cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:06:37 EDT 2018 */ cannam@167: cannam@167: #include "rdft/codelet-rdft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 5 -dit -name hf2_5 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 44 FP additions, 40 FP multiplications, cannam@167: * (or, 14 additions, 10 multiplications, 30 fused multiply/add), cannam@167: * 38 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf2_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@167: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@167: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@167: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@167: E T2, Ta, T8, T5, Tb, Tm, Tf, Tj, T9, Te; cannam@167: T2 = W[0]; cannam@167: Ta = W[3]; cannam@167: T8 = W[2]; cannam@167: T9 = T2 * T8; cannam@167: Te = T2 * Ta; cannam@167: T5 = W[1]; cannam@167: Tb = FNMS(T5, Ta, T9); cannam@167: Tm = FNMS(T5, T8, Te); cannam@167: Tf = FMA(T5, T8, Te); cannam@167: Tj = FMA(T5, Ta, T9); cannam@167: { cannam@167: E T1, TL, T7, Th, Ti, Tz, TB, TM, To, Ts, Tt, TE, TG, TN; cannam@167: T1 = cr[0]; cannam@167: TL = ci[0]; cannam@167: { cannam@167: E T3, T4, T6, Ty, Tc, Td, Tg, TA; cannam@167: T3 = cr[WS(rs, 1)]; cannam@167: T4 = T2 * T3; cannam@167: T6 = ci[WS(rs, 1)]; cannam@167: Ty = T2 * T6; cannam@167: Tc = cr[WS(rs, 4)]; cannam@167: Td = Tb * Tc; cannam@167: Tg = ci[WS(rs, 4)]; cannam@167: TA = Tb * Tg; cannam@167: T7 = FMA(T5, T6, T4); cannam@167: Th = FMA(Tf, Tg, Td); cannam@167: Ti = T7 + Th; cannam@167: Tz = FNMS(T5, T3, Ty); cannam@167: TB = FNMS(Tf, Tc, TA); cannam@167: TM = Tz + TB; cannam@167: } cannam@167: { cannam@167: E Tk, Tl, Tn, TD, Tp, Tq, Tr, TF; cannam@167: Tk = cr[WS(rs, 2)]; cannam@167: Tl = Tj * Tk; cannam@167: Tn = ci[WS(rs, 2)]; cannam@167: TD = Tj * Tn; cannam@167: Tp = cr[WS(rs, 3)]; cannam@167: Tq = T8 * Tp; cannam@167: Tr = ci[WS(rs, 3)]; cannam@167: TF = T8 * Tr; cannam@167: To = FMA(Tm, Tn, Tl); cannam@167: Ts = FMA(Ta, Tr, Tq); cannam@167: Tt = To + Ts; cannam@167: TE = FNMS(Tm, Tk, TD); cannam@167: TG = FNMS(Ta, Tp, TF); cannam@167: TN = TE + TG; cannam@167: } cannam@167: { cannam@167: E Tw, Tu, Tv, TI, TK, TC, TH, Tx, TJ; cannam@167: Tw = Ti - Tt; cannam@167: Tu = Ti + Tt; cannam@167: Tv = FNMS(KP250000000, Tu, T1); cannam@167: TC = Tz - TB; cannam@167: TH = TE - TG; cannam@167: TI = FMA(KP618033988, TH, TC); cannam@167: TK = FNMS(KP618033988, TC, TH); cannam@167: cr[0] = T1 + Tu; cannam@167: Tx = FMA(KP559016994, Tw, Tv); cannam@167: ci[0] = FNMS(KP951056516, TI, Tx); cannam@167: cr[WS(rs, 1)] = FMA(KP951056516, TI, Tx); cannam@167: TJ = FNMS(KP559016994, Tw, Tv); cannam@167: cr[WS(rs, 2)] = FNMS(KP951056516, TK, TJ); cannam@167: ci[WS(rs, 1)] = FMA(KP951056516, TK, TJ); cannam@167: } cannam@167: { cannam@167: E TQ, TO, TP, TU, TW, TS, TT, TV, TR; cannam@167: TQ = TM - TN; cannam@167: TO = TM + TN; cannam@167: TP = FNMS(KP250000000, TO, TL); cannam@167: TS = To - Ts; cannam@167: TT = Th - T7; cannam@167: TU = FMA(KP618033988, TT, TS); cannam@167: TW = FNMS(KP618033988, TS, TT); cannam@167: ci[WS(rs, 4)] = TO + TL; cannam@167: TV = FMA(KP559016994, TQ, TP); cannam@167: cr[WS(rs, 4)] = FMS(KP951056516, TW, TV); cannam@167: ci[WS(rs, 3)] = FMA(KP951056516, TW, TV); cannam@167: TR = FNMS(KP559016994, TQ, TP); cannam@167: cr[WS(rs, 3)] = FMS(KP951056516, TU, TR); cannam@167: ci[WS(rs, 2)] = FMA(KP951056516, TU, TR); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_CEXP, 1, 1}, cannam@167: {TW_CEXP, 1, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 5, "hf2_5", twinstr, &GENUS, {14, 10, 30, 0} }; cannam@167: cannam@167: void X(codelet_hf2_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hf2_5, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 5 -dit -name hf2_5 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 44 FP additions, 32 FP multiplications, cannam@167: * (or, 30 additions, 18 multiplications, 14 fused multiply/add), cannam@167: * 37 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf2_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@167: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@167: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 4); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 4, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@167: E T2, T4, T7, T9, Tb, Tl, Tf, Tj; cannam@167: { cannam@167: E T8, Te, Ta, Td; cannam@167: T2 = W[0]; cannam@167: T4 = W[1]; cannam@167: T7 = W[2]; cannam@167: T9 = W[3]; cannam@167: T8 = T2 * T7; cannam@167: Te = T4 * T7; cannam@167: Ta = T4 * T9; cannam@167: Td = T2 * T9; cannam@167: Tb = T8 - Ta; cannam@167: Tl = Td - Te; cannam@167: Tf = Td + Te; cannam@167: Tj = T8 + Ta; cannam@167: } cannam@167: { cannam@167: E T1, TI, Ty, TB, TG, TF, TJ, TK, TL, Ti, Tr, Ts; cannam@167: T1 = cr[0]; cannam@167: TI = ci[0]; cannam@167: { cannam@167: E T6, Tw, Tq, TA, Th, Tx, Tn, Tz; cannam@167: { cannam@167: E T3, T5, To, Tp; cannam@167: T3 = cr[WS(rs, 1)]; cannam@167: T5 = ci[WS(rs, 1)]; cannam@167: T6 = FMA(T2, T3, T4 * T5); cannam@167: Tw = FNMS(T4, T3, T2 * T5); cannam@167: To = cr[WS(rs, 3)]; cannam@167: Tp = ci[WS(rs, 3)]; cannam@167: Tq = FMA(T7, To, T9 * Tp); cannam@167: TA = FNMS(T9, To, T7 * Tp); cannam@167: } cannam@167: { cannam@167: E Tc, Tg, Tk, Tm; cannam@167: Tc = cr[WS(rs, 4)]; cannam@167: Tg = ci[WS(rs, 4)]; cannam@167: Th = FMA(Tb, Tc, Tf * Tg); cannam@167: Tx = FNMS(Tf, Tc, Tb * Tg); cannam@167: Tk = cr[WS(rs, 2)]; cannam@167: Tm = ci[WS(rs, 2)]; cannam@167: Tn = FMA(Tj, Tk, Tl * Tm); cannam@167: Tz = FNMS(Tl, Tk, Tj * Tm); cannam@167: } cannam@167: Ty = Tw - Tx; cannam@167: TB = Tz - TA; cannam@167: TG = Tn - Tq; cannam@167: TF = Th - T6; cannam@167: TJ = Tw + Tx; cannam@167: TK = Tz + TA; cannam@167: TL = TJ + TK; cannam@167: Ti = T6 + Th; cannam@167: Tr = Tn + Tq; cannam@167: Ts = Ti + Tr; cannam@167: } cannam@167: cr[0] = T1 + Ts; cannam@167: { cannam@167: E TC, TE, Tv, TD, Tt, Tu; cannam@167: TC = FMA(KP951056516, Ty, KP587785252 * TB); cannam@167: TE = FNMS(KP587785252, Ty, KP951056516 * TB); cannam@167: Tt = KP559016994 * (Ti - Tr); cannam@167: Tu = FNMS(KP250000000, Ts, T1); cannam@167: Tv = Tt + Tu; cannam@167: TD = Tu - Tt; cannam@167: ci[0] = Tv - TC; cannam@167: ci[WS(rs, 1)] = TD + TE; cannam@167: cr[WS(rs, 1)] = Tv + TC; cannam@167: cr[WS(rs, 2)] = TD - TE; cannam@167: } cannam@167: ci[WS(rs, 4)] = TL + TI; cannam@167: { cannam@167: E TH, TP, TO, TQ, TM, TN; cannam@167: TH = FMA(KP587785252, TF, KP951056516 * TG); cannam@167: TP = FNMS(KP587785252, TG, KP951056516 * TF); cannam@167: TM = FNMS(KP250000000, TL, TI); cannam@167: TN = KP559016994 * (TJ - TK); cannam@167: TO = TM - TN; cannam@167: TQ = TN + TM; cannam@167: cr[WS(rs, 3)] = TH - TO; cannam@167: ci[WS(rs, 3)] = TP + TQ; cannam@167: ci[WS(rs, 2)] = TH + TO; cannam@167: cr[WS(rs, 4)] = TP - TQ; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_CEXP, 1, 1}, cannam@167: {TW_CEXP, 1, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 5, "hf2_5", twinstr, &GENUS, {30, 18, 14, 0} }; cannam@167: cannam@167: void X(codelet_hf2_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hf2_5, &desc); cannam@167: } cannam@167: #endif