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 07:39:47 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_r2cf.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 16 -name r2cf_16 -include r2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 20 FP multiplications, cannam@95: * (or, 38 additions, 0 multiplications, 20 fused multiply/add), cannam@95: * 38 stack variables, 3 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "r2cf.h" cannam@95: cannam@95: static void r2cf_16(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: DK(KP414213562, +0.414213562373095048801688724209698078569671875); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(64, rs), MAKE_VOLATILE_STRIDE(64, csr), MAKE_VOLATILE_STRIDE(64, csi)) { cannam@95: E TQ, TP; cannam@95: { cannam@95: E TB, TN, Tf, T7, Te, Tv, TO, TE, Tq, TJ, Tp, TI, TT, Ty, Tm; cannam@95: E Tr, TK, Ts; cannam@95: { cannam@95: E TC, Ta, Td, TD; cannam@95: { cannam@95: E T1, T2, T4, T5; cannam@95: T1 = R0[0]; cannam@95: T2 = R0[WS(rs, 4)]; cannam@95: T4 = R0[WS(rs, 2)]; cannam@95: T5 = R0[WS(rs, 6)]; cannam@95: { cannam@95: E T8, T3, T6, T9, Tb, Tc; cannam@95: T8 = R0[WS(rs, 1)]; cannam@95: TB = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: TN = T4 - T5; cannam@95: T6 = T4 + T5; cannam@95: T9 = R0[WS(rs, 5)]; cannam@95: Tb = R0[WS(rs, 7)]; cannam@95: Tc = R0[WS(rs, 3)]; cannam@95: Tf = T3 - T6; cannam@95: T7 = T3 + T6; cannam@95: TC = T8 - T9; cannam@95: Ta = T8 + T9; cannam@95: Td = Tb + Tc; cannam@95: TD = Tb - Tc; cannam@95: } cannam@95: } cannam@95: { cannam@95: E TG, Ti, Tj, Tk, Tg, Th; cannam@95: Tg = R1[0]; cannam@95: Th = R1[WS(rs, 4)]; cannam@95: Te = Ta + Td; cannam@95: Tv = Td - Ta; cannam@95: TO = TD - TC; cannam@95: TE = TC + TD; cannam@95: TG = Tg - Th; cannam@95: Ti = Tg + Th; cannam@95: Tj = R1[WS(rs, 2)]; cannam@95: Tk = R1[WS(rs, 6)]; cannam@95: { cannam@95: E Tn, To, TH, Tl; cannam@95: Tn = R1[WS(rs, 7)]; cannam@95: To = R1[WS(rs, 3)]; cannam@95: Tq = R1[WS(rs, 1)]; cannam@95: TH = Tj - Tk; cannam@95: Tl = Tj + Tk; cannam@95: TJ = Tn - To; cannam@95: Tp = Tn + To; cannam@95: TI = FNMS(KP414213562, TH, TG); cannam@95: TT = FMA(KP414213562, TG, TH); cannam@95: Ty = Ti + Tl; cannam@95: Tm = Ti - Tl; cannam@95: Tr = R1[WS(rs, 5)]; cannam@95: } cannam@95: } cannam@95: } cannam@95: Cr[WS(csr, 4)] = T7 - Te; cannam@95: TK = Tr - Tq; cannam@95: Ts = Tq + Tr; cannam@95: { cannam@95: E Tx, TV, TF, TS, Tz, Tt, TM, TL; cannam@95: Tx = T7 + Te; cannam@95: TV = FNMS(KP707106781, TE, TB); cannam@95: TF = FMA(KP707106781, TE, TB); cannam@95: TL = FNMS(KP414213562, TK, TJ); cannam@95: TS = FMA(KP414213562, TJ, TK); cannam@95: Tz = Tp + Ts; cannam@95: Tt = Tp - Ts; cannam@95: TM = TI + TL; cannam@95: TQ = TL - TI; cannam@95: { cannam@95: E TR, TU, TW, TA, Tw, Tu; cannam@95: TP = FMA(KP707106781, TO, TN); cannam@95: TR = FNMS(KP707106781, TO, TN); cannam@95: TA = Ty + Tz; cannam@95: Ci[WS(csi, 4)] = Tz - Ty; cannam@95: Tw = Tt - Tm; cannam@95: Tu = Tm + Tt; cannam@95: Cr[WS(csr, 1)] = FMA(KP923879532, TM, TF); cannam@95: Cr[WS(csr, 7)] = FNMS(KP923879532, TM, TF); cannam@95: Cr[0] = Tx + TA; cannam@95: Cr[WS(csr, 8)] = Tx - TA; cannam@95: Ci[WS(csi, 6)] = FMS(KP707106781, Tw, Tv); cannam@95: Ci[WS(csi, 2)] = FMA(KP707106781, Tw, Tv); cannam@95: Cr[WS(csr, 2)] = FMA(KP707106781, Tu, Tf); cannam@95: Cr[WS(csr, 6)] = FNMS(KP707106781, Tu, Tf); cannam@95: TU = TS - TT; cannam@95: TW = TT + TS; cannam@95: Ci[WS(csi, 7)] = FMA(KP923879532, TU, TR); cannam@95: Ci[WS(csi, 1)] = FMS(KP923879532, TU, TR); cannam@95: Cr[WS(csr, 3)] = FMA(KP923879532, TW, TV); cannam@95: Cr[WS(csr, 5)] = FNMS(KP923879532, TW, TV); cannam@95: } cannam@95: } cannam@95: } cannam@95: Ci[WS(csi, 5)] = FMS(KP923879532, TQ, TP); cannam@95: Ci[WS(csi, 3)] = FMA(KP923879532, TQ, TP); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 16, "r2cf_16", {38, 0, 20, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cf_16) (planner *p) { cannam@95: X(kr2c_register) (p, r2cf_16, &desc); cannam@95: } cannam@95: cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 16 -name r2cf_16 -include r2cf.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 12 FP multiplications, cannam@95: * (or, 54 additions, 8 multiplications, 4 fused multiply/add), cannam@95: * 34 stack variables, 3 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "r2cf.h" cannam@95: cannam@95: static void r2cf_16(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@95: DK(KP382683432, +0.382683432365089771728459984030398866761344562); cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(64, rs), MAKE_VOLATILE_STRIDE(64, csr), MAKE_VOLATILE_STRIDE(64, csi)) { cannam@95: E T3, T6, T7, Tz, Ti, Ta, Td, Te, TA, Th, Tq, TV, TF, TP, Tx; cannam@95: E TU, TE, TM, Tg, Tf, TJ, TQ; cannam@95: { cannam@95: E T1, T2, T4, T5; cannam@95: T1 = R0[0]; cannam@95: T2 = R0[WS(rs, 4)]; cannam@95: T3 = T1 + T2; cannam@95: T4 = R0[WS(rs, 2)]; cannam@95: T5 = R0[WS(rs, 6)]; cannam@95: T6 = T4 + T5; cannam@95: T7 = T3 + T6; cannam@95: Tz = T1 - T2; cannam@95: Ti = T4 - T5; cannam@95: } cannam@95: { cannam@95: E T8, T9, Tb, Tc; cannam@95: T8 = R0[WS(rs, 1)]; cannam@95: T9 = R0[WS(rs, 5)]; cannam@95: Ta = T8 + T9; cannam@95: Tg = T8 - T9; cannam@95: Tb = R0[WS(rs, 7)]; cannam@95: Tc = R0[WS(rs, 3)]; cannam@95: Td = Tb + Tc; cannam@95: Tf = Tb - Tc; cannam@95: } cannam@95: Te = Ta + Td; cannam@95: TA = KP707106781 * (Tg + Tf); cannam@95: Th = KP707106781 * (Tf - Tg); cannam@95: { cannam@95: E Tm, TN, Tp, TO; cannam@95: { cannam@95: E Tk, Tl, Tn, To; cannam@95: Tk = R1[WS(rs, 7)]; cannam@95: Tl = R1[WS(rs, 3)]; cannam@95: Tm = Tk - Tl; cannam@95: TN = Tk + Tl; cannam@95: Tn = R1[WS(rs, 1)]; cannam@95: To = R1[WS(rs, 5)]; cannam@95: Tp = Tn - To; cannam@95: TO = Tn + To; cannam@95: } cannam@95: Tq = FNMS(KP923879532, Tp, KP382683432 * Tm); cannam@95: TV = TN + TO; cannam@95: TF = FMA(KP923879532, Tm, KP382683432 * Tp); cannam@95: TP = TN - TO; cannam@95: } cannam@95: { cannam@95: E Tt, TK, Tw, TL; cannam@95: { cannam@95: E Tr, Ts, Tu, Tv; cannam@95: Tr = R1[0]; cannam@95: Ts = R1[WS(rs, 4)]; cannam@95: Tt = Tr - Ts; cannam@95: TK = Tr + Ts; cannam@95: Tu = R1[WS(rs, 2)]; cannam@95: Tv = R1[WS(rs, 6)]; cannam@95: Tw = Tu - Tv; cannam@95: TL = Tu + Tv; cannam@95: } cannam@95: Tx = FMA(KP382683432, Tt, KP923879532 * Tw); cannam@95: TU = TK + TL; cannam@95: TE = FNMS(KP382683432, Tw, KP923879532 * Tt); cannam@95: TM = TK - TL; cannam@95: } cannam@95: Cr[WS(csr, 4)] = T7 - Te; cannam@95: Ci[WS(csi, 4)] = TV - TU; cannam@95: { cannam@95: E Tj, Ty, TD, TG; cannam@95: Tj = Th - Ti; cannam@95: Ty = Tq - Tx; cannam@95: Ci[WS(csi, 1)] = Tj + Ty; cannam@95: Ci[WS(csi, 7)] = Ty - Tj; cannam@95: TD = Tz + TA; cannam@95: TG = TE + TF; cannam@95: Cr[WS(csr, 7)] = TD - TG; cannam@95: Cr[WS(csr, 1)] = TD + TG; cannam@95: } cannam@95: { cannam@95: E TB, TC, TH, TI; cannam@95: TB = Tz - TA; cannam@95: TC = Tx + Tq; cannam@95: Cr[WS(csr, 5)] = TB - TC; cannam@95: Cr[WS(csr, 3)] = TB + TC; cannam@95: TH = Ti + Th; cannam@95: TI = TF - TE; cannam@95: Ci[WS(csi, 3)] = TH + TI; cannam@95: Ci[WS(csi, 5)] = TI - TH; cannam@95: } cannam@95: TJ = T3 - T6; cannam@95: TQ = KP707106781 * (TM + TP); cannam@95: Cr[WS(csr, 6)] = TJ - TQ; cannam@95: Cr[WS(csr, 2)] = TJ + TQ; cannam@95: { cannam@95: E TR, TS, TT, TW; cannam@95: TR = Td - Ta; cannam@95: TS = KP707106781 * (TP - TM); cannam@95: Ci[WS(csi, 2)] = TR + TS; cannam@95: Ci[WS(csi, 6)] = TS - TR; cannam@95: TT = T7 + Te; cannam@95: TW = TU + TV; cannam@95: Cr[WS(csr, 8)] = TT - TW; cannam@95: Cr[0] = TT + TW; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 16, "r2cf_16", {54, 8, 4, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cf_16) (planner *p) { cannam@95: X(kr2c_register) (p, r2cf_16, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */