Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:04:13 EDT 2018 */ Chris@82: Chris@82: #include "dft/codelet-dft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_twiddle.native -fma -compact -variables 4 -pipeline-latency 4 -n 6 -name t1_6 -include dft/scalar/t.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 32 FP multiplications, Chris@82: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), Chris@82: * 31 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "dft/scalar/t.h" Chris@82: Chris@82: static void t1_6(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: for (m = mb, W = W + (mb * 10); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@82: E T1, TX, T7, TW, Tl, TR, TB, TJ, Ty, TS, TC, TO; Chris@82: T1 = ri[0]; Chris@82: TX = ii[0]; Chris@82: { Chris@82: E T3, T6, T4, TV, T2, T5; Chris@82: T3 = ri[WS(rs, 3)]; Chris@82: T6 = ii[WS(rs, 3)]; Chris@82: T2 = W[4]; Chris@82: T4 = T2 * T3; Chris@82: TV = T2 * T6; Chris@82: T5 = W[5]; Chris@82: T7 = FMA(T5, T6, T4); Chris@82: TW = FNMS(T5, T3, TV); Chris@82: } Chris@82: { Chris@82: E Ta, Td, Tb, TF, Tg, Tj, Th, TH, T9, Tf; Chris@82: Ta = ri[WS(rs, 2)]; Chris@82: Td = ii[WS(rs, 2)]; Chris@82: T9 = W[2]; Chris@82: Tb = T9 * Ta; Chris@82: TF = T9 * Td; Chris@82: Tg = ri[WS(rs, 5)]; Chris@82: Tj = ii[WS(rs, 5)]; Chris@82: Tf = W[8]; Chris@82: Th = Tf * Tg; Chris@82: TH = Tf * Tj; Chris@82: { Chris@82: E Te, TG, Tk, TI, Tc, Ti; Chris@82: Tc = W[3]; Chris@82: Te = FMA(Tc, Td, Tb); Chris@82: TG = FNMS(Tc, Ta, TF); Chris@82: Ti = W[9]; Chris@82: Tk = FMA(Ti, Tj, Th); Chris@82: TI = FNMS(Ti, Tg, TH); Chris@82: Tl = Te - Tk; Chris@82: TR = TG + TI; Chris@82: TB = Te + Tk; Chris@82: TJ = TG - TI; Chris@82: } Chris@82: } Chris@82: { Chris@82: E Tn, Tq, To, TK, Tt, Tw, Tu, TM, Tm, Ts; Chris@82: Tn = ri[WS(rs, 4)]; Chris@82: Tq = ii[WS(rs, 4)]; Chris@82: Tm = W[6]; Chris@82: To = Tm * Tn; Chris@82: TK = Tm * Tq; Chris@82: Tt = ri[WS(rs, 1)]; Chris@82: Tw = ii[WS(rs, 1)]; Chris@82: Ts = W[0]; Chris@82: Tu = Ts * Tt; Chris@82: TM = Ts * Tw; Chris@82: { Chris@82: E Tr, TL, Tx, TN, Tp, Tv; Chris@82: Tp = W[7]; Chris@82: Tr = FMA(Tp, Tq, To); Chris@82: TL = FNMS(Tp, Tn, TK); Chris@82: Tv = W[1]; Chris@82: Tx = FMA(Tv, Tw, Tu); Chris@82: TN = FNMS(Tv, Tt, TM); Chris@82: Ty = Tr - Tx; Chris@82: TS = TL + TN; Chris@82: TC = Tr + Tx; Chris@82: TO = TL - TN; Chris@82: } Chris@82: } Chris@82: { Chris@82: E TP, T8, Tz, TE; Chris@82: TP = TJ - TO; Chris@82: T8 = T1 - T7; Chris@82: Tz = Tl + Ty; Chris@82: TE = FNMS(KP500000000, Tz, T8); Chris@82: ri[WS(rs, 3)] = T8 + Tz; Chris@82: ri[WS(rs, 1)] = FMA(KP866025403, TP, TE); Chris@82: ri[WS(rs, 5)] = FNMS(KP866025403, TP, TE); Chris@82: } Chris@82: { Chris@82: E T14, T11, T12, T13; Chris@82: T14 = Ty - Tl; Chris@82: T11 = TX - TW; Chris@82: T12 = TJ + TO; Chris@82: T13 = FNMS(KP500000000, T12, T11); Chris@82: ii[WS(rs, 1)] = FMA(KP866025403, T14, T13); Chris@82: ii[WS(rs, 3)] = T12 + T11; Chris@82: ii[WS(rs, 5)] = FNMS(KP866025403, T14, T13); Chris@82: } Chris@82: { Chris@82: E TT, TA, TD, TQ; Chris@82: TT = TR - TS; Chris@82: TA = T1 + T7; Chris@82: TD = TB + TC; Chris@82: TQ = FNMS(KP500000000, TD, TA); Chris@82: ri[0] = TA + TD; Chris@82: ri[WS(rs, 4)] = FMA(KP866025403, TT, TQ); Chris@82: ri[WS(rs, 2)] = FNMS(KP866025403, TT, TQ); Chris@82: } Chris@82: { Chris@82: E T10, TU, TY, TZ; Chris@82: T10 = TC - TB; Chris@82: TU = TR + TS; Chris@82: TY = TW + TX; Chris@82: TZ = FNMS(KP500000000, TU, TY); Chris@82: ii[0] = TU + TY; Chris@82: ii[WS(rs, 4)] = FMA(KP866025403, T10, TZ); Chris@82: ii[WS(rs, 2)] = FNMS(KP866025403, T10, TZ); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 0, 6}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 6, "t1_6", twinstr, &GENUS, {24, 10, 22, 0}, 0, 0, 0 }; Chris@82: Chris@82: void X(codelet_t1_6) (planner *p) { Chris@82: X(kdft_dit_register) (p, t1_6, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_twiddle.native -compact -variables 4 -pipeline-latency 4 -n 6 -name t1_6 -include dft/scalar/t.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 28 FP multiplications, Chris@82: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), Chris@82: * 23 stack variables, 2 constants, and 24 memory accesses Chris@82: */ Chris@82: #include "dft/scalar/t.h" Chris@82: Chris@82: static void t1_6(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT m; Chris@82: for (m = mb, W = W + (mb * 10); m < me; m = m + 1, ri = ri + ms, ii = ii + ms, W = W + 10, MAKE_VOLATILE_STRIDE(12, rs)) { Chris@82: E T7, TS, Tv, TO, Tt, TJ, Tx, TF, Ti, TI, Tw, TC; Chris@82: { Chris@82: E T1, TN, T6, TM; Chris@82: T1 = ri[0]; Chris@82: TN = ii[0]; Chris@82: { Chris@82: E T3, T5, T2, T4; Chris@82: T3 = ri[WS(rs, 3)]; Chris@82: T5 = ii[WS(rs, 3)]; Chris@82: T2 = W[4]; Chris@82: T4 = W[5]; Chris@82: T6 = FMA(T2, T3, T4 * T5); Chris@82: TM = FNMS(T4, T3, T2 * T5); Chris@82: } Chris@82: T7 = T1 - T6; Chris@82: TS = TN - TM; Chris@82: Tv = T1 + T6; Chris@82: TO = TM + TN; Chris@82: } Chris@82: { Chris@82: E Tn, TD, Ts, TE; Chris@82: { Chris@82: E Tk, Tm, Tj, Tl; Chris@82: Tk = ri[WS(rs, 4)]; Chris@82: Tm = ii[WS(rs, 4)]; Chris@82: Tj = W[6]; Chris@82: Tl = W[7]; Chris@82: Tn = FMA(Tj, Tk, Tl * Tm); Chris@82: TD = FNMS(Tl, Tk, Tj * Tm); Chris@82: } Chris@82: { Chris@82: E Tp, Tr, To, Tq; Chris@82: Tp = ri[WS(rs, 1)]; Chris@82: Tr = ii[WS(rs, 1)]; Chris@82: To = W[0]; Chris@82: Tq = W[1]; Chris@82: Ts = FMA(To, Tp, Tq * Tr); Chris@82: TE = FNMS(Tq, Tp, To * Tr); Chris@82: } Chris@82: Tt = Tn - Ts; Chris@82: TJ = TD + TE; Chris@82: Tx = Tn + Ts; Chris@82: TF = TD - TE; Chris@82: } Chris@82: { Chris@82: E Tc, TA, Th, TB; Chris@82: { Chris@82: E T9, Tb, T8, Ta; Chris@82: T9 = ri[WS(rs, 2)]; Chris@82: Tb = ii[WS(rs, 2)]; Chris@82: T8 = W[2]; Chris@82: Ta = W[3]; Chris@82: Tc = FMA(T8, T9, Ta * Tb); Chris@82: TA = FNMS(Ta, T9, T8 * Tb); Chris@82: } Chris@82: { Chris@82: E Te, Tg, Td, Tf; Chris@82: Te = ri[WS(rs, 5)]; Chris@82: Tg = ii[WS(rs, 5)]; Chris@82: Td = W[8]; Chris@82: Tf = W[9]; Chris@82: Th = FMA(Td, Te, Tf * Tg); Chris@82: TB = FNMS(Tf, Te, Td * Tg); Chris@82: } Chris@82: Ti = Tc - Th; Chris@82: TI = TA + TB; Chris@82: Tw = Tc + Th; Chris@82: TC = TA - TB; Chris@82: } Chris@82: { Chris@82: E TG, Tu, Tz, TR, TT, TU; Chris@82: TG = KP866025403 * (TC - TF); Chris@82: Tu = Ti + Tt; Chris@82: Tz = FNMS(KP500000000, Tu, T7); Chris@82: ri[WS(rs, 3)] = T7 + Tu; Chris@82: ri[WS(rs, 1)] = Tz + TG; Chris@82: ri[WS(rs, 5)] = Tz - TG; Chris@82: TR = KP866025403 * (Tt - Ti); Chris@82: TT = TC + TF; Chris@82: TU = FNMS(KP500000000, TT, TS); Chris@82: ii[WS(rs, 1)] = TR + TU; Chris@82: ii[WS(rs, 3)] = TT + TS; Chris@82: ii[WS(rs, 5)] = TU - TR; Chris@82: } Chris@82: { Chris@82: E TK, Ty, TH, TQ, TL, TP; Chris@82: TK = KP866025403 * (TI - TJ); Chris@82: Ty = Tw + Tx; Chris@82: TH = FNMS(KP500000000, Ty, Tv); Chris@82: ri[0] = Tv + Ty; Chris@82: ri[WS(rs, 4)] = TH + TK; Chris@82: ri[WS(rs, 2)] = TH - TK; Chris@82: TQ = KP866025403 * (Tx - Tw); Chris@82: TL = TI + TJ; Chris@82: TP = FNMS(KP500000000, TL, TO); Chris@82: ii[0] = TL + TO; Chris@82: ii[WS(rs, 4)] = TQ + TP; Chris@82: ii[WS(rs, 2)] = TP - TQ; Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: {TW_FULL, 0, 6}, Chris@82: {TW_NEXT, 1, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 6, "t1_6", twinstr, &GENUS, {32, 14, 14, 0}, 0, 0, 0 }; Chris@82: Chris@82: void X(codelet_t1_6) (planner *p) { Chris@82: X(kdft_dit_register) (p, t1_6, &desc); Chris@82: } Chris@82: #endif