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 07:41:12 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 -sign 1 -n 5 -dif -name hb_5 -include hb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 40 FP additions, 34 FP multiplications, cannam@95: * (or, 14 additions, 8 multiplications, 26 fused multiply/add), cannam@95: * 42 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hb.h" cannam@95: cannam@95: static void hb_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@95: E TQ, TP, TT, TR, TS, TU; cannam@95: { cannam@95: E T1, Tn, TM, Tw, Tb, T8, To, Tf, Ta, Tg, Th; cannam@95: { cannam@95: E T2, T3, T5, T6, T4, Tu; cannam@95: T1 = cr[0]; cannam@95: T2 = cr[WS(rs, 1)]; cannam@95: T3 = ci[0]; cannam@95: T5 = cr[WS(rs, 2)]; cannam@95: T6 = ci[WS(rs, 1)]; cannam@95: Tn = ci[WS(rs, 4)]; cannam@95: T4 = T2 + T3; cannam@95: Tu = T2 - T3; cannam@95: { cannam@95: E T7, Tv, Td, Te; cannam@95: T7 = T5 + T6; cannam@95: Tv = T5 - T6; cannam@95: Td = ci[WS(rs, 3)]; cannam@95: Te = cr[WS(rs, 4)]; cannam@95: TM = FNMS(KP618033988, Tu, Tv); cannam@95: Tw = FMA(KP618033988, Tv, Tu); cannam@95: Tb = T4 - T7; cannam@95: T8 = T4 + T7; cannam@95: To = Td - Te; cannam@95: Tf = Td + Te; cannam@95: Ta = FNMS(KP250000000, T8, T1); cannam@95: Tg = ci[WS(rs, 2)]; cannam@95: Th = cr[WS(rs, 3)]; cannam@95: } cannam@95: } cannam@95: cr[0] = T1 + T8; cannam@95: { cannam@95: E TG, T9, Tm, Tz, TH, TC, TA, Tk, Tt, TL, Tc, Ti, Tp, TI; cannam@95: TG = FNMS(KP559016994, Tb, Ta); cannam@95: Tc = FMA(KP559016994, Tb, Ta); cannam@95: T9 = W[0]; cannam@95: Ti = Tg + Th; cannam@95: Tp = Tg - Th; cannam@95: Tm = W[1]; cannam@95: { cannam@95: E Ts, Tj, Tr, Tq; cannam@95: Tz = W[6]; cannam@95: Ts = To - Tp; cannam@95: Tq = To + Tp; cannam@95: Tj = FMA(KP618033988, Ti, Tf); cannam@95: TH = FNMS(KP618033988, Tf, Ti); cannam@95: ci[0] = Tn + Tq; cannam@95: Tr = FNMS(KP250000000, Tq, Tn); cannam@95: TC = W[7]; cannam@95: TA = FMA(KP951056516, Tj, Tc); cannam@95: Tk = FNMS(KP951056516, Tj, Tc); cannam@95: Tt = FMA(KP559016994, Ts, Tr); cannam@95: TL = FNMS(KP559016994, Ts, Tr); cannam@95: } cannam@95: { cannam@95: E TE, TB, Ty, Tl, TD, Tx; cannam@95: TE = TC * TA; cannam@95: TB = Tz * TA; cannam@95: Ty = Tm * Tk; cannam@95: Tl = T9 * Tk; cannam@95: TD = FNMS(KP951056516, Tw, Tt); cannam@95: Tx = FMA(KP951056516, Tw, Tt); cannam@95: TI = FMA(KP951056516, TH, TG); cannam@95: TQ = FNMS(KP951056516, TH, TG); cannam@95: ci[WS(rs, 4)] = FMA(Tz, TD, TE); cannam@95: cr[WS(rs, 4)] = FNMS(TC, TD, TB); cannam@95: ci[WS(rs, 1)] = FMA(T9, Tx, Ty); cannam@95: cr[WS(rs, 1)] = FNMS(Tm, Tx, Tl); cannam@95: } cannam@95: { cannam@95: E TF, TK, TN, TJ, TO; cannam@95: TF = W[2]; cannam@95: TK = W[3]; cannam@95: TP = W[4]; cannam@95: TT = FMA(KP951056516, TM, TL); cannam@95: TN = FNMS(KP951056516, TM, TL); cannam@95: TJ = TF * TI; cannam@95: TO = TK * TI; cannam@95: TR = TP * TQ; cannam@95: TS = W[5]; cannam@95: cr[WS(rs, 2)] = FNMS(TK, TN, TJ); cannam@95: ci[WS(rs, 2)] = FMA(TF, TN, TO); cannam@95: } cannam@95: } cannam@95: } cannam@95: cr[WS(rs, 3)] = FNMS(TS, TT, TR); cannam@95: TU = TS * TQ; cannam@95: ci[WS(rs, 3)] = FMA(TP, TT, TU); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 5}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 5, "hb_5", twinstr, &GENUS, {14, 8, 26, 0} }; cannam@95: cannam@95: void X(codelet_hb_5) (planner *p) { cannam@95: X(khc2hc_register) (p, hb_5, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 5 -dif -name hb_5 -include hb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 40 FP additions, 28 FP multiplications, cannam@95: * (or, 26 additions, 14 multiplications, 14 fused multiply/add), cannam@95: * 27 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hb.h" cannam@95: cannam@95: static void hb_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@95: E T1, Tj, TG, Ts, T8, Ti, T9, Tn, TD, Tu, Tg, Tt; cannam@95: { cannam@95: E T4, Tq, T7, Tr; cannam@95: T1 = cr[0]; cannam@95: { cannam@95: E T2, T3, T5, T6; cannam@95: T2 = cr[WS(rs, 1)]; cannam@95: T3 = ci[0]; cannam@95: T4 = T2 + T3; cannam@95: Tq = T2 - T3; cannam@95: T5 = cr[WS(rs, 2)]; cannam@95: T6 = ci[WS(rs, 1)]; cannam@95: T7 = T5 + T6; cannam@95: Tr = T5 - T6; cannam@95: } cannam@95: Tj = KP559016994 * (T4 - T7); cannam@95: TG = FMA(KP951056516, Tq, KP587785252 * Tr); cannam@95: Ts = FNMS(KP951056516, Tr, KP587785252 * Tq); cannam@95: T8 = T4 + T7; cannam@95: Ti = FNMS(KP250000000, T8, T1); cannam@95: } cannam@95: { cannam@95: E Tc, Tl, Tf, Tm; cannam@95: T9 = ci[WS(rs, 4)]; cannam@95: { cannam@95: E Ta, Tb, Td, Te; cannam@95: Ta = ci[WS(rs, 3)]; cannam@95: Tb = cr[WS(rs, 4)]; cannam@95: Tc = Ta - Tb; cannam@95: Tl = Ta + Tb; cannam@95: Td = ci[WS(rs, 2)]; cannam@95: Te = cr[WS(rs, 3)]; cannam@95: Tf = Td - Te; cannam@95: Tm = Td + Te; cannam@95: } cannam@95: Tn = FNMS(KP951056516, Tm, KP587785252 * Tl); cannam@95: TD = FMA(KP951056516, Tl, KP587785252 * Tm); cannam@95: Tu = KP559016994 * (Tc - Tf); cannam@95: Tg = Tc + Tf; cannam@95: Tt = FNMS(KP250000000, Tg, T9); cannam@95: } cannam@95: cr[0] = T1 + T8; cannam@95: ci[0] = T9 + Tg; cannam@95: { cannam@95: E To, Ty, Tw, TA, Tk, Tv; cannam@95: Tk = Ti - Tj; cannam@95: To = Tk - Tn; cannam@95: Ty = Tk + Tn; cannam@95: Tv = Tt - Tu; cannam@95: Tw = Ts + Tv; cannam@95: TA = Tv - Ts; cannam@95: { cannam@95: E Th, Tp, Tx, Tz; cannam@95: Th = W[2]; cannam@95: Tp = W[3]; cannam@95: cr[WS(rs, 2)] = FNMS(Tp, Tw, Th * To); cannam@95: ci[WS(rs, 2)] = FMA(Th, Tw, Tp * To); cannam@95: Tx = W[4]; cannam@95: Tz = W[5]; cannam@95: cr[WS(rs, 3)] = FNMS(Tz, TA, Tx * Ty); cannam@95: ci[WS(rs, 3)] = FMA(Tx, TA, Tz * Ty); cannam@95: } cannam@95: } cannam@95: { cannam@95: E TE, TK, TI, TM, TC, TH; cannam@95: TC = Tj + Ti; cannam@95: TE = TC - TD; cannam@95: TK = TC + TD; cannam@95: TH = Tu + Tt; cannam@95: TI = TG + TH; cannam@95: TM = TH - TG; cannam@95: { cannam@95: E TB, TF, TJ, TL; cannam@95: TB = W[0]; cannam@95: TF = W[1]; cannam@95: cr[WS(rs, 1)] = FNMS(TF, TI, TB * TE); cannam@95: ci[WS(rs, 1)] = FMA(TB, TI, TF * TE); cannam@95: TJ = W[6]; cannam@95: TL = W[7]; cannam@95: cr[WS(rs, 4)] = FNMS(TL, TM, TJ * TK); cannam@95: ci[WS(rs, 4)] = FMA(TJ, TM, TL * TK); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 5}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2hc_desc desc = { 5, "hb_5", twinstr, &GENUS, {26, 14, 14, 0} }; cannam@95: cannam@95: void X(codelet_hb_5) (planner *p) { cannam@95: X(khc2hc_register) (p, hb_5, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */