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 07:41:57 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_hc2c.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hc2cb2_8 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 74 FP additions, 50 FP multiplications, cannam@95: * (or, 44 additions, 20 multiplications, 30 fused multiply/add), cannam@95: * 64 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: 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(32, rs)) { cannam@95: E Tf, Ti, TK, Tq, TH, TT, TX, TW, TY, TU, TI; cannam@95: { cannam@95: E Tg, Tl, Tp, Th, T1n, T1t, Tj; cannam@95: Tf = W[0]; cannam@95: Tg = W[2]; cannam@95: Tl = W[4]; cannam@95: Tp = W[5]; cannam@95: Ti = W[1]; cannam@95: Th = Tf * Tg; cannam@95: T1n = Tf * Tl; cannam@95: T1t = Tf * Tp; cannam@95: Tj = W[3]; cannam@95: { cannam@95: E T1o, T1u, Tk, T1b, To, T1e, T13, TP, T1p, T7, T1h, T1v, TZ, Tv, T1i; cannam@95: E TB, TA, TQ, Te, T1w, TE, T1j; cannam@95: { cannam@95: E Tr, T3, Ts, T1f, TO, TL, T6, Tt; cannam@95: { cannam@95: E TM, TN, T4, T5; cannam@95: { cannam@95: E T1, Tn, T2, TJ, Tm; cannam@95: T1 = Rp[0]; cannam@95: T1o = FMA(Ti, Tp, T1n); cannam@95: T1u = FNMS(Ti, Tl, T1t); cannam@95: Tk = FMA(Ti, Tj, Th); cannam@95: T1b = FNMS(Ti, Tj, Th); cannam@95: Tn = Tf * Tj; cannam@95: T2 = Rm[WS(rs, 3)]; cannam@95: TM = Ip[0]; cannam@95: TJ = Tk * Tp; cannam@95: Tm = Tk * Tl; cannam@95: To = FNMS(Ti, Tg, Tn); cannam@95: T1e = FMA(Ti, Tg, Tn); cannam@95: Tr = T1 - T2; cannam@95: T3 = T1 + T2; cannam@95: TK = FNMS(To, Tl, TJ); cannam@95: Tq = FMA(To, Tp, Tm); cannam@95: TN = Im[WS(rs, 3)]; cannam@95: } cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 1)]; cannam@95: Ts = Ip[WS(rs, 2)]; cannam@95: T1f = TM - TN; cannam@95: TO = TM + TN; cannam@95: TL = T4 - T5; cannam@95: T6 = T4 + T5; cannam@95: Tt = Im[WS(rs, 1)]; cannam@95: } cannam@95: { cannam@95: E Tw, Ta, TC, Tz, Td, TD; cannam@95: { cannam@95: E Tx, Ty, Tb, Tc; cannam@95: { cannam@95: E T8, T1g, Tu, T9; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: T13 = TO - TL; cannam@95: TP = TL + TO; cannam@95: T1p = T3 - T6; cannam@95: T7 = T3 + T6; cannam@95: T1g = Ts - Tt; cannam@95: Tu = Ts + Tt; cannam@95: T9 = Rm[WS(rs, 2)]; cannam@95: Tx = Ip[WS(rs, 1)]; cannam@95: T1h = T1f + T1g; cannam@95: T1v = T1f - T1g; cannam@95: TZ = Tr + Tu; cannam@95: Tv = Tr - Tu; cannam@95: Tw = T8 - T9; cannam@95: Ta = T8 + T9; cannam@95: Ty = Im[WS(rs, 2)]; cannam@95: } cannam@95: Tb = Rm[0]; cannam@95: Tc = Rp[WS(rs, 3)]; cannam@95: TC = Ip[WS(rs, 3)]; cannam@95: T1i = Tx - Ty; cannam@95: Tz = Tx + Ty; cannam@95: TB = Tb - Tc; cannam@95: Td = Tb + Tc; cannam@95: TD = Im[0]; cannam@95: } cannam@95: TA = Tw - Tz; cannam@95: TQ = Tw + Tz; cannam@95: Te = Ta + Td; cannam@95: T1w = Ta - Td; cannam@95: TE = TC + TD; cannam@95: T1j = TC - TD; cannam@95: } cannam@95: } cannam@95: { cannam@95: E T1x, T1k, T1r, TG, TS, T19, T15, T17, T11, T16, T12; cannam@95: { cannam@95: E T1B, T1z, T10, T1A, T1C; cannam@95: T1x = T1v - T1w; cannam@95: T1B = T1w + T1v; cannam@95: Rp[0] = T7 + Te; cannam@95: { cannam@95: E T1q, TR, TF, T14; cannam@95: T1k = T1i + T1j; cannam@95: T1q = T1j - T1i; cannam@95: TR = TB + TE; cannam@95: TF = TB - TE; cannam@95: T1r = T1p - T1q; cannam@95: T1z = T1p + T1q; cannam@95: Rm[0] = T1h + T1k; cannam@95: TG = TA + TF; cannam@95: T14 = TA - TF; cannam@95: TS = TQ - TR; cannam@95: T10 = TQ + TR; cannam@95: T1A = Tk * T1z; cannam@95: T19 = FNMS(KP707106781, T14, T13); cannam@95: T15 = FMA(KP707106781, T14, T13); cannam@95: T1C = Tk * T1B; cannam@95: } cannam@95: T17 = FMA(KP707106781, T10, TZ); cannam@95: T11 = FNMS(KP707106781, T10, TZ); cannam@95: Rp[WS(rs, 1)] = FNMS(To, T1B, T1A); cannam@95: T16 = Tg * T15; cannam@95: Rm[WS(rs, 1)] = FMA(To, T1z, T1C); cannam@95: } cannam@95: T12 = Tg * T11; cannam@95: { cannam@95: E T1l, T1a, T1c, T18; cannam@95: Im[WS(rs, 1)] = FMA(Tj, T11, T16); cannam@95: Ip[WS(rs, 1)] = FNMS(Tj, T15, T12); cannam@95: T18 = Tl * T17; cannam@95: T1l = T1h - T1k; cannam@95: T1a = Tl * T19; cannam@95: T1c = T7 - Te; cannam@95: Ip[WS(rs, 3)] = FNMS(Tp, T19, T18); cannam@95: { cannam@95: E T1s, T1m, T1d, T1y, TV; cannam@95: Im[WS(rs, 3)] = FMA(Tp, T17, T1a); cannam@95: T1m = T1e * T1c; cannam@95: T1d = T1b * T1c; cannam@95: T1s = T1o * T1r; cannam@95: Rm[WS(rs, 2)] = FMA(T1b, T1l, T1m); cannam@95: Rp[WS(rs, 2)] = FNMS(T1e, T1l, T1d); cannam@95: Rp[WS(rs, 3)] = FNMS(T1u, T1x, T1s); cannam@95: T1y = T1o * T1x; cannam@95: TV = FMA(KP707106781, TG, Tv); cannam@95: TH = FNMS(KP707106781, TG, Tv); cannam@95: TT = FNMS(KP707106781, TS, TP); cannam@95: TX = FMA(KP707106781, TS, TP); cannam@95: Rm[WS(rs, 3)] = FMA(T1u, T1r, T1y); cannam@95: TW = Tf * TV; cannam@95: TY = Ti * TV; cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: Ip[0] = FNMS(Ti, TX, TW); cannam@95: Im[0] = FMA(Tf, TX, TY); cannam@95: TU = TK * TH; cannam@95: TI = Tq * TH; cannam@95: Im[WS(rs, 2)] = FMA(Tq, TT, TU); cannam@95: Ip[WS(rs, 2)] = FNMS(TK, TT, TI); cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_CEXP, 1, 7}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cb2_8", twinstr, &GENUS, {44, 20, 30, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb2_8, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hc2cb2_8 -include hc2cb.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 74 FP additions, 44 FP multiplications, cannam@95: * (or, 56 additions, 26 multiplications, 18 fused multiply/add), cannam@95: * 46 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "hc2cb.h" cannam@95: cannam@95: static void hc2cb2_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: 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(32, rs)) { cannam@95: E Tf, Ti, Tg, Tj, Tl, Tp, TP, TR, TF, TG, TH, T15, TL, TT; cannam@95: { cannam@95: E Th, To, Tk, Tn; cannam@95: Tf = W[0]; cannam@95: Ti = W[1]; cannam@95: Tg = W[2]; cannam@95: Tj = W[3]; cannam@95: Th = Tf * Tg; cannam@95: To = Ti * Tg; cannam@95: Tk = Ti * Tj; cannam@95: Tn = Tf * Tj; cannam@95: Tl = Th - Tk; cannam@95: Tp = Tn + To; cannam@95: TP = Th + Tk; cannam@95: TR = Tn - To; cannam@95: TF = W[4]; cannam@95: TG = W[5]; cannam@95: TH = FMA(Tf, TF, Ti * TG); cannam@95: T15 = FNMS(TR, TF, TP * TG); cannam@95: TL = FNMS(Ti, TF, Tf * TG); cannam@95: TT = FMA(TP, TF, TR * TG); cannam@95: } cannam@95: { cannam@95: E T7, T1f, T1i, Tw, TI, TW, T18, TM, Te, T19, T1a, TD, TJ, TZ, T12; cannam@95: E TN, Tm, TE; cannam@95: { cannam@95: E T3, TU, Ts, T17, T6, T16, Tv, TV; cannam@95: { cannam@95: E T1, T2, Tq, Tr; cannam@95: T1 = Rp[0]; cannam@95: T2 = Rm[WS(rs, 3)]; cannam@95: T3 = T1 + T2; cannam@95: TU = T1 - T2; cannam@95: Tq = Ip[0]; cannam@95: Tr = Im[WS(rs, 3)]; cannam@95: Ts = Tq - Tr; cannam@95: T17 = Tq + Tr; cannam@95: } cannam@95: { cannam@95: E T4, T5, Tt, Tu; cannam@95: T4 = Rp[WS(rs, 2)]; cannam@95: T5 = Rm[WS(rs, 1)]; cannam@95: T6 = T4 + T5; cannam@95: T16 = T4 - T5; cannam@95: Tt = Ip[WS(rs, 2)]; cannam@95: Tu = Im[WS(rs, 1)]; cannam@95: Tv = Tt - Tu; cannam@95: TV = Tt + Tu; cannam@95: } cannam@95: T7 = T3 + T6; cannam@95: T1f = TU + TV; cannam@95: T1i = T17 - T16; cannam@95: Tw = Ts + Tv; cannam@95: TI = T3 - T6; cannam@95: TW = TU - TV; cannam@95: T18 = T16 + T17; cannam@95: TM = Ts - Tv; cannam@95: } cannam@95: { cannam@95: E Ta, TX, Tz, TY, Td, T10, TC, T11; cannam@95: { cannam@95: E T8, T9, Tx, Ty; cannam@95: T8 = Rp[WS(rs, 1)]; cannam@95: T9 = Rm[WS(rs, 2)]; cannam@95: Ta = T8 + T9; cannam@95: TX = T8 - T9; cannam@95: Tx = Ip[WS(rs, 1)]; cannam@95: Ty = Im[WS(rs, 2)]; cannam@95: Tz = Tx - Ty; cannam@95: TY = Tx + Ty; cannam@95: } cannam@95: { cannam@95: E Tb, Tc, TA, TB; cannam@95: Tb = Rm[0]; cannam@95: Tc = Rp[WS(rs, 3)]; cannam@95: Td = Tb + Tc; cannam@95: T10 = Tb - Tc; cannam@95: TA = Ip[WS(rs, 3)]; cannam@95: TB = Im[0]; cannam@95: TC = TA - TB; cannam@95: T11 = TA + TB; cannam@95: } cannam@95: Te = Ta + Td; cannam@95: T19 = TX + TY; cannam@95: T1a = T10 + T11; cannam@95: TD = Tz + TC; cannam@95: TJ = TC - Tz; cannam@95: TZ = TX - TY; cannam@95: T12 = T10 - T11; cannam@95: TN = Ta - Td; cannam@95: } cannam@95: Rp[0] = T7 + Te; cannam@95: Rm[0] = Tw + TD; cannam@95: Tm = T7 - Te; cannam@95: TE = Tw - TD; cannam@95: Rp[WS(rs, 2)] = FNMS(Tp, TE, Tl * Tm); cannam@95: Rm[WS(rs, 2)] = FMA(Tp, Tm, Tl * TE); cannam@95: { cannam@95: E TQ, TS, TK, TO; cannam@95: TQ = TI + TJ; cannam@95: TS = TN + TM; cannam@95: Rp[WS(rs, 1)] = FNMS(TR, TS, TP * TQ); cannam@95: Rm[WS(rs, 1)] = FMA(TP, TS, TR * TQ); cannam@95: TK = TI - TJ; cannam@95: TO = TM - TN; cannam@95: Rp[WS(rs, 3)] = FNMS(TL, TO, TH * TK); cannam@95: Rm[WS(rs, 3)] = FMA(TH, TO, TL * TK); cannam@95: } cannam@95: { cannam@95: E T1h, T1l, T1k, T1m, T1g, T1j; cannam@95: T1g = KP707106781 * (T19 + T1a); cannam@95: T1h = T1f - T1g; cannam@95: T1l = T1f + T1g; cannam@95: T1j = KP707106781 * (TZ - T12); cannam@95: T1k = T1i + T1j; cannam@95: T1m = T1i - T1j; cannam@95: Ip[WS(rs, 1)] = FNMS(Tj, T1k, Tg * T1h); cannam@95: Im[WS(rs, 1)] = FMA(Tg, T1k, Tj * T1h); cannam@95: Ip[WS(rs, 3)] = FNMS(TG, T1m, TF * T1l); cannam@95: Im[WS(rs, 3)] = FMA(TF, T1m, TG * T1l); cannam@95: } cannam@95: { cannam@95: E T14, T1d, T1c, T1e, T13, T1b; cannam@95: T13 = KP707106781 * (TZ + T12); cannam@95: T14 = TW - T13; cannam@95: T1d = TW + T13; cannam@95: T1b = KP707106781 * (T19 - T1a); cannam@95: T1c = T18 - T1b; cannam@95: T1e = T18 + T1b; cannam@95: Ip[WS(rs, 2)] = FNMS(T15, T1c, TT * T14); cannam@95: Im[WS(rs, 2)] = FMA(T15, T14, TT * T1c); cannam@95: Ip[0] = FNMS(Ti, T1e, Tf * T1d); cannam@95: Im[0] = FMA(Ti, T1d, Tf * T1e); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: {TW_CEXP, 1, 1}, cannam@95: {TW_CEXP, 1, 3}, cannam@95: {TW_CEXP, 1, 7}, cannam@95: {TW_NEXT, 1, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 8, "hc2cb2_8", twinstr, &GENUS, {56, 26, 18, 0} }; cannam@95: cannam@95: void X(codelet_hc2cb2_8) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cb2_8, &desc, HC2C_VIA_RDFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */