cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:42:04 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 10 -dif -name hc2cbdft_10 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 122 FP additions, 72 FP multiplications, cannam@95: * (or, 68 additions, 18 multiplications, 54 fused multiply/add), cannam@95: * 95 stack variables, 4 constants, and 40 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cbdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: E T2d, T2f; cannam@95: { cannam@95: E T1g, TQ, T1z, TZ, Tu, T23, T1p, T14, Tt, T27, T13, Tj, Tz, T1i, T18; cannam@95: E TJ, TS, T19, Ty, TA; cannam@95: { cannam@95: E Tl, T3, T7, Tm, T6, Tr, TY, T1n, Th, T8, T1, T2; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 4)]; cannam@95: { cannam@95: E Te, Tp, Td, Tf, Tb, Tc; cannam@95: Tb = Rp[WS(rs, 4)]; cannam@95: Tc = Rm[0]; cannam@95: Te = Rm[WS(rs, 3)]; cannam@95: Tl = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: Tp = Tb - Tc; cannam@95: Td = Tb + Tc; cannam@95: Tf = Rp[WS(rs, 1)]; cannam@95: { cannam@95: E T4, T5, Tq, Tg; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 2)]; cannam@95: T7 = Rm[WS(rs, 1)]; cannam@95: Tq = Te - Tf; cannam@95: Tg = Te + Tf; cannam@95: Tm = T4 - T5; cannam@95: T6 = T4 + T5; cannam@95: Tr = Tp + Tq; cannam@95: TY = Tp - Tq; cannam@95: T1n = Td - Tg; cannam@95: Th = Td + Tg; cannam@95: T8 = Rp[WS(rs, 3)]; cannam@95: } cannam@95: } cannam@95: { cannam@95: E TO, Tn, T9, TP; cannam@95: TO = Ip[0]; cannam@95: Tn = T7 - T8; cannam@95: T9 = T7 + T8; cannam@95: TP = Im[WS(rs, 4)]; cannam@95: { cannam@95: E TG, TH, TF, T16, TD, TE, Ti; cannam@95: TD = Ip[WS(rs, 4)]; cannam@95: { cannam@95: E TX, To, T1o, Ta, Ts; cannam@95: TX = Tm - Tn; cannam@95: To = Tm + Tn; cannam@95: T1o = T6 - T9; cannam@95: Ta = T6 + T9; cannam@95: T1g = TO - TP; cannam@95: TQ = TO + TP; cannam@95: T1z = FNMS(KP618033988, TX, TY); cannam@95: TZ = FMA(KP618033988, TY, TX); cannam@95: Ts = To + Tr; cannam@95: Tu = To - Tr; cannam@95: T23 = FMA(KP618033988, T1n, T1o); cannam@95: T1p = FNMS(KP618033988, T1o, T1n); cannam@95: Ti = Ta + Th; cannam@95: T14 = Ta - Th; cannam@95: Tt = FNMS(KP250000000, Ts, Tl); cannam@95: T27 = Tl + Ts; cannam@95: TE = Im[0]; cannam@95: } cannam@95: T13 = FNMS(KP250000000, Ti, T3); cannam@95: Tj = T3 + Ti; cannam@95: TG = Im[WS(rs, 3)]; cannam@95: TH = Ip[WS(rs, 1)]; cannam@95: TF = TD + TE; cannam@95: T16 = TD - TE; cannam@95: { cannam@95: E Tw, T17, TI, Tx; cannam@95: Tw = Ip[WS(rs, 2)]; cannam@95: T17 = TH - TG; cannam@95: TI = TG + TH; cannam@95: Tx = Im[WS(rs, 2)]; cannam@95: Tz = Im[WS(rs, 1)]; cannam@95: T1i = T16 + T17; cannam@95: T18 = T16 - T17; cannam@95: TJ = TF + TI; cannam@95: TS = TF - TI; cannam@95: T19 = Tw - Tx; cannam@95: Ty = Tw + Tx; cannam@95: TA = Ip[WS(rs, 3)]; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: E T26, T2y, T2a, T28, T1q, T1K, T24, T2k, T10, T1Q, T1A, T2q, T29, Tk, TN; cannam@95: E T2c, T1M, T1P, T2w, TM, T1O, T1S, T1s, T1x, T2m, T2p, T1w, T1C, T2o, T2s; cannam@95: E T12, T1f, T1G, T1J, T1I, T1E, T1e, T1U, T1W, T21, T2g, T2j, T20, T2e, T2i; cannam@95: E T2u, T1a, TB; cannam@95: T1a = TA - Tz; cannam@95: TB = Tz + TA; cannam@95: { cannam@95: E T1Y, T1c, T1u, T1t, T1N, TL, TK, Tv, T2n, T1v; cannam@95: { cannam@95: E T1l, TV, T1k, TU, T1b, T1h; cannam@95: T26 = W[9]; cannam@95: T1b = T19 - T1a; cannam@95: T1h = T19 + T1a; cannam@95: { cannam@95: E TC, TR, T1j, TT; cannam@95: TC = Ty + TB; cannam@95: TR = Ty - TB; cannam@95: T1Y = FMA(KP618033988, T18, T1b); cannam@95: T1c = FNMS(KP618033988, T1b, T18); cannam@95: T1j = T1h + T1i; cannam@95: T1l = T1h - T1i; cannam@95: T1u = FNMS(KP618033988, TC, TJ); cannam@95: TK = FMA(KP618033988, TJ, TC); cannam@95: TT = TR + TS; cannam@95: TV = TR - TS; cannam@95: T2y = T1g + T1j; cannam@95: T1k = FNMS(KP250000000, T1j, T1g); cannam@95: T2a = TQ + TT; cannam@95: TU = FNMS(KP250000000, TT, TQ); cannam@95: T28 = T26 * T27; cannam@95: } cannam@95: { cannam@95: E T22, T1m, T1y, TW; cannam@95: T22 = FMA(KP559016994, T1l, T1k); cannam@95: T1m = FNMS(KP559016994, T1l, T1k); cannam@95: T1y = FNMS(KP559016994, TV, TU); cannam@95: TW = FMA(KP559016994, TV, TU); cannam@95: T1q = FNMS(KP951056516, T1p, T1m); cannam@95: T1K = FMA(KP951056516, T1p, T1m); cannam@95: T24 = FNMS(KP951056516, T23, T22); cannam@95: T2k = FMA(KP951056516, T23, T22); cannam@95: T10 = FMA(KP951056516, TZ, TW); cannam@95: T1Q = FNMS(KP951056516, TZ, TW); cannam@95: T1A = FMA(KP951056516, T1z, T1y); cannam@95: T2q = FNMS(KP951056516, T1z, T1y); cannam@95: T29 = W[8]; cannam@95: } cannam@95: } cannam@95: Tv = FMA(KP559016994, Tu, Tt); cannam@95: T1t = FNMS(KP559016994, Tu, Tt); cannam@95: Tk = W[1]; cannam@95: TN = W[0]; cannam@95: T2c = T29 * T27; cannam@95: T1N = FMA(KP951056516, TK, Tv); cannam@95: TL = FNMS(KP951056516, TK, Tv); cannam@95: T1M = W[17]; cannam@95: T1P = W[16]; cannam@95: T2w = TN * TL; cannam@95: TM = Tk * TL; cannam@95: T1O = T1M * T1N; cannam@95: T1S = T1P * T1N; cannam@95: T2n = FMA(KP951056516, T1u, T1t); cannam@95: T1v = FNMS(KP951056516, T1u, T1t); cannam@95: T1s = W[5]; cannam@95: T1x = W[4]; cannam@95: T2m = W[13]; cannam@95: T2p = W[12]; cannam@95: T1w = T1s * T1v; cannam@95: T1C = T1x * T1v; cannam@95: T2o = T2m * T2n; cannam@95: T2s = T2p * T2n; cannam@95: { cannam@95: E T1X, T1d, T1H, T15, T2h, T1Z; cannam@95: T1X = FMA(KP559016994, T14, T13); cannam@95: T15 = FNMS(KP559016994, T14, T13); cannam@95: T12 = W[2]; cannam@95: T1f = W[3]; cannam@95: T1G = W[14]; cannam@95: T1d = FMA(KP951056516, T1c, T15); cannam@95: T1H = FNMS(KP951056516, T1c, T15); cannam@95: T1J = W[15]; cannam@95: T1I = T1G * T1H; cannam@95: T1E = T1f * T1d; cannam@95: T1e = T12 * T1d; cannam@95: T1U = T1J * T1H; cannam@95: T2h = FNMS(KP951056516, T1Y, T1X); cannam@95: T1Z = FMA(KP951056516, T1Y, T1X); cannam@95: T1W = W[6]; cannam@95: T21 = W[7]; cannam@95: T2g = W[10]; cannam@95: T2j = W[11]; cannam@95: T20 = T1W * T1Z; cannam@95: T2e = T21 * T1Z; cannam@95: T2i = T2g * T2h; cannam@95: T2u = T2j * T2h; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1D, T1F, T1L, T1R; cannam@95: { cannam@95: E T11, T2x, T1r, T1B; cannam@95: T11 = FMA(TN, T10, TM); cannam@95: T2x = FNMS(Tk, T10, T2w); cannam@95: T1r = FNMS(T1f, T1q, T1e); cannam@95: T1B = FMA(T1x, T1A, T1w); cannam@95: Rm[0] = Tj + T11; cannam@95: Rp[0] = Tj - T11; cannam@95: Ip[0] = T2x + T2y; cannam@95: Im[0] = T2x - T2y; cannam@95: Rp[WS(rs, 1)] = T1r - T1B; cannam@95: Rm[WS(rs, 1)] = T1B + T1r; cannam@95: T1D = FNMS(T1s, T1A, T1C); cannam@95: T1F = FMA(T12, T1q, T1E); cannam@95: T1L = FNMS(T1J, T1K, T1I); cannam@95: T1R = FMA(T1P, T1Q, T1O); cannam@95: } cannam@95: { cannam@95: E T1T, T1V, T2t, T2v; cannam@95: T1T = FNMS(T1M, T1Q, T1S); cannam@95: Ip[WS(rs, 1)] = T1D + T1F; cannam@95: Im[WS(rs, 1)] = T1D - T1F; cannam@95: Rm[WS(rs, 4)] = T1R + T1L; cannam@95: Rp[WS(rs, 4)] = T1L - T1R; cannam@95: T1V = FMA(T1G, T1K, T1U); cannam@95: T2t = FNMS(T2m, T2q, T2s); cannam@95: T2v = FMA(T2g, T2k, T2u); cannam@95: { cannam@95: E T2l, T2r, T25, T2b; cannam@95: T2l = FNMS(T2j, T2k, T2i); cannam@95: Ip[WS(rs, 4)] = T1T + T1V; cannam@95: Im[WS(rs, 4)] = T1T - T1V; cannam@95: Ip[WS(rs, 3)] = T2t + T2v; cannam@95: Im[WS(rs, 3)] = T2t - T2v; cannam@95: T2r = FMA(T2p, T2q, T2o); cannam@95: T25 = FNMS(T21, T24, T20); cannam@95: T2b = FMA(T29, T2a, T28); cannam@95: T2d = FNMS(T26, T2a, T2c); cannam@95: Rm[WS(rs, 3)] = T2r + T2l; cannam@95: Rp[WS(rs, 3)] = T2l - T2r; cannam@95: Rm[WS(rs, 2)] = T2b + T25; cannam@95: Rp[WS(rs, 2)] = T25 - T2b; cannam@95: T2f = FMA(T1W, T24, T2e); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: Ip[WS(rs, 2)] = T2d + T2f; cannam@95: Im[WS(rs, 2)] = T2d - T2f; cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 10}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, "hc2cbdft_10", twinstr, &GENUS, {68, 18, 54, 0} }; cannam@95: cannam@95: void X(codelet_hc2cbdft_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdft_10, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 10 -dif -name hc2cbdft_10 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 122 FP additions, 60 FP multiplications, cannam@95: * (or, 92 additions, 30 multiplications, 30 fused multiply/add), cannam@95: * 61 stack variables, 4 constants, and 40 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cbdft_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * 18); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 18, MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: E T3, TS, TR, T13, Ti, T12, TT, TU, T1g, T1T, Tr, T1s, TJ, T1h, TG; cannam@95: E T1m, TK, TL, T1k, T1l, T1b, T1P, TY, T1w; cannam@95: { cannam@95: E Td, To, Tg, Tp, Th, TQ, T6, Tl, T9, Tm, Ta, TP, T1, T2; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 4)]; cannam@95: T3 = T1 + T2; cannam@95: TS = T1 - T2; cannam@95: { cannam@95: E Tb, Tc, Te, Tf; cannam@95: Tb = Rp[WS(rs, 4)]; cannam@95: Tc = Rm[0]; cannam@95: Td = Tb + Tc; cannam@95: To = Tb - Tc; cannam@95: Te = Rm[WS(rs, 3)]; cannam@95: Tf = Rp[WS(rs, 1)]; cannam@95: Tg = Te + Tf; cannam@95: Tp = Te - Tf; cannam@95: } cannam@95: Th = Td + Tg; cannam@95: TQ = To + Tp; cannam@95: { cannam@95: E T4, T5, T7, T8; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 2)]; cannam@95: T6 = T4 + T5; cannam@95: Tl = T4 - T5; cannam@95: T7 = Rm[WS(rs, 1)]; cannam@95: T8 = Rp[WS(rs, 3)]; cannam@95: T9 = T7 + T8; cannam@95: Tm = T7 - T8; cannam@95: } cannam@95: Ta = T6 + T9; cannam@95: TP = Tl + Tm; cannam@95: TR = KP559016994 * (TP - TQ); cannam@95: T13 = KP559016994 * (Ta - Th); cannam@95: Ti = Ta + Th; cannam@95: T12 = FNMS(KP250000000, Ti, T3); cannam@95: TT = TP + TQ; cannam@95: TU = FNMS(KP250000000, TT, TS); cannam@95: { cannam@95: E T1e, T1f, Tn, Tq; cannam@95: T1e = T6 - T9; cannam@95: T1f = Td - Tg; cannam@95: T1g = FNMS(KP951056516, T1f, KP587785252 * T1e); cannam@95: T1T = FMA(KP951056516, T1e, KP587785252 * T1f); cannam@95: Tn = Tl - Tm; cannam@95: Tq = To - Tp; cannam@95: Tr = FMA(KP951056516, Tn, KP587785252 * Tq); cannam@95: T1s = FNMS(KP951056516, Tq, KP587785252 * Tn); cannam@95: } cannam@95: } cannam@95: { cannam@95: E TB, T18, TE, T19, TF, T1j, Tu, T15, Tx, T16, Ty, T1i, TH, TI; cannam@95: TH = Ip[0]; cannam@95: TI = Im[WS(rs, 4)]; cannam@95: TJ = TH + TI; cannam@95: T1h = TH - TI; cannam@95: { cannam@95: E Tz, TA, TC, TD; cannam@95: Tz = Ip[WS(rs, 4)]; cannam@95: TA = Im[0]; cannam@95: TB = Tz + TA; cannam@95: T18 = Tz - TA; cannam@95: TC = Im[WS(rs, 3)]; cannam@95: TD = Ip[WS(rs, 1)]; cannam@95: TE = TC + TD; cannam@95: T19 = TD - TC; cannam@95: } cannam@95: TF = TB - TE; cannam@95: T1j = T18 + T19; cannam@95: { cannam@95: E Ts, Tt, Tv, Tw; cannam@95: Ts = Ip[WS(rs, 2)]; cannam@95: Tt = Im[WS(rs, 2)]; cannam@95: Tu = Ts + Tt; cannam@95: T15 = Ts - Tt; cannam@95: Tv = Im[WS(rs, 1)]; cannam@95: Tw = Ip[WS(rs, 3)]; cannam@95: Tx = Tv + Tw; cannam@95: T16 = Tw - Tv; cannam@95: } cannam@95: Ty = Tu - Tx; cannam@95: T1i = T15 + T16; cannam@95: TG = KP559016994 * (Ty - TF); cannam@95: T1m = KP559016994 * (T1i - T1j); cannam@95: TK = Ty + TF; cannam@95: TL = FNMS(KP250000000, TK, TJ); cannam@95: T1k = T1i + T1j; cannam@95: T1l = FNMS(KP250000000, T1k, T1h); cannam@95: { cannam@95: E T17, T1a, TW, TX; cannam@95: T17 = T15 - T16; cannam@95: T1a = T18 - T19; cannam@95: T1b = FNMS(KP951056516, T1a, KP587785252 * T17); cannam@95: T1P = FMA(KP951056516, T17, KP587785252 * T1a); cannam@95: TW = Tu + Tx; cannam@95: TX = TB + TE; cannam@95: TY = FMA(KP951056516, TW, KP587785252 * TX); cannam@95: T1w = FNMS(KP951056516, TX, KP587785252 * TW); cannam@95: } cannam@95: } cannam@95: { cannam@95: E Tj, T2g, TN, T1H, T1U, T26, TZ, T1J, T1Q, T24, T1c, T1C, T1t, T29, T1o; cannam@95: E T1E, T1x, T2b, T20, T21, TM, T1S, TV; cannam@95: Tj = T3 + Ti; cannam@95: T2g = T1h + T1k; cannam@95: TM = TG + TL; cannam@95: TN = Tr + TM; cannam@95: T1H = TM - Tr; cannam@95: T1S = T1m + T1l; cannam@95: T1U = T1S - T1T; cannam@95: T26 = T1T + T1S; cannam@95: TV = TR + TU; cannam@95: TZ = TV - TY; cannam@95: T1J = TV + TY; cannam@95: { cannam@95: E T1O, T14, T1r, T1n, T1v; cannam@95: T1O = T13 + T12; cannam@95: T1Q = T1O + T1P; cannam@95: T24 = T1O - T1P; cannam@95: T14 = T12 - T13; cannam@95: T1c = T14 - T1b; cannam@95: T1C = T14 + T1b; cannam@95: T1r = TL - TG; cannam@95: T1t = T1r - T1s; cannam@95: T29 = T1s + T1r; cannam@95: T1n = T1l - T1m; cannam@95: T1o = T1g + T1n; cannam@95: T1E = T1n - T1g; cannam@95: T1v = TU - TR; cannam@95: T1x = T1v + T1w; cannam@95: T2b = T1v - T1w; cannam@95: { cannam@95: E T1X, T1Z, T1W, T1Y; cannam@95: T1X = TS + TT; cannam@95: T1Z = TJ + TK; cannam@95: T1W = W[9]; cannam@95: T1Y = W[8]; cannam@95: T20 = FMA(T1W, T1X, T1Y * T1Z); cannam@95: T21 = FNMS(T1W, T1Z, T1Y * T1X); cannam@95: } cannam@95: } cannam@95: { cannam@95: E T10, T2f, Tk, TO; cannam@95: Tk = W[0]; cannam@95: TO = W[1]; cannam@95: T10 = FMA(Tk, TN, TO * TZ); cannam@95: T2f = FNMS(TO, TN, Tk * TZ); cannam@95: Rp[0] = Tj - T10; cannam@95: Ip[0] = T2f + T2g; cannam@95: Rm[0] = Tj + T10; cannam@95: Im[0] = T2f - T2g; cannam@95: } cannam@95: { cannam@95: E T1V, T22, T1N, T1R; cannam@95: T1N = W[6]; cannam@95: T1R = W[7]; cannam@95: T1V = FNMS(T1R, T1U, T1N * T1Q); cannam@95: T22 = FMA(T1R, T1Q, T1N * T1U); cannam@95: Rp[WS(rs, 2)] = T1V - T20; cannam@95: Ip[WS(rs, 2)] = T21 + T22; cannam@95: Rm[WS(rs, 2)] = T20 + T1V; cannam@95: Im[WS(rs, 2)] = T21 - T22; cannam@95: } cannam@95: { cannam@95: E T1p, T1A, T1y, T1z; cannam@95: { cannam@95: E T11, T1d, T1q, T1u; cannam@95: T11 = W[2]; cannam@95: T1d = W[3]; cannam@95: T1p = FNMS(T1d, T1o, T11 * T1c); cannam@95: T1A = FMA(T1d, T1c, T11 * T1o); cannam@95: T1q = W[4]; cannam@95: T1u = W[5]; cannam@95: T1y = FMA(T1q, T1t, T1u * T1x); cannam@95: T1z = FNMS(T1u, T1t, T1q * T1x); cannam@95: } cannam@95: Rp[WS(rs, 1)] = T1p - T1y; cannam@95: Ip[WS(rs, 1)] = T1z + T1A; cannam@95: Rm[WS(rs, 1)] = T1y + T1p; cannam@95: Im[WS(rs, 1)] = T1z - T1A; cannam@95: } cannam@95: { cannam@95: E T1F, T1M, T1K, T1L; cannam@95: { cannam@95: E T1B, T1D, T1G, T1I; cannam@95: T1B = W[14]; cannam@95: T1D = W[15]; cannam@95: T1F = FNMS(T1D, T1E, T1B * T1C); cannam@95: T1M = FMA(T1D, T1C, T1B * T1E); cannam@95: T1G = W[16]; cannam@95: T1I = W[17]; cannam@95: T1K = FMA(T1G, T1H, T1I * T1J); cannam@95: T1L = FNMS(T1I, T1H, T1G * T1J); cannam@95: } cannam@95: Rp[WS(rs, 4)] = T1F - T1K; cannam@95: Ip[WS(rs, 4)] = T1L + T1M; cannam@95: Rm[WS(rs, 4)] = T1K + T1F; cannam@95: Im[WS(rs, 4)] = T1L - T1M; cannam@95: } cannam@95: { cannam@95: E T27, T2e, T2c, T2d; cannam@95: { cannam@95: E T23, T25, T28, T2a; cannam@95: T23 = W[10]; cannam@95: T25 = W[11]; cannam@95: T27 = FNMS(T25, T26, T23 * T24); cannam@95: T2e = FMA(T25, T24, T23 * T26); cannam@95: T28 = W[12]; cannam@95: T2a = W[13]; cannam@95: T2c = FMA(T28, T29, T2a * T2b); cannam@95: T2d = FNMS(T2a, T29, T28 * T2b); cannam@95: } cannam@95: Rp[WS(rs, 3)] = T27 - T2c; cannam@95: Ip[WS(rs, 3)] = T2d + T2e; cannam@95: Rm[WS(rs, 3)] = T2c + T27; cannam@95: Im[WS(rs, 3)] = T2d - T2e; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_FULL, 1, 10}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, "hc2cbdft_10", twinstr, &GENUS, {92, 30, 30, 0} }; cannam@95: cannam@95: void X(codelet_hc2cbdft_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdft_10, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */