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 07:41:08 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_r2cb.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 16 -name r2cb_16 -include r2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 32 FP multiplications, cannam@95: * (or, 26 additions, 0 multiplications, 32 fused multiply/add), cannam@95: * 47 stack variables, 4 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "r2cb.h" cannam@95: cannam@95: static void r2cb_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(KP1_847759065, +1.847759065022573512256366378793576573644833252); cannam@95: DK(KP414213562, +0.414213562373095048801688724209698078569671875); cannam@95: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); cannam@95: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(64, rs), MAKE_VOLATILE_STRIDE(64, csr), MAKE_VOLATILE_STRIDE(64, csi)) { cannam@95: E TN, TS, TF, TI; cannam@95: { cannam@95: E T8, TD, Tj, TL, T5, TM, TE, To, Td, Tq, Tc, TP, Ty, Te, Tr; cannam@95: E Ts; cannam@95: { cannam@95: E T4, Ti, T1, T2; cannam@95: T4 = Cr[WS(csr, 4)]; cannam@95: Ti = Ci[WS(csi, 4)]; cannam@95: T1 = Cr[0]; cannam@95: T2 = Cr[WS(csr, 8)]; cannam@95: { cannam@95: E Tk, Tn, T6, T7; cannam@95: T6 = Cr[WS(csr, 2)]; cannam@95: T7 = Cr[WS(csr, 6)]; cannam@95: { cannam@95: E Tl, Th, T3, Tm; cannam@95: Tl = Ci[WS(csi, 2)]; cannam@95: Th = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: Tk = T6 - T7; cannam@95: T8 = T6 + T7; cannam@95: Tm = Ci[WS(csi, 6)]; cannam@95: TD = FMA(KP2_000000000, Ti, Th); cannam@95: Tj = FNMS(KP2_000000000, Ti, Th); cannam@95: TL = FNMS(KP2_000000000, T4, T3); cannam@95: T5 = FMA(KP2_000000000, T4, T3); cannam@95: Tn = Tl + Tm; cannam@95: TM = Tl - Tm; cannam@95: } cannam@95: { cannam@95: E Ta, Tb, Tw, Tx; cannam@95: Ta = Cr[WS(csr, 1)]; cannam@95: TE = Tk + Tn; cannam@95: To = Tk - Tn; cannam@95: Tb = Cr[WS(csr, 7)]; cannam@95: Tw = Ci[WS(csi, 1)]; cannam@95: Tx = Ci[WS(csi, 7)]; cannam@95: Td = Cr[WS(csr, 5)]; cannam@95: Tq = Ta - Tb; cannam@95: Tc = Ta + Tb; cannam@95: TP = Tw - Tx; cannam@95: Ty = Tw + Tx; cannam@95: Te = Cr[WS(csr, 3)]; cannam@95: Tr = Ci[WS(csi, 5)]; cannam@95: Ts = Ci[WS(csi, 3)]; cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E TV, TG, TW, TH, TB, Tp, TA, TC, TJ, TK; cannam@95: { cannam@95: E T9, Tz, Tg, Tu, TT, TU, TO, TR; cannam@95: TV = FNMS(KP2_000000000, T8, T5); cannam@95: T9 = FMA(KP2_000000000, T8, T5); cannam@95: { cannam@95: E Tv, Tf, TQ, Tt; cannam@95: Tv = Td - Te; cannam@95: Tf = Td + Te; cannam@95: TQ = Tr - Ts; cannam@95: Tt = Tr + Ts; cannam@95: TG = Ty - Tv; cannam@95: Tz = Tv + Ty; cannam@95: TO = Tc - Tf; cannam@95: Tg = Tc + Tf; cannam@95: TW = TQ + TP; cannam@95: TR = TP - TQ; cannam@95: TH = Tq + Tt; cannam@95: Tu = Tq - Tt; cannam@95: } cannam@95: TN = FNMS(KP2_000000000, TM, TL); cannam@95: TT = FMA(KP2_000000000, TM, TL); cannam@95: TU = TO + TR; cannam@95: TS = TO - TR; cannam@95: R0[0] = FMA(KP2_000000000, Tg, T9); cannam@95: R0[WS(rs, 4)] = FNMS(KP2_000000000, Tg, T9); cannam@95: R0[WS(rs, 7)] = FMA(KP1_414213562, TU, TT); cannam@95: R0[WS(rs, 3)] = FNMS(KP1_414213562, TU, TT); cannam@95: TB = FNMS(KP1_414213562, To, Tj); cannam@95: Tp = FMA(KP1_414213562, To, Tj); cannam@95: TA = FNMS(KP414213562, Tz, Tu); cannam@95: TC = FMA(KP414213562, Tu, Tz); cannam@95: } cannam@95: R0[WS(rs, 6)] = FMA(KP2_000000000, TW, TV); cannam@95: R0[WS(rs, 2)] = FNMS(KP2_000000000, TW, TV); cannam@95: R1[0] = FMA(KP1_847759065, TA, Tp); cannam@95: R1[WS(rs, 4)] = FNMS(KP1_847759065, TA, Tp); cannam@95: TF = FNMS(KP1_414213562, TE, TD); cannam@95: TJ = FMA(KP1_414213562, TE, TD); cannam@95: TK = FMA(KP414213562, TG, TH); cannam@95: TI = FNMS(KP414213562, TH, TG); cannam@95: R1[WS(rs, 6)] = FMA(KP1_847759065, TC, TB); cannam@95: R1[WS(rs, 2)] = FNMS(KP1_847759065, TC, TB); cannam@95: R1[WS(rs, 7)] = FMA(KP1_847759065, TK, TJ); cannam@95: R1[WS(rs, 3)] = FNMS(KP1_847759065, TK, TJ); cannam@95: } cannam@95: } cannam@95: R0[WS(rs, 1)] = FMA(KP1_414213562, TS, TN); cannam@95: R0[WS(rs, 5)] = FNMS(KP1_414213562, TS, TN); cannam@95: R1[WS(rs, 5)] = FMA(KP1_847759065, TI, TF); cannam@95: R1[WS(rs, 1)] = FNMS(KP1_847759065, TI, TF); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 16, "r2cb_16", {26, 0, 32, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cb_16) (planner *p) { cannam@95: X(kr2c_register) (p, r2cb_16, &desc); cannam@95: } cannam@95: cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 16 -name r2cb_16 -include r2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 58 FP additions, 18 FP multiplications, cannam@95: * (or, 54 additions, 14 multiplications, 4 fused multiply/add), cannam@95: * 31 stack variables, 4 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "r2cb.h" cannam@95: cannam@95: static void r2cb_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(KP1_847759065, +1.847759065022573512256366378793576573644833252); cannam@95: DK(KP765366864, +0.765366864730179543456919968060797733522689125); cannam@95: DK(KP1_414213562, +1.414213562373095048801688724209698078569671875); cannam@95: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT i; cannam@95: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(64, rs), MAKE_VOLATILE_STRIDE(64, csr), MAKE_VOLATILE_STRIDE(64, csi)) { cannam@95: E T9, TS, Tl, TG, T6, TR, Ti, TD, Td, Tq, Tg, Tt, Tn, Tu, TV; cannam@95: E TU, TN, TK; cannam@95: { cannam@95: E T7, T8, TE, Tj, Tk, TF; cannam@95: T7 = Cr[WS(csr, 2)]; cannam@95: T8 = Cr[WS(csr, 6)]; cannam@95: TE = T7 - T8; cannam@95: Tj = Ci[WS(csi, 2)]; cannam@95: Tk = Ci[WS(csi, 6)]; cannam@95: TF = Tj + Tk; cannam@95: T9 = KP2_000000000 * (T7 + T8); cannam@95: TS = KP1_414213562 * (TE + TF); cannam@95: Tl = KP2_000000000 * (Tj - Tk); cannam@95: TG = KP1_414213562 * (TE - TF); cannam@95: } cannam@95: { cannam@95: E T5, TC, T3, TA; cannam@95: { cannam@95: E T4, TB, T1, T2; cannam@95: T4 = Cr[WS(csr, 4)]; cannam@95: T5 = KP2_000000000 * T4; cannam@95: TB = Ci[WS(csi, 4)]; cannam@95: TC = KP2_000000000 * TB; cannam@95: T1 = Cr[0]; cannam@95: T2 = Cr[WS(csr, 8)]; cannam@95: T3 = T1 + T2; cannam@95: TA = T1 - T2; cannam@95: } cannam@95: T6 = T3 + T5; cannam@95: TR = TA + TC; cannam@95: Ti = T3 - T5; cannam@95: TD = TA - TC; cannam@95: } cannam@95: { cannam@95: E TI, TM, TL, TJ; cannam@95: { cannam@95: E Tb, Tc, To, Tp; cannam@95: Tb = Cr[WS(csr, 1)]; cannam@95: Tc = Cr[WS(csr, 7)]; cannam@95: Td = Tb + Tc; cannam@95: TI = Tb - Tc; cannam@95: To = Ci[WS(csi, 1)]; cannam@95: Tp = Ci[WS(csi, 7)]; cannam@95: Tq = To - Tp; cannam@95: TM = To + Tp; cannam@95: } cannam@95: { cannam@95: E Te, Tf, Tr, Ts; cannam@95: Te = Cr[WS(csr, 5)]; cannam@95: Tf = Cr[WS(csr, 3)]; cannam@95: Tg = Te + Tf; cannam@95: TL = Te - Tf; cannam@95: Tr = Ci[WS(csi, 5)]; cannam@95: Ts = Ci[WS(csi, 3)]; cannam@95: Tt = Tr - Ts; cannam@95: TJ = Tr + Ts; cannam@95: } cannam@95: Tn = Td - Tg; cannam@95: Tu = Tq - Tt; cannam@95: TV = TM - TL; cannam@95: TU = TI + TJ; cannam@95: TN = TL + TM; cannam@95: TK = TI - TJ; cannam@95: } cannam@95: { cannam@95: E Ta, Th, TT, TW; cannam@95: Ta = T6 + T9; cannam@95: Th = KP2_000000000 * (Td + Tg); cannam@95: R0[WS(rs, 4)] = Ta - Th; cannam@95: R0[0] = Ta + Th; cannam@95: TT = TR - TS; cannam@95: TW = FNMS(KP1_847759065, TV, KP765366864 * TU); cannam@95: R1[WS(rs, 5)] = TT - TW; cannam@95: R1[WS(rs, 1)] = TT + TW; cannam@95: } cannam@95: { cannam@95: E TX, TY, Tm, Tv; cannam@95: TX = TR + TS; cannam@95: TY = FMA(KP1_847759065, TU, KP765366864 * TV); cannam@95: R1[WS(rs, 3)] = TX - TY; cannam@95: R1[WS(rs, 7)] = TX + TY; cannam@95: Tm = Ti - Tl; cannam@95: Tv = KP1_414213562 * (Tn - Tu); cannam@95: R0[WS(rs, 5)] = Tm - Tv; cannam@95: R0[WS(rs, 1)] = Tm + Tv; cannam@95: } cannam@95: { cannam@95: E Tw, Tx, TH, TO; cannam@95: Tw = Ti + Tl; cannam@95: Tx = KP1_414213562 * (Tn + Tu); cannam@95: R0[WS(rs, 3)] = Tw - Tx; cannam@95: R0[WS(rs, 7)] = Tw + Tx; cannam@95: TH = TD + TG; cannam@95: TO = FNMS(KP765366864, TN, KP1_847759065 * TK); cannam@95: R1[WS(rs, 4)] = TH - TO; cannam@95: R1[0] = TH + TO; cannam@95: } cannam@95: { cannam@95: E TP, TQ, Ty, Tz; cannam@95: TP = TD - TG; cannam@95: TQ = FMA(KP765366864, TK, KP1_847759065 * TN); cannam@95: R1[WS(rs, 2)] = TP - TQ; cannam@95: R1[WS(rs, 6)] = TP + TQ; cannam@95: Ty = T6 - T9; cannam@95: Tz = KP2_000000000 * (Tt + Tq); cannam@95: R0[WS(rs, 2)] = Ty - Tz; cannam@95: R0[WS(rs, 6)] = Ty + Tz; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const kr2c_desc desc = { 16, "r2cb_16", {54, 14, 4, 0}, &GENUS }; cannam@95: cannam@95: void X(codelet_r2cb_16) (planner *p) { cannam@95: X(kr2c_register) (p, r2cb_16, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */