cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:46:07 EDT 2016 */ cannam@127: cannam@127: #include "codelet-rdft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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@127: cannam@127: /* cannam@127: * This function contains 58 FP additions, 20 FP multiplications, cannam@127: * (or, 38 additions, 0 multiplications, 20 fused multiply/add), cannam@127: * 38 stack variables, 3 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "r2cf.h" cannam@127: cannam@127: 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@127: { cannam@127: DK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@127: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: DK(KP414213562, +0.414213562373095048801688724209698078569671875); cannam@127: { cannam@127: INT i; cannam@127: 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@127: E TQ, TP; cannam@127: { cannam@127: E TB, TN, Tf, T7, Te, Tv, TO, TE, Tq, TJ, Tp, TI, TT, Ty, Tm; cannam@127: E Tr, TK, Ts; cannam@127: { cannam@127: E TC, Ta, Td, TD; cannam@127: { cannam@127: E T1, T2, T4, T5; cannam@127: T1 = R0[0]; cannam@127: T2 = R0[WS(rs, 4)]; cannam@127: T4 = R0[WS(rs, 2)]; cannam@127: T5 = R0[WS(rs, 6)]; cannam@127: { cannam@127: E T8, T3, T6, T9, Tb, Tc; cannam@127: T8 = R0[WS(rs, 1)]; cannam@127: TB = T1 - T2; cannam@127: T3 = T1 + T2; cannam@127: TN = T4 - T5; cannam@127: T6 = T4 + T5; cannam@127: T9 = R0[WS(rs, 5)]; cannam@127: Tb = R0[WS(rs, 7)]; cannam@127: Tc = R0[WS(rs, 3)]; cannam@127: Tf = T3 - T6; cannam@127: T7 = T3 + T6; cannam@127: TC = T8 - T9; cannam@127: Ta = T8 + T9; cannam@127: Td = Tb + Tc; cannam@127: TD = Tb - Tc; cannam@127: } cannam@127: } cannam@127: { cannam@127: E TG, Ti, Tj, Tk, Tg, Th; cannam@127: Tg = R1[0]; cannam@127: Th = R1[WS(rs, 4)]; cannam@127: Te = Ta + Td; cannam@127: Tv = Td - Ta; cannam@127: TO = TD - TC; cannam@127: TE = TC + TD; cannam@127: TG = Tg - Th; cannam@127: Ti = Tg + Th; cannam@127: Tj = R1[WS(rs, 2)]; cannam@127: Tk = R1[WS(rs, 6)]; cannam@127: { cannam@127: E Tn, To, TH, Tl; cannam@127: Tn = R1[WS(rs, 7)]; cannam@127: To = R1[WS(rs, 3)]; cannam@127: Tq = R1[WS(rs, 1)]; cannam@127: TH = Tj - Tk; cannam@127: Tl = Tj + Tk; cannam@127: TJ = Tn - To; cannam@127: Tp = Tn + To; cannam@127: TI = FNMS(KP414213562, TH, TG); cannam@127: TT = FMA(KP414213562, TG, TH); cannam@127: Ty = Ti + Tl; cannam@127: Tm = Ti - Tl; cannam@127: Tr = R1[WS(rs, 5)]; cannam@127: } cannam@127: } cannam@127: } cannam@127: Cr[WS(csr, 4)] = T7 - Te; cannam@127: TK = Tr - Tq; cannam@127: Ts = Tq + Tr; cannam@127: { cannam@127: E Tx, TV, TF, TS, Tz, Tt, TM, TL; cannam@127: Tx = T7 + Te; cannam@127: TV = FNMS(KP707106781, TE, TB); cannam@127: TF = FMA(KP707106781, TE, TB); cannam@127: TL = FNMS(KP414213562, TK, TJ); cannam@127: TS = FMA(KP414213562, TJ, TK); cannam@127: Tz = Tp + Ts; cannam@127: Tt = Tp - Ts; cannam@127: TM = TI + TL; cannam@127: TQ = TL - TI; cannam@127: { cannam@127: E TR, TU, TW, TA, Tw, Tu; cannam@127: TP = FMA(KP707106781, TO, TN); cannam@127: TR = FNMS(KP707106781, TO, TN); cannam@127: TA = Ty + Tz; cannam@127: Ci[WS(csi, 4)] = Tz - Ty; cannam@127: Tw = Tt - Tm; cannam@127: Tu = Tm + Tt; cannam@127: Cr[WS(csr, 1)] = FMA(KP923879532, TM, TF); cannam@127: Cr[WS(csr, 7)] = FNMS(KP923879532, TM, TF); cannam@127: Cr[0] = Tx + TA; cannam@127: Cr[WS(csr, 8)] = Tx - TA; cannam@127: Ci[WS(csi, 6)] = FMS(KP707106781, Tw, Tv); cannam@127: Ci[WS(csi, 2)] = FMA(KP707106781, Tw, Tv); cannam@127: Cr[WS(csr, 2)] = FMA(KP707106781, Tu, Tf); cannam@127: Cr[WS(csr, 6)] = FNMS(KP707106781, Tu, Tf); cannam@127: TU = TS - TT; cannam@127: TW = TT + TS; cannam@127: Ci[WS(csi, 7)] = FMA(KP923879532, TU, TR); cannam@127: Ci[WS(csi, 1)] = FMS(KP923879532, TU, TR); cannam@127: Cr[WS(csr, 3)] = FMA(KP923879532, TW, TV); cannam@127: Cr[WS(csr, 5)] = FNMS(KP923879532, TW, TV); cannam@127: } cannam@127: } cannam@127: } cannam@127: Ci[WS(csi, 5)] = FMS(KP923879532, TQ, TP); cannam@127: Ci[WS(csi, 3)] = FMA(KP923879532, TQ, TP); cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const kr2c_desc desc = { 16, "r2cf_16", {38, 0, 20, 0}, &GENUS }; cannam@127: cannam@127: void X(codelet_r2cf_16) (planner *p) { cannam@127: X(kr2c_register) (p, r2cf_16, &desc); cannam@127: } cannam@127: cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 16 -name r2cf_16 -include r2cf.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 58 FP additions, 12 FP multiplications, cannam@127: * (or, 54 additions, 8 multiplications, 4 fused multiply/add), cannam@127: * 34 stack variables, 3 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "r2cf.h" cannam@127: cannam@127: 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@127: { cannam@127: DK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@127: DK(KP382683432, +0.382683432365089771728459984030398866761344562); cannam@127: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT i; cannam@127: 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@127: E T3, T6, T7, Tz, Ti, Ta, Td, Te, TA, Th, Tq, TV, TF, TP, Tx; cannam@127: E TU, TE, TM, Tg, Tf, TJ, TQ; cannam@127: { cannam@127: E T1, T2, T4, T5; cannam@127: T1 = R0[0]; cannam@127: T2 = R0[WS(rs, 4)]; cannam@127: T3 = T1 + T2; cannam@127: T4 = R0[WS(rs, 2)]; cannam@127: T5 = R0[WS(rs, 6)]; cannam@127: T6 = T4 + T5; cannam@127: T7 = T3 + T6; cannam@127: Tz = T1 - T2; cannam@127: Ti = T4 - T5; cannam@127: } cannam@127: { cannam@127: E T8, T9, Tb, Tc; cannam@127: T8 = R0[WS(rs, 1)]; cannam@127: T9 = R0[WS(rs, 5)]; cannam@127: Ta = T8 + T9; cannam@127: Tg = T8 - T9; cannam@127: Tb = R0[WS(rs, 7)]; cannam@127: Tc = R0[WS(rs, 3)]; cannam@127: Td = Tb + Tc; cannam@127: Tf = Tb - Tc; cannam@127: } cannam@127: Te = Ta + Td; cannam@127: TA = KP707106781 * (Tg + Tf); cannam@127: Th = KP707106781 * (Tf - Tg); cannam@127: { cannam@127: E Tm, TN, Tp, TO; cannam@127: { cannam@127: E Tk, Tl, Tn, To; cannam@127: Tk = R1[WS(rs, 7)]; cannam@127: Tl = R1[WS(rs, 3)]; cannam@127: Tm = Tk - Tl; cannam@127: TN = Tk + Tl; cannam@127: Tn = R1[WS(rs, 1)]; cannam@127: To = R1[WS(rs, 5)]; cannam@127: Tp = Tn - To; cannam@127: TO = Tn + To; cannam@127: } cannam@127: Tq = FNMS(KP923879532, Tp, KP382683432 * Tm); cannam@127: TV = TN + TO; cannam@127: TF = FMA(KP923879532, Tm, KP382683432 * Tp); cannam@127: TP = TN - TO; cannam@127: } cannam@127: { cannam@127: E Tt, TK, Tw, TL; cannam@127: { cannam@127: E Tr, Ts, Tu, Tv; cannam@127: Tr = R1[0]; cannam@127: Ts = R1[WS(rs, 4)]; cannam@127: Tt = Tr - Ts; cannam@127: TK = Tr + Ts; cannam@127: Tu = R1[WS(rs, 2)]; cannam@127: Tv = R1[WS(rs, 6)]; cannam@127: Tw = Tu - Tv; cannam@127: TL = Tu + Tv; cannam@127: } cannam@127: Tx = FMA(KP382683432, Tt, KP923879532 * Tw); cannam@127: TU = TK + TL; cannam@127: TE = FNMS(KP382683432, Tw, KP923879532 * Tt); cannam@127: TM = TK - TL; cannam@127: } cannam@127: Cr[WS(csr, 4)] = T7 - Te; cannam@127: Ci[WS(csi, 4)] = TV - TU; cannam@127: { cannam@127: E Tj, Ty, TD, TG; cannam@127: Tj = Th - Ti; cannam@127: Ty = Tq - Tx; cannam@127: Ci[WS(csi, 1)] = Tj + Ty; cannam@127: Ci[WS(csi, 7)] = Ty - Tj; cannam@127: TD = Tz + TA; cannam@127: TG = TE + TF; cannam@127: Cr[WS(csr, 7)] = TD - TG; cannam@127: Cr[WS(csr, 1)] = TD + TG; cannam@127: } cannam@127: { cannam@127: E TB, TC, TH, TI; cannam@127: TB = Tz - TA; cannam@127: TC = Tx + Tq; cannam@127: Cr[WS(csr, 5)] = TB - TC; cannam@127: Cr[WS(csr, 3)] = TB + TC; cannam@127: TH = Ti + Th; cannam@127: TI = TF - TE; cannam@127: Ci[WS(csi, 3)] = TH + TI; cannam@127: Ci[WS(csi, 5)] = TI - TH; cannam@127: } cannam@127: TJ = T3 - T6; cannam@127: TQ = KP707106781 * (TM + TP); cannam@127: Cr[WS(csr, 6)] = TJ - TQ; cannam@127: Cr[WS(csr, 2)] = TJ + TQ; cannam@127: { cannam@127: E TR, TS, TT, TW; cannam@127: TR = Td - Ta; cannam@127: TS = KP707106781 * (TP - TM); cannam@127: Ci[WS(csi, 2)] = TR + TS; cannam@127: Ci[WS(csi, 6)] = TS - TR; cannam@127: TT = T7 + Te; cannam@127: TW = TU + TV; cannam@127: Cr[WS(csr, 8)] = TT - TW; cannam@127: Cr[0] = TT + TW; cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const kr2c_desc desc = { 16, "r2cf_16", {54, 8, 4, 0}, &GENUS }; cannam@127: cannam@127: void X(codelet_r2cf_16) (planner *p) { cannam@127: X(kr2c_register) (p, r2cf_16, &desc); cannam@127: } cannam@127: cannam@127: #endif /* HAVE_FMA */