Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:07:40 EDT 2018 */ Chris@82: Chris@82: #include "rdft/codelet-rdft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 5 -dif -name hb2_5 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 44 FP additions, 40 FP multiplications, Chris@82: * (or, 14 additions, 10 multiplications, 30 fused multiply/add), Chris@82: * 37 stack variables, 4 constants, and 20 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb2_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@82: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@82: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@82: DK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: E T9, TB, Tz, Tm, TC, TO, TG, TJ, TA, TF; Chris@82: T9 = W[0]; Chris@82: TB = W[3]; Chris@82: Tz = W[2]; Chris@82: TA = T9 * Tz; Chris@82: TF = T9 * TB; Chris@82: Tm = W[1]; Chris@82: TC = FNMS(Tm, TB, TA); Chris@82: TO = FNMS(Tm, Tz, TF); Chris@82: TG = FMA(Tm, Tz, TF); Chris@82: TJ = FMA(Tm, TB, TA); Chris@82: { Chris@82: E T1, Tb, TQ, Tw, T8, Ta, Tn, Tj, TL, Ts, Tq, Tr; Chris@82: { Chris@82: E T4, Tu, T7, Tv; Chris@82: T1 = cr[0]; Chris@82: { Chris@82: E T2, T3, T5, T6; Chris@82: T2 = cr[WS(rs, 1)]; Chris@82: T3 = ci[0]; Chris@82: T4 = T2 + T3; Chris@82: Tu = T2 - T3; Chris@82: T5 = cr[WS(rs, 2)]; Chris@82: T6 = ci[WS(rs, 1)]; Chris@82: T7 = T5 + T6; Chris@82: Tv = T5 - T6; Chris@82: } Chris@82: Tb = T4 - T7; Chris@82: TQ = FNMS(KP618033988, Tu, Tv); Chris@82: Tw = FMA(KP618033988, Tv, Tu); Chris@82: T8 = T4 + T7; Chris@82: Ta = FNMS(KP250000000, T8, T1); Chris@82: } Chris@82: { Chris@82: E Tf, To, Ti, Tp; Chris@82: Tn = ci[WS(rs, 4)]; Chris@82: { Chris@82: E Td, Te, Tg, Th; Chris@82: Td = ci[WS(rs, 3)]; Chris@82: Te = cr[WS(rs, 4)]; Chris@82: Tf = Td + Te; Chris@82: To = Td - Te; Chris@82: Tg = ci[WS(rs, 2)]; Chris@82: Th = cr[WS(rs, 3)]; Chris@82: Ti = Tg + Th; Chris@82: Tp = Tg - Th; Chris@82: } Chris@82: Tj = FMA(KP618033988, Ti, Tf); Chris@82: TL = FNMS(KP618033988, Tf, Ti); Chris@82: Ts = To - Tp; Chris@82: Tq = To + Tp; Chris@82: Tr = FNMS(KP250000000, Tq, Tn); Chris@82: } Chris@82: cr[0] = T1 + T8; Chris@82: ci[0] = Tn + Tq; Chris@82: { Chris@82: E Tk, TD, Tx, TH, Tc, Tt; Chris@82: Tc = FMA(KP559016994, Tb, Ta); Chris@82: Tk = FNMS(KP951056516, Tj, Tc); Chris@82: TD = FMA(KP951056516, Tj, Tc); Chris@82: Tt = FMA(KP559016994, Ts, Tr); Chris@82: Tx = FMA(KP951056516, Tw, Tt); Chris@82: TH = FNMS(KP951056516, Tw, Tt); Chris@82: { Chris@82: E Tl, Ty, TE, TI; Chris@82: Tl = T9 * Tk; Chris@82: cr[WS(rs, 1)] = FNMS(Tm, Tx, Tl); Chris@82: Ty = Tm * Tk; Chris@82: ci[WS(rs, 1)] = FMA(T9, Tx, Ty); Chris@82: TE = TC * TD; Chris@82: cr[WS(rs, 4)] = FNMS(TG, TH, TE); Chris@82: TI = TG * TD; Chris@82: ci[WS(rs, 4)] = FMA(TC, TH, TI); Chris@82: } Chris@82: } Chris@82: { Chris@82: E TM, TT, TR, TV, TK, TP; Chris@82: TK = FNMS(KP559016994, Tb, Ta); Chris@82: TM = FMA(KP951056516, TL, TK); Chris@82: TT = FNMS(KP951056516, TL, TK); Chris@82: TP = FNMS(KP559016994, Ts, Tr); Chris@82: TR = FNMS(KP951056516, TQ, TP); Chris@82: TV = FMA(KP951056516, TQ, TP); Chris@82: { Chris@82: E TN, TS, TU, TW; Chris@82: TN = TJ * TM; Chris@82: cr[WS(rs, 2)] = FNMS(TO, TR, TN); Chris@82: TS = TO * TM; Chris@82: ci[WS(rs, 2)] = FMA(TJ, TR, TS); Chris@82: TU = Tz * TT; Chris@82: cr[WS(rs, 3)] = FNMS(TB, TV, TU); Chris@82: TW = TB * TT; Chris@82: ci[WS(rs, 3)] = FMA(Tz, TV, TW); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_CEXP, 1, 1}, Chris@82: {TW_CEXP, 1, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 5, "hb2_5", twinstr, &GENUS, {14, 10, 30, 0} }; Chris@82: Chris@82: void X(codelet_hb2_5) (planner *p) { Chris@82: X(khc2hc_register) (p, hb2_5, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 5 -dif -name hb2_5 -include rdft/scalar/hb.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 44 FP additions, 32 FP multiplications, Chris@82: * (or, 30 additions, 18 multiplications, 14 fused multiply/add), Chris@82: * 33 stack variables, 4 constants, and 20 memory accesses Chris@82: */ Chris@82: #include "rdft/scalar/hb.h" Chris@82: Chris@82: static void hb2_5(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@82: DK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@82: DK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@82: DK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: E Th, Tk, Ti, Tl, Tn, TP, Tx, TN; Chris@82: { Chris@82: E Tj, Tw, Tm, Tv; Chris@82: Th = W[0]; Chris@82: Tk = W[1]; Chris@82: Ti = W[2]; Chris@82: Tl = W[3]; Chris@82: Tj = Th * Ti; Chris@82: Tw = Tk * Ti; Chris@82: Tm = Tk * Tl; Chris@82: Tv = Th * Tl; Chris@82: Tn = Tj + Tm; Chris@82: TP = Tv + Tw; Chris@82: Tx = Tv - Tw; Chris@82: TN = Tj - Tm; Chris@82: } Chris@82: { Chris@82: E T1, Tp, TK, TA, T8, To, T9, Tt, TI, TC, Tg, TB; Chris@82: { Chris@82: E T4, Ty, T7, Tz; Chris@82: T1 = cr[0]; Chris@82: { Chris@82: E T2, T3, T5, T6; Chris@82: T2 = cr[WS(rs, 1)]; Chris@82: T3 = ci[0]; Chris@82: T4 = T2 + T3; Chris@82: Ty = T2 - T3; Chris@82: T5 = cr[WS(rs, 2)]; Chris@82: T6 = ci[WS(rs, 1)]; Chris@82: T7 = T5 + T6; Chris@82: Tz = T5 - T6; Chris@82: } Chris@82: Tp = KP559016994 * (T4 - T7); Chris@82: TK = FMA(KP951056516, Ty, KP587785252 * Tz); Chris@82: TA = FNMS(KP951056516, Tz, KP587785252 * Ty); Chris@82: T8 = T4 + T7; Chris@82: To = FNMS(KP250000000, T8, T1); Chris@82: } Chris@82: { Chris@82: E Tc, Tr, Tf, Ts; Chris@82: T9 = ci[WS(rs, 4)]; Chris@82: { Chris@82: E Ta, Tb, Td, Te; Chris@82: Ta = ci[WS(rs, 3)]; Chris@82: Tb = cr[WS(rs, 4)]; Chris@82: Tc = Ta - Tb; Chris@82: Tr = Ta + Tb; Chris@82: Td = ci[WS(rs, 2)]; Chris@82: Te = cr[WS(rs, 3)]; Chris@82: Tf = Td - Te; Chris@82: Ts = Td + Te; Chris@82: } Chris@82: Tt = FNMS(KP951056516, Ts, KP587785252 * Tr); Chris@82: TI = FMA(KP951056516, Tr, KP587785252 * Ts); Chris@82: TC = KP559016994 * (Tc - Tf); Chris@82: Tg = Tc + Tf; Chris@82: TB = FNMS(KP250000000, Tg, T9); Chris@82: } Chris@82: cr[0] = T1 + T8; Chris@82: ci[0] = T9 + Tg; Chris@82: { Chris@82: E Tu, TF, TE, TG, Tq, TD; Chris@82: Tq = To - Tp; Chris@82: Tu = Tq - Tt; Chris@82: TF = Tq + Tt; Chris@82: TD = TB - TC; Chris@82: TE = TA + TD; Chris@82: TG = TD - TA; Chris@82: cr[WS(rs, 2)] = FNMS(Tx, TE, Tn * Tu); Chris@82: ci[WS(rs, 2)] = FMA(Tn, TE, Tx * Tu); Chris@82: cr[WS(rs, 3)] = FNMS(Tl, TG, Ti * TF); Chris@82: ci[WS(rs, 3)] = FMA(Ti, TG, Tl * TF); Chris@82: } Chris@82: { Chris@82: E TJ, TO, TM, TQ, TH, TL; Chris@82: TH = Tp + To; Chris@82: TJ = TH - TI; Chris@82: TO = TH + TI; Chris@82: TL = TC + TB; Chris@82: TM = TK + TL; Chris@82: TQ = TL - TK; Chris@82: cr[WS(rs, 1)] = FNMS(Tk, TM, Th * TJ); Chris@82: ci[WS(rs, 1)] = FMA(Th, TM, Tk * TJ); Chris@82: cr[WS(rs, 4)] = FNMS(TP, TQ, TN * TO); Chris@82: ci[WS(rs, 4)] = FMA(TN, TQ, TP * TO); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_CEXP, 1, 1}, Chris@82: {TW_CEXP, 1, 3}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const hc2hc_desc desc = { 5, "hb2_5", twinstr, &GENUS, {30, 18, 14, 0} }; Chris@82: Chris@82: void X(codelet_hb2_5) (planner *p) { Chris@82: X(khc2hc_register) (p, hb2_5, &desc); Chris@82: } Chris@82: #endif