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 08:08:00 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 -sign 1 -n 4 -dif -name hc2cbdft2_4 -include rdft/scalar/hc2cb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 30 FP additions, 12 FP multiplications, cannam@167: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), cannam@167: * 23 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cb.h" cannam@167: cannam@167: static void hc2cbdft2_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) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T3, Tm, T6, Tn, Td, Tk, TB, Ty, Tv, Ts; cannam@167: { cannam@167: E Tg, Tc, T9, Tj; cannam@167: { cannam@167: E T1, T2, Ta, Tb; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 1)]; cannam@167: T3 = T1 + T2; cannam@167: Tg = T1 - T2; cannam@167: Ta = Ip[0]; cannam@167: Tb = Im[WS(rs, 1)]; cannam@167: Tc = Ta + Tb; cannam@167: Tm = Ta - Tb; 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: T9 = T4 - T5; cannam@167: Th = Ip[WS(rs, 1)]; cannam@167: Ti = Im[0]; cannam@167: Tj = Th + Ti; cannam@167: Tn = Th - Ti; cannam@167: } cannam@167: Td = T9 + Tc; cannam@167: Tk = Tg - Tj; cannam@167: TB = Tg + Tj; cannam@167: Ty = Tc - T9; cannam@167: Tv = Tm - Tn; cannam@167: Ts = T3 - T6; cannam@167: } cannam@167: { cannam@167: E T7, To, Te, Tp, T8, Tl, Tq, Tf; cannam@167: T7 = T3 + T6; cannam@167: To = Tm + Tn; cannam@167: T8 = W[0]; cannam@167: Te = T8 * Td; cannam@167: Tp = T8 * Tk; cannam@167: Tf = W[1]; cannam@167: Tl = FMA(Tf, Tk, Te); cannam@167: Tq = FNMS(Tf, Td, Tp); cannam@167: Rp[0] = T7 - Tl; cannam@167: Ip[0] = To + Tq; cannam@167: Rm[0] = T7 + Tl; cannam@167: Im[0] = Tq - To; cannam@167: } cannam@167: { cannam@167: E Tr, Tt, Tu, TD, Tz, TF, Tx; cannam@167: Tr = W[2]; cannam@167: Tt = Tr * Ts; cannam@167: Tu = W[3]; cannam@167: TD = Tu * Ts; cannam@167: Tx = W[4]; cannam@167: Tz = Tx * Ty; cannam@167: TF = Tx * TB; cannam@167: { cannam@167: E Tw, TE, TC, TG, TA; cannam@167: Tw = FNMS(Tu, Tv, Tt); cannam@167: TE = FMA(Tr, Tv, TD); cannam@167: TA = W[5]; cannam@167: TC = FMA(TA, TB, Tz); cannam@167: TG = FNMS(TA, Ty, TF); cannam@167: Rp[WS(rs, 1)] = Tw - TC; cannam@167: Ip[WS(rs, 1)] = TE + TG; cannam@167: Rm[WS(rs, 1)] = Tw + TC; cannam@167: Im[WS(rs, 1)] = TG - TE; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 4}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 4, "hc2cbdft2_4", twinstr, &GENUS, {24, 6, 6, 0} }; cannam@167: cannam@167: void X(codelet_hc2cbdft2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cbdft2_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 -sign 1 -n 4 -dif -name hc2cbdft2_4 -include rdft/scalar/hc2cb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 30 FP additions, 12 FP multiplications, cannam@167: * (or, 24 additions, 6 multiplications, 6 fused multiply/add), cannam@167: * 19 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cb.h" cannam@167: cannam@167: static void hc2cbdft2_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) * 6); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E T3, Tl, T6, Tm, Td, Tj, Tx, Tv, Ts, Tq; cannam@167: { cannam@167: E Tf, Tc, T9, Ti; cannam@167: { cannam@167: E T1, T2, Ta, Tb; cannam@167: T1 = Rp[0]; cannam@167: T2 = Rm[WS(rs, 1)]; cannam@167: T3 = T1 + T2; cannam@167: Tf = T1 - T2; cannam@167: Ta = Ip[0]; cannam@167: Tb = Im[WS(rs, 1)]; cannam@167: Tc = Ta + Tb; cannam@167: Tl = Ta - Tb; cannam@167: } cannam@167: { cannam@167: E T4, T5, Tg, Th; cannam@167: T4 = Rp[WS(rs, 1)]; cannam@167: T5 = Rm[0]; cannam@167: T6 = T4 + T5; cannam@167: T9 = T4 - T5; cannam@167: Tg = Ip[WS(rs, 1)]; cannam@167: Th = Im[0]; cannam@167: Ti = Tg + Th; cannam@167: Tm = Tg - Th; cannam@167: } cannam@167: Td = T9 + Tc; cannam@167: Tj = Tf - Ti; cannam@167: Tx = Tf + Ti; cannam@167: Tv = Tc - T9; cannam@167: Ts = Tl - Tm; cannam@167: Tq = T3 - T6; cannam@167: } cannam@167: { cannam@167: E T7, Tn, Tk, To, T8, Te; cannam@167: T7 = T3 + T6; cannam@167: Tn = Tl + Tm; cannam@167: T8 = W[0]; cannam@167: Te = W[1]; cannam@167: Tk = FMA(T8, Td, Te * Tj); cannam@167: To = FNMS(Te, Td, T8 * Tj); cannam@167: Rp[0] = T7 - Tk; cannam@167: Ip[0] = Tn + To; cannam@167: Rm[0] = T7 + Tk; cannam@167: Im[0] = To - Tn; cannam@167: } cannam@167: { cannam@167: E Tt, Tz, Ty, TA; cannam@167: { cannam@167: E Tp, Tr, Tu, Tw; cannam@167: Tp = W[2]; cannam@167: Tr = W[3]; cannam@167: Tt = FNMS(Tr, Ts, Tp * Tq); cannam@167: Tz = FMA(Tr, Tq, Tp * Ts); cannam@167: Tu = W[4]; cannam@167: Tw = W[5]; cannam@167: Ty = FMA(Tu, Tv, Tw * Tx); cannam@167: TA = FNMS(Tw, Tv, Tu * Tx); cannam@167: } cannam@167: Rp[WS(rs, 1)] = Tt - Ty; cannam@167: Ip[WS(rs, 1)] = Tz + TA; cannam@167: Rm[WS(rs, 1)] = Tt + Ty; cannam@167: Im[WS(rs, 1)] = TA - Tz; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 4}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 4, "hc2cbdft2_4", twinstr, &GENUS, {24, 6, 6, 0} }; cannam@167: cannam@167: void X(codelet_hc2cbdft2_4) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cbdft2_4, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #endif