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 07:39:26 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 8 -name t2sv_8 -include ts.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 "ts.h" cannam@95: cannam@95: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 6); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 6), MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: V T1m, T1l, T1k, T1u, T1n, T1o; cannam@95: { cannam@95: V T2, T3, Tl, Tn, T5, T6; cannam@95: T2 = LDW(&(W[0])); cannam@95: T3 = LDW(&(W[TWVL * 2])); cannam@95: Tl = LDW(&(W[TWVL * 4])); cannam@95: Tn = LDW(&(W[TWVL * 5])); cannam@95: T5 = LDW(&(W[TWVL * 1])); cannam@95: T6 = LDW(&(W[TWVL * 3])); cannam@95: { cannam@95: V T1, T1s, TK, T1r, Td, Tk, TG, TC, TY, Tu, TW, TL, TM, TO, TQ; cannam@95: V Tx, Tz, TD, TH; cannam@95: { cannam@95: V T8, T4, Tm, Tr, Tc, Ta; cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: T1s = LD(&(ii[0]), ms, &(ii[0])); cannam@95: T8 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@95: T4 = VMUL(T2, T3); cannam@95: Tm = VMUL(T2, Tl); cannam@95: Tr = VMUL(T2, Tn); cannam@95: Tc = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@95: Ta = VMUL(T2, T6); cannam@95: { cannam@95: V Tp, Tt, Tg, T7, Tf, To, Ts, Ti, Tb, Tj; cannam@95: Tp = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@95: Tt = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@95: Tg = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: T7 = VFNMS(T5, T6, T4); cannam@95: Tf = VFMA(T5, T6, T4); cannam@95: To = VFMA(T5, Tn, Tm); cannam@95: Ts = VFNMS(T5, Tl, Tr); cannam@95: Ti = VFNMS(T5, T3, Ta); cannam@95: Tb = VFMA(T5, T3, Ta); cannam@95: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: TK = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); cannam@95: { cannam@95: V T1q, T9, Th, TF; cannam@95: T1q = VMUL(T7, Tc); cannam@95: T9 = VMUL(T7, T8); cannam@95: Th = VMUL(Tf, Tg); cannam@95: TF = VMUL(Tf, Tn); cannam@95: { cannam@95: V TB, TX, Tq, TV; cannam@95: TB = VMUL(Tf, Tl); cannam@95: TX = VMUL(To, Tt); cannam@95: Tq = VMUL(To, Tp); cannam@95: TV = VMUL(Tf, Tj); cannam@95: T1r = VFNMS(Tb, T8, T1q); cannam@95: Td = VFMA(Tb, Tc, T9); cannam@95: Tk = VFMA(Ti, Tj, Th); cannam@95: TG = VFNMS(Ti, Tl, TF); cannam@95: TC = VFMA(Ti, Tn, TB); cannam@95: TY = VFNMS(Ts, Tp, TX); cannam@95: Tu = VFMA(Ts, Tt, Tq); cannam@95: TW = VFNMS(Ti, Tg, TV); cannam@95: TL = VMUL(Tl, TK); cannam@95: } cannam@95: } cannam@95: TM = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); cannam@95: TO = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: TQ = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tx = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: Tz = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: TD = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@95: TH = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: { cannam@95: V Te, T1p, T1g, T10, TS, T18, T1d, T1t, T1x, T1y, Tv, TJ, T11, T16; cannam@95: { cannam@95: V TN, T1a, TR, T1c, TA, T13, TI, T15; cannam@95: { cannam@95: V TU, T19, TP, T1b, Ty, T12, TE, T14, TZ; cannam@95: TU = VSUB(T1, Td); cannam@95: Te = VADD(T1, Td); cannam@95: TN = VFMA(Tn, TM, TL); cannam@95: T19 = VMUL(Tl, TM); cannam@95: TP = VMUL(T3, TO); cannam@95: T1b = VMUL(T3, TQ); cannam@95: Ty = VMUL(T2, Tx); cannam@95: T12 = VMUL(T2, Tz); cannam@95: TE = VMUL(TC, TD); cannam@95: T14 = VMUL(TC, TH); cannam@95: T1p = VADD(TW, TY); cannam@95: TZ = VSUB(TW, TY); cannam@95: T1a = VFNMS(Tn, TK, T19); cannam@95: TR = VFMA(T6, TQ, TP); cannam@95: T1c = VFNMS(T6, TO, T1b); cannam@95: TA = VFMA(T5, Tz, Ty); cannam@95: T13 = VFNMS(T5, Tx, T12); cannam@95: TI = VFMA(TG, TH, TE); cannam@95: T15 = VFNMS(TG, TD, T14); cannam@95: T1g = VSUB(TU, TZ); cannam@95: T10 = VADD(TU, TZ); cannam@95: } cannam@95: TS = VADD(TN, TR); cannam@95: T18 = VSUB(TN, TR); cannam@95: T1d = VSUB(T1a, T1c); cannam@95: T1m = VADD(T1a, T1c); cannam@95: T1t = VADD(T1r, T1s); cannam@95: T1x = VSUB(T1s, T1r); cannam@95: T1y = VSUB(Tk, Tu); cannam@95: Tv = VADD(Tk, Tu); cannam@95: TJ = VADD(TA, TI); cannam@95: T11 = VSUB(TA, TI); cannam@95: T16 = VSUB(T13, T15); cannam@95: T1l = VADD(T13, T15); cannam@95: } cannam@95: { cannam@95: V Tw, T1w, T1v, TT; cannam@95: { cannam@95: V T1i, T1e, T1B, T1z, T1h, T17; cannam@95: T1i = VADD(T18, T1d); cannam@95: T1e = VSUB(T18, T1d); cannam@95: T1B = VADD(T1y, T1x); cannam@95: T1z = VSUB(T1x, T1y); cannam@95: T1h = VSUB(T16, T11); cannam@95: T17 = VADD(T11, T16); cannam@95: T1k = VSUB(Te, Tv); cannam@95: Tw = VADD(Te, Tv); cannam@95: { cannam@95: V T1A, T1j, T1C, T1f; cannam@95: T1A = VADD(T1h, T1i); cannam@95: T1j = VSUB(T1h, T1i); cannam@95: T1C = VSUB(T1e, T17); cannam@95: T1f = VADD(T17, T1e); cannam@95: T1w = VSUB(T1t, T1p); cannam@95: T1u = VADD(T1p, T1t); cannam@95: T1v = VSUB(TS, TJ); cannam@95: TT = VADD(TJ, TS); cannam@95: ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: ST(&(ri[WS(rs, 4)]), VSUB(Tw, TT), ms, &(ri[0])); cannam@95: ST(&(ri[0]), VADD(Tw, TT), ms, &(ri[0])); cannam@95: ST(&(ii[WS(rs, 6)]), VSUB(T1w, T1v), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 2)]), VADD(T1v, T1w), ms, &(ii[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: T1n = VSUB(T1l, T1m); cannam@95: T1o = VADD(T1l, T1m); cannam@95: ST(&(ii[0]), VADD(T1o, T1u), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 4)]), VSUB(T1u, T1o), ms, &(ii[0])); cannam@95: ST(&(ri[WS(rs, 2)]), VADD(T1k, T1n), ms, &(ri[0])); cannam@95: ST(&(ri[WS(rs, 6)]), VSUB(T1k, T1n), ms, &(ri[0])); cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 3), cannam@95: VTW(0, 7), cannam@95: {TW_NEXT, (2 * VL), 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {44, 20, 30, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t2sv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2sv_8, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 8 -name t2sv_8 -include ts.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: * 42 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "ts.h" cannam@95: cannam@95: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 6); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 6), MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: V T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx; cannam@95: { cannam@95: V T4, Tb, T7, Ta; cannam@95: T2 = LDW(&(W[0])); cannam@95: T5 = LDW(&(W[TWVL * 1])); cannam@95: T3 = LDW(&(W[TWVL * 2])); cannam@95: T6 = LDW(&(W[TWVL * 3])); cannam@95: T4 = VMUL(T2, T3); cannam@95: Tb = VMUL(T5, T3); cannam@95: T7 = VMUL(T5, T6); cannam@95: Ta = VMUL(T2, T6); cannam@95: T8 = VSUB(T4, T7); cannam@95: Tc = VADD(Ta, Tb); cannam@95: Tg = VADD(T4, T7); cannam@95: Ti = VSUB(Ta, Tb); cannam@95: Tl = LDW(&(W[TWVL * 4])); cannam@95: Tm = LDW(&(W[TWVL * 5])); cannam@95: Tn = VFMA(T2, Tl, VMUL(T5, Tm)); cannam@95: Tz = VFNMS(Ti, Tl, VMUL(Tg, Tm)); cannam@95: Tp = VFNMS(T5, Tl, VMUL(T2, Tm)); cannam@95: Tx = VFMA(Tg, Tl, VMUL(Ti, Tm)); cannam@95: } cannam@95: { cannam@95: V Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ; cannam@95: V TT; cannam@95: { cannam@95: V T1, T1c, Te, T1b, T9, Td; cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: T1c = LD(&(ii[0]), ms, &(ii[0])); cannam@95: T9 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@95: Td = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@95: Te = VFMA(T8, T9, VMUL(Tc, Td)); cannam@95: T1b = VFNMS(Tc, T9, VMUL(T8, Td)); cannam@95: Tf = VADD(T1, Te); cannam@95: T1i = VSUB(T1c, T1b); cannam@95: TL = VSUB(T1, Te); cannam@95: T1d = VADD(T1b, T1c); cannam@95: } cannam@95: { cannam@95: V TF, TW, TI, TX; cannam@95: { cannam@95: V TD, TE, TG, TH; cannam@95: TD = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); cannam@95: TE = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); cannam@95: TF = VFMA(Tl, TD, VMUL(Tm, TE)); cannam@95: TW = VFNMS(Tm, TD, VMUL(Tl, TE)); cannam@95: TG = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: TH = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: TI = VFMA(T3, TG, VMUL(T6, TH)); cannam@95: TX = VFNMS(T6, TG, VMUL(T3, TH)); cannam@95: } cannam@95: TJ = VADD(TF, TI); cannam@95: T17 = VADD(TW, TX); cannam@95: TV = VSUB(TF, TI); cannam@95: TY = VSUB(TW, TX); cannam@95: } cannam@95: { cannam@95: V Tk, TM, Tr, TN; cannam@95: { cannam@95: V Th, Tj, To, Tq; cannam@95: Th = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: Tk = VFMA(Tg, Th, VMUL(Ti, Tj)); cannam@95: TM = VFNMS(Ti, Th, VMUL(Tg, Tj)); cannam@95: To = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@95: Tq = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@95: Tr = VFMA(Tn, To, VMUL(Tp, Tq)); cannam@95: TN = VFNMS(Tp, To, VMUL(Tn, Tq)); cannam@95: } cannam@95: Ts = VADD(Tk, Tr); cannam@95: T1j = VSUB(Tk, Tr); cannam@95: TO = VSUB(TM, TN); cannam@95: T1a = VADD(TM, TN); cannam@95: } cannam@95: { cannam@95: V Tw, TR, TB, TS; cannam@95: { cannam@95: V Tu, Tv, Ty, TA; cannam@95: Tu = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: Tv = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tw = VFMA(T2, Tu, VMUL(T5, Tv)); cannam@95: TR = VFNMS(T5, Tu, VMUL(T2, Tv)); cannam@95: Ty = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@95: TA = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@95: TB = VFMA(Tx, Ty, VMUL(Tz, TA)); cannam@95: TS = VFNMS(Tz, Ty, VMUL(Tx, TA)); cannam@95: } cannam@95: TC = VADD(Tw, TB); cannam@95: T16 = VADD(TR, TS); cannam@95: TQ = VSUB(Tw, TB); cannam@95: TT = VSUB(TR, TS); cannam@95: } cannam@95: { cannam@95: V Tt, TK, T1f, T1g; cannam@95: Tt = VADD(Tf, Ts); cannam@95: TK = VADD(TC, TJ); cannam@95: ST(&(ri[WS(rs, 4)]), VSUB(Tt, TK), ms, &(ri[0])); cannam@95: ST(&(ri[0]), VADD(Tt, TK), ms, &(ri[0])); cannam@95: { cannam@95: V T19, T1e, T15, T18; cannam@95: T19 = VADD(T16, T17); cannam@95: T1e = VADD(T1a, T1d); cannam@95: ST(&(ii[0]), VADD(T19, T1e), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 4)]), VSUB(T1e, T19), ms, &(ii[0])); cannam@95: T15 = VSUB(Tf, Ts); cannam@95: T18 = VSUB(T16, T17); cannam@95: ST(&(ri[WS(rs, 6)]), VSUB(T15, T18), ms, &(ri[0])); cannam@95: ST(&(ri[WS(rs, 2)]), VADD(T15, T18), ms, &(ri[0])); cannam@95: } cannam@95: T1f = VSUB(TJ, TC); cannam@95: T1g = VSUB(T1d, T1a); cannam@95: ST(&(ii[WS(rs, 2)]), VADD(T1f, T1g), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 6)]), VSUB(T1g, T1f), ms, &(ii[0])); cannam@95: { cannam@95: V T11, T1k, T14, T1h, T12, T13; cannam@95: T11 = VSUB(TL, TO); cannam@95: T1k = VSUB(T1i, T1j); cannam@95: T12 = VSUB(TT, TQ); cannam@95: T13 = VADD(TV, TY); cannam@95: T14 = VMUL(LDK(KP707106781), VSUB(T12, T13)); cannam@95: T1h = VMUL(LDK(KP707106781), VADD(T12, T13)); cannam@95: ST(&(ri[WS(rs, 7)]), VSUB(T11, T14), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 5)]), VSUB(T1k, T1h), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 3)]), VADD(T11, T14), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 1)]), VADD(T1h, T1k), ms, &(ii[WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V TP, T1m, T10, T1l, TU, TZ; cannam@95: TP = VADD(TL, TO); cannam@95: T1m = VADD(T1j, T1i); cannam@95: TU = VADD(TQ, TT); cannam@95: TZ = VSUB(TV, TY); cannam@95: T10 = VMUL(LDK(KP707106781), VADD(TU, TZ)); cannam@95: T1l = VMUL(LDK(KP707106781), VSUB(TZ, TU)); cannam@95: ST(&(ri[WS(rs, 5)]), VSUB(TP, T10), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 7)]), VSUB(T1m, T1l), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 1)]), VADD(TP, T10), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 3)]), VADD(T1l, T1m), ms, &(ii[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 3), cannam@95: VTW(0, 7), cannam@95: {TW_NEXT, (2 * VL), 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {56, 26, 18, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t2sv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2sv_8, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */