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 08:07:10 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_hc2cdft.native -fma -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cfdft_8 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 82 FP additions, 52 FP multiplications, cannam@167: * (or, 60 additions, 30 multiplications, 22 fused multiply/add), cannam@167: * 31 stack variables, 2 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft_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: DK(KP500000000, +0.500000000000000000000000000000000000000000000); 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 Ty, T14, TO, T1o, Tv, T16, TG, T1m, Ta, T19, TV, T1h, Tk, T1b, T11; cannam@167: E T1j; cannam@167: { cannam@167: E Tw, Tx, TN, TI, TJ, TK; cannam@167: Tw = Ip[0]; cannam@167: Tx = Im[0]; cannam@167: TN = Tw + Tx; cannam@167: TI = Rm[0]; cannam@167: TJ = Rp[0]; cannam@167: TK = TI - TJ; cannam@167: Ty = Tw - Tx; cannam@167: T14 = TJ + TI; cannam@167: { cannam@167: E TH, TL, TM, T1n; cannam@167: TH = W[0]; cannam@167: TL = TH * TK; cannam@167: TM = W[1]; cannam@167: T1n = TM * TK; cannam@167: TO = FNMS(TM, TN, TL); cannam@167: T1o = FMA(TH, TN, T1n); cannam@167: } cannam@167: } cannam@167: { cannam@167: E Tp, TF, Tu, TC; cannam@167: { cannam@167: E Tn, To, Ts, Tt; cannam@167: Tn = Ip[WS(rs, 2)]; cannam@167: To = Im[WS(rs, 2)]; cannam@167: Tp = Tn - To; cannam@167: TF = Tn + To; cannam@167: Ts = Rp[WS(rs, 2)]; cannam@167: Tt = Rm[WS(rs, 2)]; cannam@167: Tu = Ts + Tt; cannam@167: TC = Tt - Ts; cannam@167: } cannam@167: { cannam@167: E Tq, T15, Tm, Tr; cannam@167: Tm = W[6]; cannam@167: Tq = Tm * Tp; cannam@167: T15 = Tm * Tu; cannam@167: Tr = W[7]; cannam@167: Tv = FNMS(Tr, Tu, Tq); cannam@167: T16 = FMA(Tr, Tp, T15); cannam@167: } cannam@167: { cannam@167: E TB, TD, TE, T1l; cannam@167: TB = W[8]; cannam@167: TD = TB * TC; cannam@167: TE = W[9]; cannam@167: T1l = TE * TC; cannam@167: TG = FNMS(TE, TF, TD); cannam@167: T1m = FMA(TB, TF, T1l); cannam@167: } cannam@167: } cannam@167: { cannam@167: E T4, TU, T9, TR; cannam@167: { cannam@167: E T2, T3, T7, T8; cannam@167: T2 = Ip[WS(rs, 1)]; cannam@167: T3 = Im[WS(rs, 1)]; cannam@167: T4 = T2 - T3; cannam@167: TU = T2 + T3; cannam@167: T7 = Rp[WS(rs, 1)]; cannam@167: T8 = Rm[WS(rs, 1)]; cannam@167: T9 = T7 + T8; cannam@167: TR = T7 - T8; cannam@167: } cannam@167: { cannam@167: E T5, T18, T1, T6; cannam@167: T1 = W[2]; cannam@167: T5 = T1 * T4; cannam@167: T18 = T1 * T9; cannam@167: T6 = W[3]; cannam@167: Ta = FNMS(T6, T9, T5); cannam@167: T19 = FMA(T6, T4, T18); cannam@167: } cannam@167: { cannam@167: E TS, T1g, TQ, TT; cannam@167: TQ = W[4]; cannam@167: TS = TQ * TR; cannam@167: T1g = TQ * TU; cannam@167: TT = W[5]; cannam@167: TV = FMA(TT, TU, TS); cannam@167: T1h = FNMS(TT, TR, T1g); cannam@167: } cannam@167: } cannam@167: { cannam@167: E Te, T10, Tj, TX; cannam@167: { cannam@167: E Tc, Td, Th, Ti; cannam@167: Tc = Ip[WS(rs, 3)]; cannam@167: Td = Im[WS(rs, 3)]; cannam@167: Te = Tc - Td; cannam@167: T10 = Tc + Td; cannam@167: Th = Rp[WS(rs, 3)]; cannam@167: Ti = Rm[WS(rs, 3)]; cannam@167: Tj = Th + Ti; cannam@167: TX = Th - Ti; cannam@167: } cannam@167: { cannam@167: E Tf, T1a, Tb, Tg; cannam@167: Tb = W[10]; cannam@167: Tf = Tb * Te; cannam@167: T1a = Tb * Tj; cannam@167: Tg = W[11]; cannam@167: Tk = FNMS(Tg, Tj, Tf); cannam@167: T1b = FMA(Tg, Te, T1a); cannam@167: } cannam@167: { cannam@167: E TY, T1i, TW, TZ; cannam@167: TW = W[12]; cannam@167: TY = TW * TX; cannam@167: T1i = TW * T10; cannam@167: TZ = W[13]; cannam@167: T11 = FMA(TZ, T10, TY); cannam@167: T1j = FNMS(TZ, TX, T1i); cannam@167: } cannam@167: } cannam@167: { cannam@167: E TA, T1f, T1q, T1s, T13, T1e, T1d, T1r; cannam@167: { cannam@167: E Tl, Tz, T1k, T1p; cannam@167: Tl = Ta + Tk; cannam@167: Tz = Tv + Ty; cannam@167: TA = Tl + Tz; cannam@167: T1f = Tz - Tl; cannam@167: T1k = T1h + T1j; cannam@167: T1p = T1m + T1o; cannam@167: T1q = T1k - T1p; cannam@167: T1s = T1k + T1p; cannam@167: } cannam@167: { cannam@167: E TP, T12, T17, T1c; cannam@167: TP = TG + TO; cannam@167: T12 = TV + T11; cannam@167: T13 = TP - T12; cannam@167: T1e = T12 + TP; cannam@167: T17 = T14 + T16; cannam@167: T1c = T19 + T1b; cannam@167: T1d = T17 - T1c; cannam@167: T1r = T17 + T1c; cannam@167: } cannam@167: Ip[0] = KP500000000 * (TA + T13); cannam@167: Rp[0] = KP500000000 * (T1r + T1s); cannam@167: Im[WS(rs, 3)] = KP500000000 * (T13 - TA); cannam@167: Rm[WS(rs, 3)] = KP500000000 * (T1r - T1s); cannam@167: Rm[WS(rs, 1)] = KP500000000 * (T1d - T1e); cannam@167: Im[WS(rs, 1)] = KP500000000 * (T1q - T1f); cannam@167: Rp[WS(rs, 2)] = KP500000000 * (T1d + T1e); cannam@167: Ip[WS(rs, 2)] = KP500000000 * (T1f + T1q); cannam@167: } cannam@167: { cannam@167: E T1v, T1H, T1F, T1L, T1y, T1I, T1B, T1J; cannam@167: { cannam@167: E T1t, T1u, T1D, T1E; cannam@167: T1t = Ty - Tv; cannam@167: T1u = T19 - T1b; cannam@167: T1v = T1t - T1u; cannam@167: T1H = T1u + T1t; cannam@167: T1D = T14 - T16; cannam@167: T1E = Ta - Tk; cannam@167: T1F = T1D - T1E; cannam@167: T1L = T1D + T1E; cannam@167: } cannam@167: { cannam@167: E T1w, T1x, T1z, T1A; cannam@167: T1w = T1j - T1h; cannam@167: T1x = TV - T11; cannam@167: T1y = T1w + T1x; cannam@167: T1I = T1w - T1x; cannam@167: T1z = TO - TG; cannam@167: T1A = T1o - T1m; cannam@167: T1B = T1z - T1A; cannam@167: T1J = T1z + T1A; cannam@167: } cannam@167: { cannam@167: E T1C, T1M, T1G, T1K; cannam@167: T1C = T1y + T1B; cannam@167: Ip[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1C, T1v)); cannam@167: Im[WS(rs, 2)] = -(KP500000000 * (FNMS(KP707106781, T1C, T1v))); cannam@167: T1M = T1I + T1J; cannam@167: Rm[WS(rs, 2)] = KP500000000 * (FNMS(KP707106781, T1M, T1L)); cannam@167: Rp[WS(rs, 1)] = KP500000000 * (FMA(KP707106781, T1M, T1L)); cannam@167: T1G = T1B - T1y; cannam@167: Rm[0] = KP500000000 * (FNMS(KP707106781, T1G, T1F)); cannam@167: Rp[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1G, T1F)); cannam@167: T1K = T1I - T1J; cannam@167: Ip[WS(rs, 3)] = KP500000000 * (FMA(KP707106781, T1K, T1H)); cannam@167: Im[0] = -(KP500000000 * (FNMS(KP707106781, T1K, T1H))); 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, "hc2cfdft_8", twinstr, &GENUS, {60, 30, 22, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft_8) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft_8, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2cdft.native -compact -variables 4 -pipeline-latency 4 -n 8 -dit -name hc2cfdft_8 -include rdft/scalar/hc2cf.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 82 FP additions, 44 FP multiplications, cannam@167: * (or, 68 additions, 30 multiplications, 14 fused multiply/add), cannam@167: * 39 stack variables, 2 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hc2cf.h" cannam@167: cannam@167: static void hc2cfdft_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(KP353553390, +0.353553390593273762200422181052424519642417969); cannam@167: DK(KP500000000, +0.500000000000000000000000000000000000000000000); 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 Tv, TX, Ts, TY, TE, T1a, TJ, T19, T1l, T1m, T9, T10, Ti, T11, TP; cannam@167: E T16, TU, T17, T1i, T1j; cannam@167: { cannam@167: E Tt, Tu, TD, Tz, TA, TB, Tn, TI, Tr, TG, Tk, To; cannam@167: Tt = Ip[0]; cannam@167: Tu = Im[0]; cannam@167: TD = Tt + Tu; cannam@167: Tz = Rm[0]; cannam@167: TA = Rp[0]; cannam@167: TB = Tz - TA; cannam@167: { cannam@167: E Tl, Tm, Tp, Tq; cannam@167: Tl = Ip[WS(rs, 2)]; cannam@167: Tm = Im[WS(rs, 2)]; cannam@167: Tn = Tl - Tm; cannam@167: TI = Tl + Tm; cannam@167: Tp = Rp[WS(rs, 2)]; cannam@167: Tq = Rm[WS(rs, 2)]; cannam@167: Tr = Tp + Tq; cannam@167: TG = Tp - Tq; cannam@167: } cannam@167: Tv = Tt - Tu; cannam@167: TX = TA + Tz; cannam@167: Tk = W[6]; cannam@167: To = W[7]; cannam@167: Ts = FNMS(To, Tr, Tk * Tn); cannam@167: TY = FMA(Tk, Tr, To * Tn); cannam@167: { cannam@167: E Ty, TC, TF, TH; cannam@167: Ty = W[0]; cannam@167: TC = W[1]; cannam@167: TE = FNMS(TC, TD, Ty * TB); cannam@167: T1a = FMA(TC, TB, Ty * TD); cannam@167: TF = W[8]; cannam@167: TH = W[9]; cannam@167: TJ = FMA(TF, TG, TH * TI); cannam@167: T19 = FNMS(TH, TG, TF * TI); cannam@167: } cannam@167: T1l = TJ + TE; cannam@167: T1m = T1a - T19; cannam@167: } cannam@167: { cannam@167: E T4, TO, T8, TM, Td, TT, Th, TR; cannam@167: { cannam@167: E T2, T3, T6, T7; cannam@167: T2 = Ip[WS(rs, 1)]; cannam@167: T3 = Im[WS(rs, 1)]; cannam@167: T4 = T2 - T3; cannam@167: TO = T2 + T3; cannam@167: T6 = Rp[WS(rs, 1)]; cannam@167: T7 = Rm[WS(rs, 1)]; cannam@167: T8 = T6 + T7; cannam@167: TM = T6 - T7; cannam@167: } cannam@167: { cannam@167: E Tb, Tc, Tf, Tg; cannam@167: Tb = Ip[WS(rs, 3)]; cannam@167: Tc = Im[WS(rs, 3)]; cannam@167: Td = Tb - Tc; cannam@167: TT = Tb + Tc; cannam@167: Tf = Rp[WS(rs, 3)]; cannam@167: Tg = Rm[WS(rs, 3)]; cannam@167: Th = Tf + Tg; cannam@167: TR = Tf - Tg; cannam@167: } cannam@167: { cannam@167: E T1, T5, Ta, Te; cannam@167: T1 = W[2]; cannam@167: T5 = W[3]; cannam@167: T9 = FNMS(T5, T8, T1 * T4); cannam@167: T10 = FMA(T1, T8, T5 * T4); cannam@167: Ta = W[10]; cannam@167: Te = W[11]; cannam@167: Ti = FNMS(Te, Th, Ta * Td); cannam@167: T11 = FMA(Ta, Th, Te * Td); cannam@167: { cannam@167: E TL, TN, TQ, TS; cannam@167: TL = W[4]; cannam@167: TN = W[5]; cannam@167: TP = FMA(TL, TM, TN * TO); cannam@167: T16 = FNMS(TN, TM, TL * TO); cannam@167: TQ = W[12]; cannam@167: TS = W[13]; cannam@167: TU = FMA(TQ, TR, TS * TT); cannam@167: T17 = FNMS(TS, TR, TQ * TT); cannam@167: } cannam@167: T1i = T17 - T16; cannam@167: T1j = TP - TU; cannam@167: } cannam@167: } cannam@167: { cannam@167: E T1h, T1t, T1w, T1y, T1o, T1s, T1r, T1x; cannam@167: { cannam@167: E T1f, T1g, T1u, T1v; cannam@167: T1f = Tv - Ts; cannam@167: T1g = T10 - T11; cannam@167: T1h = KP500000000 * (T1f - T1g); cannam@167: T1t = KP500000000 * (T1g + T1f); cannam@167: T1u = T1i - T1j; cannam@167: T1v = T1l + T1m; cannam@167: T1w = KP353553390 * (T1u - T1v); cannam@167: T1y = KP353553390 * (T1u + T1v); cannam@167: } cannam@167: { cannam@167: E T1k, T1n, T1p, T1q; cannam@167: T1k = T1i + T1j; cannam@167: T1n = T1l - T1m; cannam@167: T1o = KP353553390 * (T1k + T1n); cannam@167: T1s = KP353553390 * (T1n - T1k); cannam@167: T1p = TX - TY; cannam@167: T1q = T9 - Ti; cannam@167: T1r = KP500000000 * (T1p - T1q); cannam@167: T1x = KP500000000 * (T1p + T1q); cannam@167: } cannam@167: Ip[WS(rs, 1)] = T1h + T1o; cannam@167: Rp[WS(rs, 1)] = T1x + T1y; cannam@167: Im[WS(rs, 2)] = T1o - T1h; cannam@167: Rm[WS(rs, 2)] = T1x - T1y; cannam@167: Rm[0] = T1r - T1s; cannam@167: Im[0] = T1w - T1t; cannam@167: Rp[WS(rs, 3)] = T1r + T1s; cannam@167: Ip[WS(rs, 3)] = T1t + T1w; cannam@167: } cannam@167: { cannam@167: E Tx, T15, T1c, T1e, TW, T14, T13, T1d; cannam@167: { cannam@167: E Tj, Tw, T18, T1b; cannam@167: Tj = T9 + Ti; cannam@167: Tw = Ts + Tv; cannam@167: Tx = Tj + Tw; cannam@167: T15 = Tw - Tj; cannam@167: T18 = T16 + T17; cannam@167: T1b = T19 + T1a; cannam@167: T1c = T18 - T1b; cannam@167: T1e = T18 + T1b; cannam@167: } cannam@167: { cannam@167: E TK, TV, TZ, T12; cannam@167: TK = TE - TJ; cannam@167: TV = TP + TU; cannam@167: TW = TK - TV; cannam@167: T14 = TV + TK; cannam@167: TZ = TX + TY; cannam@167: T12 = T10 + T11; cannam@167: T13 = TZ - T12; cannam@167: T1d = TZ + T12; cannam@167: } cannam@167: Ip[0] = KP500000000 * (Tx + TW); cannam@167: Rp[0] = KP500000000 * (T1d + T1e); cannam@167: Im[WS(rs, 3)] = KP500000000 * (TW - Tx); cannam@167: Rm[WS(rs, 3)] = KP500000000 * (T1d - T1e); cannam@167: Rm[WS(rs, 1)] = KP500000000 * (T13 - T14); cannam@167: Im[WS(rs, 1)] = KP500000000 * (T1c - T15); cannam@167: Rp[WS(rs, 2)] = KP500000000 * (T13 + T14); cannam@167: Ip[WS(rs, 2)] = KP500000000 * (T15 + T1c); 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, "hc2cfdft_8", twinstr, &GENUS, {68, 30, 14, 0} }; cannam@167: cannam@167: void X(codelet_hc2cfdft_8) (planner *p) { cannam@167: X(khc2c_register) (p, hc2cfdft_8, &desc, HC2C_VIA_DFT); cannam@167: } cannam@167: #endif