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 07:40:42 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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cf2_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 74 FP additions, 50 FP multiplications, cannam@95: * (or, 44 additions, 20 multiplications, 30 fused multiply/add), cannam@95: * 64 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cf2_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) * 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 TS, T1m, TJ, T1l, T1k, Tw, T1w, T1u; cannam@95: { cannam@95: E T2, T3, Tl, Tn, T5, T4, Tm, Tr, T6; cannam@95: T2 = W[0]; cannam@95: T3 = W[2]; cannam@95: Tl = W[4]; cannam@95: Tn = W[5]; cannam@95: T5 = W[1]; cannam@95: T4 = T2 * T3; cannam@95: Tm = T2 * Tl; cannam@95: Tr = T2 * Tn; cannam@95: T6 = W[3]; cannam@95: { cannam@95: E T1, T1s, TG, Td, T1r, Tu, TY, Tk, TW, T18, T1d, TD, TH, TA, T13; cannam@95: E TE, T14; cannam@95: { cannam@95: E To, Ts, Tf, T7, T8, Ti, Tb, T9, Tc, TC, Ta, TF, TB, Tg, Th; cannam@95: E Tj; cannam@95: T1 = Rp[0]; cannam@95: To = FMA(T5, Tn, Tm); cannam@95: Ts = FNMS(T5, Tl, Tr); cannam@95: Tf = FMA(T5, T6, T4); cannam@95: T7 = FNMS(T5, T6, T4); cannam@95: Ta = T2 * T6; cannam@95: T1s = Rm[0]; cannam@95: T8 = Rp[WS(rs, 2)]; cannam@95: TF = Tf * Tn; cannam@95: TB = Tf * Tl; cannam@95: Ti = FNMS(T5, T3, Ta); cannam@95: Tb = FMA(T5, T3, Ta); cannam@95: T9 = T7 * T8; cannam@95: Tc = Rm[WS(rs, 2)]; cannam@95: TG = FNMS(Ti, Tl, TF); cannam@95: TC = FMA(Ti, Tn, TB); cannam@95: { cannam@95: E Tp, T1q, Tt, Tq, TX; cannam@95: Tp = Rp[WS(rs, 3)]; cannam@95: Td = FMA(Tb, Tc, T9); cannam@95: T1q = T7 * Tc; cannam@95: Tt = Rm[WS(rs, 3)]; cannam@95: Tq = To * Tp; cannam@95: Tg = Rp[WS(rs, 1)]; cannam@95: T1r = FNMS(Tb, T8, T1q); cannam@95: TX = To * Tt; cannam@95: Tu = FMA(Ts, Tt, Tq); cannam@95: Th = Tf * Tg; cannam@95: Tj = Rm[WS(rs, 1)]; cannam@95: TY = FNMS(Ts, Tp, TX); cannam@95: } cannam@95: { cannam@95: E TO, TQ, TN, TP, T1a, T1b; cannam@95: { cannam@95: E TK, TM, TL, T19, TV; cannam@95: TK = Ip[WS(rs, 3)]; cannam@95: TM = Im[WS(rs, 3)]; cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: TV = Tf * Tj; cannam@95: TL = Tl * TK; cannam@95: T19 = Tl * TM; cannam@95: TO = Ip[WS(rs, 1)]; cannam@95: TW = FNMS(Ti, Tg, TV); cannam@95: TQ = Im[WS(rs, 1)]; cannam@95: TN = FMA(Tn, TM, TL); cannam@95: TP = T3 * TO; cannam@95: T1a = FNMS(Tn, TK, T19); cannam@95: T1b = T3 * TQ; cannam@95: } cannam@95: { cannam@95: E Tx, Tz, Ty, T12, T1c, TR; cannam@95: Tx = Ip[0]; cannam@95: TR = FMA(T6, TQ, TP); cannam@95: Tz = Im[0]; cannam@95: T1c = FNMS(T6, TO, T1b); cannam@95: Ty = T2 * Tx; cannam@95: T18 = TN - TR; cannam@95: TS = TN + TR; cannam@95: T12 = T2 * Tz; cannam@95: T1d = T1a - T1c; cannam@95: T1m = T1a + T1c; cannam@95: TD = Ip[WS(rs, 2)]; cannam@95: TH = Im[WS(rs, 2)]; cannam@95: TA = FMA(T5, Tz, Ty); cannam@95: T13 = FNMS(T5, Tx, T12); cannam@95: TE = TC * TD; cannam@95: T14 = TC * TH; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E Te, T1p, T1t, Tv; cannam@95: { cannam@95: E T1g, T10, T1z, T1B, T1A, T1j, T1C, T1f; cannam@95: { cannam@95: E T1x, T11, T16, T1y; cannam@95: { cannam@95: E TU, TZ, TI, T15; cannam@95: Te = T1 + Td; cannam@95: TU = T1 - Td; cannam@95: TZ = TW - TY; cannam@95: T1p = TW + TY; cannam@95: TI = FMA(TG, TH, TE); cannam@95: T15 = FNMS(TG, TD, T14); cannam@95: T1t = T1r + T1s; cannam@95: T1x = T1s - T1r; cannam@95: T1g = TU - TZ; cannam@95: T10 = TU + TZ; cannam@95: T11 = TA - TI; cannam@95: TJ = TA + TI; cannam@95: T1l = T13 + T15; cannam@95: T16 = T13 - T15; cannam@95: T1y = Tk - Tu; cannam@95: Tv = Tk + Tu; cannam@95: } cannam@95: { cannam@95: E T1i, T1e, T17, T1h; cannam@95: T1i = T18 + T1d; cannam@95: T1e = T18 - T1d; cannam@95: T17 = T11 + T16; cannam@95: T1h = T16 - T11; cannam@95: T1z = T1x - T1y; cannam@95: T1B = T1y + T1x; cannam@95: T1A = T1h + T1i; cannam@95: T1j = T1h - T1i; cannam@95: T1C = T1e - T17; cannam@95: T1f = T17 + T1e; cannam@95: } cannam@95: } cannam@95: Rm[0] = FNMS(KP707106781, T1j, T1g); cannam@95: Im[0] = FMS(KP707106781, T1C, T1B); cannam@95: Rp[WS(rs, 1)] = FMA(KP707106781, T1f, T10); cannam@95: Rm[WS(rs, 2)] = FNMS(KP707106781, T1f, T10); cannam@95: Ip[WS(rs, 1)] = FMA(KP707106781, T1A, T1z); cannam@95: Im[WS(rs, 2)] = FMS(KP707106781, T1A, T1z); cannam@95: Rp[WS(rs, 3)] = FMA(KP707106781, T1j, T1g); cannam@95: Ip[WS(rs, 3)] = FMA(KP707106781, T1C, T1B); cannam@95: } cannam@95: T1k = Te - Tv; cannam@95: Tw = Te + Tv; cannam@95: T1w = T1t - T1p; cannam@95: T1u = T1p + T1t; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TT, T1v, T1n, T1o; cannam@95: TT = TJ + TS; cannam@95: T1v = TS - TJ; cannam@95: T1n = T1l - T1m; cannam@95: T1o = T1l + T1m; cannam@95: Ip[WS(rs, 2)] = T1v + T1w; cannam@95: Im[WS(rs, 1)] = T1v - T1w; cannam@95: Rp[0] = Tw + TT; cannam@95: Rm[WS(rs, 3)] = Tw - TT; cannam@95: Ip[0] = T1o + T1u; cannam@95: Im[WS(rs, 3)] = T1o - T1u; cannam@95: Rp[WS(rs, 2)] = T1k + T1n; cannam@95: Rm[WS(rs, 1)] = T1k - T1n; 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, "hc2cf2_8", twinstr, &GENUS, {44, 20, 30, 0} }; cannam@95: cannam@95: void X(codelet_hc2cf2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cf2_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 -twiddle-log3 -precompute-twiddles -n 8 -dit -name hc2cf2_8 -include hc2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 74 FP additions, 44 FP multiplications, cannam@95: * (or, 56 additions, 26 multiplications, 18 fused multiply/add), cannam@95: * 42 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cf.h" cannam@95: cannam@95: static void hc2cf2_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) * 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 T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx; cannam@95: { cannam@95: E T4, Tb, T7, Ta; cannam@95: T2 = W[0]; cannam@95: T5 = W[1]; cannam@95: T3 = W[2]; cannam@95: T6 = W[3]; cannam@95: T4 = T2 * T3; cannam@95: Tb = T5 * T3; cannam@95: T7 = T5 * T6; cannam@95: Ta = T2 * T6; cannam@95: T8 = T4 - T7; cannam@95: Tc = Ta + Tb; cannam@95: Tg = T4 + T7; cannam@95: Ti = Ta - Tb; cannam@95: Tl = W[4]; cannam@95: Tm = W[5]; cannam@95: Tn = FMA(T2, Tl, T5 * Tm); cannam@95: Tz = FNMS(Ti, Tl, Tg * Tm); cannam@95: Tp = FNMS(T5, Tl, T2 * Tm); cannam@95: Tx = FMA(Tg, Tl, Ti * Tm); cannam@95: } cannam@95: { cannam@95: E Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ; cannam@95: E TT; cannam@95: { cannam@95: E T1, T1c, Te, T1b, T9, Td; cannam@95: T1 = Rp[0]; cannam@95: T1c = Rm[0]; cannam@95: T9 = Rp[WS(rs, 2)]; cannam@95: Td = Rm[WS(rs, 2)]; cannam@95: Te = FMA(T8, T9, Tc * Td); cannam@95: T1b = FNMS(Tc, T9, T8 * Td); cannam@95: Tf = T1 + Te; cannam@95: T1i = T1c - T1b; cannam@95: TL = T1 - Te; cannam@95: T1d = T1b + T1c; cannam@95: } cannam@95: { cannam@95: E TF, TW, TI, TX; cannam@95: { cannam@95: E TD, TE, TG, TH; cannam@95: TD = Ip[WS(rs, 3)]; cannam@95: TE = Im[WS(rs, 3)]; cannam@95: TF = FMA(Tl, TD, Tm * TE); cannam@95: TW = FNMS(Tm, TD, Tl * TE); cannam@95: TG = Ip[WS(rs, 1)]; cannam@95: TH = Im[WS(rs, 1)]; cannam@95: TI = FMA(T3, TG, T6 * TH); cannam@95: TX = FNMS(T6, TG, T3 * TH); cannam@95: } cannam@95: TJ = TF + TI; cannam@95: T17 = TW + TX; cannam@95: TV = TF - TI; cannam@95: TY = TW - TX; cannam@95: } cannam@95: { cannam@95: E Tk, TM, Tr, TN; cannam@95: { cannam@95: E Th, Tj, To, Tq; cannam@95: Th = Rp[WS(rs, 1)]; cannam@95: Tj = Rm[WS(rs, 1)]; cannam@95: Tk = FMA(Tg, Th, Ti * Tj); cannam@95: TM = FNMS(Ti, Th, Tg * Tj); cannam@95: To = Rp[WS(rs, 3)]; cannam@95: Tq = Rm[WS(rs, 3)]; cannam@95: Tr = FMA(Tn, To, Tp * Tq); cannam@95: TN = FNMS(Tp, To, Tn * Tq); cannam@95: } cannam@95: Ts = Tk + Tr; cannam@95: T1j = Tk - Tr; cannam@95: TO = TM - TN; cannam@95: T1a = TM + TN; cannam@95: } cannam@95: { cannam@95: E Tw, TR, TB, TS; cannam@95: { cannam@95: E Tu, Tv, Ty, TA; cannam@95: Tu = Ip[0]; cannam@95: Tv = Im[0]; cannam@95: Tw = FMA(T2, Tu, T5 * Tv); cannam@95: TR = FNMS(T5, Tu, T2 * Tv); cannam@95: Ty = Ip[WS(rs, 2)]; cannam@95: TA = Im[WS(rs, 2)]; cannam@95: TB = FMA(Tx, Ty, Tz * TA); cannam@95: TS = FNMS(Tz, Ty, Tx * TA); cannam@95: } cannam@95: TC = Tw + TB; cannam@95: T16 = TR + TS; cannam@95: TQ = Tw - TB; cannam@95: TT = TR - TS; cannam@95: } cannam@95: { cannam@95: E Tt, TK, T1f, T1g; cannam@95: Tt = Tf + Ts; cannam@95: TK = TC + TJ; cannam@95: Rm[WS(rs, 3)] = Tt - TK; cannam@95: Rp[0] = Tt + TK; cannam@95: { cannam@95: E T19, T1e, T15, T18; cannam@95: T19 = T16 + T17; cannam@95: T1e = T1a + T1d; cannam@95: Im[WS(rs, 3)] = T19 - T1e; cannam@95: Ip[0] = T19 + T1e; cannam@95: T15 = Tf - Ts; cannam@95: T18 = T16 - T17; cannam@95: Rm[WS(rs, 1)] = T15 - T18; cannam@95: Rp[WS(rs, 2)] = T15 + T18; cannam@95: } cannam@95: T1f = TJ - TC; cannam@95: T1g = T1d - T1a; cannam@95: Im[WS(rs, 1)] = T1f - T1g; cannam@95: Ip[WS(rs, 2)] = T1f + T1g; cannam@95: { cannam@95: E T11, T1k, T14, T1h, T12, T13; cannam@95: T11 = TL - TO; cannam@95: T1k = T1i - T1j; cannam@95: T12 = TT - TQ; cannam@95: T13 = TV + TY; cannam@95: T14 = KP707106781 * (T12 - T13); cannam@95: T1h = KP707106781 * (T12 + T13); cannam@95: Rm[0] = T11 - T14; cannam@95: Ip[WS(rs, 1)] = T1h + T1k; cannam@95: Rp[WS(rs, 3)] = T11 + T14; cannam@95: Im[WS(rs, 2)] = T1h - T1k; cannam@95: } cannam@95: { cannam@95: E TP, T1m, T10, T1l, TU, TZ; cannam@95: TP = TL + TO; cannam@95: T1m = T1j + T1i; cannam@95: TU = TQ + TT; cannam@95: TZ = TV - TY; cannam@95: T10 = KP707106781 * (TU + TZ); cannam@95: T1l = KP707106781 * (TZ - TU); cannam@95: Rm[WS(rs, 2)] = TP - T10; cannam@95: Ip[WS(rs, 3)] = T1l + T1m; cannam@95: Rp[WS(rs, 1)] = TP + T10; cannam@95: Im[0] = T1l - T1m; cannam@95: } 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, "hc2cf2_8", twinstr, &GENUS, {56, 26, 18, 0} }; cannam@95: cannam@95: void X(codelet_hc2cf2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cf2_8, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */