: T1t = T11 - T12; Chris@10: T1u = Th - Ts; Chris@10: T1v = KP500000000 * (T1t - T1u); Chris@10: T1B = KP500000000 * (T1t + T1u); Chris@10: } Chris@10: Ip[WS(rs, 1)] = T1l + T1s; Chris@10: Rp[WS(rs, 1)] = T1B + T1C; Chris@10: Im[WS(rs, 2)] = T1s - T1l; Chris@10: Rm[WS(rs, 2)] = T1B - T1C; Chris@10: Rm[0] = T1v - T1w; Chris@10: Im[0] = T1A - T1x; Chris@10: Rp[WS(rs, 3)] = T1v + T1w; Chris@10: Ip[WS(rs, 3)] = T1x + T1A; Chris@10: } Chris@10: { Chris@10: E TH, T19, T1g, T1i, T10, T18, T17, T1h; Chris@10: { Chris@10: E Tt, TG, T1c, T1f; Chris@10: Tt = Th + Ts; Chris@10: TG = TC + TF; Chris@10: TH = Tt + TG; Chris@10: T19 = TG - Tt; Chris@10: T1c = T1a + T1b; Chris@10: T1f = T1d + T1e; Chris@10: T1g = T1c - T1f; Chris@10: T1i = T1c + T1f; Chris@10: } Chris@10: { Chris@10: E TS, TZ, T13, T16; Chris@10: TS = TM - TR; Chris@10: TZ = TV + TY; Chris@10: T10 = TS - TZ; Chris@10: T18 = TZ + TS; Chris@10: T13 = T11 + T12; Chris@10: T16 = T14 + T15; Chris@10: T17 = T13 - T16; Chris@10: T1h = T13 + T16; Chris@10: } Chris@10: Ip[0] = KP500000000 * (TH + T10); Chris@10: Rp[0] = KP500000000 * (T1h + T1i); Chris@10: Im[WS(rs, 3)] = KP500000000 * (T10 - TH); Chris@10: Rm[WS(rs, 3)] = KP500000000 * (T1h - T1i); Chris@10: Rm[WS(rs, 1)] = KP500000000 * (T17 - T18); Chris@10: Im[WS(rs, 1)] = KP500000000 * (T1g - T19); Chris@10: Rp[WS(rs, 2)] = KP500000000 * (T17 + T18); Chris@10: Ip[WS(rs, 2)] = KP500000000 * (T19 + T1g); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_CEXP, 1, 1}, Chris@10: {TW_CEXP, 1, 3}, Chris@10: {TW_CEXP, 1, 7}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 8, "hc2cfdft2_8", twinstr, &GENUS, {72, 38, 18, 0} }; Chris@10: Chris@10: void X(codelet_hc2cfdft2_8) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cfdft2_8, &desc, HC2C_VIA_DFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */