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 07:41:52 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_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cb_8 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 36 FP multiplications, cannam@95: * (or, 44 additions, 14 multiplications, 22 fused multiply/add), cannam@95: * 52 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb_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: { 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 Tw, TH, Tf, Ty, Tx, TI; cannam@95: { cannam@95: E TV, TD, T1i, T7, T1b, T1n, TQ, Tk, Tp, TE, Te, T1o, T1e, T1j, Tu; cannam@95: E TF; cannam@95: { cannam@95: E T4, Tg, T3, T19, TC, T5, Th, Ti; cannam@95: { cannam@95: E T1, T2, TA, TB; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 3)]; cannam@95: TA = Ip[0]; cannam@95: TB = Im[WS(rs, 3)]; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: Tg = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: T19 = TA - TB; cannam@95: TC = TA + TB; cannam@95: T5 = Rm[WS(rs, 1)]; cannam@95: Th = Ip[WS(rs, 2)]; cannam@95: Ti = Im[WS(rs, 1)]; cannam@95: } cannam@95: { cannam@95: E Tb, Tl, Ta, T1c, To, Tc, Tr, Ts; cannam@95: { cannam@95: E T8, T9, Tm, Tn; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: { cannam@95: E Tz, T6, T1a, Tj; cannam@95: Tz = T4 - T5; cannam@95: T6 = T4 + T5; cannam@95: T1a = Th - Ti; cannam@95: Tj = Th + Ti; cannam@95: TV = TC - Tz; cannam@95: TD = Tz + TC; cannam@95: T1i = T3 - T6; cannam@95: T7 = T3 + T6; cannam@95: T1b = T19 + T1a; cannam@95: T1n = T19 - T1a; cannam@95: TQ = Tg + Tj; cannam@95: Tk = Tg - Tj; cannam@95: T9 = Rm[WS(rs, 2)]; cannam@95: } cannam@95: Tm = Ip[WS(rs, 1)]; cannam@95: Tn = Im[WS(rs, 2)]; cannam@95: Tb = Rm[0]; cannam@95: Tl = T8 - T9; cannam@95: Ta = T8 + T9; cannam@95: T1c = Tm - Tn; cannam@95: To = Tm + Tn; cannam@95: Tc = Rp[WS(rs, 3)]; cannam@95: Tr = Ip[WS(rs, 3)]; cannam@95: Ts = Im[0]; cannam@95: } cannam@95: { cannam@95: E Tq, Td, T1d, Tt; cannam@95: Tp = Tl - To; cannam@95: TE = Tl + To; cannam@95: Tq = Tb - Tc; cannam@95: Td = Tb + Tc; cannam@95: T1d = Tr - Ts; cannam@95: Tt = Tr + Ts; cannam@95: Te = Ta + Td; cannam@95: T1o = Ta - Td; cannam@95: T1e = T1c + T1d; cannam@95: T1j = T1d - T1c; cannam@95: Tu = Tq - Tt; cannam@95: TF = Tq + Tt; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TG, Tv, T10, T13, T1s, T1k, T1p, T1v, T1u, T1w, T1t, TR, TW; cannam@95: Rp[0] = T7 + Te; cannam@95: Rm[0] = T1b + T1e; cannam@95: TG = TE - TF; cannam@95: TR = TE + TF; cannam@95: TW = Tp - Tu; cannam@95: Tv = Tp + Tu; cannam@95: { cannam@95: E TP, TS, TX, TU, T1r, TT, TY; cannam@95: TP = W[4]; cannam@95: T10 = FMA(KP707106781, TR, TQ); cannam@95: TS = FNMS(KP707106781, TR, TQ); cannam@95: TX = FMA(KP707106781, TW, TV); cannam@95: T13 = FNMS(KP707106781, TW, TV); cannam@95: TU = W[5]; cannam@95: T1s = T1i + T1j; cannam@95: T1k = T1i - T1j; cannam@95: TT = TP * TS; cannam@95: TY = TP * TX; cannam@95: T1p = T1n - T1o; cannam@95: T1v = T1o + T1n; cannam@95: T1r = W[2]; cannam@95: Ip[WS(rs, 1)] = FNMS(TU, TX, TT); cannam@95: Im[WS(rs, 1)] = FMA(TU, TS, TY); cannam@95: T1u = W[3]; cannam@95: T1w = T1r * T1v; cannam@95: T1t = T1r * T1s; cannam@95: } cannam@95: { cannam@95: E T1f, T15, T18, T17, T1g, T1h, T1m; cannam@95: { cannam@95: E TZ, T12, T16, T14, T11; cannam@95: Rm[WS(rs, 1)] = FMA(T1u, T1s, T1w); cannam@95: Rp[WS(rs, 1)] = FNMS(T1u, T1v, T1t); cannam@95: TZ = W[12]; cannam@95: T12 = W[13]; cannam@95: T1f = T1b - T1e; cannam@95: T16 = T7 - Te; cannam@95: T14 = TZ * T13; cannam@95: T11 = TZ * T10; cannam@95: T15 = W[6]; cannam@95: T18 = W[7]; cannam@95: Im[WS(rs, 3)] = FMA(T12, T10, T14); cannam@95: Ip[WS(rs, 3)] = FNMS(T12, T13, T11); cannam@95: T17 = T15 * T16; cannam@95: T1g = T18 * T16; cannam@95: } cannam@95: Rp[WS(rs, 2)] = FNMS(T18, T1f, T17); cannam@95: Rm[WS(rs, 2)] = FMA(T15, T1f, T1g); cannam@95: T1h = W[10]; cannam@95: T1m = W[11]; cannam@95: { cannam@95: E TN, TJ, TM, TL, TO, TK, T1q, T1l; cannam@95: Tw = FNMS(KP707106781, Tv, Tk); cannam@95: TK = FMA(KP707106781, Tv, Tk); cannam@95: T1q = T1h * T1p; cannam@95: T1l = T1h * T1k; cannam@95: TN = FMA(KP707106781, TG, TD); cannam@95: TH = FNMS(KP707106781, TG, TD); cannam@95: Rm[WS(rs, 3)] = FMA(T1m, T1k, T1q); cannam@95: Rp[WS(rs, 3)] = FNMS(T1m, T1p, T1l); cannam@95: TJ = W[0]; cannam@95: TM = W[1]; cannam@95: Tf = W[8]; cannam@95: TL = TJ * TK; cannam@95: TO = TM * TK; cannam@95: Ty = W[9]; cannam@95: Tx = Tf * Tw; cannam@95: Ip[0] = FNMS(TM, TN, TL); cannam@95: Im[0] = FMA(TJ, TN, TO); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: Ip[WS(rs, 2)] = FNMS(Ty, TH, Tx); cannam@95: TI = Ty * Tw; cannam@95: Im[WS(rs, 2)] = FMA(Tf, TH, TI); 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, "hc2cb_8", twinstr, &GENUS, {44, 14, 22, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb_8, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 8 -dif -name hc2cb_8 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 32 FP multiplications, cannam@95: * (or, 52 additions, 18 multiplications, 14 fused multiply/add), cannam@95: * 30 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb_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: { 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 T7, T18, T1c, To, Ty, TM, TY, TC, Te, TZ, T10, Tv, Tz, TP, TS; cannam@95: E TD; cannam@95: { cannam@95: E T3, TK, Tk, TX, T6, TW, Tn, TL; cannam@95: { cannam@95: E T1, T2, Ti, Tj; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 3)]; cannam@95: T3 = T1 + T2; cannam@95: TK = T1 - T2; cannam@95: Ti = Ip[0]; cannam@95: Tj = Im[WS(rs, 3)]; cannam@95: Tk = Ti - Tj; cannam@95: TX = Ti + Tj; cannam@95: } cannam@95: { cannam@95: E T4, T5, Tl, Tm; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 1)]; cannam@95: T6 = T4 + T5; cannam@95: TW = T4 - T5; cannam@95: Tl = Ip[WS(rs, 2)]; cannam@95: Tm = Im[WS(rs, 1)]; cannam@95: Tn = Tl - Tm; cannam@95: TL = Tl + Tm; cannam@95: } cannam@95: T7 = T3 + T6; cannam@95: T18 = TK + TL; cannam@95: T1c = TX - TW; cannam@95: To = Tk + Tn; cannam@95: Ty = T3 - T6; cannam@95: TM = TK - TL; cannam@95: TY = TW + TX; cannam@95: TC = Tk - Tn; cannam@95: } cannam@95: { cannam@95: E Ta, TN, Tr, TO, Td, TQ, Tu, TR; cannam@95: { cannam@95: E T8, T9, Tp, Tq; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: T9 = Rm[WS(rs, 2)]; cannam@95: Ta = T8 + T9; cannam@95: TN = T8 - T9; cannam@95: Tp = Ip[WS(rs, 1)]; cannam@95: Tq = Im[WS(rs, 2)]; cannam@95: Tr = Tp - Tq; cannam@95: TO = Tp + Tq; cannam@95: } cannam@95: { cannam@95: E Tb, Tc, Ts, Tt; cannam@95: Tb = Rm[0]; cannam@95: Tc = Rp[WS(rs, 3)]; cannam@95: Td = Tb + Tc; cannam@95: TQ = Tb - Tc; cannam@95: Ts = Ip[WS(rs, 3)]; cannam@95: Tt = Im[0]; cannam@95: Tu = Ts - Tt; cannam@95: TR = Ts + Tt; cannam@95: } cannam@95: Te = Ta + Td; cannam@95: TZ = TN + TO; cannam@95: T10 = TQ + TR; cannam@95: Tv = Tr + Tu; cannam@95: Tz = Tu - Tr; cannam@95: TP = TN - TO; cannam@95: TS = TQ - TR; cannam@95: TD = Ta - Td; cannam@95: } cannam@95: Rp[0] = T7 + Te; cannam@95: Rm[0] = To + Tv; cannam@95: { cannam@95: E Tg, Tw, Tf, Th; cannam@95: Tg = T7 - Te; cannam@95: Tw = To - Tv; cannam@95: Tf = W[6]; cannam@95: Th = W[7]; cannam@95: Rp[WS(rs, 2)] = FNMS(Th, Tw, Tf * Tg); cannam@95: Rm[WS(rs, 2)] = FMA(Th, Tg, Tf * Tw); cannam@95: } cannam@95: { cannam@95: E TG, TI, TF, TH; cannam@95: TG = Ty + Tz; cannam@95: TI = TD + TC; cannam@95: TF = W[2]; cannam@95: TH = W[3]; cannam@95: Rp[WS(rs, 1)] = FNMS(TH, TI, TF * TG); cannam@95: Rm[WS(rs, 1)] = FMA(TF, TI, TH * TG); cannam@95: } cannam@95: { cannam@95: E TA, TE, Tx, TB; cannam@95: TA = Ty - Tz; cannam@95: TE = TC - TD; cannam@95: Tx = W[10]; cannam@95: TB = W[11]; cannam@95: Rp[WS(rs, 3)] = FNMS(TB, TE, Tx * TA); cannam@95: Rm[WS(rs, 3)] = FMA(Tx, TE, TB * TA); cannam@95: } cannam@95: { cannam@95: E T1a, T1g, T1e, T1i, T19, T1d; cannam@95: T19 = KP707106781 * (TZ + T10); cannam@95: T1a = T18 - T19; cannam@95: T1g = T18 + T19; cannam@95: T1d = KP707106781 * (TP - TS); cannam@95: T1e = T1c + T1d; cannam@95: T1i = T1c - T1d; cannam@95: { cannam@95: E T17, T1b, T1f, T1h; cannam@95: T17 = W[4]; cannam@95: T1b = W[5]; cannam@95: Ip[WS(rs, 1)] = FNMS(T1b, T1e, T17 * T1a); cannam@95: Im[WS(rs, 1)] = FMA(T17, T1e, T1b * T1a); cannam@95: T1f = W[12]; cannam@95: T1h = W[13]; cannam@95: Ip[WS(rs, 3)] = FNMS(T1h, T1i, T1f * T1g); cannam@95: Im[WS(rs, 3)] = FMA(T1f, T1i, T1h * T1g); cannam@95: } cannam@95: } cannam@95: { cannam@95: E TU, T14, T12, T16, TT, T11; cannam@95: TT = KP707106781 * (TP + TS); cannam@95: TU = TM - TT; cannam@95: T14 = TM + TT; cannam@95: T11 = KP707106781 * (TZ - T10); cannam@95: T12 = TY - T11; cannam@95: T16 = TY + T11; cannam@95: { cannam@95: E TJ, TV, T13, T15; cannam@95: TJ = W[8]; cannam@95: TV = W[9]; cannam@95: Ip[WS(rs, 2)] = FNMS(TV, T12, TJ * TU); cannam@95: Im[WS(rs, 2)] = FMA(TV, TU, TJ * T12); cannam@95: T13 = W[0]; cannam@95: T15 = W[1]; cannam@95: Ip[0] = FNMS(T15, T16, T13 * T14); cannam@95: Im[0] = FMA(T15, T14, T13 * T16); 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, "hc2cb_8", twinstr, &GENUS, {52, 18, 14, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb_8, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */