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 08:07:16 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_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 32 FP additions, 24 FP multiplications, cannam@167: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), cannam@167: * 37 stack variables, 1 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T1, T5, T2, T4, T6, Tc, T3, Tb; cannam@167: T1 = W[0]; cannam@167: T5 = W[3]; cannam@167: T2 = W[2]; cannam@167: T3 = T1 * T2; cannam@167: Tb = T1 * T5; cannam@167: T4 = W[1]; cannam@167: T6 = FMA(T4, T5, T3); cannam@167: Tc = FNMS(T4, T2, Tb); cannam@167: { cannam@167: E Tj, Tp, To, TE, Tw, T9, Tt, Ta, TC, Tf, Tr, Ts, Tx; cannam@167: { cannam@167: E Th, Ti, Tl, Tm, Tn; cannam@167: Th = Ip[0]; cannam@167: Ti = Im[0]; cannam@167: Tj = Th - Ti; cannam@167: Tp = Th + Ti; cannam@167: Tl = Rm[0]; cannam@167: Tm = Rp[0]; cannam@167: Tn = Tl - Tm; cannam@167: To = T1 * Tn; cannam@167: TE = T4 * Tn; cannam@167: Tw = Tm + Tl; cannam@167: } cannam@167: { cannam@167: E T7, T8, Td, Te; cannam@167: T7 = Ip[WS(rs, 1)]; cannam@167: T8 = Im[WS(rs, 1)]; cannam@167: T9 = T7 - T8; cannam@167: Tt = T7 + T8; cannam@167: Ta = T6 * T9; cannam@167: TC = T2 * Tt; cannam@167: Td = Rp[WS(rs, 1)]; cannam@167: Te = Rm[WS(rs, 1)]; cannam@167: Tf = Td + Te; cannam@167: Tr = Td - Te; cannam@167: Ts = T2 * Tr; cannam@167: Tx = T6 * Tf; cannam@167: } cannam@167: { cannam@167: E Tk, TB, Tz, TH, Tv, TA, TG, TI, Tg, Ty; cannam@167: Tg = FNMS(Tc, Tf, Ta); cannam@167: Tk = Tg + Tj; cannam@167: TB = Tj - Tg; cannam@167: Ty = FMA(Tc, T9, Tx); cannam@167: Tz = Tw - Ty; cannam@167: TH = Tw + Ty; cannam@167: { cannam@167: E Tq, Tu, TD, TF; cannam@167: Tq = FNMS(T4, Tp, To); cannam@167: Tu = FMA(T5, Tt, Ts); cannam@167: Tv = Tq - Tu; cannam@167: TA = Tu + Tq; cannam@167: TD = FNMS(T5, Tr, TC); cannam@167: TF = FMA(T1, Tp, TE); cannam@167: TG = TD - TF; cannam@167: TI = TD + TF; cannam@167: } cannam@167: Ip[0] = KP500000000 * (Tk + Tv); cannam@167: Rp[0] = KP500000000 * (TH + TI); cannam@167: Im[WS(rs, 1)] = KP500000000 * (Tv - Tk); cannam@167: Rm[WS(rs, 1)] = KP500000000 * (TH - TI); cannam@167: Rm[0] = KP500000000 * (Tz - TA); cannam@167: Im[0] = KP500000000 * (TG - TB); cannam@167: Rp[WS(rs, 1)] = KP500000000 * (Tz + TA); cannam@167: Ip[WS(rs, 1)] = KP500000000 * (TB + TG); 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, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -twiddle-log3 -precompute-twiddles -n 4 -dit -name hc2cfdft2_4 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 32 FP additions, 24 FP multiplications, cannam@167: * (or, 24 additions, 16 multiplications, 8 fused multiply/add), cannam@167: * 24 stack variables, 1 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft2_4(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { 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, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 4, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T1, T3, T2, T4, T5, T9; cannam@167: T1 = W[0]; cannam@167: T3 = W[1]; cannam@167: T2 = W[2]; cannam@167: T4 = W[3]; cannam@167: T5 = FMA(T1, T2, T3 * T4); cannam@167: T9 = FNMS(T3, T2, T1 * T4); cannam@167: { cannam@167: E Tg, Tr, Tm, Tx, Td, Tw, Tp, Ts; cannam@167: { cannam@167: E Te, Tf, Tl, Ti, Tj, Tk; cannam@167: Te = Ip[0]; cannam@167: Tf = Im[0]; cannam@167: Tl = Te + Tf; cannam@167: Ti = Rm[0]; cannam@167: Tj = Rp[0]; cannam@167: Tk = Ti - Tj; cannam@167: Tg = Te - Tf; cannam@167: Tr = Tj + Ti; cannam@167: Tm = FNMS(T3, Tl, T1 * Tk); cannam@167: Tx = FMA(T3, Tk, T1 * Tl); cannam@167: } cannam@167: { cannam@167: E T8, To, Tc, Tn; cannam@167: { cannam@167: E T6, T7, Ta, Tb; cannam@167: T6 = Ip[WS(rs, 1)]; cannam@167: T7 = Im[WS(rs, 1)]; cannam@167: T8 = T6 - T7; cannam@167: To = T6 + T7; cannam@167: Ta = Rp[WS(rs, 1)]; cannam@167: Tb = Rm[WS(rs, 1)]; cannam@167: Tc = Ta + Tb; cannam@167: Tn = Ta - Tb; cannam@167: } cannam@167: Td = FNMS(T9, Tc, T5 * T8); cannam@167: Tw = FNMS(T4, Tn, T2 * To); cannam@167: Tp = FMA(T2, Tn, T4 * To); cannam@167: Ts = FMA(T5, Tc, T9 * T8); cannam@167: } cannam@167: { cannam@167: E Th, Tq, Tz, TA; cannam@167: Th = Td + Tg; cannam@167: Tq = Tm - Tp; cannam@167: Ip[0] = KP500000000 * (Th + Tq); cannam@167: Im[WS(rs, 1)] = KP500000000 * (Tq - Th); cannam@167: Tz = Tr + Ts; cannam@167: TA = Tw + Tx; cannam@167: Rm[WS(rs, 1)] = KP500000000 * (Tz - TA); cannam@167: Rp[0] = KP500000000 * (Tz + TA); cannam@167: } cannam@167: { cannam@167: E Tt, Tu, Tv, Ty; cannam@167: Tt = Tr - Ts; cannam@167: Tu = Tp + Tm; cannam@167: Rm[0] = KP500000000 * (Tt - Tu); cannam@167: Rp[WS(rs, 1)] = KP500000000 * (Tt + Tu); cannam@167: Tv = Tg - Td; cannam@167: Ty = Tw - Tx; cannam@167: Ip[WS(rs, 1)] = KP500000000 * (Tv + Ty); cannam@167: Im[0] = KP500000000 * (Ty - Tv); 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, "hc2cfdft2_4", twinstr, &GENUS, {24, 16, 8, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft2_4, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #endif