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