Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:06:11 EDT 2018 */ Chris@82: Chris@82: #include "dft/codelet-dft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* 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 */ Chris@82: Chris@82: /* Chris@82: * This function contains 74 FP additions, 50 FP multiplications, Chris@82: * (or, 44 additions, 20 multiplications, 30 fused multiply/add), Chris@82: * 48 stack variables, 1 constants, and 32 memory accesses Chris@82: */ Chris@82: #include "dft/simd/ts.h" Chris@82: Chris@82: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: V T2, T3, Tl, Tn, T5, T6, Tf, T7, Ts, Tb, To, Ti, TC, TG; Chris@82: { Chris@82: V T4, Tm, Tr, Ta, TB, TF; Chris@82: T2 = LDW(&(W[0])); Chris@82: T3 = LDW(&(W[TWVL * 2])); Chris@82: T4 = VMUL(T2, T3); Chris@82: Tl = LDW(&(W[TWVL * 4])); Chris@82: Tm = VMUL(T2, Tl); Chris@82: Tn = LDW(&(W[TWVL * 5])); Chris@82: Tr = VMUL(T2, Tn); Chris@82: T5 = LDW(&(W[TWVL * 1])); Chris@82: T6 = LDW(&(W[TWVL * 3])); Chris@82: Ta = VMUL(T2, T6); Chris@82: Tf = VFMA(T5, T6, T4); Chris@82: T7 = VFNMS(T5, T6, T4); Chris@82: Ts = VFNMS(T5, Tl, Tr); Chris@82: Tb = VFMA(T5, T3, Ta); Chris@82: To = VFMA(T5, Tn, Tm); Chris@82: TB = VMUL(Tf, Tl); Chris@82: TF = VMUL(Tf, Tn); Chris@82: Ti = VFNMS(T5, T3, Ta); Chris@82: TC = VFMA(Ti, Tn, TB); Chris@82: TG = VFNMS(Ti, Tl, TF); Chris@82: } Chris@82: { Chris@82: V T1, T1s, Td, T1r, Tu, TY, Tk, TW, TN, TR, T18, T1a, T1c, T1d, TA; Chris@82: V TI, T11, T13, T15, T16; Chris@82: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@82: T1s = LD(&(ii[0]), ms, &(ii[0])); Chris@82: { Chris@82: V T8, T9, Tc, T1q; Chris@82: T8 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); Chris@82: T9 = VMUL(T7, T8); Chris@82: Tc = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); Chris@82: T1q = VMUL(T7, Tc); Chris@82: Td = VFMA(Tb, Tc, T9); Chris@82: T1r = VFNMS(Tb, T8, T1q); Chris@82: } Chris@82: { Chris@82: V Tp, Tq, Tt, TX; Chris@82: Tp = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); Chris@82: Tq = VMUL(To, Tp); Chris@82: Tt = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); Chris@82: TX = VMUL(To, Tt); Chris@82: Tu = VFMA(Ts, Tt, Tq); Chris@82: TY = VFNMS(Ts, Tp, TX); Chris@82: } Chris@82: { Chris@82: V Tg, Th, Tj, TV; Chris@82: Tg = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@82: Th = VMUL(Tf, Tg); Chris@82: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@82: TV = VMUL(Tf, Tj); Chris@82: Tk = VFMA(Ti, Tj, Th); Chris@82: TW = VFNMS(Ti, Tg, TV); Chris@82: } Chris@82: { Chris@82: V TK, TL, TM, T19, TO, TP, TQ, T1b; Chris@82: TK = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); Chris@82: TL = VMUL(Tl, TK); Chris@82: TM = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); Chris@82: T19 = VMUL(Tl, TM); Chris@82: TO = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Chris@82: TP = VMUL(T3, TO); Chris@82: TQ = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Chris@82: T1b = VMUL(T3, TQ); Chris@82: TN = VFMA(Tn, TM, TL); Chris@82: TR = VFMA(T6, TQ, TP); Chris@82: T18 = VSUB(TN, TR); Chris@82: T1a = VFNMS(Tn, TK, T19); Chris@82: T1c = VFNMS(T6, TO, T1b); Chris@82: T1d = VSUB(T1a, T1c); Chris@82: } Chris@82: { Chris@82: V Tx, Ty, Tz, T12, TD, TE, TH, T14; Chris@82: Tx = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Chris@82: Ty = VMUL(T2, Tx); Chris@82: Tz = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Chris@82: T12 = VMUL(T2, Tz); Chris@82: TD = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); Chris@82: TE = VMUL(TC, TD); Chris@82: TH = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); Chris@82: T14 = VMUL(TC, TH); Chris@82: TA = VFMA(T5, Tz, Ty); Chris@82: TI = VFMA(TG, TH, TE); Chris@82: T11 = VSUB(TA, TI); Chris@82: T13 = VFNMS(T5, Tx, T12); Chris@82: T15 = VFNMS(TG, TD, T14); Chris@82: T16 = VSUB(T13, T15); Chris@82: } Chris@82: { Chris@82: V T10, T1g, T1z, T1B, T1f, T1C, T1j, T1A; Chris@82: { Chris@82: V TU, TZ, T1x, T1y; Chris@82: TU = VSUB(T1, Td); Chris@82: TZ = VSUB(TW, TY); Chris@82: T10 = VADD(TU, TZ); Chris@82: T1g = VSUB(TU, TZ); Chris@82: T1x = VSUB(T1s, T1r); Chris@82: T1y = VSUB(Tk, Tu); Chris@82: T1z = VSUB(T1x, T1y); Chris@82: T1B = VADD(T1y, T1x); Chris@82: } Chris@82: { Chris@82: V T17, T1e, T1h, T1i; Chris@82: T17 = VADD(T11, T16); Chris@82: T1e = VSUB(T18, T1d); Chris@82: T1f = VADD(T17, T1e); Chris@82: T1C = VSUB(T1e, T17); Chris@82: T1h = VSUB(T16, T11); Chris@82: T1i = VADD(T18, T1d); Chris@82: T1j = VSUB(T1h, T1i); Chris@82: T1A = VADD(T1h, T1i); Chris@82: } Chris@82: ST(&(ri[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); Chris@82: ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP707106781), T1f, T10), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP707106781), T1A, T1z), ms, &(ii[WS(rs, 1)])); Chris@82: ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); Chris@82: ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP707106781), T1j, T1g), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP707106781), T1C, T1B), ms, &(ii[WS(rs, 1)])); Chris@82: } Chris@82: { Chris@82: V Tw, T1k, T1u, T1w, TT, T1v, T1n, T1o; Chris@82: { Chris@82: V Te, Tv, T1p, T1t; Chris@82: Te = VADD(T1, Td); Chris@82: Tv = VADD(Tk, Tu); Chris@82: Tw = VADD(Te, Tv); Chris@82: T1k = VSUB(Te, Tv); Chris@82: T1p = VADD(TW, TY); Chris@82: T1t = VADD(T1r, T1s); Chris@82: T1u = VADD(T1p, T1t); Chris@82: T1w = VSUB(T1t, T1p); Chris@82: } Chris@82: { Chris@82: V TJ, TS, T1l, T1m; Chris@82: TJ = VADD(TA, TI); Chris@82: TS = VADD(TN, TR); Chris@82: TT = VADD(TJ, TS); Chris@82: T1v = VSUB(TS, TJ); Chris@82: T1l = VADD(T13, T15); Chris@82: T1m = VADD(T1a, T1c); Chris@82: T1n = VSUB(T1l, T1m); Chris@82: T1o = VADD(T1l, T1m); Chris@82: } Chris@82: ST(&(ri[WS(rs, 4)]), VSUB(Tw, TT), ms, &(ri[0])); Chris@82: ST(&(ii[WS(rs, 4)]), VSUB(T1u, T1o), ms, &(ii[0])); Chris@82: ST(&(ri[0]), VADD(Tw, TT), ms, &(ri[0])); Chris@82: ST(&(ii[0]), VADD(T1o, T1u), ms, &(ii[0])); Chris@82: ST(&(ri[WS(rs, 6)]), VSUB(T1k, T1n), ms, &(ri[0])); Chris@82: ST(&(ii[WS(rs, 6)]), VSUB(T1w, T1v), ms, &(ii[0])); Chris@82: ST(&(ri[WS(rs, 2)]), VADD(T1k, T1n), ms, &(ri[0])); Chris@82: ST(&(ii[WS(rs, 2)]), VADD(T1v, T1w), ms, &(ii[0])); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: VTW(0, 1), Chris@82: VTW(0, 3), Chris@82: VTW(0, 7), Chris@82: {TW_NEXT, (2 * VL), 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {44, 20, 30, 0}, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_t2sv_8) (planner *p) { Chris@82: X(kdft_dit_register) (p, t2sv_8, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* 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 */ Chris@82: Chris@82: /* Chris@82: * This function contains 74 FP additions, 44 FP multiplications, Chris@82: * (or, 56 additions, 26 multiplications, 18 fused multiply/add), Chris@82: * 42 stack variables, 1 constants, and 32 memory accesses Chris@82: */ Chris@82: #include "dft/simd/ts.h" Chris@82: Chris@82: static void t2sv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@82: { Chris@82: INT m; Chris@82: 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)) { Chris@82: V T2, T5, T3, T6, T8, Tc, Tg, Ti, Tl, Tm, Tn, Tz, Tp, Tx; Chris@82: { Chris@82: V T4, Tb, T7, Ta; Chris@82: T2 = LDW(&(W[0])); Chris@82: T5 = LDW(&(W[TWVL * 1])); Chris@82: T3 = LDW(&(W[TWVL * 2])); Chris@82: T6 = LDW(&(W[TWVL * 3])); Chris@82: T4 = VMUL(T2, T3); Chris@82: Tb = VMUL(T5, T3); Chris@82: T7 = VMUL(T5, T6); Chris@82: Ta = VMUL(T2, T6); Chris@82: T8 = VSUB(T4, T7); Chris@82: Tc = VADD(Ta, Tb); Chris@82: Tg = VADD(T4, T7); Chris@82: Ti = VSUB(Ta, Tb); Chris@82: Tl = LDW(&(W[TWVL * 4])); Chris@82: Tm = LDW(&(W[TWVL * 5])); Chris@82: Tn = VFMA(T2, Tl, VMUL(T5, Tm)); Chris@82: Tz = VFNMS(Ti, Tl, VMUL(Tg, Tm)); Chris@82: Tp = VFNMS(T5, Tl, VMUL(T2, Tm)); Chris@82: Tx = VFMA(Tg, Tl, VMUL(Ti, Tm)); Chris@82: } Chris@82: { Chris@82: V Tf, T1i, TL, T1d, TJ, T17, TV, TY, Ts, T1j, TO, T1a, TC, T16, TQ; Chris@82: V TT; Chris@82: { Chris@82: V T1, T1c, Te, T1b, T9, Td; Chris@82: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@82: T1c = LD(&(ii[0]), ms, &(ii[0])); Chris@82: T9 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); Chris@82: Td = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); Chris@82: Te = VFMA(T8, T9, VMUL(Tc, Td)); Chris@82: T1b = VFNMS(Tc, T9, VMUL(T8, Td)); Chris@82: Tf = VADD(T1, Te); Chris@82: T1i = VSUB(T1c, T1b); Chris@82: TL = VSUB(T1, Te); Chris@82: T1d = VADD(T1b, T1c); Chris@82: } Chris@82: { Chris@82: V TF, TW, TI, TX; Chris@82: { Chris@82: V TD, TE, TG, TH; Chris@82: TD = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); Chris@82: TE = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); Chris@82: TF = VFMA(Tl, TD, VMUL(Tm, TE)); Chris@82: TW = VFNMS(Tm, TD, VMUL(Tl, TE)); Chris@82: TG = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Chris@82: TH = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Chris@82: TI = VFMA(T3, TG, VMUL(T6, TH)); Chris@82: TX = VFNMS(T6, TG, VMUL(T3, TH)); Chris@82: } Chris@82: TJ = VADD(TF, TI); Chris@82: T17 = VADD(TW, TX); Chris@82: TV = VSUB(TF, TI); Chris@82: TY = VSUB(TW, TX); Chris@82: } Chris@82: { Chris@82: V Tk, TM, Tr, TN; Chris@82: { Chris@82: V Th, Tj, To, Tq; Chris@82: Th = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@82: Tj = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@82: Tk = VFMA(Tg, Th, VMUL(Ti, Tj)); Chris@82: TM = VFNMS(Ti, Th, VMUL(Tg, Tj)); Chris@82: To = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); Chris@82: Tq = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); Chris@82: Tr = VFMA(Tn, To, VMUL(Tp, Tq)); Chris@82: TN = VFNMS(Tp, To, VMUL(Tn, Tq)); Chris@82: } Chris@82: Ts = VADD(Tk, Tr); Chris@82: T1j = VSUB(Tk, Tr); Chris@82: TO = VSUB(TM, TN); Chris@82: T1a = VADD(TM, TN); Chris@82: } Chris@82: { Chris@82: V Tw, TR, TB, TS; Chris@82: { Chris@82: V Tu, Tv, Ty, TA; Chris@82: Tu = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Chris@82: Tv = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Chris@82: Tw = VFMA(T2, Tu, VMUL(T5, Tv)); Chris@82: TR = VFNMS(T5, Tu, VMUL(T2, Tv)); Chris@82: Ty = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); Chris@82: TA = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); Chris@82: TB = VFMA(Tx, Ty, VMUL(Tz, TA)); Chris@82: TS = VFNMS(Tz, Ty, VMUL(Tx, TA)); Chris@82: } Chris@82: TC = VADD(Tw, TB); Chris@82: T16 = VADD(TR, TS); Chris@82: TQ = VSUB(Tw, TB); Chris@82: TT = VSUB(TR, TS); Chris@82: } Chris@82: { Chris@82: V Tt, TK, T1f, T1g; Chris@82: Tt = VADD(Tf, Ts); Chris@82: TK = VADD(TC, TJ); Chris@82: ST(&(ri[WS(rs, 4)]), VSUB(Tt, TK), ms, &(ri[0])); Chris@82: ST(&(ri[0]), VADD(Tt, TK), ms, &(ri[0])); Chris@82: { Chris@82: V T19, T1e, T15, T18; Chris@82: T19 = VADD(T16, T17); Chris@82: T1e = VADD(T1a, T1d); Chris@82: ST(&(ii[0]), VADD(T19, T1e), ms, &(ii[0])); Chris@82: ST(&(ii[WS(rs, 4)]), VSUB(T1e, T19), ms, &(ii[0])); Chris@82: T15 = VSUB(Tf, Ts); Chris@82: T18 = VSUB(T16, T17); Chris@82: ST(&(ri[WS(rs, 6)]), VSUB(T15, T18), ms, &(ri[0])); Chris@82: ST(&(ri[WS(rs, 2)]), VADD(T15, T18), ms, &(ri[0])); Chris@82: } Chris@82: T1f = VSUB(TJ, TC); Chris@82: T1g = VSUB(T1d, T1a); Chris@82: ST(&(ii[WS(rs, 2)]), VADD(T1f, T1g), ms, &(ii[0])); Chris@82: ST(&(ii[WS(rs, 6)]), VSUB(T1g, T1f), ms, &(ii[0])); Chris@82: { Chris@82: V T11, T1k, T14, T1h, T12, T13; Chris@82: T11 = VSUB(TL, TO); Chris@82: T1k = VSUB(T1i, T1j); Chris@82: T12 = VSUB(TT, TQ); Chris@82: T13 = VADD(TV, TY); Chris@82: T14 = VMUL(LDK(KP707106781), VSUB(T12, T13)); Chris@82: T1h = VMUL(LDK(KP707106781), VADD(T12, T13)); Chris@82: ST(&(ri[WS(rs, 7)]), VSUB(T11, T14), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 5)]), VSUB(T1k, T1h), ms, &(ii[WS(rs, 1)])); Chris@82: ST(&(ri[WS(rs, 3)]), VADD(T11, T14), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 1)]), VADD(T1h, T1k), ms, &(ii[WS(rs, 1)])); Chris@82: } Chris@82: { Chris@82: V TP, T1m, T10, T1l, TU, TZ; Chris@82: TP = VADD(TL, TO); Chris@82: T1m = VADD(T1j, T1i); Chris@82: TU = VADD(TQ, TT); Chris@82: TZ = VSUB(TV, TY); Chris@82: T10 = VMUL(LDK(KP707106781), VADD(TU, TZ)); Chris@82: T1l = VMUL(LDK(KP707106781), VSUB(TZ, TU)); Chris@82: ST(&(ri[WS(rs, 5)]), VSUB(TP, T10), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 7)]), VSUB(T1m, T1l), ms, &(ii[WS(rs, 1)])); Chris@82: ST(&(ri[WS(rs, 1)]), VADD(TP, T10), ms, &(ri[WS(rs, 1)])); Chris@82: ST(&(ii[WS(rs, 3)]), VADD(T1l, T1m), ms, &(ii[WS(rs, 1)])); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const tw_instr twinstr[] = { Chris@82: VTW(0, 1), Chris@82: VTW(0, 3), Chris@82: VTW(0, 7), Chris@82: {TW_NEXT, (2 * VL), 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 8, XSIMD_STRING("t2sv_8"), twinstr, &GENUS, {56, 26, 18, 0}, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_t2sv_8) (planner *p) { Chris@82: X(kdft_dit_register) (p, t2sv_8, &desc); Chris@82: } Chris@82: #endif