Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:40:30 EST 2012 */ Chris@10: Chris@10: #include "codelet-rdft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cf_6 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 32 FP multiplications, Chris@10: * (or, 24 additions, 10 multiplications, 22 fused multiply/add), Chris@10: * 47 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cf_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: E TY, TU, T10, TZ; Chris@10: { Chris@10: E T1, TX, TW, T7, Tn, Tq, TJ, TS, TB, Tl, To, TK, Tt, Tw, Ts; Chris@10: E Tp, Tv; Chris@10: T1 = Rp[0]; Chris@10: TX = Rm[0]; Chris@10: { Chris@10: E T3, T6, T2, T5; Chris@10: T3 = Ip[WS(rs, 1)]; Chris@10: T6 = Im[WS(rs, 1)]; Chris@10: T2 = W[4]; Chris@10: T5 = W[5]; Chris@10: { Chris@10: E Ta, Td, Tg, TF, Tb, Tj, Tf, Tc, Ti, TV, T4, T9; Chris@10: Ta = Rp[WS(rs, 1)]; Chris@10: Td = Rm[WS(rs, 1)]; Chris@10: TV = T2 * T6; Chris@10: T4 = T2 * T3; Chris@10: T9 = W[2]; Chris@10: Tg = Ip[WS(rs, 2)]; Chris@10: TW = FNMS(T5, T3, TV); Chris@10: T7 = FMA(T5, T6, T4); Chris@10: TF = T9 * Td; Chris@10: Tb = T9 * Ta; Chris@10: Tj = Im[WS(rs, 2)]; Chris@10: Tf = W[8]; Chris@10: Tc = W[3]; Chris@10: Ti = W[9]; Chris@10: { Chris@10: E TG, Te, TI, Tk, TH, Th, Tm; Chris@10: Tn = Rp[WS(rs, 2)]; Chris@10: TH = Tf * Tj; Chris@10: Th = Tf * Tg; Chris@10: TG = FNMS(Tc, Ta, TF); Chris@10: Te = FMA(Tc, Td, Tb); Chris@10: TI = FNMS(Ti, Tg, TH); Chris@10: Tk = FMA(Ti, Tj, Th); Chris@10: Tq = Rm[WS(rs, 2)]; Chris@10: Tm = W[6]; Chris@10: TJ = TG + TI; Chris@10: TS = TI - TG; Chris@10: TB = Te + Tk; Chris@10: Tl = Te - Tk; Chris@10: To = Tm * Tn; Chris@10: TK = Tm * Tq; Chris@10: } Chris@10: Tt = Ip[0]; Chris@10: Tw = Im[0]; Chris@10: Ts = W[0]; Chris@10: Tp = W[7]; Chris@10: Tv = W[1]; Chris@10: } Chris@10: } Chris@10: { Chris@10: E TA, T8, TL, Tr, TN, Tx, T12, TM, Tu; Chris@10: TA = T1 + T7; Chris@10: T8 = T1 - T7; Chris@10: TM = Ts * Tw; Chris@10: Tu = Ts * Tt; Chris@10: TL = FNMS(Tp, Tn, TK); Chris@10: Tr = FMA(Tp, Tq, To); Chris@10: TN = FNMS(Tv, Tt, TM); Chris@10: Tx = FMA(Tv, Tw, Tu); Chris@10: T12 = TX - TW; Chris@10: TY = TW + TX; Chris@10: { Chris@10: E TP, TT, TD, TQ, TE, Tz, T14, T13; Chris@10: { Chris@10: E TO, TR, TC, Ty, T11; Chris@10: TO = TL + TN; Chris@10: TR = TN - TL; Chris@10: TC = Tr + Tx; Chris@10: Ty = Tr - Tx; Chris@10: TP = TJ - TO; Chris@10: TU = TJ + TO; Chris@10: TT = TR - TS; Chris@10: T11 = TS + TR; Chris@10: Tz = Tl + Ty; Chris@10: T14 = Ty - Tl; Chris@10: Im[WS(rs, 2)] = T11 - T12; Chris@10: T13 = FMA(KP500000000, T11, T12); Chris@10: T10 = TB - TC; Chris@10: TD = TB + TC; Chris@10: } Chris@10: Rm[WS(rs, 2)] = T8 + Tz; Chris@10: TQ = FNMS(KP500000000, Tz, T8); Chris@10: Im[0] = FMS(KP866025403, T14, T13); Chris@10: Ip[WS(rs, 1)] = FMA(KP866025403, T14, T13); Chris@10: TE = FNMS(KP500000000, TD, TA); Chris@10: Rm[0] = FNMS(KP866025403, TT, TQ); Chris@10: Rp[WS(rs, 1)] = FMA(KP866025403, TT, TQ); Chris@10: Rp[0] = TA + TD; Chris@10: Rm[WS(rs, 1)] = FMA(KP866025403, TP, TE); Chris@10: Rp[WS(rs, 2)] = FNMS(KP866025403, TP, TE); Chris@10: } Chris@10: } Chris@10: } Chris@10: Ip[0] = TU + TY; Chris@10: TZ = FNMS(KP500000000, TU, TY); Chris@10: Im[WS(rs, 1)] = FMS(KP866025403, T10, TZ); Chris@10: Ip[WS(rs, 2)] = FMA(KP866025403, T10, TZ); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, "hc2cf_6", twinstr, &GENUS, {24, 10, 22, 0} }; Chris@10: Chris@10: void X(codelet_hc2cf_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cf_6, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -n 6 -dit -name hc2cf_6 -include hc2cf.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 28 FP multiplications, Chris@10: * (or, 32 additions, 14 multiplications, 14 fused multiply/add), Chris@10: * 23 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "hc2cf.h" Chris@10: Chris@10: static void hc2cf_6(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: for (m = mb, W = W + ((mb - 1) * 10); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 10, MAKE_VOLATILE_STRIDE(24, rs)) { Chris@10: E T7, TS, Tv, TO, Tt, TJ, Tx, TF, Ti, TI, Tw, TC; Chris@10: { Chris@10: E T1, TN, T6, TM; Chris@10: T1 = Rp[0]; Chris@10: TN = Rm[0]; Chris@10: { Chris@10: E T3, T5, T2, T4; Chris@10: T3 = Ip[WS(rs, 1)]; Chris@10: T5 = Im[WS(rs, 1)]; Chris@10: T2 = W[4]; Chris@10: T4 = W[5]; Chris@10: T6 = FMA(T2, T3, T4 * T5); Chris@10: TM = FNMS(T4, T3, T2 * T5); Chris@10: } Chris@10: T7 = T1 - T6; Chris@10: TS = TN - TM; Chris@10: Tv = T1 + T6; Chris@10: TO = TM + TN; Chris@10: } Chris@10: { Chris@10: E Tn, TD, Ts, TE; Chris@10: { Chris@10: E Tk, Tm, Tj, Tl; Chris@10: Tk = Rp[WS(rs, 2)]; Chris@10: Tm = Rm[WS(rs, 2)]; Chris@10: Tj = W[6]; Chris@10: Tl = W[7]; Chris@10: Tn = FMA(Tj, Tk, Tl * Tm); Chris@10: TD = FNMS(Tl, Tk, Tj * Tm); Chris@10: } Chris@10: { Chris@10: E Tp, Tr, To, Tq; Chris@10: Tp = Ip[0]; Chris@10: Tr = Im[0]; Chris@10: To = W[0]; Chris@10: Tq = W[1]; Chris@10: Ts = FMA(To, Tp, Tq * Tr); Chris@10: TE = FNMS(Tq, Tp, To * Tr); Chris@10: } Chris@10: Tt = Tn - Ts; Chris@10: TJ = TE - TD; Chris@10: Tx = Tn + Ts; Chris@10: TF = TD + TE; Chris@10: } Chris@10: { Chris@10: E Tc, TA, Th, TB; Chris@10: { Chris@10: E T9, Tb, T8, Ta; Chris@10: T9 = Rp[WS(rs, 1)]; Chris@10: Tb = Rm[WS(rs, 1)]; Chris@10: T8 = W[2]; Chris@10: Ta = W[3]; Chris@10: Tc = FMA(T8, T9, Ta * Tb); Chris@10: TA = FNMS(Ta, T9, T8 * Tb); Chris@10: } Chris@10: { Chris@10: E Te, Tg, Td, Tf; Chris@10: Te = Ip[WS(rs, 2)]; Chris@10: Tg = Im[WS(rs, 2)]; Chris@10: Td = W[8]; Chris@10: Tf = W[9]; Chris@10: Th = FMA(Td, Te, Tf * Tg); Chris@10: TB = FNMS(Tf, Te, Td * Tg); Chris@10: } Chris@10: Ti = Tc - Th; Chris@10: TI = TA - TB; Chris@10: Tw = Tc + Th; Chris@10: TC = TA + TB; Chris@10: } Chris@10: { Chris@10: E TK, Tu, TH, TT, TR, TU; Chris@10: TK = KP866025403 * (TI + TJ); Chris@10: Tu = Ti + Tt; Chris@10: TH = FNMS(KP500000000, Tu, T7); Chris@10: Rm[WS(rs, 2)] = T7 + Tu; Chris@10: Rp[WS(rs, 1)] = TH + TK; Chris@10: Rm[0] = TH - TK; Chris@10: TT = KP866025403 * (Tt - Ti); Chris@10: TR = TJ - TI; Chris@10: TU = FMA(KP500000000, TR, TS); Chris@10: Im[WS(rs, 2)] = TR - TS; Chris@10: Ip[WS(rs, 1)] = TT + TU; Chris@10: Im[0] = TT - TU; Chris@10: } Chris@10: { Chris@10: E TG, Ty, Tz, TP, TL, TQ; Chris@10: TG = KP866025403 * (TC - TF); Chris@10: Ty = Tw + Tx; Chris@10: Tz = FNMS(KP500000000, Ty, Tv); Chris@10: Rp[0] = Tv + Ty; Chris@10: Rm[WS(rs, 1)] = Tz + TG; Chris@10: Rp[WS(rs, 2)] = Tz - TG; Chris@10: TP = KP866025403 * (Tw - Tx); Chris@10: TL = TC + TF; Chris@10: TQ = FNMS(KP500000000, TL, TO); Chris@10: Ip[0] = TL + TO; Chris@10: Ip[WS(rs, 2)] = TP + TQ; Chris@10: Im[WS(rs, 1)] = TP - TQ; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: {TW_FULL, 1, 6}, Chris@10: {TW_NEXT, 1, 0} Chris@10: }; Chris@10: Chris@10: static const hc2c_desc desc = { 6, "hc2cf_6", twinstr, &GENUS, {32, 14, 14, 0} }; Chris@10: Chris@10: void X(codelet_hc2cf_6) (planner *p) { Chris@10: X(khc2c_register) (p, hc2cf_6, &desc, HC2C_VIA_RDFT); Chris@10: } Chris@10: #endif /* HAVE_FMA */