cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:06:55 EDT 2018 */ cannam@167: cannam@167: #include "rdft/codelet-rdft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2c.native -fma -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cf_8 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 66 FP additions, 36 FP multiplications, cannam@167: * (or, 44 additions, 14 multiplications, 22 fused multiply/add), cannam@167: * 34 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cf_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) { cannam@167: E T1, T1m, T7, T1l, Tk, TS, Te, TQ, TF, T14, TL, T16, T12, T17, Ts; cannam@167: E TX, Ty, TZ, TV, T10; cannam@167: T1 = Rp[0]; cannam@167: T1m = Rm[0]; cannam@167: { cannam@167: E T3, T6, T4, T1k, T2, T5; cannam@167: T3 = Rp[WS(rs, 2)]; cannam@167: T6 = Rm[WS(rs, 2)]; cannam@167: T2 = W[6]; cannam@167: T4 = T2 * T3; cannam@167: T1k = T2 * T6; cannam@167: T5 = W[7]; cannam@167: T7 = FMA(T5, T6, T4); cannam@167: T1l = FNMS(T5, T3, T1k); cannam@167: } cannam@167: { cannam@167: E Tg, Tj, Th, TR, Tf, Ti; cannam@167: Tg = Rp[WS(rs, 3)]; cannam@167: Tj = Rm[WS(rs, 3)]; cannam@167: Tf = W[10]; cannam@167: Th = Tf * Tg; cannam@167: TR = Tf * Tj; cannam@167: Ti = W[11]; cannam@167: Tk = FMA(Ti, Tj, Th); cannam@167: TS = FNMS(Ti, Tg, TR); cannam@167: } cannam@167: { cannam@167: E Ta, Td, Tb, TP, T9, Tc; cannam@167: Ta = Rp[WS(rs, 1)]; cannam@167: Td = Rm[WS(rs, 1)]; cannam@167: T9 = W[2]; cannam@167: Tb = T9 * Ta; cannam@167: TP = T9 * Td; cannam@167: Tc = W[3]; cannam@167: Te = FMA(Tc, Td, Tb); cannam@167: TQ = FNMS(Tc, Ta, TP); cannam@167: } cannam@167: { cannam@167: E TB, TE, TC, T13, TH, TK, TI, T15, TA, TG, TD, TJ; cannam@167: TB = Ip[WS(rs, 3)]; cannam@167: TE = Im[WS(rs, 3)]; cannam@167: TA = W[12]; cannam@167: TC = TA * TB; cannam@167: T13 = TA * TE; cannam@167: TH = Ip[WS(rs, 1)]; cannam@167: TK = Im[WS(rs, 1)]; cannam@167: TG = W[4]; cannam@167: TI = TG * TH; cannam@167: T15 = TG * TK; cannam@167: TD = W[13]; cannam@167: TF = FMA(TD, TE, TC); cannam@167: T14 = FNMS(TD, TB, T13); cannam@167: TJ = W[5]; cannam@167: TL = FMA(TJ, TK, TI); cannam@167: T16 = FNMS(TJ, TH, T15); cannam@167: T12 = TF - TL; cannam@167: T17 = T14 - T16; cannam@167: } cannam@167: { cannam@167: E To, Tr, Tp, TW, Tu, Tx, Tv, TY, Tn, Tt, Tq, Tw; cannam@167: To = Ip[0]; cannam@167: Tr = Im[0]; cannam@167: Tn = W[0]; cannam@167: Tp = Tn * To; cannam@167: TW = Tn * Tr; cannam@167: Tu = Ip[WS(rs, 2)]; cannam@167: Tx = Im[WS(rs, 2)]; cannam@167: Tt = W[8]; cannam@167: Tv = Tt * Tu; cannam@167: TY = Tt * Tx; cannam@167: Tq = W[1]; cannam@167: Ts = FMA(Tq, Tr, Tp); cannam@167: TX = FNMS(Tq, To, TW); cannam@167: Tw = W[9]; cannam@167: Ty = FMA(Tw, Tx, Tv); cannam@167: TZ = FNMS(Tw, Tu, TY); cannam@167: TV = Ts - Ty; cannam@167: T10 = TX - TZ; cannam@167: } cannam@167: { cannam@167: E TU, T1a, T1t, T1v, T19, T1w, T1d, T1u; cannam@167: { cannam@167: E TO, TT, T1r, T1s; cannam@167: TO = T1 - T7; cannam@167: TT = TQ - TS; cannam@167: TU = TO + TT; cannam@167: T1a = TO - TT; cannam@167: T1r = T1m - T1l; cannam@167: T1s = Te - Tk; cannam@167: T1t = T1r - T1s; cannam@167: T1v = T1s + T1r; cannam@167: } cannam@167: { cannam@167: E T11, T18, T1b, T1c; cannam@167: T11 = TV + T10; cannam@167: T18 = T12 - T17; cannam@167: T19 = T11 + T18; cannam@167: T1w = T18 - T11; cannam@167: T1b = T10 - TV; cannam@167: T1c = T12 + T17; cannam@167: T1d = T1b - T1c; cannam@167: T1u = T1b + T1c; cannam@167: } cannam@167: Rm[WS(rs, 2)] = FNMS(KP707106781, T19, TU); cannam@167: Im[WS(rs, 2)] = FMS(KP707106781, T1u, T1t); cannam@167: Rp[WS(rs, 1)] = FMA(KP707106781, T19, TU); cannam@167: Ip[WS(rs, 1)] = FMA(KP707106781, T1u, T1t); cannam@167: Rm[0] = FNMS(KP707106781, T1d, T1a); cannam@167: Im[0] = FMS(KP707106781, T1w, T1v); cannam@167: Rp[WS(rs, 3)] = FMA(KP707106781, T1d, T1a); cannam@167: Ip[WS(rs, 3)] = FMA(KP707106781, T1w, T1v); cannam@167: } cannam@167: { cannam@167: E Tm, T1e, T1o, T1q, TN, T1p, T1h, T1i; cannam@167: { cannam@167: E T8, Tl, T1j, T1n; cannam@167: T8 = T1 + T7; cannam@167: Tl = Te + Tk; cannam@167: Tm = T8 + Tl; cannam@167: T1e = T8 - Tl; cannam@167: T1j = TQ + TS; cannam@167: T1n = T1l + T1m; cannam@167: T1o = T1j + T1n; cannam@167: T1q = T1n - T1j; cannam@167: } cannam@167: { cannam@167: E Tz, TM, T1f, T1g; cannam@167: Tz = Ts + Ty; cannam@167: TM = TF + TL; cannam@167: TN = Tz + TM; cannam@167: T1p = TM - Tz; cannam@167: T1f = TX + TZ; cannam@167: T1g = T14 + T16; cannam@167: T1h = T1f - T1g; cannam@167: T1i = T1f + T1g; cannam@167: } cannam@167: Rm[WS(rs, 3)] = Tm - TN; cannam@167: Im[WS(rs, 3)] = T1i - T1o; cannam@167: Rp[0] = Tm + TN; cannam@167: Ip[0] = T1i + T1o; cannam@167: Rm[WS(rs, 1)] = T1e - T1h; cannam@167: Im[WS(rs, 1)] = T1p - T1q; cannam@167: Rp[WS(rs, 2)] = T1e + T1h; cannam@167: Ip[WS(rs, 2)] = T1p + T1q; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 8}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 8, "hc2cf_8", twinstr, &GENUS, {44, 14, 22, 0} }; cannam@167: cannam@167: void X(codelet_hc2cf_8) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cf_8, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2c.native -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cf_8 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 66 FP additions, 32 FP multiplications, cannam@167: * (or, 52 additions, 18 multiplications, 14 fused multiply/add), cannam@167: * 28 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cf_8(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@167: { cannam@167: INT m; cannam@167: for (m = mb, W = W + ((mb - 1) * 14); m < me; m = m + 1, Rp = Rp + ms, Ip = Ip + ms, Rm = Rm - ms, Im = Im - ms, W = W + 14, MAKE_VOLATILE_STRIDE(32, rs)) { cannam@167: E T7, T1e, TH, T19, TF, T13, TR, TU, Ti, T1f, TK, T16, Tu, T12, TM; cannam@167: E TP; cannam@167: { cannam@167: E T1, T18, T6, T17; cannam@167: T1 = Rp[0]; cannam@167: T18 = Rm[0]; cannam@167: { cannam@167: E T3, T5, T2, T4; cannam@167: T3 = Rp[WS(rs, 2)]; cannam@167: T5 = Rm[WS(rs, 2)]; cannam@167: T2 = W[6]; cannam@167: T4 = W[7]; cannam@167: T6 = FMA(T2, T3, T4 * T5); cannam@167: T17 = FNMS(T4, T3, T2 * T5); cannam@167: } cannam@167: T7 = T1 + T6; cannam@167: T1e = T18 - T17; cannam@167: TH = T1 - T6; cannam@167: T19 = T17 + T18; cannam@167: } cannam@167: { cannam@167: E Tz, TS, TE, TT; cannam@167: { cannam@167: E Tw, Ty, Tv, Tx; cannam@167: Tw = Ip[WS(rs, 3)]; cannam@167: Ty = Im[WS(rs, 3)]; cannam@167: Tv = W[12]; cannam@167: Tx = W[13]; cannam@167: Tz = FMA(Tv, Tw, Tx * Ty); cannam@167: TS = FNMS(Tx, Tw, Tv * Ty); cannam@167: } cannam@167: { cannam@167: E TB, TD, TA, TC; cannam@167: TB = Ip[WS(rs, 1)]; cannam@167: TD = Im[WS(rs, 1)]; cannam@167: TA = W[4]; cannam@167: TC = W[5]; cannam@167: TE = FMA(TA, TB, TC * TD); cannam@167: TT = FNMS(TC, TB, TA * TD); cannam@167: } cannam@167: TF = Tz + TE; cannam@167: T13 = TS + TT; cannam@167: TR = Tz - TE; cannam@167: TU = TS - TT; cannam@167: } cannam@167: { cannam@167: E Tc, TI, Th, TJ; cannam@167: { cannam@167: E T9, Tb, T8, Ta; cannam@167: T9 = Rp[WS(rs, 1)]; cannam@167: Tb = Rm[WS(rs, 1)]; cannam@167: T8 = W[2]; cannam@167: Ta = W[3]; cannam@167: Tc = FMA(T8, T9, Ta * Tb); cannam@167: TI = FNMS(Ta, T9, T8 * Tb); cannam@167: } cannam@167: { cannam@167: E Te, Tg, Td, Tf; cannam@167: Te = Rp[WS(rs, 3)]; cannam@167: Tg = Rm[WS(rs, 3)]; cannam@167: Td = W[10]; cannam@167: Tf = W[11]; cannam@167: Th = FMA(Td, Te, Tf * Tg); cannam@167: TJ = FNMS(Tf, Te, Td * Tg); cannam@167: } cannam@167: Ti = Tc + Th; cannam@167: T1f = Tc - Th; cannam@167: TK = TI - TJ; cannam@167: T16 = TI + TJ; cannam@167: } cannam@167: { cannam@167: E To, TN, Tt, TO; cannam@167: { cannam@167: E Tl, Tn, Tk, Tm; cannam@167: Tl = Ip[0]; cannam@167: Tn = Im[0]; cannam@167: Tk = W[0]; cannam@167: Tm = W[1]; cannam@167: To = FMA(Tk, Tl, Tm * Tn); cannam@167: TN = FNMS(Tm, Tl, Tk * Tn); cannam@167: } cannam@167: { cannam@167: E Tq, Ts, Tp, Tr; cannam@167: Tq = Ip[WS(rs, 2)]; cannam@167: Ts = Im[WS(rs, 2)]; cannam@167: Tp = W[8]; cannam@167: Tr = W[9]; cannam@167: Tt = FMA(Tp, Tq, Tr * Ts); cannam@167: TO = FNMS(Tr, Tq, Tp * Ts); cannam@167: } cannam@167: Tu = To + Tt; cannam@167: T12 = TN + TO; cannam@167: TM = To - Tt; cannam@167: TP = TN - TO; cannam@167: } cannam@167: { cannam@167: E Tj, TG, T1b, T1c; cannam@167: Tj = T7 + Ti; cannam@167: TG = Tu + TF; cannam@167: Rm[WS(rs, 3)] = Tj - TG; cannam@167: Rp[0] = Tj + TG; cannam@167: { cannam@167: E T15, T1a, T11, T14; cannam@167: T15 = T12 + T13; cannam@167: T1a = T16 + T19; cannam@167: Im[WS(rs, 3)] = T15 - T1a; cannam@167: Ip[0] = T15 + T1a; cannam@167: T11 = T7 - Ti; cannam@167: T14 = T12 - T13; cannam@167: Rm[WS(rs, 1)] = T11 - T14; cannam@167: Rp[WS(rs, 2)] = T11 + T14; cannam@167: } cannam@167: T1b = TF - Tu; cannam@167: T1c = T19 - T16; cannam@167: Im[WS(rs, 1)] = T1b - T1c; cannam@167: Ip[WS(rs, 2)] = T1b + T1c; cannam@167: { cannam@167: E TX, T1g, T10, T1d, TY, TZ; cannam@167: TX = TH - TK; cannam@167: T1g = T1e - T1f; cannam@167: TY = TP - TM; cannam@167: TZ = TR + TU; cannam@167: T10 = KP707106781 * (TY - TZ); cannam@167: T1d = KP707106781 * (TY + TZ); cannam@167: Rm[0] = TX - T10; cannam@167: Ip[WS(rs, 1)] = T1d + T1g; cannam@167: Rp[WS(rs, 3)] = TX + T10; cannam@167: Im[WS(rs, 2)] = T1d - T1g; cannam@167: } cannam@167: { cannam@167: E TL, T1i, TW, T1h, TQ, TV; cannam@167: TL = TH + TK; cannam@167: T1i = T1f + T1e; cannam@167: TQ = TM + TP; cannam@167: TV = TR - TU; cannam@167: TW = KP707106781 * (TQ + TV); cannam@167: T1h = KP707106781 * (TV - TQ); cannam@167: Rm[WS(rs, 2)] = TL - TW; cannam@167: Ip[WS(rs, 3)] = T1h + T1i; cannam@167: Rp[WS(rs, 1)] = TL + TW; cannam@167: Im[0] = T1h - T1i; cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_FULL, 1, 8}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2c_desc desc = { 8, "hc2cf_8", twinstr, &GENUS, {52, 18, 14, 0} }; cannam@167: cannam@167: void X(codelet_hc2cf_8) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cf_8, &desc, HC2C_VIA_RDFT); cannam@167: } cannam@167: #endif