cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:50:10 EDT 2016 */ cannam@127: cannam@127: #include "codelet-rdft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2hc.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hb2_4 -include hb.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 33 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hb.h" cannam@127: cannam@127: static void hb2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: E Tg, Tc, Te, To, Tn; cannam@127: { cannam@127: E T7, Tb, T8, Ta; cannam@127: T7 = W[0]; cannam@127: Tb = W[3]; cannam@127: T8 = W[2]; cannam@127: Ta = W[1]; cannam@127: { cannam@127: E Tj, Tm, T3, T6, Tx, Tr, Tz, Tv, Td; cannam@127: { cannam@127: E Tu, T4, Tq, T5, Tp, Tt; cannam@127: { cannam@127: E Tk, Tl, T1, T2; cannam@127: { cannam@127: E Th, Tf, T9, Ti; cannam@127: Th = ci[WS(rs, 3)]; cannam@127: Tf = T7 * Tb; cannam@127: T9 = T7 * T8; cannam@127: Ti = cr[WS(rs, 2)]; cannam@127: Tk = ci[WS(rs, 2)]; cannam@127: Tg = FNMS(Ta, T8, Tf); cannam@127: Tc = FMA(Ta, Tb, T9); cannam@127: Tu = Th + Ti; cannam@127: Tj = Th - Ti; cannam@127: Tl = cr[WS(rs, 3)]; cannam@127: } cannam@127: T1 = cr[0]; cannam@127: T2 = ci[WS(rs, 1)]; cannam@127: T4 = cr[WS(rs, 1)]; cannam@127: Tm = Tk - Tl; cannam@127: Tq = Tk + Tl; cannam@127: T5 = ci[0]; cannam@127: T3 = T1 + T2; cannam@127: Tp = T1 - T2; cannam@127: } cannam@127: Tt = T4 - T5; cannam@127: T6 = T4 + T5; cannam@127: Tx = Tp + Tq; cannam@127: Tr = Tp - Tq; cannam@127: Tz = Tu - Tt; cannam@127: Tv = Tt + Tu; cannam@127: Td = T3 - T6; cannam@127: } cannam@127: { cannam@127: E Ts, Tw, TA, Ty; cannam@127: cr[0] = T3 + T6; cannam@127: Ts = T7 * Tr; cannam@127: ci[0] = Tj + Tm; cannam@127: Tw = T7 * Tv; cannam@127: TA = T8 * Tz; cannam@127: cr[WS(rs, 1)] = FNMS(Ta, Tv, Ts); cannam@127: Ty = T8 * Tx; cannam@127: ci[WS(rs, 1)] = FMA(Ta, Tr, Tw); cannam@127: ci[WS(rs, 3)] = FMA(Tb, Tx, TA); cannam@127: Te = Tc * Td; cannam@127: cr[WS(rs, 3)] = FNMS(Tb, Tz, Ty); cannam@127: To = Tg * Td; cannam@127: Tn = Tj - Tm; cannam@127: } cannam@127: } cannam@127: } cannam@127: ci[WS(rs, 2)] = FMA(Tc, Tn, To); cannam@127: cr[WS(rs, 2)] = FNMS(Tg, Tn, Te); cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 1, 1}, cannam@127: {TW_CEXP, 1, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2hc_desc desc = { 4, "hb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@127: cannam@127: void X(codelet_hb2_4) (planner *p) { cannam@127: X(khc2hc_register) (p, hb2_4, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hb2_4 -include hb.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 21 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "hb.h" cannam@127: cannam@127: static void hb2_4(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: E T7, T9, T8, Ta, Tb, Td; cannam@127: T7 = W[0]; cannam@127: T9 = W[1]; cannam@127: T8 = W[2]; cannam@127: Ta = W[3]; cannam@127: Tb = FMA(T7, T8, T9 * Ta); cannam@127: Td = FNMS(T9, T8, T7 * Ta); cannam@127: { cannam@127: E T3, Tl, T6, To, Tg, Tp, Tj, Tm, Tc, Tk; cannam@127: { cannam@127: E T1, T2, T4, T5; cannam@127: T1 = cr[0]; cannam@127: T2 = ci[WS(rs, 1)]; cannam@127: T3 = T1 + T2; cannam@127: Tl = T1 - T2; cannam@127: T4 = cr[WS(rs, 1)]; cannam@127: T5 = ci[0]; cannam@127: T6 = T4 + T5; cannam@127: To = T4 - T5; cannam@127: } cannam@127: { cannam@127: E Te, Tf, Th, Ti; cannam@127: Te = ci[WS(rs, 3)]; cannam@127: Tf = cr[WS(rs, 2)]; cannam@127: Tg = Te - Tf; cannam@127: Tp = Te + Tf; cannam@127: Th = ci[WS(rs, 2)]; cannam@127: Ti = cr[WS(rs, 3)]; cannam@127: Tj = Th - Ti; cannam@127: Tm = Th + Ti; cannam@127: } cannam@127: cr[0] = T3 + T6; cannam@127: ci[0] = Tg + Tj; cannam@127: Tc = T3 - T6; cannam@127: Tk = Tg - Tj; cannam@127: cr[WS(rs, 2)] = FNMS(Td, Tk, Tb * Tc); cannam@127: ci[WS(rs, 2)] = FMA(Td, Tc, Tb * Tk); cannam@127: { cannam@127: E Tn, Tq, Tr, Ts; cannam@127: Tn = Tl - Tm; cannam@127: Tq = To + Tp; cannam@127: cr[WS(rs, 1)] = FNMS(T9, Tq, T7 * Tn); cannam@127: ci[WS(rs, 1)] = FMA(T7, Tq, T9 * Tn); cannam@127: Tr = Tl + Tm; cannam@127: Ts = Tp - To; cannam@127: cr[WS(rs, 3)] = FNMS(Ta, Ts, T8 * Tr); cannam@127: ci[WS(rs, 3)] = FMA(T8, Ts, Ta * Tr); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: {TW_CEXP, 1, 1}, cannam@127: {TW_CEXP, 1, 3}, cannam@127: {TW_NEXT, 1, 0} cannam@127: }; cannam@127: cannam@127: static const hc2hc_desc desc = { 4, "hb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@127: cannam@127: void X(codelet_hb2_4) (planner *p) { cannam@127: X(khc2hc_register) (p, hb2_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */