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 07:41:44 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_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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 94 FP additions, 56 FP multiplications, Chris@10: * (or, 58 additions, 20 multiplications, 36 fused multiply/add), Chris@10: * 59 stack variables, 6 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "r2cbIII.h" Chris@10: Chris@10: static void r2cbIII_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: { Chris@10: INT i; Chris@10: 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)) { Chris@10: E TZ, TD, TW, Tw, Tt, TF, T1f, T1b; Chris@10: { Chris@10: E T1l, Tk, T9, Tj, Ta, TV, TI, Ts, TU, T1t, T11, Tx, T13, TC, T1a; Chris@10: E T1i, Th, Tv, Ty; Chris@10: { Chris@10: E TQ, TS, Tr, Tm, Tn; Chris@10: { Chris@10: E T1, T5, T6, T2, T3, T7, TY; Chris@10: T1 = Cr[WS(csr, 2)]; Chris@10: T5 = Cr[WS(csr, 9)]; Chris@10: T6 = Cr[WS(csr, 5)]; Chris@10: T2 = Cr[WS(csr, 6)]; Chris@10: T3 = Cr[WS(csr, 1)]; Chris@10: TQ = Ci[WS(csi, 2)]; Chris@10: T7 = T5 + T6; Chris@10: TY = T5 - T6; Chris@10: { Chris@10: E T4, TX, T8, Tp, Tq; Chris@10: T4 = T2 + T3; Chris@10: TX = T2 - T3; Chris@10: Tp = Ci[WS(csi, 5)]; Chris@10: Tq = Ci[WS(csi, 9)]; Chris@10: T1l = FNMS(KP618033988, TX, TY); Chris@10: TZ = FMA(KP618033988, TY, TX); Chris@10: Tk = T4 - T7; Chris@10: T8 = T4 + T7; Chris@10: TS = Tp + Tq; Chris@10: Tr = Tp - Tq; Chris@10: T9 = T1 + T8; Chris@10: Tj = FNMS(KP250000000, T8, T1); Chris@10: Tm = Ci[WS(csi, 6)]; Chris@10: Tn = Ci[WS(csi, 1)]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tb, T19, Tg, Tc; Chris@10: Ta = Cr[WS(csr, 7)]; Chris@10: { Chris@10: E Te, Tf, To, TR, TT; Chris@10: Te = Cr[0]; Chris@10: Tf = Cr[WS(csr, 4)]; Chris@10: To = Tm + Tn; Chris@10: TR = Tm - Tn; Chris@10: Tb = Cr[WS(csr, 3)]; Chris@10: T19 = Te - Tf; Chris@10: Tg = Te + Tf; Chris@10: TT = TR - TS; Chris@10: TV = TR + TS; Chris@10: TI = FNMS(KP618033988, To, Tr); Chris@10: Ts = FMA(KP618033988, Tr, To); Chris@10: TU = FNMS(KP250000000, TT, TQ); Chris@10: T1t = TT + TQ; Chris@10: Tc = Cr[WS(csr, 8)]; Chris@10: } Chris@10: T11 = Ci[WS(csi, 7)]; Chris@10: { Chris@10: E TA, TB, Td, T18; Chris@10: TA = Ci[WS(csi, 4)]; Chris@10: TB = Ci[0]; Chris@10: Td = Tb + Tc; Chris@10: T18 = Tb - Tc; Chris@10: Tx = Ci[WS(csi, 3)]; Chris@10: T13 = TB + TA; Chris@10: TC = TA - TB; Chris@10: T1a = FMA(KP618033988, T19, T18); Chris@10: T1i = FNMS(KP618033988, T18, T19); Chris@10: Th = Td + Tg; Chris@10: Tv = Td - Tg; Chris@10: Ty = Ci[WS(csi, 8)]; Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tu, T1w, T16, TL, T15, T1u; Chris@10: { Chris@10: E Ti, T12, Tz, T14; Chris@10: Tu = FNMS(KP250000000, Th, Ta); Chris@10: Ti = Ta + Th; Chris@10: T12 = Tx - Ty; Chris@10: Tz = Tx + Ty; Chris@10: T1w = T9 - Ti; Chris@10: T14 = T12 - T13; Chris@10: T16 = T12 + T13; Chris@10: TL = FNMS(KP618033988, Tz, TC); Chris@10: TD = FMA(KP618033988, TC, Tz); Chris@10: T15 = FNMS(KP250000000, T14, T11); Chris@10: T1u = T14 + T11; Chris@10: R0[0] = KP2_000000000 * (T9 + Ti); Chris@10: } Chris@10: { Chris@10: E Tl, TJ, TN, T1q, T1m, TK, T1h, T17, TH, T1k, T1v; Chris@10: Tl = FMA(KP559016994, Tk, Tj); Chris@10: TH = FNMS(KP559016994, Tk, Tj); Chris@10: T1k = FNMS(KP559016994, TV, TU); Chris@10: TW = FMA(KP559016994, TV, TU); Chris@10: R0[WS(rs, 5)] = KP2_000000000 * (T1u - T1t); Chris@10: T1v = T1t + T1u; Chris@10: TJ = FNMS(KP951056516, TI, TH); Chris@10: TN = FMA(KP951056516, TI, TH); Chris@10: T1q = FMA(KP951056516, T1l, T1k); Chris@10: T1m = FNMS(KP951056516, T1l, T1k); Chris@10: R1[WS(rs, 7)] = KP1_414213562 * (T1w + T1v); Chris@10: R1[WS(rs, 2)] = KP1_414213562 * (T1v - T1w); Chris@10: Tw = FMA(KP559016994, Tv, Tu); Chris@10: TK = FNMS(KP559016994, Tv, Tu); Chris@10: T1h = FNMS(KP559016994, T16, T15); Chris@10: T17 = FMA(KP559016994, T16, T15); Chris@10: { Chris@10: E TM, TO, T1j, T1r; Chris@10: TM = FMA(KP951056516, TL, TK); Chris@10: TO = FNMS(KP951056516, TL, TK); Chris@10: T1j = FMA(KP951056516, T1i, T1h); Chris@10: T1r = FNMS(KP951056516, T1i, T1h); Chris@10: Tt = FNMS(KP951056516, Ts, Tl); Chris@10: TF = FMA(KP951056516, Ts, Tl); Chris@10: { Chris@10: E T1n, T1p, T1s, T1o; Chris@10: T1n = TN - TO; Chris@10: R0[WS(rs, 6)] = -(KP2_000000000 * (TN + TO)); Chris@10: T1p = TM - TJ; Chris@10: R0[WS(rs, 4)] = KP2_000000000 * (TJ + TM); Chris@10: T1s = T1q + T1r; Chris@10: R0[WS(rs, 9)] = KP2_000000000 * (T1r - T1q); Chris@10: T1o = T1m + T1j; Chris@10: R0[WS(rs, 1)] = KP2_000000000 * (T1j - T1m); Chris@10: R1[WS(rs, 6)] = KP1_414213562 * (T1p + T1s); Chris@10: R1[WS(rs, 1)] = KP1_414213562 * (T1p - T1s); Chris@10: R1[WS(rs, 3)] = KP1_414213562 * (T1n + T1o); Chris@10: R1[WS(rs, 8)] = KP1_414213562 * (T1n - T1o); Chris@10: T1f = FMA(KP951056516, T1a, T17); Chris@10: T1b = FNMS(KP951056516, T1a, T17); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: E TE, TG, T10, T1e; Chris@10: TE = FMA(KP951056516, TD, Tw); Chris@10: TG = FNMS(KP951056516, TD, Tw); Chris@10: T10 = FMA(KP951056516, TZ, TW); Chris@10: T1e = FNMS(KP951056516, TZ, TW); Chris@10: { Chris@10: E T1d, TP, T1g, T1c; Chris@10: T1d = TF - TG; Chris@10: R0[WS(rs, 2)] = -(KP2_000000000 * (TF + TG)); Chris@10: TP = Tt - TE; Chris@10: R0[WS(rs, 8)] = KP2_000000000 * (Tt + TE); Chris@10: T1g = T1e + T1f; Chris@10: R0[WS(rs, 7)] = KP2_000000000 * (T1e - T1f); Chris@10: T1c = T10 + T1b; Chris@10: R0[WS(rs, 3)] = KP2_000000000 * (T10 - T1b); Chris@10: R1[WS(rs, 9)] = -(KP1_414213562 * (T1d + T1g)); Chris@10: R1[WS(rs, 4)] = KP1_414213562 * (T1d - T1g); Chris@10: R1[WS(rs, 5)] = -(KP1_414213562 * (TP + T1c)); Chris@10: R1[0] = KP1_414213562 * (TP - T1c); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 20, "r2cbIII_20", {58, 20, 36, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cbIII_20) (planner *p) { Chris@10: X(kr2c_register) (p, r2cbIII_20, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 20 -name r2cbIII_20 -dft-III -include r2cbIII.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 94 FP additions, 44 FP multiplications, Chris@10: * (or, 82 additions, 32 multiplications, 12 fused multiply/add), Chris@10: * 43 stack variables, 6 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "r2cbIII.h" Chris@10: Chris@10: static void r2cbIII_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); Chris@10: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); Chris@10: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: { Chris@10: INT i; Chris@10: 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)) { Chris@10: E T1, Tj, T1k, T13, T8, Tk, T17, Ts, T16, TI, T18, T19, Ta, Tu, T1i; Chris@10: E TS, Th, Tv, TX, TD, TV, TL, TW, TY; Chris@10: { Chris@10: E T7, T12, T4, T11; Chris@10: T1 = Cr[WS(csr, 2)]; Chris@10: { Chris@10: E T5, T6, T2, T3; Chris@10: T5 = Cr[WS(csr, 9)]; Chris@10: T6 = Cr[WS(csr, 5)]; Chris@10: T7 = T5 + T6; Chris@10: T12 = T5 - T6; Chris@10: T2 = Cr[WS(csr, 6)]; Chris@10: T3 = Cr[WS(csr, 1)]; Chris@10: T4 = T2 + T3; Chris@10: T11 = T2 - T3; Chris@10: } Chris@10: Tj = KP559016994 * (T4 - T7); Chris@10: T1k = FNMS(KP951056516, T12, KP587785252 * T11); Chris@10: T13 = FMA(KP951056516, T11, KP587785252 * T12); Chris@10: T8 = T4 + T7; Chris@10: Tk = FNMS(KP250000000, T8, T1); Chris@10: } Chris@10: { Chris@10: E Tr, T15, To, T14; Chris@10: T17 = Ci[WS(csi, 2)]; Chris@10: { Chris@10: E Tp, Tq, Tm, Tn; Chris@10: Tp = Ci[WS(csi, 5)]; Chris@10: Tq = Ci[WS(csi, 9)]; Chris@10: Tr = Tp - Tq; Chris@10: T15 = Tp + Tq; Chris@10: Tm = Ci[WS(csi, 6)]; Chris@10: Tn = Ci[WS(csi, 1)]; Chris@10: To = Tm + Tn; Chris@10: T14 = Tm - Tn; Chris@10: } Chris@10: Ts = FMA(KP951056516, To, KP587785252 * Tr); Chris@10: T16 = KP559016994 * (T14 + T15); Chris@10: TI = FNMS(KP951056516, Tr, KP587785252 * To); Chris@10: T18 = T14 - T15; Chris@10: T19 = FNMS(KP250000000, T18, T17); Chris@10: } Chris@10: { Chris@10: E Tg, TR, Td, TQ; Chris@10: Ta = Cr[WS(csr, 7)]; Chris@10: { Chris@10: E Te, Tf, Tb, Tc; Chris@10: Te = Cr[0]; Chris@10: Tf = Cr[WS(csr, 4)]; Chris@10: Tg = Te + Tf; Chris@10: TR = Te - Tf; Chris@10: Tb = Cr[WS(csr, 3)]; Chris@10: Tc = Cr[WS(csr, 8)]; Chris@10: Td = Tb + Tc; Chris@10: TQ = Tb - Tc; Chris@10: } Chris@10: Tu = KP559016994 * (Td - Tg); Chris@10: T1i = FNMS(KP951056516, TR, KP587785252 * TQ); Chris@10: TS = FMA(KP951056516, TQ, KP587785252 * TR); Chris@10: Th = Td + Tg; Chris@10: Tv = FNMS(KP250000000, Th, Ta); Chris@10: } Chris@10: { Chris@10: E TC, TU, Tz, TT; Chris@10: TX = Ci[WS(csi, 7)]; Chris@10: { Chris@10: E TA, TB, Tx, Ty; Chris@10: TA = Ci[WS(csi, 4)]; Chris@10: TB = Ci[0]; Chris@10: TC = TA - TB; Chris@10: TU = TB + TA; Chris@10: Tx = Ci[WS(csi, 3)]; Chris@10: Ty = Ci[WS(csi, 8)]; Chris@10: Tz = Tx + Ty; Chris@10: TT = Ty - Tx; Chris@10: } Chris@10: TD = FMA(KP951056516, Tz, KP587785252 * TC); Chris@10: TV = KP559016994 * (TT - TU); Chris@10: TL = FNMS(KP587785252, Tz, KP951056516 * TC); Chris@10: TW = TT + TU; Chris@10: TY = FMA(KP250000000, TW, TX); Chris@10: } Chris@10: { Chris@10: E T9, Ti, T1w, T1t, T1u, T1v; Chris@10: T9 = T1 + T8; Chris@10: Ti = Ta + Th; Chris@10: T1w = T9 - Ti; Chris@10: T1t = T18 + T17; Chris@10: T1u = TX - TW; Chris@10: T1v = T1t + T1u; Chris@10: R0[0] = KP2_000000000 * (T9 + Ti); Chris@10: R0[WS(rs, 5)] = KP2_000000000 * (T1u - T1t); Chris@10: R1[WS(rs, 2)] = KP1_414213562 * (T1v - T1w); Chris@10: R1[WS(rs, 7)] = KP1_414213562 * (T1w + T1v); Chris@10: } Chris@10: { Chris@10: E TJ, TO, T1m, T1q, TM, TN, T1j, T1r; Chris@10: { Chris@10: E TH, T1l, TK, T1h; Chris@10: TH = Tk - Tj; Chris@10: TJ = TH + TI; Chris@10: TO = TH - TI; Chris@10: T1l = T19 - T16; Chris@10: T1m = T1k + T1l; Chris@10: T1q = T1l - T1k; Chris@10: TK = Tv - Tu; Chris@10: TM = TK + TL; Chris@10: TN = TL - TK; Chris@10: T1h = TV + TY; Chris@10: T1j = T1h - T1i; Chris@10: T1r = T1i + T1h; Chris@10: } Chris@10: R0[WS(rs, 4)] = KP2_000000000 * (TJ + TM); Chris@10: R0[WS(rs, 6)] = KP2_000000000 * (TN - TO); Chris@10: R0[WS(rs, 9)] = KP2_000000000 * (T1r - T1q); Chris@10: R0[WS(rs, 1)] = KP2_000000000 * (T1j - T1m); Chris@10: { Chris@10: E T1p, T1s, T1n, T1o; Chris@10: T1p = TM - TJ; Chris@10: T1s = T1q + T1r; Chris@10: R1[WS(rs, 1)] = KP1_414213562 * (T1p - T1s); Chris@10: R1[WS(rs, 6)] = KP1_414213562 * (T1p + T1s); Chris@10: T1n = TO + TN; Chris@10: T1o = T1m + T1j; Chris@10: R1[WS(rs, 8)] = KP1_414213562 * (T1n - T1o); Chris@10: R1[WS(rs, 3)] = KP1_414213562 * (T1n + T1o); Chris@10: } Chris@10: } Chris@10: { Chris@10: E Tt, TG, T1b, T1f, TE, TF, T10, T1e; Chris@10: { Chris@10: E Tl, T1a, Tw, TZ; Chris@10: Tl = Tj + Tk; Chris@10: Tt = Tl - Ts; Chris@10: TG = Tl + Ts; Chris@10: T1a = T16 + T19; Chris@10: T1b = T13 + T1a; Chris@10: T1f = T1a - T13; Chris@10: Tw = Tu + Tv; Chris@10: TE = Tw + TD; Chris@10: TF = TD - Tw; Chris@10: TZ = TV - TY; Chris@10: T10 = TS + TZ; Chris@10: T1e = TZ - TS; Chris@10: } Chris@10: R0[WS(rs, 8)] = KP2_000000000 * (Tt + TE); Chris@10: R0[WS(rs, 2)] = KP2_000000000 * (TF - TG); Chris@10: R0[WS(rs, 7)] = KP2_000000000 * (T1f + T1e); Chris@10: R0[WS(rs, 3)] = KP2_000000000 * (T1b + T10); Chris@10: { Chris@10: E T1d, T1g, TP, T1c; Chris@10: T1d = TG + TF; Chris@10: T1g = T1e - T1f; Chris@10: R1[WS(rs, 4)] = KP1_414213562 * (T1d + T1g); Chris@10: R1[WS(rs, 9)] = KP1_414213562 * (T1g - T1d); Chris@10: TP = Tt - TE; Chris@10: T1c = T10 - T1b; Chris@10: R1[0] = KP1_414213562 * (TP + T1c); Chris@10: R1[WS(rs, 5)] = KP1_414213562 * (T1c - TP); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 20, "r2cbIII_20", {82, 32, 12, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cbIII_20) (planner *p) { Chris@10: X(kr2c_register) (p, r2cbIII_20, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */