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 08:06:28 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 -n 5 -dit -name hf_5 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 40 FP additions, 34 FP multiplications, cannam@167: * (or, 14 additions, 8 multiplications, 26 fused multiply/add), cannam@167: * 31 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf_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) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@167: E T1, TJ, T7, Tx, Td, Tz, Te, TK, Tk, TC, Tq, TE, Tr, TL; cannam@167: T1 = cr[0]; cannam@167: TJ = ci[0]; cannam@167: { cannam@167: E T3, T6, T4, Tw, T9, Tc, Ta, Ty, T2, T8, T5, Tb; cannam@167: T3 = cr[WS(rs, 1)]; cannam@167: T6 = ci[WS(rs, 1)]; cannam@167: T2 = W[0]; cannam@167: T4 = T2 * T3; cannam@167: Tw = T2 * T6; cannam@167: T9 = cr[WS(rs, 4)]; cannam@167: Tc = ci[WS(rs, 4)]; cannam@167: T8 = W[6]; cannam@167: Ta = T8 * T9; cannam@167: Ty = T8 * Tc; cannam@167: T5 = W[1]; cannam@167: T7 = FMA(T5, T6, T4); cannam@167: Tx = FNMS(T5, T3, Tw); cannam@167: Tb = W[7]; cannam@167: Td = FMA(Tb, Tc, Ta); cannam@167: Tz = FNMS(Tb, T9, Ty); cannam@167: Te = T7 + Td; cannam@167: TK = Tx + Tz; cannam@167: } cannam@167: { cannam@167: E Tg, Tj, Th, TB, Tm, Tp, Tn, TD, Tf, Tl, Ti, To; cannam@167: Tg = cr[WS(rs, 2)]; cannam@167: Tj = ci[WS(rs, 2)]; cannam@167: Tf = W[2]; cannam@167: Th = Tf * Tg; cannam@167: TB = Tf * Tj; cannam@167: Tm = cr[WS(rs, 3)]; cannam@167: Tp = ci[WS(rs, 3)]; cannam@167: Tl = W[4]; cannam@167: Tn = Tl * Tm; cannam@167: TD = Tl * Tp; cannam@167: Ti = W[3]; cannam@167: Tk = FMA(Ti, Tj, Th); cannam@167: TC = FNMS(Ti, Tg, TB); cannam@167: To = W[5]; cannam@167: Tq = FMA(To, Tp, Tn); cannam@167: TE = FNMS(To, Tm, TD); cannam@167: Tr = Tk + Tq; cannam@167: TL = TC + TE; cannam@167: } cannam@167: { cannam@167: E Tu, Ts, Tt, TG, TI, TA, TF, Tv, TH; cannam@167: Tu = Te - Tr; cannam@167: Ts = Te + Tr; cannam@167: Tt = FNMS(KP250000000, Ts, T1); cannam@167: TA = Tx - Tz; cannam@167: TF = TC - TE; cannam@167: TG = FMA(KP618033988, TF, TA); cannam@167: TI = FNMS(KP618033988, TA, TF); cannam@167: cr[0] = T1 + Ts; cannam@167: Tv = FMA(KP559016994, Tu, Tt); cannam@167: ci[0] = FNMS(KP951056516, TG, Tv); cannam@167: cr[WS(rs, 1)] = FMA(KP951056516, TG, Tv); cannam@167: TH = FNMS(KP559016994, Tu, Tt); cannam@167: cr[WS(rs, 2)] = FNMS(KP951056516, TI, TH); cannam@167: ci[WS(rs, 1)] = FMA(KP951056516, TI, TH); cannam@167: } cannam@167: { cannam@167: E TO, TM, TN, TS, TU, TQ, TR, TT, TP; cannam@167: TO = TK - TL; cannam@167: TM = TK + TL; cannam@167: TN = FNMS(KP250000000, TM, TJ); cannam@167: TQ = Tk - Tq; cannam@167: TR = Td - T7; cannam@167: TS = FMA(KP618033988, TR, TQ); cannam@167: TU = FNMS(KP618033988, TQ, TR); cannam@167: ci[WS(rs, 4)] = TM + TJ; cannam@167: TT = FMA(KP559016994, TO, TN); cannam@167: cr[WS(rs, 4)] = FMS(KP951056516, TU, TT); cannam@167: ci[WS(rs, 3)] = FMA(KP951056516, TU, TT); cannam@167: TP = FNMS(KP559016994, TO, TN); cannam@167: cr[WS(rs, 3)] = FMS(KP951056516, TS, TP); cannam@167: ci[WS(rs, 2)] = FMA(KP951056516, TS, TP); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 5}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {14, 8, 26, 0} }; cannam@167: cannam@167: void X(codelet_hf_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hf_5, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 5 -dit -name hf_5 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 40 FP additions, 28 FP multiplications, cannam@167: * (or, 26 additions, 14 multiplications, 14 fused multiply/add), cannam@167: * 29 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf_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) * 8); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 8, MAKE_VOLATILE_STRIDE(10, rs)) { cannam@167: E T1, TE, Tu, Tx, TC, TB, TF, TG, TH, Tc, Tn, To; cannam@167: T1 = cr[0]; cannam@167: TE = ci[0]; cannam@167: { cannam@167: E T6, Ts, Tm, Tw, Tb, Tt, Th, Tv; cannam@167: { cannam@167: E T3, T5, T2, T4; cannam@167: T3 = cr[WS(rs, 1)]; cannam@167: T5 = ci[WS(rs, 1)]; cannam@167: T2 = W[0]; cannam@167: T4 = W[1]; cannam@167: T6 = FMA(T2, T3, T4 * T5); cannam@167: Ts = FNMS(T4, T3, T2 * T5); cannam@167: } cannam@167: { cannam@167: E Tj, Tl, Ti, Tk; cannam@167: Tj = cr[WS(rs, 3)]; cannam@167: Tl = ci[WS(rs, 3)]; cannam@167: Ti = W[4]; cannam@167: Tk = W[5]; cannam@167: Tm = FMA(Ti, Tj, Tk * Tl); cannam@167: Tw = FNMS(Tk, Tj, Ti * Tl); cannam@167: } cannam@167: { cannam@167: E T8, Ta, T7, T9; cannam@167: T8 = cr[WS(rs, 4)]; cannam@167: Ta = ci[WS(rs, 4)]; cannam@167: T7 = W[6]; cannam@167: T9 = W[7]; cannam@167: Tb = FMA(T7, T8, T9 * Ta); cannam@167: Tt = FNMS(T9, T8, T7 * Ta); cannam@167: } cannam@167: { cannam@167: E Te, Tg, Td, Tf; cannam@167: Te = cr[WS(rs, 2)]; cannam@167: Tg = ci[WS(rs, 2)]; cannam@167: Td = W[2]; cannam@167: Tf = W[3]; cannam@167: Th = FMA(Td, Te, Tf * Tg); cannam@167: Tv = FNMS(Tf, Te, Td * Tg); cannam@167: } cannam@167: Tu = Ts - Tt; cannam@167: Tx = Tv - Tw; cannam@167: TC = Th - Tm; cannam@167: TB = Tb - T6; cannam@167: TF = Ts + Tt; cannam@167: TG = Tv + Tw; cannam@167: TH = TF + TG; cannam@167: Tc = T6 + Tb; cannam@167: Tn = Th + Tm; cannam@167: To = Tc + Tn; cannam@167: } cannam@167: cr[0] = T1 + To; cannam@167: { cannam@167: E Ty, TA, Tr, Tz, Tp, Tq; cannam@167: Ty = FMA(KP951056516, Tu, KP587785252 * Tx); cannam@167: TA = FNMS(KP587785252, Tu, KP951056516 * Tx); cannam@167: Tp = KP559016994 * (Tc - Tn); cannam@167: Tq = FNMS(KP250000000, To, T1); cannam@167: Tr = Tp + Tq; cannam@167: Tz = Tq - Tp; cannam@167: ci[0] = Tr - Ty; cannam@167: ci[WS(rs, 1)] = Tz + TA; cannam@167: cr[WS(rs, 1)] = Tr + Ty; cannam@167: cr[WS(rs, 2)] = Tz - TA; cannam@167: } cannam@167: ci[WS(rs, 4)] = TH + TE; cannam@167: { cannam@167: E TD, TL, TK, TM, TI, TJ; cannam@167: TD = FMA(KP587785252, TB, KP951056516 * TC); cannam@167: TL = FNMS(KP587785252, TC, KP951056516 * TB); cannam@167: TI = FNMS(KP250000000, TH, TE); cannam@167: TJ = KP559016994 * (TF - TG); cannam@167: TK = TI - TJ; cannam@167: TM = TJ + TI; cannam@167: cr[WS(rs, 3)] = TD - TK; cannam@167: ci[WS(rs, 3)] = TL + TM; cannam@167: ci[WS(rs, 2)] = TD + TK; cannam@167: cr[WS(rs, 4)] = TL - TM; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 5}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 5, "hf_5", twinstr, &GENUS, {26, 14, 14, 0} }; cannam@167: cannam@167: void X(codelet_hf_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hf_5, &desc); cannam@167: } cannam@167: #endif