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 3 -dit -name hf_3 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 16 FP additions, 14 FP multiplications, cannam@167: * (or, 6 additions, 4 multiplications, 10 fused multiply/add), cannam@167: * 15 stack variables, 2 constants, and 12 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); 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(6, rs)) { cannam@167: E T1, Tl, T7, Th, Td, Tj; cannam@167: T1 = cr[0]; cannam@167: Tl = ci[0]; cannam@167: { cannam@167: E T3, T6, T4, Tg, T2, T5; 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: Tg = T2 * T6; cannam@167: T5 = W[1]; cannam@167: T7 = FMA(T5, T6, T4); cannam@167: Th = FNMS(T5, T3, Tg); cannam@167: } cannam@167: { cannam@167: E T9, Tc, Ta, Ti, T8, Tb; cannam@167: T9 = cr[WS(rs, 2)]; cannam@167: Tc = ci[WS(rs, 2)]; cannam@167: T8 = W[2]; cannam@167: Ta = T8 * T9; cannam@167: Ti = T8 * Tc; cannam@167: Tb = W[3]; cannam@167: Td = FMA(Tb, Tc, Ta); cannam@167: Tj = FNMS(Tb, T9, Ti); cannam@167: } cannam@167: { cannam@167: E Tk, Te, Tf, To, Tm, Tn; cannam@167: Tk = Th - Tj; cannam@167: Te = T7 + Td; cannam@167: Tf = FNMS(KP500000000, Te, T1); cannam@167: cr[0] = T1 + Te; cannam@167: ci[0] = FNMS(KP866025403, Tk, Tf); cannam@167: cr[WS(rs, 1)] = FMA(KP866025403, Tk, Tf); cannam@167: To = Td - T7; cannam@167: Tm = Th + Tj; cannam@167: Tn = FNMS(KP500000000, Tm, Tl); cannam@167: cr[WS(rs, 2)] = FMS(KP866025403, To, Tn); cannam@167: ci[WS(rs, 2)] = Tm + Tl; cannam@167: ci[WS(rs, 1)] = FMA(KP866025403, To, Tn); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {6, 4, 10, 0} }; cannam@167: cannam@167: void X(codelet_hf_3) (planner *p) { cannam@167: X(khc2hc_register) (p, hf_3, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -n 3 -dit -name hf_3 -include rdft/scalar/hf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 16 FP additions, 12 FP multiplications, cannam@167: * (or, 10 additions, 6 multiplications, 6 fused multiply/add), cannam@167: * 15 stack variables, 2 constants, and 12 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hf.h" cannam@167: cannam@167: static void hf_3(R *cr, R *ci, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); 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(6, rs)) { cannam@167: E T1, Ti, T6, Te, Tb, Tf, Tc, Tj; cannam@167: T1 = cr[0]; cannam@167: Ti = ci[0]; 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: Te = FNMS(T4, T3, T2 * T5); cannam@167: } cannam@167: { cannam@167: E T8, Ta, T7, T9; cannam@167: T8 = cr[WS(rs, 2)]; cannam@167: Ta = ci[WS(rs, 2)]; cannam@167: T7 = W[2]; cannam@167: T9 = W[3]; cannam@167: Tb = FMA(T7, T8, T9 * Ta); cannam@167: Tf = FNMS(T9, T8, T7 * Ta); cannam@167: } cannam@167: Tc = T6 + Tb; cannam@167: Tj = Te + Tf; cannam@167: { cannam@167: E Td, Tg, Th, Tk; cannam@167: cr[0] = T1 + Tc; cannam@167: Td = FNMS(KP500000000, Tc, T1); cannam@167: Tg = KP866025403 * (Te - Tf); cannam@167: ci[0] = Td - Tg; cannam@167: cr[WS(rs, 1)] = Td + Tg; cannam@167: ci[WS(rs, 2)] = Tj + Ti; cannam@167: Th = KP866025403 * (Tb - T6); cannam@167: Tk = FNMS(KP500000000, Tj, Ti); cannam@167: cr[WS(rs, 2)] = Th - Tk; cannam@167: ci[WS(rs, 1)] = Th + Tk; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 3}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 3, "hf_3", twinstr, &GENUS, {10, 6, 6, 0} }; cannam@167: cannam@167: void X(codelet_hf_3) (planner *p) { cannam@167: X(khc2hc_register) (p, hf_3, &desc); cannam@167: } cannam@167: #endif