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 08:06:11 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle.native -fma -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 8 -name t2sv_8 -include dft/simd/ts.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: * 48 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "dft/simd/ts.h" cannam@167: cannam@167: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@167: { cannam@167: INT m; cannam@167: 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@167: V T2, T3, Tl, Tn, T5, T6, Tf, T7, Ts, Tb, To, Ti, TC, TG; cannam@167: { cannam@167: V T4, Tm, Tr, Ta, TB, TF; cannam@167: T2 = LDW(&(W[0])); cannam@167: T3 = LDW(&(W[TWVL * 2])); cannam@167: T4 = VMUL(T2, T3); cannam@167: Tl = LDW(&(W[TWVL * 4])); cannam@167: Tm = VMUL(T2, Tl); cannam@167: Tn = LDW(&(W[TWVL * 5])); cannam@167: Tr = VMUL(T2, Tn); cannam@167: T5 = LDW(&(W[TWVL * 1])); cannam@167: T6 = LDW(&(W[TWVL * 3])); cannam@167: Ta = VMUL(T2, T6); cannam@167: Tf = VFMA(T5, T6, T4); cannam@167: T7 = VFNMS(T5, T6, T4); cannam@167: Ts = VFNMS(T5, Tl, Tr); cannam@167: Tb = VFMA(T5, T3, Ta); cannam@167: To = VFMA(T5, Tn, Tm); cannam@167: TB = VMUL(Tf, Tl); cannam@167: TF = VMUL(Tf, Tn); cannam@167: Ti = VFNMS(T5, T3, Ta); cannam@167: TC = VFMA(Ti, Tn, TB); cannam@167: TG = VFNMS(Ti, Tl, TF); cannam@167: } cannam@167: { cannam@167: V T1, T1s, Td, T1r, Tu, TY, Tk, TW, TN, TR, T18, T1a, T1c, T1d, TA; cannam@167: V TI, T11, T13, T15, T16; cannam@167: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@167: T1s = LD(&(ii[0]), ms, &(ii[0])); cannam@167: { cannam@167: V T8, T9, Tc, T1q; cannam@167: T8 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@167: T9 = VMUL(T7, T8); cannam@167: Tc = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@167: T1q = VMUL(T7, Tc); cannam@167: Td = VFMA(Tb, Tc, T9); cannam@167: T1r = VFNMS(Tb, T8, T1q); cannam@167: } cannam@167: { cannam@167: V Tp, Tq, Tt, TX; cannam@167: Tp = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@167: Tq = VMUL(To, Tp); cannam@167: Tt = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@167: TX = VMUL(To, Tt); cannam@167: Tu = VFMA(Ts, Tt, Tq); cannam@167: TY = VFNMS(Ts, Tp, TX); cannam@167: } cannam@167: { cannam@167: V Tg, Th, Tj, TV; cannam@167: Tg = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@167: Th = VMUL(Tf, Tg); cannam@167: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@167: TV = VMUL(Tf, Tj); cannam@167: Tk = VFMA(Ti, Tj, Th); cannam@167: TW = VFNMS(Ti, Tg, TV); cannam@167: } cannam@167: { cannam@167: V TK, TL, TM, T19, TO, TP, TQ, T1b; cannam@167: TK = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); cannam@167: TL = VMUL(Tl, TK); cannam@167: TM = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); cannam@167: T19 = VMUL(Tl, TM); cannam@167: TO = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@167: TP = VMUL(T3, TO); cannam@167: TQ = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@167: T1b = VMUL(T3, TQ); cannam@167: TN = VFMA(Tn, TM, TL); cannam@167: TR = VFMA(T6, TQ, TP); cannam@167: T18 = VSUB(TN, TR); cannam@167: T1a = VFNMS(Tn, TK, T19); cannam@167: T1c = VFNMS(T6, TO, T1b); cannam@167: T1d = VSUB(T1a, T1c); cannam@167: } cannam@167: { cannam@167: V Tx, Ty, Tz, T12, TD, TE, TH, T14; cannam@167: Tx = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@167: Ty = VMUL(T2, Tx); cannam@167: Tz = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@167: T12 = VMUL(T2, Tz); cannam@167: TD = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@167: TE = VMUL(TC, TD); cannam@167: TH = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@167: T14 = VMUL(TC, TH); cannam@167: TA = VFMA(T5, Tz, Ty); cannam@167: TI = VFMA(TG, TH, TE); cannam@167: T11 = VSUB(TA, TI); cannam@167: T13 = VFNMS(T5, Tx, T12); cannam@167: T15 = VFNMS(TG, TD, T14); cannam@167: T16 = VSUB(T13, T15); cannam@167: } cannam@167: { cannam@167: V T10, T1g, T1z, T1B, T1f, T1C, T1j, T1A; cannam@167: { cannam@167: V TU, TZ, T1x, T1y; cannam@167: TU = VSUB(T1, Td); cannam@167: TZ = VSUB(TW, TY); cannam@167: T10 = VADD(TU, TZ); cannam@167: T1g = VSUB(TU, TZ); cannam@167: T1x = VSUB(T1s, T1r); cannam@167: T1y = VSUB(Tk, Tu); cannam@167: T1z = VSUB(T1x, T1y); cannam@167: T1B = VADD(T1y, T1x); cannam@167: } cannam@167: { cannam@167: V T17, T1e, T1h, T1i; cannam@167: T17 = VADD(T11, T16); cannam@167: T1e = VSUB(T18, T1d); cannam@167: T1f = VADD(T17, T1e); cannam@167: T1C = VSUB(T1e, T17); cannam@167: T1h = VSUB(T16, T11); cannam@167: T1i = VADD(T18, T1d); cannam@167: T1j = VSUB(T1h, T1i); cannam@167: T1A = VADD(T1h, T1i); cannam@167: } cannam@167: ST(&(ri[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V Tw, T1k, T1u, T1w, TT, T1v, T1n, T1o; cannam@167: { cannam@167: V Te, Tv, T1p, T1t; cannam@167: Te = VADD(T1, Td); cannam@167: Tv = VADD(Tk, Tu); cannam@167: Tw = VADD(Te, Tv); cannam@167: T1k = VSUB(Te, Tv); cannam@167: T1p = VADD(TW, TY); cannam@167: T1t = VADD(T1r, T1s); cannam@167: T1u = VADD(T1p, T1t); cannam@167: T1w = VSUB(T1t, T1p); cannam@167: } cannam@167: { cannam@167: V TJ, TS, T1l, T1m; cannam@167: TJ = VADD(TA, TI); cannam@167: TS = VADD(TN, TR); cannam@167: TT = VADD(TJ, TS); cannam@167: T1v = VSUB(TS, TJ); cannam@167: T1l = VADD(T13, T15); cannam@167: T1m = VADD(T1a, T1c); cannam@167: T1n = VSUB(T1l, T1m); cannam@167: T1o = VADD(T1l, T1m); cannam@167: } cannam@167: ST(&(ri[WS(rs, 4)]), VSUB(Tw, TT), ms, &(ri[0])); cannam@167: ST(&(ii[WS(rs, 4)]), VSUB(T1u, T1o), ms, &(ii[0])); cannam@167: ST(&(ri[0]), VADD(Tw, TT), ms, &(ri[0])); cannam@167: ST(&(ii[0]), VADD(T1o, T1u), ms, &(ii[0])); cannam@167: ST(&(ri[WS(rs, 6)]), VSUB(T1k, T1n), ms, &(ri[0])); cannam@167: ST(&(ii[WS(rs, 6)]), VSUB(T1w, T1v), ms, &(ii[0])); cannam@167: ST(&(ri[WS(rs, 2)]), VADD(T1k, T1n), ms, &(ri[0])); cannam@167: ST(&(ii[WS(rs, 2)]), VADD(T1v, T1w), ms, &(ii[0])); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 3), cannam@167: VTW(0, 7), cannam@167: {TW_NEXT, (2 * VL), 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {44, 20, 30, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t2sv_8) (planner *p) { cannam@167: X(kdft_dit_register) (p, t2sv_8, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 8 -name t2sv_8 -include dft/simd/ts.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: * 42 stack variables, 1 constants, and 32 memory accesses cannam@167: */ cannam@167: #include "dft/simd/ts.h" cannam@167: cannam@167: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@167: { cannam@167: INT m; cannam@167: 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@167: V T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx; cannam@167: { cannam@167: V T4, Tb, T7, Ta; cannam@167: T2 = LDW(&(W[0])); cannam@167: T5 = LDW(&(W[TWVL * 1])); cannam@167: T3 = LDW(&(W[TWVL * 2])); cannam@167: T6 = LDW(&(W[TWVL * 3])); cannam@167: T4 = VMUL(T2, T3); cannam@167: Tb = VMUL(T5, T3); cannam@167: T7 = VMUL(T5, T6); cannam@167: Ta = VMUL(T2, T6); cannam@167: T8 = VSUB(T4, T7); cannam@167: Tc = VADD(Ta, Tb); cannam@167: Tg = VADD(T4, T7); cannam@167: Ti = VSUB(Ta, Tb); cannam@167: Tl = LDW(&(W[TWVL * 4])); cannam@167: Tm = LDW(&(W[TWVL * 5])); cannam@167: Tn = VFMA(T2, Tl, VMUL(T5, Tm)); cannam@167: Tz = VFNMS(Ti, Tl, VMUL(Tg, Tm)); cannam@167: Tp = VFNMS(T5, Tl, VMUL(T2, Tm)); cannam@167: Tx = VFMA(Tg, Tl, VMUL(Ti, Tm)); cannam@167: } cannam@167: { cannam@167: V Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ; cannam@167: V TT; cannam@167: { cannam@167: V T1, T1c, Te, T1b, T9, Td; cannam@167: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@167: T1c = LD(&(ii[0]), ms, &(ii[0])); cannam@167: T9 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@167: Td = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@167: Te = VFMA(T8, T9, VMUL(Tc, Td)); cannam@167: T1b = VFNMS(Tc, T9, VMUL(T8, Td)); cannam@167: Tf = VADD(T1, Te); cannam@167: T1i = VSUB(T1c, T1b); cannam@167: TL = VSUB(T1, Te); cannam@167: T1d = VADD(T1b, T1c); cannam@167: } cannam@167: { cannam@167: V TF, TW, TI, TX; cannam@167: { cannam@167: V TD, TE, TG, TH; cannam@167: TD = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); cannam@167: TE = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); cannam@167: TF = VFMA(Tl, TD, VMUL(Tm, TE)); cannam@167: TW = VFNMS(Tm, TD, VMUL(Tl, TE)); cannam@167: TG = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@167: TH = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@167: TI = VFMA(T3, TG, VMUL(T6, TH)); cannam@167: TX = VFNMS(T6, TG, VMUL(T3, TH)); cannam@167: } cannam@167: TJ = VADD(TF, TI); cannam@167: T17 = VADD(TW, TX); cannam@167: TV = VSUB(TF, TI); cannam@167: TY = VSUB(TW, TX); cannam@167: } cannam@167: { cannam@167: V Tk, TM, Tr, TN; cannam@167: { cannam@167: V Th, Tj, To, Tq; cannam@167: Th = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@167: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@167: Tk = VFMA(Tg, Th, VMUL(Ti, Tj)); cannam@167: TM = VFNMS(Ti, Th, VMUL(Tg, Tj)); cannam@167: To = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@167: Tq = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@167: Tr = VFMA(Tn, To, VMUL(Tp, Tq)); cannam@167: TN = VFNMS(Tp, To, VMUL(Tn, Tq)); cannam@167: } cannam@167: Ts = VADD(Tk, Tr); cannam@167: T1j = VSUB(Tk, Tr); cannam@167: TO = VSUB(TM, TN); cannam@167: T1a = VADD(TM, TN); cannam@167: } cannam@167: { cannam@167: V Tw, TR, TB, TS; cannam@167: { cannam@167: V Tu, Tv, Ty, TA; cannam@167: Tu = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@167: Tv = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@167: Tw = VFMA(T2, Tu, VMUL(T5, Tv)); cannam@167: TR = VFNMS(T5, Tu, VMUL(T2, Tv)); cannam@167: Ty = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@167: TA = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@167: TB = VFMA(Tx, Ty, VMUL(Tz, TA)); cannam@167: TS = VFNMS(Tz, Ty, VMUL(Tx, TA)); cannam@167: } cannam@167: TC = VADD(Tw, TB); cannam@167: T16 = VADD(TR, TS); cannam@167: TQ = VSUB(Tw, TB); cannam@167: TT = VSUB(TR, TS); cannam@167: } cannam@167: { cannam@167: V Tt, TK, T1f, T1g; cannam@167: Tt = VADD(Tf, Ts); cannam@167: TK = VADD(TC, TJ); cannam@167: ST(&(ri[WS(rs, 4)]), VSUB(Tt, TK), ms, &(ri[0])); cannam@167: ST(&(ri[0]), VADD(Tt, TK), ms, &(ri[0])); cannam@167: { cannam@167: V T19, T1e, T15, T18; cannam@167: T19 = VADD(T16, T17); cannam@167: T1e = VADD(T1a, T1d); cannam@167: ST(&(ii[0]), VADD(T19, T1e), ms, &(ii[0])); cannam@167: ST(&(ii[WS(rs, 4)]), VSUB(T1e, T19), ms, &(ii[0])); cannam@167: T15 = VSUB(Tf, Ts); cannam@167: T18 = VSUB(T16, T17); cannam@167: ST(&(ri[WS(rs, 6)]), VSUB(T15, T18), ms, &(ri[0])); cannam@167: ST(&(ri[WS(rs, 2)]), VADD(T15, T18), ms, &(ri[0])); cannam@167: } cannam@167: T1f = VSUB(TJ, TC); cannam@167: T1g = VSUB(T1d, T1a); cannam@167: ST(&(ii[WS(rs, 2)]), VADD(T1f, T1g), ms, &(ii[0])); cannam@167: ST(&(ii[WS(rs, 6)]), VSUB(T1g, T1f), ms, &(ii[0])); cannam@167: { cannam@167: V T11, T1k, T14, T1h, T12, T13; cannam@167: T11 = VSUB(TL, TO); cannam@167: T1k = VSUB(T1i, T1j); cannam@167: T12 = VSUB(TT, TQ); cannam@167: T13 = VADD(TV, TY); cannam@167: T14 = VMUL(LDK(KP707106781), VSUB(T12, T13)); cannam@167: T1h = VMUL(LDK(KP707106781), VADD(T12, T13)); cannam@167: ST(&(ri[WS(rs, 7)]), VSUB(T11, T14), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 5)]), VSUB(T1k, T1h), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 3)]), VADD(T11, T14), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 1)]), VADD(T1h, T1k), ms, &(ii[WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V TP, T1m, T10, T1l, TU, TZ; cannam@167: TP = VADD(TL, TO); cannam@167: T1m = VADD(T1j, T1i); cannam@167: TU = VADD(TQ, TT); cannam@167: TZ = VSUB(TV, TY); cannam@167: T10 = VMUL(LDK(KP707106781), VADD(TU, TZ)); cannam@167: T1l = VMUL(LDK(KP707106781), VSUB(TZ, TU)); cannam@167: ST(&(ri[WS(rs, 5)]), VSUB(TP, T10), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 7)]), VSUB(T1m, T1l), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 1)]), VADD(TP, T10), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 3)]), VADD(T1l, T1m), ms, &(ii[WS(rs, 1)])); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 3), cannam@167: VTW(0, 7), cannam@167: {TW_NEXT, (2 * VL), 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {56, 26, 18, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t2sv_8) (planner *p) { cannam@167: X(kdft_dit_register) (p, t2sv_8, &desc); cannam@167: } cannam@167: #endif