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 07:39:50 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_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hf_6 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 32 FP multiplications, Chris@10: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), Chris@10: * 47 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@10: E T11, T12, T14, T13; Chris@10: { Chris@10: E T1, TV, TX, T7, Tn, Tq, TO, TR, TB, Tl, To, TH, Tt, Tw, Ts; Chris@10: E Tp, Tv; Chris@10: T1 = cr[0]; Chris@10: TV = ci[0]; Chris@10: { Chris@10: E T3, T6, T2, T5; Chris@10: T3 = cr[WS(rs, 3)]; Chris@10: T6 = ci[WS(rs, 3)]; Chris@10: T2 = W[4]; Chris@10: T5 = W[5]; Chris@10: { Chris@10: E Ta, Td, Tg, TM, Tb, Tj, Tf, Tc, Ti, TW, T4, T9; Chris@10: Ta = cr[WS(rs, 2)]; Chris@10: Td = ci[WS(rs, 2)]; Chris@10: TW = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T9 = W[2]; Chris@10: Tg = cr[WS(rs, 5)]; Chris@10: TX = FNMS(T5, T3, TW); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: TM = T9 * Td; Chris@10: Tb = T9 * Ta; Chris@10: Tj = ci[WS(rs, 5)]; Chris@10: Tf = W[8]; Chris@10: Tc = W[3]; Chris@10: Ti = W[9]; Chris@10: { Chris@10: E TN, Te, TL, Tk, TK, Th, Tm; Chris@10: Tn = cr[WS(rs, 4)]; Chris@10: TK = Tf * Tj; Chris@10: Th = Tf * Tg; Chris@10: TN = FNMS(Tc, Ta, TM); Chris@10: Te = FMA(Tc, Td, Tb); Chris@10: TL = FNMS(Ti, Tg, TK); Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: Tq = ci[WS(rs, 4)]; Chris@10: Tm = W[6]; Chris@10: TO = TL - TN; Chris@10: TR = TN + TL; Chris@10: TB = Te + Tk; Chris@10: Tl = Te - Tk; Chris@10: To = Tm * Tn; Chris@10: TH = Tm * Tq; Chris@10: } Chris@10: Tt = cr[WS(rs, 1)]; Chris@10: Tw = ci[WS(rs, 1)]; Chris@10: Ts = W[0]; Chris@10: Tp = W[7]; Chris@10: Tv = W[1]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TA, T8, TI, Tr, TG, Tx, TF, Tu; Chris@10: TA = T1 + T7; Chris@10: T8 = T1 - T7; Chris@10: TF = Ts * Tw; Chris@10: Tu = Ts * Tt; Chris@10: TI = FNMS(Tp, Tn, TH); Chris@10: Tr = FMA(Tp, Tq, To); Chris@10: TG = FNMS(Tv, Tt, TF); Chris@10: Tx = FMA(Tv, Tw, Tu); Chris@10: { Chris@10: E TY, TU, TP, TT, TD, T10, Tz, TZ, TQ, TE; Chris@10: T11 = TX + TV; Chris@10: TY = TV - TX; Chris@10: { Chris@10: E TJ, TS, TC, Ty; Chris@10: TJ = TG - TI; Chris@10: TS = TI + TG; Chris@10: TC = Tr + Tx; Chris@10: Ty = Tr - Tx; Chris@10: TU = TO + TJ; Chris@10: TP = TJ - TO; Chris@10: TT = TR - TS; Chris@10: T12 = TR + TS; Chris@10: T14 = TB - TC; Chris@10: TD = TB + TC; Chris@10: T10 = Ty - Tl; Chris@10: Tz = Tl + Ty; Chris@10: TZ = FMA(KP500000000, TU, TY); Chris@10: } Chris@10: cr[0] = TA + TD; Chris@10: TQ = FNMS(KP500000000, TD, TA); Chris@10: ci[WS(rs, 2)] = T8 + Tz; Chris@10: TE = FNMS(KP500000000, Tz, T8); Chris@10: cr[WS(rs, 3)] = TU - TY; Chris@10: cr[WS(rs, 2)] = FNMS(KP866025403, TT, TQ); Chris@10: ci[WS(rs, 1)] = FMA(KP866025403, TT, TQ); Chris@10: ci[0] = FNMS(KP866025403, TP, TE); Chris@10: cr[WS(rs, 1)] = FMA(KP866025403, TP, TE); Chris@10: ci[WS(rs, 4)] = FMA(KP866025403, T10, TZ); Chris@10: cr[WS(rs, 5)] = FMS(KP866025403, T10, TZ); Chris@10: } Chris@10: } Chris@10: } Chris@10: ci[WS(rs, 5)] = T12 + T11; Chris@10: T13 = FNMS(KP500000000, T12, T11); Chris@10: ci[WS(rs, 3)] = FMA(KP866025403, T14, T13); Chris@10: cr[WS(rs, 4)] = FMS(KP866025403, T14, T13); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 6, "hf_6", twinstr, &GENUS, {24, 10, 22, 0} }; Chris@10: Chris@10: void X(codelet_hf_6) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_6, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hf_6 -include hf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 28 FP multiplications, Chris@10: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), Chris@10: * 23 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hf.h" Chris@10: Chris@10: static void hf_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@10: E T7, TS, Tv, TO, Tt, TJ, Tx, TF, Ti, TI, Tw, TC; Chris@10: { Chris@10: E T1, TM, T6, TN; Chris@10: T1 = cr[0]; Chris@10: TM = ci[0]; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = cr[WS(rs, 3)]; Chris@10: T5 = ci[WS(rs, 3)]; Chris@10: T2 = W[4]; Chris@10: T4 = W[5]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: TN = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: T7 = T1 - T6; Chris@10: TS = TN + TM; Chris@10: Tv = T1 + T6; Chris@10: TO = TM - TN; Chris@10: } Chris@10: { Chris@10: E Tn, TE, Ts, TD; Chris@10: { Chris@10: E Tk, Tm, Tj, Tl; Chris@10: Tk = cr[WS(rs, 4)]; Chris@10: Tm = ci[WS(rs, 4)]; Chris@10: Tj = W[6]; Chris@10: Tl = W[7]; Chris@10: Tn = FMA(Tj, Tk, Tl * Tm); Chris@10: TE = FNMS(Tl, Tk, Tj * Tm); Chris@10: } Chris@10: { Chris@10: E Tp, Tr, To, Tq; Chris@10: Tp = cr[WS(rs, 1)]; Chris@10: Tr = ci[WS(rs, 1)]; Chris@10: To = W[0]; Chris@10: Tq = W[1]; Chris@10: Ts = FMA(To, Tp, Tq * Tr); Chris@10: TD = FNMS(Tq, Tp, To * Tr); Chris@10: } Chris@10: Tt = Tn - Ts; Chris@10: TJ = TE + TD; Chris@10: Tx = Tn + Ts; Chris@10: TF = TD - TE; Chris@10: } Chris@10: { Chris@10: E Tc, TA, Th, TB; Chris@10: { Chris@10: E T9, Tb, T8, Ta; Chris@10: T9 = cr[WS(rs, 2)]; Chris@10: Tb = ci[WS(rs, 2)]; Chris@10: T8 = W[2]; Chris@10: Ta = W[3]; Chris@10: Tc = FMA(T8, T9, Ta * Tb); Chris@10: TA = FNMS(Ta, T9, T8 * Tb); Chris@10: } Chris@10: { Chris@10: E Te, Tg, Td, Tf; Chris@10: Te = cr[WS(rs, 5)]; Chris@10: Tg = ci[WS(rs, 5)]; Chris@10: Td = W[8]; Chris@10: Tf = W[9]; Chris@10: Th = FMA(Td, Te, Tf * Tg); Chris@10: TB = FNMS(Tf, Te, Td * Tg); Chris@10: } Chris@10: Ti = Tc - Th; Chris@10: TI = TA + TB; Chris@10: Tw = Tc + Th; Chris@10: TC = TA - TB; Chris@10: } Chris@10: { Chris@10: E TG, Tu, Tz, TK, Ty, TH; Chris@10: TG = KP866025403 * (TC + TF); Chris@10: Tu = Ti + Tt; Chris@10: Tz = FNMS(KP500000000, Tu, T7); Chris@10: ci[WS(rs, 2)] = T7 + Tu; Chris@10: cr[WS(rs, 1)] = Tz + TG; Chris@10: ci[0] = Tz - TG; Chris@10: TK = KP866025403 * (TI - TJ); Chris@10: Ty = Tw + Tx; Chris@10: TH = FNMS(KP500000000, Ty, Tv); Chris@10: cr[0] = Tv + Ty; Chris@10: ci[WS(rs, 1)] = TH + TK; Chris@10: cr[WS(rs, 2)] = TH - TK; Chris@10: } Chris@10: { Chris@10: E TP, TL, TQ, TR, TT, TU; Chris@10: TP = KP866025403 * (Tt - Ti); Chris@10: TL = TF - TC; Chris@10: TQ = FMA(KP500000000, TL, TO); Chris@10: cr[WS(rs, 3)] = TL - TO; Chris@10: ci[WS(rs, 4)] = TP + TQ; Chris@10: cr[WS(rs, 5)] = TP - TQ; Chris@10: TR = KP866025403 * (Tw - Tx); Chris@10: TT = TI + TJ; Chris@10: TU = FNMS(KP500000000, TT, TS); Chris@10: cr[WS(rs, 4)] = TR - TU; Chris@10: ci[WS(rs, 5)] = TT + TS; Chris@10: ci[WS(rs, 3)] = TR + TU; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2hc_desc desc = { 6, "hf_6", twinstr, &GENUS, {32, 14, 14, 0} }; Chris@10: Chris@10: void X(codelet_hf_6) (planner *p) { Chris@10: X(khc2hc_register) (p, hf_6, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */