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 08:06:35 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 4 -dit -name hf2_4 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 24 FP additions, 16 FP multiplications, cannam@167: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@167: * 21 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { 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(8, rs)) { cannam@167: E T2, T6, T3, T5, T7, Tb, T4, Ta; cannam@167: T2 = W[0]; cannam@167: T6 = W[3]; cannam@167: T3 = W[2]; cannam@167: T4 = T2 * T3; cannam@167: Ta = T2 * T6; cannam@167: T5 = W[1]; cannam@167: T7 = FMA(T5, T6, T4); cannam@167: Tb = FNMS(T5, T3, Ta); cannam@167: { cannam@167: E T1, Tx, Td, Tw, Ti, Tq, Tm, Ts; cannam@167: T1 = cr[0]; cannam@167: Tx = ci[0]; cannam@167: { cannam@167: E T8, T9, Tc, Tv; cannam@167: T8 = cr[WS(rs, 2)]; cannam@167: T9 = T7 * T8; cannam@167: Tc = ci[WS(rs, 2)]; cannam@167: Tv = T7 * Tc; cannam@167: Td = FMA(Tb, Tc, T9); cannam@167: Tw = FNMS(Tb, T8, Tv); cannam@167: } cannam@167: { cannam@167: E Tf, Tg, Th, Tp; cannam@167: Tf = cr[WS(rs, 1)]; cannam@167: Tg = T2 * Tf; cannam@167: Th = ci[WS(rs, 1)]; cannam@167: Tp = T2 * Th; cannam@167: Ti = FMA(T5, Th, Tg); cannam@167: Tq = FNMS(T5, Tf, Tp); cannam@167: } cannam@167: { cannam@167: E Tj, Tk, Tl, Tr; cannam@167: Tj = cr[WS(rs, 3)]; cannam@167: Tk = T3 * Tj; cannam@167: Tl = ci[WS(rs, 3)]; cannam@167: Tr = T3 * Tl; cannam@167: Tm = FMA(T6, Tl, Tk); cannam@167: Ts = FNMS(T6, Tj, Tr); cannam@167: } cannam@167: { cannam@167: E Te, Tn, To, Tt; cannam@167: Te = T1 + Td; cannam@167: Tn = Ti + Tm; cannam@167: ci[WS(rs, 1)] = Te - Tn; cannam@167: cr[0] = Te + Tn; cannam@167: To = T1 - Td; cannam@167: Tt = Tq - Ts; cannam@167: ci[0] = To - Tt; cannam@167: cr[WS(rs, 1)] = To + Tt; cannam@167: } cannam@167: { cannam@167: E Tu, Ty, Tz, TA; cannam@167: Tu = Tq + Ts; cannam@167: Ty = Tw + Tx; cannam@167: cr[WS(rs, 2)] = Tu - Ty; cannam@167: ci[WS(rs, 3)] = Tu + Ty; cannam@167: Tz = Tm - Ti; cannam@167: TA = Tx - Tw; cannam@167: cr[WS(rs, 3)] = Tz - TA; cannam@167: ci[WS(rs, 2)] = Tz + TA; 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 = { 4, "hf2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@167: cannam@167: void X(codelet_hf2_4) (planner *p) { cannam@167: X(khc2hc_register) (p, hf2_4, &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 4 -dit -name hf2_4 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 24 FP additions, 16 FP multiplications, cannam@167: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@167: * 21 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { 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(8, rs)) { cannam@167: E T2, T4, T3, T5, T6, T8; cannam@167: T2 = W[0]; cannam@167: T4 = W[1]; cannam@167: T3 = W[2]; cannam@167: T5 = W[3]; cannam@167: T6 = FMA(T2, T3, T4 * T5); cannam@167: T8 = FNMS(T4, T3, T2 * T5); cannam@167: { cannam@167: E T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; cannam@167: T1 = cr[0]; cannam@167: Tp = ci[0]; cannam@167: T7 = cr[WS(rs, 2)]; cannam@167: T9 = ci[WS(rs, 2)]; cannam@167: Ta = FMA(T6, T7, T8 * T9); cannam@167: To = FNMS(T8, T7, T6 * T9); cannam@167: { cannam@167: E Tc, Td, Tf, Tg; cannam@167: Tc = cr[WS(rs, 1)]; cannam@167: Td = ci[WS(rs, 1)]; cannam@167: Te = FMA(T2, Tc, T4 * Td); cannam@167: Tk = FNMS(T4, Tc, T2 * Td); cannam@167: Tf = cr[WS(rs, 3)]; cannam@167: Tg = ci[WS(rs, 3)]; cannam@167: Th = FMA(T3, Tf, T5 * Tg); cannam@167: Tl = FNMS(T5, Tf, T3 * Tg); cannam@167: } cannam@167: { cannam@167: E Tb, Ti, Tj, Tm; cannam@167: Tb = T1 + Ta; cannam@167: Ti = Te + Th; cannam@167: ci[WS(rs, 1)] = Tb - Ti; cannam@167: cr[0] = Tb + Ti; cannam@167: Tj = T1 - Ta; cannam@167: Tm = Tk - Tl; cannam@167: ci[0] = Tj - Tm; cannam@167: cr[WS(rs, 1)] = Tj + Tm; cannam@167: } cannam@167: { cannam@167: E Tn, Tq, Tr, Ts; cannam@167: Tn = Tk + Tl; cannam@167: Tq = To + Tp; cannam@167: cr[WS(rs, 2)] = Tn - Tq; cannam@167: ci[WS(rs, 3)] = Tn + Tq; cannam@167: Tr = Th - Te; cannam@167: Ts = Tp - To; cannam@167: cr[WS(rs, 3)] = Tr - Ts; cannam@167: ci[WS(rs, 2)] = Tr + Ts; 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 = { 4, "hf2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@167: cannam@167: void X(codelet_hf2_4) (planner *p) { cannam@167: X(khc2hc_register) (p, hf2_4, &desc); cannam@167: } cannam@167: #endif