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 08:07:37 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_hc2hc.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hb2_8 -include rdft/scalar/hb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 74 FP additions, 50 FP multiplications, cannam@167: * (or, 44 additions, 20 multiplications, 30 fused multiply/add), cannam@167: * 47 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hb.h" cannam@167: cannam@167: static void hb2_8(R *cr, R *ci, 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) * 6); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E Tf, Tg, Tl, Tp, Ti, Tj, Tk, T1b, T1u, T1e, T1o, To, Tq, TK; cannam@167: { cannam@167: E Th, T1n, T1t, Tn, Tm, TJ; cannam@167: Tf = W[0]; cannam@167: Tg = W[2]; cannam@167: Th = Tf * Tg; cannam@167: Tl = W[4]; cannam@167: T1n = Tf * Tl; cannam@167: Tp = W[5]; cannam@167: T1t = Tf * Tp; cannam@167: Ti = W[1]; cannam@167: Tj = W[3]; cannam@167: Tn = Tf * Tj; cannam@167: Tk = FMA(Ti, Tj, Th); cannam@167: T1b = FNMS(Ti, Tj, Th); cannam@167: T1u = FNMS(Ti, Tl, T1t); cannam@167: T1e = FMA(Ti, Tg, Tn); cannam@167: T1o = FMA(Ti, Tp, T1n); cannam@167: Tm = Tk * Tl; cannam@167: TJ = Tk * Tp; cannam@167: To = FNMS(Ti, Tg, Tn); cannam@167: Tq = FMA(To, Tp, Tm); cannam@167: TK = FNMS(To, Tl, TJ); cannam@167: } cannam@167: { cannam@167: E T7, T1p, T1v, Tv, TP, T13, T1h, TZ, Te, T1k, T1w, T1q, TQ, TR, T10; cannam@167: E TG, T14; cannam@167: { cannam@167: E T3, Tr, TO, T1f, T6, TL, Tu, T1g; cannam@167: { cannam@167: E T1, T2, TM, TN; cannam@167: T1 = cr[0]; cannam@167: T2 = ci[WS(rs, 3)]; cannam@167: T3 = T1 + T2; cannam@167: Tr = T1 - T2; cannam@167: TM = ci[WS(rs, 7)]; cannam@167: TN = cr[WS(rs, 4)]; cannam@167: TO = TM + TN; cannam@167: T1f = TM - TN; cannam@167: } cannam@167: { cannam@167: E T4, T5, Ts, Tt; cannam@167: T4 = cr[WS(rs, 2)]; cannam@167: T5 = ci[WS(rs, 1)]; cannam@167: T6 = T4 + T5; cannam@167: TL = T4 - T5; cannam@167: Ts = ci[WS(rs, 5)]; cannam@167: Tt = cr[WS(rs, 6)]; cannam@167: Tu = Ts + Tt; cannam@167: T1g = Ts - Tt; cannam@167: } cannam@167: T7 = T3 + T6; cannam@167: T1p = T3 - T6; cannam@167: T1v = T1f - T1g; cannam@167: Tv = Tr - Tu; cannam@167: TP = TL + TO; cannam@167: T13 = TO - TL; cannam@167: T1h = T1f + T1g; cannam@167: TZ = Tr + Tu; cannam@167: } cannam@167: { cannam@167: E Ta, Tw, TE, T1j, Td, TB, Tz, T1i, TA, TF; cannam@167: { cannam@167: E T8, T9, TC, TD; cannam@167: T8 = cr[WS(rs, 1)]; cannam@167: T9 = ci[WS(rs, 2)]; cannam@167: Ta = T8 + T9; cannam@167: Tw = T8 - T9; cannam@167: TC = ci[WS(rs, 4)]; cannam@167: TD = cr[WS(rs, 7)]; cannam@167: TE = TC + TD; cannam@167: T1j = TC - TD; cannam@167: } cannam@167: { cannam@167: E Tb, Tc, Tx, Ty; cannam@167: Tb = ci[0]; cannam@167: Tc = cr[WS(rs, 3)]; cannam@167: Td = Tb + Tc; cannam@167: TB = Tb - Tc; cannam@167: Tx = ci[WS(rs, 6)]; cannam@167: Ty = cr[WS(rs, 5)]; cannam@167: Tz = Tx + Ty; cannam@167: T1i = Tx - Ty; cannam@167: } cannam@167: Te = Ta + Td; cannam@167: T1k = T1i + T1j; cannam@167: T1w = Ta - Td; cannam@167: T1q = T1j - T1i; cannam@167: TQ = Tw + Tz; cannam@167: TR = TB + TE; cannam@167: T10 = TQ + TR; cannam@167: TA = Tw - Tz; cannam@167: TF = TB - TE; cannam@167: TG = TA + TF; cannam@167: T14 = TA - TF; cannam@167: } cannam@167: cr[0] = T7 + Te; cannam@167: ci[0] = T1h + T1k; cannam@167: { cannam@167: E T11, T12, T15, T16; cannam@167: T11 = FNMS(KP707106781, T10, TZ); cannam@167: T12 = Tg * T11; cannam@167: T15 = FMA(KP707106781, T14, T13); cannam@167: T16 = Tg * T15; cannam@167: cr[WS(rs, 3)] = FNMS(Tj, T15, T12); cannam@167: ci[WS(rs, 3)] = FMA(Tj, T11, T16); cannam@167: } cannam@167: { cannam@167: E T1z, T1A, T1B, T1C; cannam@167: T1z = T1p + T1q; cannam@167: T1A = Tk * T1z; cannam@167: T1B = T1w + T1v; cannam@167: T1C = Tk * T1B; cannam@167: cr[WS(rs, 2)] = FNMS(To, T1B, T1A); cannam@167: ci[WS(rs, 2)] = FMA(To, T1z, T1C); cannam@167: } cannam@167: { cannam@167: E T17, T18, T19, T1a; cannam@167: T17 = FMA(KP707106781, T10, TZ); cannam@167: T18 = Tl * T17; cannam@167: T19 = FNMS(KP707106781, T14, T13); cannam@167: T1a = Tl * T19; cannam@167: cr[WS(rs, 7)] = FNMS(Tp, T19, T18); cannam@167: ci[WS(rs, 7)] = FMA(Tp, T17, T1a); cannam@167: } cannam@167: { cannam@167: E T1l, T1d, T1m, T1c; cannam@167: T1l = T1h - T1k; cannam@167: T1c = T7 - Te; cannam@167: T1d = T1b * T1c; cannam@167: T1m = T1e * T1c; cannam@167: cr[WS(rs, 4)] = FNMS(T1e, T1l, T1d); cannam@167: ci[WS(rs, 4)] = FMA(T1b, T1l, T1m); cannam@167: } cannam@167: { cannam@167: E T1r, T1s, T1x, T1y; cannam@167: T1r = T1p - T1q; cannam@167: T1s = T1o * T1r; cannam@167: T1x = T1v - T1w; cannam@167: T1y = T1o * T1x; cannam@167: cr[WS(rs, 6)] = FNMS(T1u, T1x, T1s); cannam@167: ci[WS(rs, 6)] = FMA(T1u, T1r, T1y); cannam@167: } cannam@167: { cannam@167: E TT, TX, TW, TY, TI, TU, TS, TV, TH; cannam@167: TS = TQ - TR; cannam@167: TT = FNMS(KP707106781, TS, TP); cannam@167: TX = FMA(KP707106781, TS, TP); cannam@167: TV = FMA(KP707106781, TG, Tv); cannam@167: TW = Tf * TV; cannam@167: TY = Ti * TV; cannam@167: TH = FNMS(KP707106781, TG, Tv); cannam@167: TI = Tq * TH; cannam@167: TU = TK * TH; cannam@167: cr[WS(rs, 5)] = FNMS(TK, TT, TI); cannam@167: ci[WS(rs, 5)] = FMA(Tq, TT, TU); cannam@167: cr[WS(rs, 1)] = FNMS(Ti, TX, TW); cannam@167: ci[WS(rs, 1)] = FMA(Tf, TX, TY); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_CEXP, 1, 1}, cannam@167: {TW_CEXP, 1, 3}, cannam@167: {TW_CEXP, 1, 7}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 8, "hb2_8", twinstr, &GENUS, {44, 20, 30, 0} }; cannam@167: cannam@167: void X(codelet_hb2_8) (planner *p) { cannam@167: X(khc2hc_register) (p, hb2_8, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_hc2hc.native -compact -variables 4 -pipeline-latency 4 -sign 1 -twiddle-log3 -precompute-twiddles -n 8 -dif -name hb2_8 -include rdft/scalar/hb.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 74 FP additions, 44 FP multiplications, cannam@167: * (or, 56 additions, 26 multiplications, 18 fused multiply/add), cannam@167: * 46 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "rdft/scalar/hb.h" cannam@167: cannam@167: static void hb2_8(R *cr, R *ci, 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) * 6); m < me; m = m + 1, cr = cr + ms, ci = ci - ms, W = W + 6, MAKE_VOLATILE_STRIDE(16, rs)) { cannam@167: E Tf, Ti, Tg, Tj, Tl, Tp, TP, TR, TF, TG, TH, T15, TL, TT; cannam@167: { cannam@167: E Th, To, Tk, Tn; cannam@167: Tf = W[0]; cannam@167: Ti = W[1]; cannam@167: Tg = W[2]; cannam@167: Tj = W[3]; cannam@167: Th = Tf * Tg; cannam@167: To = Ti * Tg; cannam@167: Tk = Ti * Tj; cannam@167: Tn = Tf * Tj; cannam@167: Tl = Th - Tk; cannam@167: Tp = Tn + To; cannam@167: TP = Th + Tk; cannam@167: TR = Tn - To; cannam@167: TF = W[4]; cannam@167: TG = W[5]; cannam@167: TH = FMA(Tf, TF, Ti * TG); cannam@167: T15 = FNMS(TR, TF, TP * TG); cannam@167: TL = FNMS(Ti, TF, Tf * TG); cannam@167: TT = FMA(TP, TF, TR * TG); cannam@167: } cannam@167: { cannam@167: E T7, T1f, T1i, Tw, TI, TW, T18, TM, Te, T19, T1a, TD, TJ, TZ, T12; cannam@167: E TN, Tm, TE; cannam@167: { cannam@167: E T3, TU, Tv, TV, T6, T16, Ts, T17; cannam@167: { cannam@167: E T1, T2, Tt, Tu; cannam@167: T1 = cr[0]; cannam@167: T2 = ci[WS(rs, 3)]; cannam@167: T3 = T1 + T2; cannam@167: TU = T1 - T2; cannam@167: Tt = ci[WS(rs, 5)]; cannam@167: Tu = cr[WS(rs, 6)]; cannam@167: Tv = Tt - Tu; cannam@167: TV = Tt + Tu; cannam@167: } cannam@167: { cannam@167: E T4, T5, Tq, Tr; cannam@167: T4 = cr[WS(rs, 2)]; cannam@167: T5 = ci[WS(rs, 1)]; cannam@167: T6 = T4 + T5; cannam@167: T16 = T4 - T5; cannam@167: Tq = ci[WS(rs, 7)]; cannam@167: Tr = cr[WS(rs, 4)]; cannam@167: Ts = Tq - Tr; cannam@167: T17 = Tq + Tr; cannam@167: } cannam@167: T7 = T3 + T6; cannam@167: T1f = TU + TV; cannam@167: T1i = T17 - T16; cannam@167: Tw = Ts + Tv; cannam@167: TI = T3 - T6; cannam@167: TW = TU - TV; cannam@167: T18 = T16 + T17; cannam@167: TM = Ts - Tv; cannam@167: } cannam@167: { cannam@167: E Ta, TX, TC, T11, Td, T10, Tz, TY; cannam@167: { cannam@167: E T8, T9, TA, TB; cannam@167: T8 = cr[WS(rs, 1)]; cannam@167: T9 = ci[WS(rs, 2)]; cannam@167: Ta = T8 + T9; cannam@167: TX = T8 - T9; cannam@167: TA = ci[WS(rs, 4)]; cannam@167: TB = cr[WS(rs, 7)]; cannam@167: TC = TA - TB; cannam@167: T11 = TA + TB; cannam@167: } cannam@167: { cannam@167: E Tb, Tc, Tx, Ty; cannam@167: Tb = ci[0]; cannam@167: Tc = cr[WS(rs, 3)]; cannam@167: Td = Tb + Tc; cannam@167: T10 = Tb - Tc; cannam@167: Tx = ci[WS(rs, 6)]; cannam@167: Ty = cr[WS(rs, 5)]; cannam@167: Tz = Tx - Ty; cannam@167: TY = Tx + Ty; cannam@167: } cannam@167: Te = Ta + Td; cannam@167: T19 = TX + TY; cannam@167: T1a = T10 + T11; cannam@167: TD = Tz + TC; cannam@167: TJ = TC - Tz; cannam@167: TZ = TX - TY; cannam@167: T12 = T10 - T11; cannam@167: TN = Ta - Td; cannam@167: } cannam@167: cr[0] = T7 + Te; cannam@167: ci[0] = Tw + TD; cannam@167: Tm = T7 - Te; cannam@167: TE = Tw - TD; cannam@167: cr[WS(rs, 4)] = FNMS(Tp, TE, Tl * Tm); cannam@167: ci[WS(rs, 4)] = FMA(Tp, Tm, Tl * TE); cannam@167: { cannam@167: E TQ, TS, TK, TO; cannam@167: TQ = TI + TJ; cannam@167: TS = TN + TM; cannam@167: cr[WS(rs, 2)] = FNMS(TR, TS, TP * TQ); cannam@167: ci[WS(rs, 2)] = FMA(TP, TS, TR * TQ); cannam@167: TK = TI - TJ; cannam@167: TO = TM - TN; cannam@167: cr[WS(rs, 6)] = FNMS(TL, TO, TH * TK); cannam@167: ci[WS(rs, 6)] = FMA(TH, TO, TL * TK); cannam@167: } cannam@167: { cannam@167: E T1h, T1l, T1k, T1m, T1g, T1j; cannam@167: T1g = KP707106781 * (T19 + T1a); cannam@167: T1h = T1f - T1g; cannam@167: T1l = T1f + T1g; cannam@167: T1j = KP707106781 * (TZ - T12); cannam@167: T1k = T1i + T1j; cannam@167: T1m = T1i - T1j; cannam@167: cr[WS(rs, 3)] = FNMS(Tj, T1k, Tg * T1h); cannam@167: ci[WS(rs, 3)] = FMA(Tg, T1k, Tj * T1h); cannam@167: cr[WS(rs, 7)] = FNMS(TG, T1m, TF * T1l); cannam@167: ci[WS(rs, 7)] = FMA(TF, T1m, TG * T1l); cannam@167: } cannam@167: { cannam@167: E T14, T1d, T1c, T1e, T13, T1b; cannam@167: T13 = KP707106781 * (TZ + T12); cannam@167: T14 = TW - T13; cannam@167: T1d = TW + T13; cannam@167: T1b = KP707106781 * (T19 - T1a); cannam@167: T1c = T18 - T1b; cannam@167: T1e = T18 + T1b; cannam@167: cr[WS(rs, 5)] = FNMS(T15, T1c, TT * T14); cannam@167: ci[WS(rs, 5)] = FMA(T15, T14, TT * T1c); cannam@167: cr[WS(rs, 1)] = FNMS(Ti, T1e, Tf * T1d); cannam@167: ci[WS(rs, 1)] = FMA(Ti, T1d, Tf * T1e); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: {TW_CEXP, 1, 1}, cannam@167: {TW_CEXP, 1, 3}, cannam@167: {TW_CEXP, 1, 7}, cannam@167: {TW_NEXT, 1, 0} cannam@167: }; cannam@167: cannam@167: static const hc2hc_desc desc = { 8, "hb2_8", twinstr, &GENUS, {56, 26, 18, 0} }; cannam@167: cannam@167: void X(codelet_hb2_8) (planner *p) { cannam@167: X(khc2hc_register) (p, hb2_8, &desc); cannam@167: } cannam@167: #endif