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 07:41:44 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_r2cb.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 20 -name r2cbIII_20 -dft-III -include r2cbIII.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 94 FP additions, 56 FP multiplications, cannam@95: * (or, 58 additions, 20 multiplications, 36 fused multiply/add), cannam@95: * 59 stack variables, 6 constants, and 40 memory accesses cannam@95: */ cannam@95: #include "r2cbIII.h" cannam@95: cannam@95: static void r2cbIII_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(80, rs), MAKE_VOLATILE_STRIDE(80, csr), MAKE_VOLATILE_STRIDE(80, csi)) { cannam@95: E TZ, TD, TW, Tw, Tt, TF, T1f, T1b; cannam@95: { cannam@95: E T1l, Tk, T9, Tj, Ta, TV, TI, Ts, TU, T1t, T11, Tx, T13, TC, T1a; cannam@95: E T1i, Th, Tv, Ty; cannam@95: { cannam@95: E TQ, TS, Tr, Tm, Tn; cannam@95: { cannam@95: E T1, T5, T6, T2, T3, T7, TY; cannam@95: T1 = Cr[WS(csr, 2)]; cannam@95: T5 = Cr[WS(csr, 9)]; cannam@95: T6 = Cr[WS(csr, 5)]; cannam@95: T2 = Cr[WS(csr, 6)]; cannam@95: T3 = Cr[WS(csr, 1)]; cannam@95: TQ = Ci[WS(csi, 2)]; cannam@95: T7 = T5 + T6; cannam@95: TY = T5 - T6; cannam@95: { cannam@95: E T4, TX, T8, Tp, Tq; cannam@95: T4 = T2 + T3; cannam@95: TX = T2 - T3; cannam@95: Tp = Ci[WS(csi, 5)]; cannam@95: Tq = Ci[WS(csi, 9)]; cannam@95: T1l = FNMS(KP618033988, TX, TY); cannam@95: TZ = FMA(KP618033988, TY, TX); cannam@95: Tk = T4 - T7; cannam@95: T8 = T4 + T7; cannam@95: TS = Tp + Tq; cannam@95: Tr = Tp - Tq; cannam@95: T9 = T1 + T8; cannam@95: Tj = FNMS(KP250000000, T8, T1); cannam@95: Tm = Ci[WS(csi, 6)]; cannam@95: Tn = Ci[WS(csi, 1)]; cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tb, T19, Tg, Tc; cannam@95: Ta = Cr[WS(csr, 7)]; cannam@95: { cannam@95: E Te, Tf, To, TR, TT; cannam@95: Te = Cr[0]; cannam@95: Tf = Cr[WS(csr, 4)]; cannam@95: To = Tm + Tn; cannam@95: TR = Tm - Tn; cannam@95: Tb = Cr[WS(csr, 3)]; cannam@95: T19 = Te - Tf; cannam@95: Tg = Te + Tf; cannam@95: TT = TR - TS; cannam@95: TV = TR + TS; cannam@95: TI = FNMS(KP618033988, To, Tr); cannam@95: Ts = FMA(KP618033988, Tr, To); cannam@95: TU = FNMS(KP250000000, TT, TQ); cannam@95: T1t = TT + TQ; cannam@95: Tc = Cr[WS(csr, 8)]; cannam@95: } cannam@95: T11 = Ci[WS(csi, 7)]; cannam@95: { cannam@95: E TA, TB, Td, T18; cannam@95: TA = Ci[WS(csi, 4)]; cannam@95: TB = Ci[0]; cannam@95: Td = Tb + Tc; cannam@95: T18 = Tb - Tc; cannam@95: Tx = Ci[WS(csi, 3)]; cannam@95: T13 = TB + TA; cannam@95: TC = TA - TB; cannam@95: T1a = FMA(KP618033988, T19, T18); cannam@95: T1i = FNMS(KP618033988, T18, T19); cannam@95: Th = Td + Tg; cannam@95: Tv = Td - Tg; cannam@95: Ty = Ci[WS(csi, 8)]; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tu, T1w, T16, TL, T15, T1u; cannam@95: { cannam@95: E Ti, T12, Tz, T14; cannam@95: Tu = FNMS(KP250000000, Th, Ta); cannam@95: Ti = Ta + Th; cannam@95: T12 = Tx - Ty; cannam@95: Tz = Tx + Ty; cannam@95: T1w = T9 - Ti; cannam@95: T14 = T12 - T13; cannam@95: T16 = T12 + T13; cannam@95: TL = FNMS(KP618033988, Tz, TC); cannam@95: TD = FMA(KP618033988, TC, Tz); cannam@95: T15 = FNMS(KP250000000, T14, T11); cannam@95: T1u = T14 + T11; cannam@95: R0[0] = KP2_000000000 * (T9 + Ti); cannam@95: } cannam@95: { cannam@95: E Tl, TJ, TN, T1q, T1m, TK, T1h, T17, TH, T1k, T1v; cannam@95: Tl = FMA(KP559016994, Tk, Tj); cannam@95: TH = FNMS(KP559016994, Tk, Tj); cannam@95: T1k = FNMS(KP559016994, TV, TU); cannam@95: TW = FMA(KP559016994, TV, TU); cannam@95: R0[WS(rs, 5)] = KP2_000000000 * (T1u - T1t); cannam@95: T1v = T1t + T1u; cannam@95: TJ = FNMS(KP951056516, TI, TH); cannam@95: TN = FMA(KP951056516, TI, TH); cannam@95: T1q = FMA(KP951056516, T1l, T1k); cannam@95: T1m = FNMS(KP951056516, T1l, T1k); cannam@95: R1[WS(rs, 7)] = KP1_414213562 * (T1w + T1v); cannam@95: R1[WS(rs, 2)] = KP1_414213562 * (T1v - T1w); cannam@95: Tw = FMA(KP559016994, Tv, Tu); cannam@95: TK = FNMS(KP559016994, Tv, Tu); cannam@95: T1h = FNMS(KP559016994, T16, T15); cannam@95: T17 = FMA(KP559016994, T16, T15); cannam@95: { cannam@95: E TM, TO, T1j, T1r; cannam@95: TM = FMA(KP951056516, TL, TK); cannam@95: TO = FNMS(KP951056516, TL, TK); cannam@95: T1j = FMA(KP951056516, T1i, T1h); cannam@95: T1r = FNMS(KP951056516, T1i, T1h); cannam@95: Tt = FNMS(KP951056516, Ts, Tl); cannam@95: TF = FMA(KP951056516, Ts, Tl); cannam@95: { cannam@95: E T1n, T1p, T1s, T1o; cannam@95: T1n = TN - TO; cannam@95: R0[WS(rs, 6)] = -(KP2_000000000 * (TN + TO)); cannam@95: T1p = TM - TJ; cannam@95: R0[WS(rs, 4)] = KP2_000000000 * (TJ + TM); cannam@95: T1s = T1q + T1r; cannam@95: R0[WS(rs, 9)] = KP2_000000000 * (T1r - T1q); cannam@95: T1o = T1m + T1j; cannam@95: R0[WS(rs, 1)] = KP2_000000000 * (T1j - T1m); cannam@95: R1[WS(rs, 6)] = KP1_414213562 * (T1p + T1s); cannam@95: R1[WS(rs, 1)] = KP1_414213562 * (T1p - T1s); cannam@95: R1[WS(rs, 3)] = KP1_414213562 * (T1n + T1o); cannam@95: R1[WS(rs, 8)] = KP1_414213562 * (T1n - T1o); cannam@95: T1f = FMA(KP951056516, T1a, T17); cannam@95: T1b = FNMS(KP951056516, T1a, T17); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TE, TG, T10, T1e; cannam@95: TE = FMA(KP951056516, TD, Tw); cannam@95: TG = FNMS(KP951056516, TD, Tw); cannam@95: T10 = FMA(KP951056516, TZ, TW); cannam@95: T1e = FNMS(KP951056516, TZ, TW); cannam@95: { cannam@95: E T1d, TP, T1g, T1c; cannam@95: T1d = TF - TG; cannam@95: R0[WS(rs, 2)] = -(KP2_000000000 * (TF + TG)); cannam@95: TP = Tt - TE; cannam@95: R0[WS(rs, 8)] = KP2_000000000 * (Tt + TE); cannam@95: T1g = T1e + T1f; cannam@95: R0[WS(rs, 7)] = KP2_000000000 * (T1e - T1f); cannam@95: T1c = T10 + T1b; cannam@95: R0[WS(rs, 3)] = KP2_000000000 * (T10 - T1b); cannam@95: R1[WS(rs, 9)] = -(KP1_414213562 * (T1d + T1g)); cannam@95: R1[WS(rs, 4)] = KP1_414213562 * (T1d - T1g); cannam@95: R1[WS(rs, 5)] = -(KP1_414213562 * (TP + T1c)); cannam@95: R1[0] = KP1_414213562 * (TP - T1c); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 20, "r2cbIII_20", {58, 20, 36, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cbIII_20) (planner *p) { cannam@95: X(kr2c_register) (p, r2cbIII_20, &desc); cannam@95: } cannam@95: cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 20 -name r2cbIII_20 -dft-III -include r2cbIII.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 94 FP additions, 44 FP multiplications, cannam@95: * (or, 82 additions, 32 multiplications, 12 fused multiply/add), cannam@95: * 43 stack variables, 6 constants, and 40 memory accesses cannam@95: */ cannam@95: #include "r2cbIII.h" cannam@95: cannam@95: static void r2cbIII_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); cannam@95: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(80, rs), MAKE_VOLATILE_STRIDE(80, csr), MAKE_VOLATILE_STRIDE(80, csi)) { cannam@95: E T1, Tj, T1k, T13, T8, Tk, T17, Ts, T16, TI, T18, T19, Ta, Tu, T1i; cannam@95: E TS, Th, Tv, TX, TD, TV, TL, TW, TY; cannam@95: { cannam@95: E T7, T12, T4, T11; cannam@95: T1 = Cr[WS(csr, 2)]; cannam@95: { cannam@95: E T5, T6, T2, T3; cannam@95: T5 = Cr[WS(csr, 9)]; cannam@95: T6 = Cr[WS(csr, 5)]; cannam@95: T7 = T5 + T6; cannam@95: T12 = T5 - T6; cannam@95: T2 = Cr[WS(csr, 6)]; cannam@95: T3 = Cr[WS(csr, 1)]; cannam@95: T4 = T2 + T3; cannam@95: T11 = T2 - T3; cannam@95: } cannam@95: Tj = KP559016994 * (T4 - T7); cannam@95: T1k = FNMS(KP951056516, T12, KP587785252 * T11); cannam@95: T13 = FMA(KP951056516, T11, KP587785252 * T12); cannam@95: T8 = T4 + T7; cannam@95: Tk = FNMS(KP250000000, T8, T1); cannam@95: } cannam@95: { cannam@95: E Tr, T15, To, T14; cannam@95: T17 = Ci[WS(csi, 2)]; cannam@95: { cannam@95: E Tp, Tq, Tm, Tn; cannam@95: Tp = Ci[WS(csi, 5)]; cannam@95: Tq = Ci[WS(csi, 9)]; cannam@95: Tr = Tp - Tq; cannam@95: T15 = Tp + Tq; cannam@95: Tm = Ci[WS(csi, 6)]; cannam@95: Tn = Ci[WS(csi, 1)]; cannam@95: To = Tm + Tn; cannam@95: T14 = Tm - Tn; cannam@95: } cannam@95: Ts = FMA(KP951056516, To, KP587785252 * Tr); cannam@95: T16 = KP559016994 * (T14 + T15); cannam@95: TI = FNMS(KP951056516, Tr, KP587785252 * To); cannam@95: T18 = T14 - T15; cannam@95: T19 = FNMS(KP250000000, T18, T17); cannam@95: } cannam@95: { cannam@95: E Tg, TR, Td, TQ; cannam@95: Ta = Cr[WS(csr, 7)]; cannam@95: { cannam@95: E Te, Tf, Tb, Tc; cannam@95: Te = Cr[0]; cannam@95: Tf = Cr[WS(csr, 4)]; cannam@95: Tg = Te + Tf; cannam@95: TR = Te - Tf; cannam@95: Tb = Cr[WS(csr, 3)]; cannam@95: Tc = Cr[WS(csr, 8)]; cannam@95: Td = Tb + Tc; cannam@95: TQ = Tb - Tc; cannam@95: } cannam@95: Tu = KP559016994 * (Td - Tg); cannam@95: T1i = FNMS(KP951056516, TR, KP587785252 * TQ); cannam@95: TS = FMA(KP951056516, TQ, KP587785252 * TR); cannam@95: Th = Td + Tg; cannam@95: Tv = FNMS(KP250000000, Th, Ta); cannam@95: } cannam@95: { cannam@95: E TC, TU, Tz, TT; cannam@95: TX = Ci[WS(csi, 7)]; cannam@95: { cannam@95: E TA, TB, Tx, Ty; cannam@95: TA = Ci[WS(csi, 4)]; cannam@95: TB = Ci[0]; cannam@95: TC = TA - TB; cannam@95: TU = TB + TA; cannam@95: Tx = Ci[WS(csi, 3)]; cannam@95: Ty = Ci[WS(csi, 8)]; cannam@95: Tz = Tx + Ty; cannam@95: TT = Ty - Tx; cannam@95: } cannam@95: TD = FMA(KP951056516, Tz, KP587785252 * TC); cannam@95: TV = KP559016994 * (TT - TU); cannam@95: TL = FNMS(KP587785252, Tz, KP951056516 * TC); cannam@95: TW = TT + TU; cannam@95: TY = FMA(KP250000000, TW, TX); cannam@95: } cannam@95: { cannam@95: E T9, Ti, T1w, T1t, T1u, T1v; cannam@95: T9 = T1 + T8; cannam@95: Ti = Ta + Th; cannam@95: T1w = T9 - Ti; cannam@95: T1t = T18 + T17; cannam@95: T1u = TX - TW; cannam@95: T1v = T1t + T1u; cannam@95: R0[0] = KP2_000000000 * (T9 + Ti); cannam@95: R0[WS(rs, 5)] = KP2_000000000 * (T1u - T1t); cannam@95: R1[WS(rs, 2)] = KP1_414213562 * (T1v - T1w); cannam@95: R1[WS(rs, 7)] = KP1_414213562 * (T1w + T1v); cannam@95: } cannam@95: { cannam@95: E TJ, TO, T1m, T1q, TM, TN, T1j, T1r; cannam@95: { cannam@95: E TH, T1l, TK, T1h; cannam@95: TH = Tk - Tj; cannam@95: TJ = TH + TI; cannam@95: TO = TH - TI; cannam@95: T1l = T19 - T16; cannam@95: T1m = T1k + T1l; cannam@95: T1q = T1l - T1k; cannam@95: TK = Tv - Tu; cannam@95: TM = TK + TL; cannam@95: TN = TL - TK; cannam@95: T1h = TV + TY; cannam@95: T1j = T1h - T1i; cannam@95: T1r = T1i + T1h; cannam@95: } cannam@95: R0[WS(rs, 4)] = KP2_000000000 * (TJ + TM); cannam@95: R0[WS(rs, 6)] = KP2_000000000 * (TN - TO); cannam@95: R0[WS(rs, 9)] = KP2_000000000 * (T1r - T1q); cannam@95: R0[WS(rs, 1)] = KP2_000000000 * (T1j - T1m); cannam@95: { cannam@95: E T1p, T1s, T1n, T1o; cannam@95: T1p = TM - TJ; cannam@95: T1s = T1q + T1r; cannam@95: R1[WS(rs, 1)] = KP1_414213562 * (T1p - T1s); cannam@95: R1[WS(rs, 6)] = KP1_414213562 * (T1p + T1s); cannam@95: T1n = TO + TN; cannam@95: T1o = T1m + T1j; cannam@95: R1[WS(rs, 8)] = KP1_414213562 * (T1n - T1o); cannam@95: R1[WS(rs, 3)] = KP1_414213562 * (T1n + T1o); cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tt, TG, T1b, T1f, TE, TF, T10, T1e; cannam@95: { cannam@95: E Tl, T1a, Tw, TZ; cannam@95: Tl = Tj + Tk; cannam@95: Tt = Tl - Ts; cannam@95: TG = Tl + Ts; cannam@95: T1a = T16 + T19; cannam@95: T1b = T13 + T1a; cannam@95: T1f = T1a - T13; cannam@95: Tw = Tu + Tv; cannam@95: TE = Tw + TD; cannam@95: TF = TD - Tw; cannam@95: TZ = TV - TY; cannam@95: T10 = TS + TZ; cannam@95: T1e = TZ - TS; cannam@95: } cannam@95: R0[WS(rs, 8)] = KP2_000000000 * (Tt + TE); cannam@95: R0[WS(rs, 2)] = KP2_000000000 * (TF - TG); cannam@95: R0[WS(rs, 7)] = KP2_000000000 * (T1f + T1e); cannam@95: R0[WS(rs, 3)] = KP2_000000000 * (T1b + T10); cannam@95: { cannam@95: E T1d, T1g, TP, T1c; cannam@95: T1d = TG + TF; cannam@95: T1g = T1e - T1f; cannam@95: R1[WS(rs, 4)] = KP1_414213562 * (T1d + T1g); cannam@95: R1[WS(rs, 9)] = KP1_414213562 * (T1g - T1d); cannam@95: TP = Tt - TE; cannam@95: T1c = T10 - T1b; cannam@95: R1[0] = KP1_414213562 * (TP + T1c); cannam@95: R1[WS(rs, 5)] = KP1_414213562 * (T1c - TP); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 20, "r2cbIII_20", {82, 32, 12, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cbIII_20) (planner *p) { cannam@95: X(kr2c_register) (p, r2cbIII_20, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */