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 07:35:43 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_notw.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 12 -name n1_12 -include n.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 96 FP additions, 24 FP multiplications, cannam@95: * (or, 72 additions, 0 multiplications, 24 fused multiply/add), cannam@95: * 63 stack variables, 2 constants, and 48 memory accesses cannam@95: */ cannam@95: #include "n.h" cannam@95: cannam@95: static void n1_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(48, is), MAKE_VOLATILE_STRIDE(48, os)) { cannam@95: E TT, TW, TF, T1q, TY, TQ, TX, T1n; cannam@95: { cannam@95: E TA, TS, TR, T5, Ts, Tz, TD, TV, TU, Ta, Tx, TC, T1d, Th, TJ; cannam@95: E TG, Tg, T1u, T1c, T1f, TM, TN, Tk, T1i; cannam@95: { cannam@95: E T6, Tt, Tu, Tv, T9; cannam@95: { cannam@95: E T1, To, Tp, Tq, T4, T2, T3, T7, T8, Tr; cannam@95: T1 = ri[0]; cannam@95: T2 = ri[WS(is, 4)]; cannam@95: T3 = ri[WS(is, 8)]; cannam@95: To = ii[0]; cannam@95: Tp = ii[WS(is, 4)]; cannam@95: Tq = ii[WS(is, 8)]; cannam@95: T4 = T2 + T3; cannam@95: TA = T3 - T2; cannam@95: T6 = ri[WS(is, 6)]; cannam@95: TS = Tp - Tq; cannam@95: Tr = Tp + Tq; cannam@95: TR = FNMS(KP500000000, T4, T1); cannam@95: T5 = T1 + T4; cannam@95: T7 = ri[WS(is, 10)]; cannam@95: Ts = To + Tr; cannam@95: Tz = FNMS(KP500000000, Tr, To); cannam@95: T8 = ri[WS(is, 2)]; cannam@95: Tt = ii[WS(is, 6)]; cannam@95: Tu = ii[WS(is, 10)]; cannam@95: Tv = ii[WS(is, 2)]; cannam@95: T9 = T7 + T8; cannam@95: TD = T8 - T7; cannam@95: } cannam@95: { cannam@95: E Tc, T1a, TH, TI, Tf, Td, Te, Tw, Ti, Tj, T1b; cannam@95: Tc = ri[WS(is, 3)]; cannam@95: TV = Tu - Tv; cannam@95: Tw = Tu + Tv; cannam@95: TU = FNMS(KP500000000, T9, T6); cannam@95: Ta = T6 + T9; cannam@95: Td = ri[WS(is, 7)]; cannam@95: Tx = Tt + Tw; cannam@95: TC = FNMS(KP500000000, Tw, Tt); cannam@95: Te = ri[WS(is, 11)]; cannam@95: T1a = ii[WS(is, 3)]; cannam@95: TH = ii[WS(is, 7)]; cannam@95: TI = ii[WS(is, 11)]; cannam@95: Tf = Td + Te; cannam@95: T1d = Te - Td; cannam@95: Th = ri[WS(is, 9)]; cannam@95: TJ = TH - TI; cannam@95: T1b = TH + TI; cannam@95: TG = FNMS(KP500000000, Tf, Tc); cannam@95: Tg = Tc + Tf; cannam@95: Ti = ri[WS(is, 1)]; cannam@95: T1u = T1a + T1b; cannam@95: T1c = FNMS(KP500000000, T1b, T1a); cannam@95: Tj = ri[WS(is, 5)]; cannam@95: T1f = ii[WS(is, 9)]; cannam@95: TM = ii[WS(is, 1)]; cannam@95: TN = ii[WS(is, 5)]; cannam@95: Tk = Ti + Tj; cannam@95: T1i = Tj - Ti; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1t, TO, TL, T1h, T1w, Tb, T1g, Tl; cannam@95: T1t = T5 - Ta; cannam@95: Tb = T5 + Ta; cannam@95: TO = TM - TN; cannam@95: T1g = TM + TN; cannam@95: TL = FNMS(KP500000000, Tk, Th); cannam@95: Tl = Th + Tk; cannam@95: { cannam@95: E T1x, Ty, T1v, Tn, Tm, T1y; cannam@95: T1x = Ts + Tx; cannam@95: Ty = Ts - Tx; cannam@95: T1v = T1f + T1g; cannam@95: T1h = FNMS(KP500000000, T1g, T1f); cannam@95: Tn = Tg - Tl; cannam@95: Tm = Tg + Tl; cannam@95: T1y = T1u + T1v; cannam@95: T1w = T1u - T1v; cannam@95: ro[0] = Tb + Tm; cannam@95: ro[WS(os, 6)] = Tb - Tm; cannam@95: io[WS(os, 3)] = Tn + Ty; cannam@95: io[0] = T1x + T1y; cannam@95: io[WS(os, 6)] = T1x - T1y; cannam@95: io[WS(os, 9)] = Ty - Tn; cannam@95: } cannam@95: { cannam@95: E TB, TE, T1o, T11, T1p, TK, TP, T15, T1k, T18, T14, T16, T1l, T1m; cannam@95: { cannam@95: E T1e, T1j, TZ, T10, T12, T13; cannam@95: TB = FNMS(KP866025403, TA, Tz); cannam@95: TZ = FMA(KP866025403, TA, Tz); cannam@95: T10 = FMA(KP866025403, TD, TC); cannam@95: TE = FNMS(KP866025403, TD, TC); cannam@95: T1o = FNMS(KP866025403, T1d, T1c); cannam@95: T1e = FMA(KP866025403, T1d, T1c); cannam@95: ro[WS(os, 9)] = T1t + T1w; cannam@95: ro[WS(os, 3)] = T1t - T1w; cannam@95: T1l = TZ + T10; cannam@95: T11 = TZ - T10; cannam@95: T1j = FMA(KP866025403, T1i, T1h); cannam@95: T1p = FNMS(KP866025403, T1i, T1h); cannam@95: TK = FNMS(KP866025403, TJ, TG); cannam@95: T12 = FMA(KP866025403, TJ, TG); cannam@95: T13 = FMA(KP866025403, TO, TL); cannam@95: TP = FNMS(KP866025403, TO, TL); cannam@95: TT = FNMS(KP866025403, TS, TR); cannam@95: T15 = FMA(KP866025403, TS, TR); cannam@95: T1m = T1e + T1j; cannam@95: T1k = T1e - T1j; cannam@95: T18 = T12 + T13; cannam@95: T14 = T12 - T13; cannam@95: T16 = FMA(KP866025403, TV, TU); cannam@95: TW = FNMS(KP866025403, TV, TU); cannam@95: } cannam@95: io[WS(os, 10)] = T1l - T1m; cannam@95: io[WS(os, 4)] = T1l + T1m; cannam@95: io[WS(os, 7)] = T11 + T14; cannam@95: io[WS(os, 1)] = T11 - T14; cannam@95: { cannam@95: E T17, T19, T1r, T1s; cannam@95: T17 = T15 + T16; cannam@95: T19 = T15 - T16; cannam@95: ro[WS(os, 7)] = T19 - T1k; cannam@95: ro[WS(os, 1)] = T19 + T1k; cannam@95: ro[WS(os, 4)] = T17 + T18; cannam@95: ro[WS(os, 10)] = T17 - T18; cannam@95: T1r = TB + TE; cannam@95: TF = TB - TE; cannam@95: T1s = T1o + T1p; cannam@95: T1q = T1o - T1p; cannam@95: TY = TK + TP; cannam@95: TQ = TK - TP; cannam@95: io[WS(os, 2)] = T1r - T1s; cannam@95: io[WS(os, 8)] = T1r + T1s; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: io[WS(os, 11)] = TF + TQ; cannam@95: io[WS(os, 5)] = TF - TQ; cannam@95: TX = TT + TW; cannam@95: T1n = TT - TW; cannam@95: ro[WS(os, 11)] = T1n - T1q; cannam@95: ro[WS(os, 5)] = T1n + T1q; cannam@95: ro[WS(os, 8)] = TX + TY; cannam@95: ro[WS(os, 2)] = TX - TY; cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 12, "n1_12", {72, 0, 24, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_n1_12) (planner *p) { cannam@95: X(kdft_register) (p, n1_12, &desc); cannam@95: } cannam@95: cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_notw.native -compact -variables 4 -pipeline-latency 4 -n 12 -name n1_12 -include n.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 96 FP additions, 16 FP multiplications, cannam@95: * (or, 88 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 43 stack variables, 2 constants, and 48 memory accesses cannam@95: */ cannam@95: #include "n.h" cannam@95: cannam@95: static void n1_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, ri = ri + ivs, ii = ii + ivs, ro = ro + ovs, io = io + ovs, MAKE_VOLATILE_STRIDE(48, is), MAKE_VOLATILE_STRIDE(48, os)) { cannam@95: E T5, TR, TA, Ts, TS, Tz, Ta, TU, TD, Tx, TV, TC, Tg, T1a, TG; cannam@95: E TJ, T1u, T1d, Tl, T1f, TL, TO, T1v, T1i; cannam@95: { cannam@95: E T1, T2, T3, T4; cannam@95: T1 = ri[0]; cannam@95: T2 = ri[WS(is, 4)]; cannam@95: T3 = ri[WS(is, 8)]; cannam@95: T4 = T2 + T3; cannam@95: T5 = T1 + T4; cannam@95: TR = FNMS(KP500000000, T4, T1); cannam@95: TA = KP866025403 * (T3 - T2); cannam@95: } cannam@95: { cannam@95: E To, Tp, Tq, Tr; cannam@95: To = ii[0]; cannam@95: Tp = ii[WS(is, 4)]; cannam@95: Tq = ii[WS(is, 8)]; cannam@95: Tr = Tp + Tq; cannam@95: Ts = To + Tr; cannam@95: TS = KP866025403 * (Tp - Tq); cannam@95: Tz = FNMS(KP500000000, Tr, To); cannam@95: } cannam@95: { cannam@95: E T6, T7, T8, T9; cannam@95: T6 = ri[WS(is, 6)]; cannam@95: T7 = ri[WS(is, 10)]; cannam@95: T8 = ri[WS(is, 2)]; cannam@95: T9 = T7 + T8; cannam@95: Ta = T6 + T9; cannam@95: TU = FNMS(KP500000000, T9, T6); cannam@95: TD = KP866025403 * (T8 - T7); cannam@95: } cannam@95: { cannam@95: E Tt, Tu, Tv, Tw; cannam@95: Tt = ii[WS(is, 6)]; cannam@95: Tu = ii[WS(is, 10)]; cannam@95: Tv = ii[WS(is, 2)]; cannam@95: Tw = Tu + Tv; cannam@95: Tx = Tt + Tw; cannam@95: TV = KP866025403 * (Tu - Tv); cannam@95: TC = FNMS(KP500000000, Tw, Tt); cannam@95: } cannam@95: { cannam@95: E Tc, Td, Te, Tf; cannam@95: Tc = ri[WS(is, 3)]; cannam@95: Td = ri[WS(is, 7)]; cannam@95: Te = ri[WS(is, 11)]; cannam@95: Tf = Td + Te; cannam@95: Tg = Tc + Tf; cannam@95: T1a = KP866025403 * (Te - Td); cannam@95: TG = FNMS(KP500000000, Tf, Tc); cannam@95: } cannam@95: { cannam@95: E T1b, TH, TI, T1c; cannam@95: T1b = ii[WS(is, 3)]; cannam@95: TH = ii[WS(is, 7)]; cannam@95: TI = ii[WS(is, 11)]; cannam@95: T1c = TH + TI; cannam@95: TJ = KP866025403 * (TH - TI); cannam@95: T1u = T1b + T1c; cannam@95: T1d = FNMS(KP500000000, T1c, T1b); cannam@95: } cannam@95: { cannam@95: E Th, Ti, Tj, Tk; cannam@95: Th = ri[WS(is, 9)]; cannam@95: Ti = ri[WS(is, 1)]; cannam@95: Tj = ri[WS(is, 5)]; cannam@95: Tk = Ti + Tj; cannam@95: Tl = Th + Tk; cannam@95: T1f = KP866025403 * (Tj - Ti); cannam@95: TL = FNMS(KP500000000, Tk, Th); cannam@95: } cannam@95: { cannam@95: E T1g, TM, TN, T1h; cannam@95: T1g = ii[WS(is, 9)]; cannam@95: TM = ii[WS(is, 1)]; cannam@95: TN = ii[WS(is, 5)]; cannam@95: T1h = TM + TN; cannam@95: TO = KP866025403 * (TM - TN); cannam@95: T1v = T1g + T1h; cannam@95: T1i = FNMS(KP500000000, T1h, T1g); cannam@95: } cannam@95: { cannam@95: E Tb, Tm, T1t, T1w; cannam@95: Tb = T5 + Ta; cannam@95: Tm = Tg + Tl; cannam@95: ro[WS(os, 6)] = Tb - Tm; cannam@95: ro[0] = Tb + Tm; cannam@95: { cannam@95: E T1x, T1y, Tn, Ty; cannam@95: T1x = Ts + Tx; cannam@95: T1y = T1u + T1v; cannam@95: io[WS(os, 6)] = T1x - T1y; cannam@95: io[0] = T1x + T1y; cannam@95: Tn = Tg - Tl; cannam@95: Ty = Ts - Tx; cannam@95: io[WS(os, 3)] = Tn + Ty; cannam@95: io[WS(os, 9)] = Ty - Tn; cannam@95: } cannam@95: T1t = T5 - Ta; cannam@95: T1w = T1u - T1v; cannam@95: ro[WS(os, 3)] = T1t - T1w; cannam@95: ro[WS(os, 9)] = T1t + T1w; cannam@95: { cannam@95: E T11, T1l, T1k, T1m, T14, T18, T17, T19; cannam@95: { cannam@95: E TZ, T10, T1e, T1j; cannam@95: TZ = TA + Tz; cannam@95: T10 = TD + TC; cannam@95: T11 = TZ - T10; cannam@95: T1l = TZ + T10; cannam@95: T1e = T1a + T1d; cannam@95: T1j = T1f + T1i; cannam@95: T1k = T1e - T1j; cannam@95: T1m = T1e + T1j; cannam@95: } cannam@95: { cannam@95: E T12, T13, T15, T16; cannam@95: T12 = TG + TJ; cannam@95: T13 = TL + TO; cannam@95: T14 = T12 - T13; cannam@95: T18 = T12 + T13; cannam@95: T15 = TR + TS; cannam@95: T16 = TU + TV; cannam@95: T17 = T15 + T16; cannam@95: T19 = T15 - T16; cannam@95: } cannam@95: io[WS(os, 1)] = T11 - T14; cannam@95: ro[WS(os, 1)] = T19 + T1k; cannam@95: io[WS(os, 7)] = T11 + T14; cannam@95: ro[WS(os, 7)] = T19 - T1k; cannam@95: ro[WS(os, 10)] = T17 - T18; cannam@95: io[WS(os, 10)] = T1l - T1m; cannam@95: ro[WS(os, 4)] = T17 + T18; cannam@95: io[WS(os, 4)] = T1l + T1m; cannam@95: } cannam@95: { cannam@95: E TF, T1r, T1q, T1s, TQ, TY, TX, T1n; cannam@95: { cannam@95: E TB, TE, T1o, T1p; cannam@95: TB = Tz - TA; cannam@95: TE = TC - TD; cannam@95: TF = TB - TE; cannam@95: T1r = TB + TE; cannam@95: T1o = T1d - T1a; cannam@95: T1p = T1i - T1f; cannam@95: T1q = T1o - T1p; cannam@95: T1s = T1o + T1p; cannam@95: } cannam@95: { cannam@95: E TK, TP, TT, TW; cannam@95: TK = TG - TJ; cannam@95: TP = TL - TO; cannam@95: TQ = TK - TP; cannam@95: TY = TK + TP; cannam@95: TT = TR - TS; cannam@95: TW = TU - TV; cannam@95: TX = TT + TW; cannam@95: T1n = TT - TW; cannam@95: } cannam@95: io[WS(os, 5)] = TF - TQ; cannam@95: ro[WS(os, 5)] = T1n + T1q; cannam@95: io[WS(os, 11)] = TF + TQ; cannam@95: ro[WS(os, 11)] = T1n - T1q; cannam@95: ro[WS(os, 2)] = TX - TY; cannam@95: io[WS(os, 2)] = T1r - T1s; cannam@95: ro[WS(os, 8)] = TX + TY; cannam@95: io[WS(os, 8)] = T1r + T1s; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 12, "n1_12", {88, 8, 8, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void X(codelet_n1_12) (planner *p) { cannam@95: X(kdft_register) (p, n1_12, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */