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 08:07:54 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_hc2c.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hc2cb2_4 -include rdft/scalar/hc2cb.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: * 33 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cb.h" cannam@167: cannam@167: static void hc2cb2_4(R *Rp, R *Ip, R *Rm, R *Im, 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T7, Tb, T8, Ta, Tc, Tg, T9, Tf; cannam@167: T7 = W[0]; cannam@167: Tb = W[3]; cannam@167: T8 = W[2]; cannam@167: T9 = T7 * T8; cannam@167: Tf = T7 * Tb; cannam@167: Ta = W[1]; cannam@167: Tc = FMA(Ta, Tb, T9); cannam@167: Tg = FNMS(Ta, T8, Tf); cannam@167: { cannam@167: E T3, T6, Td, Tj, Tz, Tx, Tr, Tm, Tv, Ts, Tw, TA; cannam@167: { cannam@167: E Th, Ti, Tu, Tp, Tk, Tl, Tq, Tt; cannam@167: { cannam@167: E T1, T2, T4, T5; cannam@167: Th = Ip[0]; cannam@167: Ti = Im[WS(rs, 1)]; cannam@167: Tu = Th + Ti; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 1)]; cannam@167: T3 = T1 + T2; cannam@167: Tp = T1 - T2; cannam@167: Tk = Ip[WS(rs, 1)]; cannam@167: Tl = Im[0]; cannam@167: Tq = Tk + Tl; cannam@167: T4 = Rp[WS(rs, 1)]; cannam@167: T5 = Rm[0]; cannam@167: T6 = T4 + T5; cannam@167: Tt = T4 - T5; cannam@167: } cannam@167: Td = T3 - T6; cannam@167: Tj = Th - Ti; cannam@167: Tz = Tu - Tt; cannam@167: Tx = Tp + Tq; cannam@167: Tr = Tp - Tq; cannam@167: Tm = Tk - Tl; cannam@167: Tv = Tt + Tu; cannam@167: } cannam@167: Rp[0] = T3 + T6; cannam@167: Rm[0] = Tj + Tm; cannam@167: Ts = T7 * Tr; cannam@167: Ip[0] = FNMS(Ta, Tv, Ts); cannam@167: Tw = T7 * Tv; cannam@167: Im[0] = FMA(Ta, Tr, Tw); cannam@167: TA = T8 * Tz; cannam@167: Im[WS(rs, 1)] = FMA(Tb, Tx, TA); cannam@167: { cannam@167: E Ty, Te, To, Tn; cannam@167: Ty = T8 * Tx; cannam@167: Ip[WS(rs, 1)] = FNMS(Tb, Tz, Ty); cannam@167: Te = Tc * Td; cannam@167: To = Tg * Td; cannam@167: Tn = Tj - Tm; cannam@167: Rp[WS(rs, 1)] = FNMS(Tg, Tn, Te); cannam@167: Rm[WS(rs, 1)] = FMA(Tc, Tn, 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 hc2c_desc desc = { 4, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@167: cannam@167: void X(codelet_hc2cb2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 4 -dif -name hc2cb2_4 -include rdft/scalar/hc2cb.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/hc2cb.h" cannam@167: cannam@167: static void hc2cb2_4(R *Rp, R *Ip, R *Rm, R *Im, 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T7, T9, T8, Ta, Tb, Td; cannam@167: T7 = W[0]; cannam@167: T9 = W[1]; cannam@167: T8 = W[2]; cannam@167: Ta = W[3]; cannam@167: Tb = FMA(T7, T8, T9 * Ta); cannam@167: Td = FNMS(T9, T8, T7 * Ta); cannam@167: { cannam@167: E T3, Tl, Tg, Tp, T6, To, Tj, Tm, Tc, Tk; cannam@167: { cannam@167: E T1, T2, Te, Tf; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 1)]; cannam@167: T3 = T1 + T2; cannam@167: Tl = T1 - T2; cannam@167: Te = Ip[0]; cannam@167: Tf = Im[WS(rs, 1)]; cannam@167: Tg = Te - Tf; cannam@167: Tp = Te + Tf; cannam@167: } cannam@167: { cannam@167: E T4, T5, Th, Ti; cannam@167: T4 = Rp[WS(rs, 1)]; cannam@167: T5 = Rm[0]; cannam@167: T6 = T4 + T5; cannam@167: To = T4 - T5; cannam@167: Th = Ip[WS(rs, 1)]; cannam@167: Ti = Im[0]; cannam@167: Tj = Th - Ti; cannam@167: Tm = Th + Ti; cannam@167: } cannam@167: Rp[0] = T3 + T6; cannam@167: Rm[0] = Tg + Tj; cannam@167: Tc = T3 - T6; cannam@167: Tk = Tg - Tj; cannam@167: Rp[WS(rs, 1)] = FNMS(Td, Tk, Tb * Tc); cannam@167: Rm[WS(rs, 1)] = FMA(Td, Tc, Tb * Tk); cannam@167: { cannam@167: E Tn, Tq, Tr, Ts; cannam@167: Tn = Tl - Tm; cannam@167: Tq = To + Tp; cannam@167: Ip[0] = FNMS(T9, Tq, T7 * Tn); cannam@167: Im[0] = FMA(T7, Tq, T9 * Tn); cannam@167: Tr = Tl + Tm; cannam@167: Ts = Tp - To; cannam@167: Ip[WS(rs, 1)] = FNMS(Ta, Ts, T8 * Tr); cannam@167: Im[WS(rs, 1)] = FMA(T8, Ts, Ta * Tr); 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 hc2c_desc desc = { 4, "hc2cb2_4", twinstr, &GENUS, {16, 8, 8, 0} }; cannam@167: cannam@167: void X(codelet_hc2cb2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cb2_4, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #endif