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 07:41:09 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_r2cb.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 20 -name r2cb_20 -include r2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 86 FP additions, 44 FP multiplications, Chris@10: * (or, 42 additions, 0 multiplications, 44 fused multiply/add), Chris@10: * 69 stack variables, 5 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "r2cb.h" Chris@10: Chris@10: static void r2cb_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP1_902113032, +1.902113032590307144232878666758764286811397268); Chris@10: DK(KP1_118033988, +1.118033988749894848204586834365638117720309180); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(80, rs), MAKE_VOLATILE_STRIDE(80, csr), MAKE_VOLATILE_STRIDE(80, csi)) { Chris@10: E TY, T1o, T1m, T14, T12, TX, T1n, T1j, TZ, T13; Chris@10: { Chris@10: E Tr, TD, Tl, T5, T1a, T1l, T1d, T1k, TT, T10, TO, T11, TE, TF, Tk; Chris@10: E TI, TC, T1i, To, TG, T16; Chris@10: { Chris@10: E T4, Tq, T1, T2; Chris@10: T4 = Cr[WS(csr, 5)]; Chris@10: Tq = Ci[WS(csi, 5)]; Chris@10: T1 = Cr[0]; Chris@10: T2 = Cr[WS(csr, 10)]; Chris@10: { Chris@10: E Ts, T8, T19, TR, T18, Tb, TS, Tv, Tx, Tf, Ty, T1c, TM, T1b, Ti; Chris@10: E Tz, Tt, Tu, TN, TA; Chris@10: { Chris@10: E TP, TQ, T9, Ta; Chris@10: { Chris@10: E T6, T7, Tp, T3; Chris@10: T6 = Cr[WS(csr, 4)]; Chris@10: T7 = Cr[WS(csr, 6)]; Chris@10: TP = Ci[WS(csi, 4)]; Chris@10: Tp = T1 - T2; Chris@10: T3 = T1 + T2; Chris@10: Ts = T6 - T7; Chris@10: T8 = T6 + T7; Chris@10: Tr = FMA(KP2_000000000, Tq, Tp); Chris@10: TD = FNMS(KP2_000000000, Tq, Tp); Chris@10: Tl = FMA(KP2_000000000, T4, T3); Chris@10: T5 = FNMS(KP2_000000000, T4, T3); Chris@10: TQ = Ci[WS(csi, 6)]; Chris@10: } Chris@10: T9 = Cr[WS(csr, 9)]; Chris@10: Ta = Cr[WS(csr, 1)]; Chris@10: Tt = Ci[WS(csi, 9)]; Chris@10: T19 = TP + TQ; Chris@10: TR = TP - TQ; Chris@10: T18 = T9 - Ta; Chris@10: Tb = T9 + Ta; Chris@10: Tu = Ci[WS(csi, 1)]; Chris@10: } Chris@10: { Chris@10: E TK, TL, Td, Te, Tg, Th; Chris@10: Td = Cr[WS(csr, 8)]; Chris@10: Te = Cr[WS(csr, 2)]; Chris@10: TK = Ci[WS(csi, 8)]; Chris@10: TS = Tt - Tu; Chris@10: Tv = Tt + Tu; Chris@10: Tx = Td - Te; Chris@10: Tf = Td + Te; Chris@10: TL = Ci[WS(csi, 2)]; Chris@10: Tg = Cr[WS(csr, 7)]; Chris@10: Th = Cr[WS(csr, 3)]; Chris@10: Ty = Ci[WS(csi, 7)]; Chris@10: T1c = TK + TL; Chris@10: TM = TK - TL; Chris@10: T1b = Tg - Th; Chris@10: Ti = Tg + Th; Chris@10: Tz = Ci[WS(csi, 3)]; Chris@10: } Chris@10: T1a = T18 + T19; Chris@10: T1l = T19 - T18; Chris@10: T1d = T1b + T1c; Chris@10: T1k = T1c - T1b; Chris@10: TT = TR - TS; Chris@10: T10 = TS + TR; Chris@10: TN = Tz - Ty; Chris@10: TA = Ty + Tz; Chris@10: TO = TM - TN; Chris@10: T11 = TN + TM; Chris@10: { Chris@10: E Tm, Tc, Tj, Tn, Tw, TB; Chris@10: Tm = T8 + Tb; Chris@10: Tc = T8 - Tb; Chris@10: Tj = Tf - Ti; Chris@10: Tn = Tf + Ti; Chris@10: TE = Ts - Tv; Chris@10: Tw = Ts + Tv; Chris@10: TB = Tx - TA; Chris@10: TF = Tx + TA; Chris@10: Tk = Tc + Tj; Chris@10: TI = Tc - Tj; Chris@10: TC = Tw + TB; Chris@10: T1i = Tw - TB; Chris@10: TY = Tm - Tn; Chris@10: To = Tm + Tn; Chris@10: } Chris@10: } Chris@10: } Chris@10: R0[WS(rs, 5)] = FMA(KP2_000000000, Tk, T5); Chris@10: R1[WS(rs, 7)] = FMA(KP2_000000000, TC, Tr); Chris@10: TG = TE + TF; Chris@10: T16 = TE - TF; Chris@10: R0[0] = FMA(KP2_000000000, To, Tl); Chris@10: { Chris@10: E TU, TW, T1g, T1e, T15, TV, TJ, TH, T1h, T1f, T17; Chris@10: TU = FNMS(KP618033988, TT, TO); Chris@10: TW = FMA(KP618033988, TO, TT); Chris@10: R1[WS(rs, 2)] = FMA(KP2_000000000, TG, TD); Chris@10: TH = FNMS(KP500000000, Tk, T5); Chris@10: T1g = FNMS(KP618033988, T1a, T1d); Chris@10: T1e = FMA(KP618033988, T1d, T1a); Chris@10: T15 = FNMS(KP500000000, TG, TD); Chris@10: TV = FMA(KP1_118033988, TI, TH); Chris@10: TJ = FNMS(KP1_118033988, TI, TH); Chris@10: T1o = FMA(KP618033988, T1k, T1l); Chris@10: T1m = FNMS(KP618033988, T1l, T1k); Chris@10: R0[WS(rs, 3)] = FNMS(KP1_902113032, TW, TV); Chris@10: R0[WS(rs, 7)] = FMA(KP1_902113032, TW, TV); Chris@10: R0[WS(rs, 1)] = FMA(KP1_902113032, TU, TJ); Chris@10: R0[WS(rs, 9)] = FNMS(KP1_902113032, TU, TJ); Chris@10: T1f = FNMS(KP1_118033988, T16, T15); Chris@10: T17 = FMA(KP1_118033988, T16, T15); Chris@10: T1h = FNMS(KP500000000, TC, Tr); Chris@10: R1[WS(rs, 6)] = FNMS(KP1_902113032, T1g, T1f); Chris@10: R1[WS(rs, 8)] = FMA(KP1_902113032, T1g, T1f); Chris@10: R1[WS(rs, 4)] = FMA(KP1_902113032, T1e, T17); Chris@10: R1[0] = FNMS(KP1_902113032, T1e, T17); Chris@10: T14 = FNMS(KP618033988, T10, T11); Chris@10: T12 = FMA(KP618033988, T11, T10); Chris@10: TX = FNMS(KP500000000, To, Tl); Chris@10: T1n = FMA(KP1_118033988, T1i, T1h); Chris@10: T1j = FNMS(KP1_118033988, T1i, T1h); Chris@10: } Chris@10: } Chris@10: R1[WS(rs, 5)] = FNMS(KP1_902113032, T1o, T1n); Chris@10: R1[WS(rs, 9)] = FMA(KP1_902113032, T1o, T1n); Chris@10: R1[WS(rs, 3)] = FMA(KP1_902113032, T1m, T1j); Chris@10: R1[WS(rs, 1)] = FNMS(KP1_902113032, T1m, T1j); Chris@10: TZ = FMA(KP1_118033988, TY, TX); Chris@10: T13 = FNMS(KP1_118033988, TY, TX); Chris@10: R0[WS(rs, 4)] = FNMS(KP1_902113032, T14, T13); Chris@10: R0[WS(rs, 6)] = FMA(KP1_902113032, T14, T13); Chris@10: R0[WS(rs, 2)] = FMA(KP1_902113032, T12, TZ); Chris@10: R0[WS(rs, 8)] = FNMS(KP1_902113032, T12, TZ); Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 20, "r2cb_20", {42, 0, 44, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cb_20) (planner *p) { Chris@10: X(kr2c_register) (p, r2cb_20, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 20 -name r2cb_20 -include r2cb.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 86 FP additions, 30 FP multiplications, Chris@10: * (or, 70 additions, 14 multiplications, 16 fused multiply/add), Chris@10: * 50 stack variables, 5 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "r2cb.h" Chris@10: Chris@10: static void r2cb_20(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DK(KP1_118033988, +1.118033988749894848204586834365638117720309180); Chris@10: DK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DK(KP1_902113032, +1.902113032590307144232878666758764286811397268); Chris@10: DK(KP1_175570504, +1.175570504584946258337411909278145537195304875); Chris@10: DK(KP2_000000000, +2.000000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(80, rs), MAKE_VOLATILE_STRIDE(80, csr), MAKE_VOLATILE_STRIDE(80, csi)) { Chris@10: E T6, TF, Tm, Tt, TQ, T1n, T1f, T12, T1m, TV, T13, T1c, Td, Tk, Tl; Chris@10: E Ty, TD, TE, Tn, To, Tp, TG, TH, TI; Chris@10: { Chris@10: E T5, Ts, T3, Tq; Chris@10: { Chris@10: E T4, Tr, T1, T2; Chris@10: T4 = Cr[WS(csr, 5)]; Chris@10: T5 = KP2_000000000 * T4; Chris@10: Tr = Ci[WS(csi, 5)]; Chris@10: Ts = KP2_000000000 * Tr; Chris@10: T1 = Cr[0]; Chris@10: T2 = Cr[WS(csr, 10)]; Chris@10: T3 = T1 + T2; Chris@10: Tq = T1 - T2; Chris@10: } Chris@10: T6 = T3 - T5; Chris@10: TF = Tq - Ts; Chris@10: Tm = T3 + T5; Chris@10: Tt = Tq + Ts; Chris@10: } Chris@10: { Chris@10: E T9, Tu, TO, T1b, Tc, T1a, Tx, TP, Tg, Tz, TT, T1e, Tj, T1d, TC; Chris@10: E TU; Chris@10: { Chris@10: E T7, T8, TM, TN; Chris@10: T7 = Cr[WS(csr, 4)]; Chris@10: T8 = Cr[WS(csr, 6)]; Chris@10: T9 = T7 + T8; Chris@10: Tu = T7 - T8; Chris@10: TM = Ci[WS(csi, 4)]; Chris@10: TN = Ci[WS(csi, 6)]; Chris@10: TO = TM - TN; Chris@10: T1b = TM + TN; Chris@10: } Chris@10: { Chris@10: E Ta, Tb, Tv, Tw; Chris@10: Ta = Cr[WS(csr, 9)]; Chris@10: Tb = Cr[WS(csr, 1)]; Chris@10: Tc = Ta + Tb; Chris@10: T1a = Ta - Tb; Chris@10: Tv = Ci[WS(csi, 9)]; Chris@10: Tw = Ci[WS(csi, 1)]; Chris@10: Tx = Tv + Tw; Chris@10: TP = Tv - Tw; Chris@10: } Chris@10: { Chris@10: E Te, Tf, TR, TS; Chris@10: Te = Cr[WS(csr, 8)]; Chris@10: Tf = Cr[WS(csr, 2)]; Chris@10: Tg = Te + Tf; Chris@10: Tz = Te - Tf; Chris@10: TR = Ci[WS(csi, 8)]; Chris@10: TS = Ci[WS(csi, 2)]; Chris@10: TT = TR - TS; Chris@10: T1e = TR + TS; Chris@10: } Chris@10: { Chris@10: E Th, Ti, TA, TB; Chris@10: Th = Cr[WS(csr, 7)]; Chris@10: Ti = Cr[WS(csr, 3)]; Chris@10: Tj = Th + Ti; Chris@10: T1d = Th - Ti; Chris@10: TA = Ci[WS(csi, 7)]; Chris@10: TB = Ci[WS(csi, 3)]; Chris@10: TC = TA + TB; Chris@10: TU = TB - TA; Chris@10: } Chris@10: TQ = TO - TP; Chris@10: T1n = T1e - T1d; Chris@10: T1f = T1d + T1e; Chris@10: T12 = TP + TO; Chris@10: T1m = T1b - T1a; Chris@10: TV = TT - TU; Chris@10: T13 = TU + TT; Chris@10: T1c = T1a + T1b; Chris@10: Td = T9 - Tc; Chris@10: Tk = Tg - Tj; Chris@10: Tl = Td + Tk; Chris@10: Ty = Tu + Tx; Chris@10: TD = Tz - TC; Chris@10: TE = Ty + TD; Chris@10: Tn = T9 + Tc; Chris@10: To = Tg + Tj; Chris@10: Tp = Tn + To; Chris@10: TG = Tu - Tx; Chris@10: TH = Tz + TC; Chris@10: TI = TG + TH; Chris@10: } Chris@10: R0[WS(rs, 5)] = FMA(KP2_000000000, Tl, T6); Chris@10: R1[WS(rs, 7)] = FMA(KP2_000000000, TE, Tt); Chris@10: R1[WS(rs, 2)] = FMA(KP2_000000000, TI, TF); Chris@10: R0[0] = FMA(KP2_000000000, Tp, Tm); Chris@10: { Chris@10: E TW, TY, TL, TX, TJ, TK; Chris@10: TW = FNMS(KP1_902113032, TV, KP1_175570504 * TQ); Chris@10: TY = FMA(KP1_902113032, TQ, KP1_175570504 * TV); Chris@10: TJ = FNMS(KP500000000, Tl, T6); Chris@10: TK = KP1_118033988 * (Td - Tk); Chris@10: TL = TJ - TK; Chris@10: TX = TK + TJ; Chris@10: R0[WS(rs, 1)] = TL - TW; Chris@10: R0[WS(rs, 7)] = TX + TY; Chris@10: R0[WS(rs, 9)] = TL + TW; Chris@10: R0[WS(rs, 3)] = TX - TY; Chris@10: } Chris@10: { Chris@10: E T1g, T1i, T19, T1h, T17, T18; Chris@10: T1g = FNMS(KP1_902113032, T1f, KP1_175570504 * T1c); Chris@10: T1i = FMA(KP1_902113032, T1c, KP1_175570504 * T1f); Chris@10: T17 = FNMS(KP500000000, TI, TF); Chris@10: T18 = KP1_118033988 * (TG - TH); Chris@10: T19 = T17 - T18; Chris@10: T1h = T18 + T17; Chris@10: R1[WS(rs, 8)] = T19 - T1g; Chris@10: R1[WS(rs, 4)] = T1h + T1i; Chris@10: R1[WS(rs, 6)] = T19 + T1g; Chris@10: R1[0] = T1h - T1i; Chris@10: } Chris@10: { Chris@10: E T1o, T1q, T1l, T1p, T1j, T1k; Chris@10: T1o = FNMS(KP1_902113032, T1n, KP1_175570504 * T1m); Chris@10: T1q = FMA(KP1_902113032, T1m, KP1_175570504 * T1n); Chris@10: T1j = FNMS(KP500000000, TE, Tt); Chris@10: T1k = KP1_118033988 * (Ty - TD); Chris@10: T1l = T1j - T1k; Chris@10: T1p = T1k + T1j; Chris@10: R1[WS(rs, 3)] = T1l - T1o; Chris@10: R1[WS(rs, 9)] = T1p + T1q; Chris@10: R1[WS(rs, 1)] = T1l + T1o; Chris@10: R1[WS(rs, 5)] = T1p - T1q; Chris@10: } Chris@10: { Chris@10: E T14, T16, T11, T15, TZ, T10; Chris@10: T14 = FNMS(KP1_902113032, T13, KP1_175570504 * T12); Chris@10: T16 = FMA(KP1_902113032, T12, KP1_175570504 * T13); Chris@10: TZ = FNMS(KP500000000, Tp, Tm); Chris@10: T10 = KP1_118033988 * (Tn - To); Chris@10: T11 = TZ - T10; Chris@10: T15 = T10 + TZ; Chris@10: R0[WS(rs, 6)] = T11 - T14; Chris@10: R0[WS(rs, 2)] = T15 + T16; Chris@10: R0[WS(rs, 4)] = T11 + T14; Chris@10: R0[WS(rs, 8)] = T15 - T16; Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: Chris@10: static const kr2c_desc desc = { 20, "r2cb_20", {70, 14, 16, 0}, &GENUS }; Chris@10: Chris@10: void X(codelet_r2cb_20) (planner *p) { Chris@10: X(kr2c_register) (p, r2cb_20, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */