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 07:39:50 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_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hf_6 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 46 FP additions, 32 FP multiplications, cannam@95: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), cannam@95: * 47 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T11, T12, T14, T13; cannam@95: { cannam@95: E T1, TV, TX, T7, Tn, Tq, TO, TR, TB, Tl, To, TH, Tt, Tw, Ts; cannam@95: E Tp, Tv; cannam@95: T1 = cr[0]; cannam@95: TV = ci[0]; cannam@95: { cannam@95: E T3, T6, T2, T5; cannam@95: T3 = cr[WS(rs, 3)]; cannam@95: T6 = ci[WS(rs, 3)]; cannam@95: T2 = W[4]; cannam@95: T5 = W[5]; cannam@95: { cannam@95: E Ta, Td, Tg, TM, Tb, Tj, Tf, Tc, Ti, TW, T4, T9; cannam@95: Ta = cr[WS(rs, 2)]; cannam@95: Td = ci[WS(rs, 2)]; cannam@95: TW = T2 * T6; cannam@95: T4 = T2 * T3; cannam@95: T9 = W[2]; cannam@95: Tg = cr[WS(rs, 5)]; cannam@95: TX = FNMS(T5, T3, TW); cannam@95: T7 = FMA(T5, T6, T4); cannam@95: TM = T9 * Td; cannam@95: Tb = T9 * Ta; cannam@95: Tj = ci[WS(rs, 5)]; cannam@95: Tf = W[8]; cannam@95: Tc = W[3]; cannam@95: Ti = W[9]; cannam@95: { cannam@95: E TN, Te, TL, Tk, TK, Th, Tm; cannam@95: Tn = cr[WS(rs, 4)]; cannam@95: TK = Tf * Tj; cannam@95: Th = Tf * Tg; cannam@95: TN = FNMS(Tc, Ta, TM); cannam@95: Te = FMA(Tc, Td, Tb); cannam@95: TL = FNMS(Ti, Tg, TK); cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: Tq = ci[WS(rs, 4)]; cannam@95: Tm = W[6]; cannam@95: TO = TL - TN; cannam@95: TR = TN + TL; cannam@95: TB = Te + Tk; cannam@95: Tl = Te - Tk; cannam@95: To = Tm * Tn; cannam@95: TH = Tm * Tq; cannam@95: } cannam@95: Tt = cr[WS(rs, 1)]; cannam@95: Tw = ci[WS(rs, 1)]; cannam@95: Ts = W[0]; cannam@95: Tp = W[7]; cannam@95: Tv = W[1]; cannam@95: } cannam@95: } cannam@95: { cannam@95: E TA, T8, TI, Tr, TG, Tx, TF, Tu; cannam@95: TA = T1 + T7; cannam@95: T8 = T1 - T7; cannam@95: TF = Ts * Tw; cannam@95: Tu = Ts * Tt; cannam@95: TI = FNMS(Tp, Tn, TH); cannam@95: Tr = FMA(Tp, Tq, To); cannam@95: TG = FNMS(Tv, Tt, TF); cannam@95: Tx = FMA(Tv, Tw, Tu); cannam@95: { cannam@95: E TY, TU, TP, TT, TD, T10, Tz, TZ, TQ, TE; cannam@95: T11 = TX + TV; cannam@95: TY = TV - TX; cannam@95: { cannam@95: E TJ, TS, TC, Ty; cannam@95: TJ = TG - TI; cannam@95: TS = TI + TG; cannam@95: TC = Tr + Tx; cannam@95: Ty = Tr - Tx; cannam@95: TU = TO + TJ; cannam@95: TP = TJ - TO; cannam@95: TT = TR - TS; cannam@95: T12 = TR + TS; cannam@95: T14 = TB - TC; cannam@95: TD = TB + TC; cannam@95: T10 = Ty - Tl; cannam@95: Tz = Tl + Ty; cannam@95: TZ = FMA(KP500000000, TU, TY); cannam@95: } cannam@95: cr[0] = TA + TD; cannam@95: TQ = FNMS(KP500000000, TD, TA); cannam@95: ci[WS(rs, 2)] = T8 + Tz; cannam@95: TE = FNMS(KP500000000, Tz, T8); cannam@95: cr[WS(rs, 3)] = TU - TY; cannam@95: cr[WS(rs, 2)] = FNMS(KP866025403, TT, TQ); cannam@95: ci[WS(rs, 1)] = FMA(KP866025403, TT, TQ); cannam@95: ci[0] = FNMS(KP866025403, TP, TE); cannam@95: cr[WS(rs, 1)] = FMA(KP866025403, TP, TE); cannam@95: ci[WS(rs, 4)] = FMA(KP866025403, T10, TZ); cannam@95: cr[WS(rs, 5)] = FMS(KP866025403, T10, TZ); cannam@95: } cannam@95: } cannam@95: } cannam@95: ci[WS(rs, 5)] = T12 + T11; cannam@95: T13 = FNMS(KP500000000, T12, T11); cannam@95: ci[WS(rs, 3)] = FMA(KP866025403, T14, T13); cannam@95: cr[WS(rs, 4)] = FMS(KP866025403, T14, T13); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 6}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 6, "hf_6", twinstr, &GENUS, {24, 10, 22, 0} }; cannam@95: cannam@95: void X(codelet_hf_6) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_6, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hf_6 -include hf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 46 FP additions, 28 FP multiplications, cannam@95: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), cannam@95: * 23 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "hf.h" cannam@95: cannam@95: static void hf_6(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT m; cannam@95: 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)) { cannam@95: E T7, TS, Tv, TO, Tt, TJ, Tx, TF, Ti, TI, Tw, TC; cannam@95: { cannam@95: E T1, TM, T6, TN; cannam@95: T1 = cr[0]; cannam@95: TM = ci[0]; cannam@95: { cannam@95: E T3, T5, T2, T4; cannam@95: T3 = cr[WS(rs, 3)]; cannam@95: T5 = ci[WS(rs, 3)]; cannam@95: T2 = W[4]; cannam@95: T4 = W[5]; cannam@95: T6 = FMA(T2, T3, T4 * T5); cannam@95: TN = FNMS(T4, T3, T2 * T5); cannam@95: } cannam@95: T7 = T1 - T6; cannam@95: TS = TN + TM; cannam@95: Tv = T1 + T6; cannam@95: TO = TM - TN; cannam@95: } cannam@95: { cannam@95: E Tn, TE, Ts, TD; cannam@95: { cannam@95: E Tk, Tm, Tj, Tl; cannam@95: Tk = cr[WS(rs, 4)]; cannam@95: Tm = ci[WS(rs, 4)]; cannam@95: Tj = W[6]; cannam@95: Tl = W[7]; cannam@95: Tn = FMA(Tj, Tk, Tl * Tm); cannam@95: TE = FNMS(Tl, Tk, Tj * Tm); cannam@95: } cannam@95: { cannam@95: E Tp, Tr, To, Tq; cannam@95: Tp = cr[WS(rs, 1)]; cannam@95: Tr = ci[WS(rs, 1)]; cannam@95: To = W[0]; cannam@95: Tq = W[1]; cannam@95: Ts = FMA(To, Tp, Tq * Tr); cannam@95: TD = FNMS(Tq, Tp, To * Tr); cannam@95: } cannam@95: Tt = Tn - Ts; cannam@95: TJ = TE + TD; cannam@95: Tx = Tn + Ts; cannam@95: TF = TD - TE; cannam@95: } cannam@95: { cannam@95: E Tc, TA, Th, TB; cannam@95: { cannam@95: E T9, Tb, T8, Ta; cannam@95: T9 = cr[WS(rs, 2)]; cannam@95: Tb = ci[WS(rs, 2)]; cannam@95: T8 = W[2]; cannam@95: Ta = W[3]; cannam@95: Tc = FMA(T8, T9, Ta * Tb); cannam@95: TA = FNMS(Ta, T9, T8 * Tb); cannam@95: } cannam@95: { cannam@95: E Te, Tg, Td, Tf; cannam@95: Te = cr[WS(rs, 5)]; cannam@95: Tg = ci[WS(rs, 5)]; cannam@95: Td = W[8]; cannam@95: Tf = W[9]; cannam@95: Th = FMA(Td, Te, Tf * Tg); cannam@95: TB = FNMS(Tf, Te, Td * Tg); cannam@95: } cannam@95: Ti = Tc - Th; cannam@95: TI = TA + TB; cannam@95: Tw = Tc + Th; cannam@95: TC = TA - TB; cannam@95: } cannam@95: { cannam@95: E TG, Tu, Tz, TK, Ty, TH; cannam@95: TG = KP866025403 * (TC + TF); cannam@95: Tu = Ti + Tt; cannam@95: Tz = FNMS(KP500000000, Tu, T7); cannam@95: ci[WS(rs, 2)] = T7 + Tu; cannam@95: cr[WS(rs, 1)] = Tz + TG; cannam@95: ci[0] = Tz - TG; cannam@95: TK = KP866025403 * (TI - TJ); cannam@95: Ty = Tw + Tx; cannam@95: TH = FNMS(KP500000000, Ty, Tv); cannam@95: cr[0] = Tv + Ty; cannam@95: ci[WS(rs, 1)] = TH + TK; cannam@95: cr[WS(rs, 2)] = TH - TK; cannam@95: } cannam@95: { cannam@95: E TP, TL, TQ, TR, TT, TU; cannam@95: TP = KP866025403 * (Tt - Ti); cannam@95: TL = TF - TC; cannam@95: TQ = FMA(KP500000000, TL, TO); cannam@95: cr[WS(rs, 3)] = TL - TO; cannam@95: ci[WS(rs, 4)] = TP + TQ; cannam@95: cr[WS(rs, 5)] = TP - TQ; cannam@95: TR = KP866025403 * (Tw - Tx); cannam@95: TT = TI + TJ; cannam@95: TU = FNMS(KP500000000, TT, TS); cannam@95: cr[WS(rs, 4)] = TR - TU; cannam@95: ci[WS(rs, 5)] = TT + TS; cannam@95: ci[WS(rs, 3)] = TR + TU; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 6}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 6, "hf_6", twinstr, &GENUS, {32, 14, 14, 0} }; cannam@95: cannam@95: void X(codelet_hf_6) (planner *p) { cannam@95: X(khc2hc_register) (p, hf_6, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */