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 08:07:40 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 -sign 1 -twiddle-log3 -precompute-twiddles -n 5 -dif -name hb2_5 -include rdft/scalar/hb.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: * 37 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hb.h" cannam@167: cannam@167: static void hb2_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(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: DK(KP618033988, +0.618033988749894848204586834365638117720309180); 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 T9, TB, Tz, Tm, TC, TO, TG, TJ, TA, TF; cannam@167: T9 = W[0]; cannam@167: TB = W[3]; cannam@167: Tz = W[2]; cannam@167: TA = T9 * Tz; cannam@167: TF = T9 * TB; cannam@167: Tm = W[1]; cannam@167: TC = FNMS(Tm, TB, TA); cannam@167: TO = FNMS(Tm, Tz, TF); cannam@167: TG = FMA(Tm, Tz, TF); cannam@167: TJ = FMA(Tm, TB, TA); cannam@167: { cannam@167: E T1, Tb, TQ, Tw, T8, Ta, Tn, Tj, TL, Ts, Tq, Tr; cannam@167: { cannam@167: E T4, Tu, T7, Tv; cannam@167: T1 = cr[0]; cannam@167: { cannam@167: E T2, T3, T5, T6; cannam@167: T2 = cr[WS(rs, 1)]; cannam@167: T3 = ci[0]; cannam@167: T4 = T2 + T3; cannam@167: Tu = T2 - T3; cannam@167: T5 = cr[WS(rs, 2)]; cannam@167: T6 = ci[WS(rs, 1)]; cannam@167: T7 = T5 + T6; cannam@167: Tv = T5 - T6; cannam@167: } cannam@167: Tb = T4 - T7; cannam@167: TQ = FNMS(KP618033988, Tu, Tv); cannam@167: Tw = FMA(KP618033988, Tv, Tu); cannam@167: T8 = T4 + T7; cannam@167: Ta = FNMS(KP250000000, T8, T1); cannam@167: } cannam@167: { cannam@167: E Tf, To, Ti, Tp; cannam@167: Tn = ci[WS(rs, 4)]; cannam@167: { cannam@167: E Td, Te, Tg, Th; cannam@167: Td = ci[WS(rs, 3)]; cannam@167: Te = cr[WS(rs, 4)]; cannam@167: Tf = Td + Te; cannam@167: To = Td - Te; cannam@167: Tg = ci[WS(rs, 2)]; cannam@167: Th = cr[WS(rs, 3)]; cannam@167: Ti = Tg + Th; cannam@167: Tp = Tg - Th; cannam@167: } cannam@167: Tj = FMA(KP618033988, Ti, Tf); cannam@167: TL = FNMS(KP618033988, Tf, Ti); cannam@167: Ts = To - Tp; cannam@167: Tq = To + Tp; cannam@167: Tr = FNMS(KP250000000, Tq, Tn); cannam@167: } cannam@167: cr[0] = T1 + T8; cannam@167: ci[0] = Tn + Tq; cannam@167: { cannam@167: E Tk, TD, Tx, TH, Tc, Tt; cannam@167: Tc = FMA(KP559016994, Tb, Ta); cannam@167: Tk = FNMS(KP951056516, Tj, Tc); cannam@167: TD = FMA(KP951056516, Tj, Tc); cannam@167: Tt = FMA(KP559016994, Ts, Tr); cannam@167: Tx = FMA(KP951056516, Tw, Tt); cannam@167: TH = FNMS(KP951056516, Tw, Tt); cannam@167: { cannam@167: E Tl, Ty, TE, TI; cannam@167: Tl = T9 * Tk; cannam@167: cr[WS(rs, 1)] = FNMS(Tm, Tx, Tl); cannam@167: Ty = Tm * Tk; cannam@167: ci[WS(rs, 1)] = FMA(T9, Tx, Ty); cannam@167: TE = TC * TD; cannam@167: cr[WS(rs, 4)] = FNMS(TG, TH, TE); cannam@167: TI = TG * TD; cannam@167: ci[WS(rs, 4)] = FMA(TC, TH, TI); cannam@167: } cannam@167: } cannam@167: { cannam@167: E TM, TT, TR, TV, TK, TP; cannam@167: TK = FNMS(KP559016994, Tb, Ta); cannam@167: TM = FMA(KP951056516, TL, TK); cannam@167: TT = FNMS(KP951056516, TL, TK); cannam@167: TP = FNMS(KP559016994, Ts, Tr); cannam@167: TR = FNMS(KP951056516, TQ, TP); cannam@167: TV = FMA(KP951056516, TQ, TP); cannam@167: { cannam@167: E TN, TS, TU, TW; cannam@167: TN = TJ * TM; cannam@167: cr[WS(rs, 2)] = FNMS(TO, TR, TN); cannam@167: TS = TO * TM; cannam@167: ci[WS(rs, 2)] = FMA(TJ, TR, TS); cannam@167: TU = Tz * TT; cannam@167: cr[WS(rs, 3)] = FNMS(TB, TV, TU); cannam@167: TW = TB * TT; cannam@167: ci[WS(rs, 3)] = FMA(Tz, TV, TW); cannam@167: } 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, "hb2_5", twinstr, &GENUS, {14, 10, 30, 0} }; cannam@167: cannam@167: void X(codelet_hb2_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hb2_5, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* 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 */ 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: * 33 stack variables, 4 constants, and 20 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hb.h" cannam@167: cannam@167: static void hb2_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(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@167: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@167: DK(KP559016994, +0.559016994374947424102293417182819058860154590); 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 Th, Tk, Ti, Tl, Tn, TP, Tx, TN; cannam@167: { cannam@167: E Tj, Tw, Tm, Tv; cannam@167: Th = W[0]; cannam@167: Tk = W[1]; cannam@167: Ti = W[2]; cannam@167: Tl = W[3]; cannam@167: Tj = Th * Ti; cannam@167: Tw = Tk * Ti; cannam@167: Tm = Tk * Tl; cannam@167: Tv = Th * Tl; cannam@167: Tn = Tj + Tm; cannam@167: TP = Tv + Tw; cannam@167: Tx = Tv - Tw; cannam@167: TN = Tj - Tm; cannam@167: } cannam@167: { cannam@167: E T1, Tp, TK, TA, T8, To, T9, Tt, TI, TC, Tg, TB; cannam@167: { cannam@167: E T4, Ty, T7, Tz; cannam@167: T1 = cr[0]; cannam@167: { cannam@167: E T2, T3, T5, T6; cannam@167: T2 = cr[WS(rs, 1)]; cannam@167: T3 = ci[0]; cannam@167: T4 = T2 + T3; cannam@167: Ty = T2 - T3; cannam@167: T5 = cr[WS(rs, 2)]; cannam@167: T6 = ci[WS(rs, 1)]; cannam@167: T7 = T5 + T6; cannam@167: Tz = T5 - T6; cannam@167: } cannam@167: Tp = KP559016994 * (T4 - T7); cannam@167: TK = FMA(KP951056516, Ty, KP587785252 * Tz); cannam@167: TA = FNMS(KP951056516, Tz, KP587785252 * Ty); cannam@167: T8 = T4 + T7; cannam@167: To = FNMS(KP250000000, T8, T1); cannam@167: } cannam@167: { cannam@167: E Tc, Tr, Tf, Ts; cannam@167: T9 = ci[WS(rs, 4)]; cannam@167: { cannam@167: E Ta, Tb, Td, Te; cannam@167: Ta = ci[WS(rs, 3)]; cannam@167: Tb = cr[WS(rs, 4)]; cannam@167: Tc = Ta - Tb; cannam@167: Tr = Ta + Tb; cannam@167: Td = ci[WS(rs, 2)]; cannam@167: Te = cr[WS(rs, 3)]; cannam@167: Tf = Td - Te; cannam@167: Ts = Td + Te; cannam@167: } cannam@167: Tt = FNMS(KP951056516, Ts, KP587785252 * Tr); cannam@167: TI = FMA(KP951056516, Tr, KP587785252 * Ts); cannam@167: TC = KP559016994 * (Tc - Tf); cannam@167: Tg = Tc + Tf; cannam@167: TB = FNMS(KP250000000, Tg, T9); cannam@167: } cannam@167: cr[0] = T1 + T8; cannam@167: ci[0] = T9 + Tg; cannam@167: { cannam@167: E Tu, TF, TE, TG, Tq, TD; cannam@167: Tq = To - Tp; cannam@167: Tu = Tq - Tt; cannam@167: TF = Tq + Tt; cannam@167: TD = TB - TC; cannam@167: TE = TA + TD; cannam@167: TG = TD - TA; cannam@167: cr[WS(rs, 2)] = FNMS(Tx, TE, Tn * Tu); cannam@167: ci[WS(rs, 2)] = FMA(Tn, TE, Tx * Tu); cannam@167: cr[WS(rs, 3)] = FNMS(Tl, TG, Ti * TF); cannam@167: ci[WS(rs, 3)] = FMA(Ti, TG, Tl * TF); cannam@167: } cannam@167: { cannam@167: E TJ, TO, TM, TQ, TH, TL; cannam@167: TH = Tp + To; cannam@167: TJ = TH - TI; cannam@167: TO = TH + TI; cannam@167: TL = TC + TB; cannam@167: TM = TK + TL; cannam@167: TQ = TL - TK; cannam@167: cr[WS(rs, 1)] = FNMS(Tk, TM, Th * TJ); cannam@167: ci[WS(rs, 1)] = FMA(Th, TM, Tk * TJ); cannam@167: cr[WS(rs, 4)] = FNMS(TP, TQ, TN * TO); cannam@167: ci[WS(rs, 4)] = FMA(TN, TQ, TP * TO); 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, "hb2_5", twinstr, &GENUS, {30, 18, 14, 0} }; cannam@167: cannam@167: void X(codelet_hb2_5) (planner *p) { cannam@167: X(khc2hc_register) (p, hb2_5, &desc); cannam@167: } cannam@167: #endif