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 07:39:24 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 -n 8 -name t1sv_8 -include ts.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 36 FP multiplications, cannam@95: * (or, 44 additions, 14 multiplications, 22 fused multiply/add), cannam@95: * 59 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "ts.h" cannam@95: cannam@95: static void t1sv_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 * 14); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 14), MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: V T1, T1m, T1l, T7, TS, Tk, TQ, Te, To, Tr, Tu, T14, TF, Tx, T16; cannam@95: V TL, Tt, TW, Tp, Tq, Tw; cannam@95: { cannam@95: V T3, T6, T2, T5; cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: T1m = LD(&(ii[0]), ms, &(ii[0])); cannam@95: T3 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@95: T6 = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@95: T2 = LDW(&(W[TWVL * 6])); cannam@95: T5 = LDW(&(W[TWVL * 7])); cannam@95: { cannam@95: V Tg, Tj, Ti, Ta, Td, T1k, T4, T9, Tc, TR, Th, Tf; cannam@95: Tg = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@95: Tj = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@95: Tf = LDW(&(W[TWVL * 10])); cannam@95: Ti = LDW(&(W[TWVL * 11])); cannam@95: Ta = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: Td = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: T1k = VMUL(T2, T6); cannam@95: T4 = VMUL(T2, T3); cannam@95: T9 = LDW(&(W[TWVL * 2])); cannam@95: Tc = LDW(&(W[TWVL * 3])); cannam@95: TR = VMUL(Tf, Tj); cannam@95: Th = VMUL(Tf, Tg); cannam@95: { cannam@95: V TB, TE, TH, TK, TG, TD, TJ, T13, TC, TA, TP, Tb, T15, TI, Tn; cannam@95: TB = 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: T1l = VFNMS(T5, T3, T1k); cannam@95: T7 = VFMA(T5, T6, T4); cannam@95: TP = VMUL(T9, Td); cannam@95: Tb = VMUL(T9, Ta); cannam@95: TS = VFNMS(Ti, Tg, TR); cannam@95: Tk = VFMA(Ti, Tj, Th); cannam@95: TA = LDW(&(W[TWVL * 12])); cannam@95: TH = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: TK = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: TG = LDW(&(W[TWVL * 4])); cannam@95: TQ = VFNMS(Tc, Ta, TP); cannam@95: Te = VFMA(Tc, Td, Tb); cannam@95: TD = LDW(&(W[TWVL * 13])); cannam@95: TJ = LDW(&(W[TWVL * 5])); cannam@95: T13 = VMUL(TA, TE); cannam@95: TC = VMUL(TA, TB); cannam@95: To = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: T15 = VMUL(TG, TK); cannam@95: TI = VMUL(TG, TH); cannam@95: Tr = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tn = LDW(&(W[0])); cannam@95: Tu = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@95: T14 = VFNMS(TD, TB, T13); cannam@95: TF = VFMA(TD, TE, TC); cannam@95: Tx = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@95: T16 = VFNMS(TJ, TH, T15); cannam@95: TL = VFMA(TJ, TK, TI); cannam@95: Tt = LDW(&(W[TWVL * 8])); cannam@95: TW = VMUL(Tn, Tr); cannam@95: Tp = VMUL(Tn, To); cannam@95: Tq = LDW(&(W[TWVL * 1])); cannam@95: Tw = LDW(&(W[TWVL * 9])); cannam@95: } cannam@95: } cannam@95: } cannam@95: { cannam@95: V T8, T1g, TM, T1j, TX, Ts, T1n, T1r, T1s, Tl, T1c, T18, TZ, Ty, T1a; cannam@95: V TU; cannam@95: { cannam@95: V TO, T17, T12, TY, Tv, TT; cannam@95: T8 = VADD(T1, T7); cannam@95: TO = VSUB(T1, T7); cannam@95: T17 = VSUB(T14, T16); cannam@95: T1g = VADD(T14, T16); cannam@95: TM = VADD(TF, TL); cannam@95: T12 = VSUB(TF, TL); cannam@95: TY = VMUL(Tt, Tx); cannam@95: Tv = VMUL(Tt, Tu); cannam@95: TT = VSUB(TQ, TS); cannam@95: T1j = VADD(TQ, TS); cannam@95: TX = VFNMS(Tq, To, TW); cannam@95: Ts = VFMA(Tq, Tr, Tp); cannam@95: T1n = VADD(T1l, T1m); cannam@95: T1r = VSUB(T1m, T1l); cannam@95: T1s = VSUB(Te, Tk); cannam@95: Tl = VADD(Te, Tk); cannam@95: T1c = VADD(T12, T17); cannam@95: T18 = VSUB(T12, T17); cannam@95: TZ = VFNMS(Tw, Tu, TY); cannam@95: Ty = VFMA(Tw, Tx, Tv); cannam@95: T1a = VSUB(TO, TT); cannam@95: TU = VADD(TO, TT); cannam@95: } cannam@95: { cannam@95: V T1v, T1t, Tm, T1e, T1o, T1q, TN, T1p, T1d, T1u, T19, T1w, T1i, T1h; cannam@95: { cannam@95: V T10, T1f, Tz, TV, T11, T1b; cannam@95: T1v = VADD(T1s, T1r); cannam@95: T1t = VSUB(T1r, T1s); cannam@95: T10 = VSUB(TX, TZ); cannam@95: T1f = VADD(TX, TZ); cannam@95: Tz = VADD(Ts, Ty); cannam@95: TV = VSUB(Ts, Ty); cannam@95: T11 = VADD(TV, T10); cannam@95: T1b = VSUB(T10, TV); cannam@95: Tm = VADD(T8, Tl); cannam@95: T1e = VSUB(T8, Tl); cannam@95: T1o = VADD(T1j, T1n); cannam@95: T1q = VSUB(T1n, T1j); cannam@95: TN = VADD(Tz, TM); cannam@95: T1p = VSUB(TM, Tz); cannam@95: T1d = VSUB(T1b, T1c); cannam@95: T1u = VADD(T1b, T1c); cannam@95: T19 = VADD(T11, T18); cannam@95: T1w = VSUB(T18, T11); cannam@95: T1i = VADD(T1f, T1g); cannam@95: T1h = VSUB(T1f, T1g); cannam@95: } cannam@95: ST(&(ii[WS(rs, 6)]), VSUB(T1q, T1p), ms, &(ii[0])); cannam@95: ST(&(ri[0]), VADD(Tm, TN), ms, &(ri[0])); cannam@95: ST(&(ri[WS(rs, 4)]), VSUB(Tm, TN), ms, &(ri[0])); cannam@95: ST(&(ii[WS(rs, 1)]), VFMA(LDK(KP707106781), T1u, T1t), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 5)]), VFNMS(LDK(KP707106781), T1u, T1t), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 3)]), VFMA(LDK(KP707106781), T1d, T1a), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1d, T1a), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 3)]), VFMA(LDK(KP707106781), T1w, T1v), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 7)]), VFNMS(LDK(KP707106781), T1w, T1v), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 1)]), VFMA(LDK(KP707106781), T19, TU), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 5)]), VFNMS(LDK(KP707106781), T19, TU), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 6)]), VSUB(T1e, T1h), ms, &(ri[0])); cannam@95: ST(&(ii[0]), VADD(T1i, T1o), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 4)]), VSUB(T1o, T1i), ms, &(ii[0])); cannam@95: ST(&(ri[WS(rs, 2)]), VADD(T1e, T1h), ms, &(ri[0])); cannam@95: ST(&(ii[WS(rs, 2)]), VADD(T1p, T1q), ms, &(ii[0])); 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, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), 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("t1sv_8"), twinstr, &GENUS, {44, 14, 22, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1sv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1sv_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 -n 8 -name t1sv_8 -include ts.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 66 FP additions, 32 FP multiplications, cannam@95: * (or, 52 additions, 18 multiplications, 14 fused multiply/add), cannam@95: * 28 stack variables, 1 constants, and 32 memory accesses cannam@95: */ cannam@95: #include "ts.h" cannam@95: cannam@95: static void t1sv_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 * 14); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 14), MAKE_VOLATILE_STRIDE(16, rs)) { cannam@95: V T7, T1e, TH, T19, TF, T13, TR, TU, Ti, T1f, TK, T16, Tu, T12, TM; cannam@95: V TP; cannam@95: { cannam@95: V T1, T18, T6, T17; cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: T18 = LD(&(ii[0]), ms, &(ii[0])); cannam@95: { cannam@95: V T3, T5, T2, T4; cannam@95: T3 = LD(&(ri[WS(rs, 4)]), ms, &(ri[0])); cannam@95: T5 = LD(&(ii[WS(rs, 4)]), ms, &(ii[0])); cannam@95: T2 = LDW(&(W[TWVL * 6])); cannam@95: T4 = LDW(&(W[TWVL * 7])); cannam@95: T6 = VFMA(T2, T3, VMUL(T4, T5)); cannam@95: T17 = VFNMS(T4, T3, VMUL(T2, T5)); cannam@95: } cannam@95: T7 = VADD(T1, T6); cannam@95: T1e = VSUB(T18, T17); cannam@95: TH = VSUB(T1, T6); cannam@95: T19 = VADD(T17, T18); cannam@95: } cannam@95: { cannam@95: V Tz, TS, TE, TT; cannam@95: { cannam@95: V Tw, Ty, Tv, Tx; cannam@95: Tw = LD(&(ri[WS(rs, 7)]), ms, &(ri[WS(rs, 1)])); cannam@95: Ty = LD(&(ii[WS(rs, 7)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tv = LDW(&(W[TWVL * 12])); cannam@95: Tx = LDW(&(W[TWVL * 13])); cannam@95: Tz = VFMA(Tv, Tw, VMUL(Tx, Ty)); cannam@95: TS = VFNMS(Tx, Tw, VMUL(Tv, Ty)); cannam@95: } cannam@95: { cannam@95: V TB, TD, TA, TC; cannam@95: TB = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: TD = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: TA = LDW(&(W[TWVL * 4])); cannam@95: TC = LDW(&(W[TWVL * 5])); cannam@95: TE = VFMA(TA, TB, VMUL(TC, TD)); cannam@95: TT = VFNMS(TC, TB, VMUL(TA, TD)); cannam@95: } cannam@95: TF = VADD(Tz, TE); cannam@95: T13 = VADD(TS, TT); cannam@95: TR = VSUB(Tz, TE); cannam@95: TU = VSUB(TS, TT); cannam@95: } cannam@95: { cannam@95: V Tc, TI, Th, TJ; cannam@95: { cannam@95: V T9, Tb, T8, Ta; cannam@95: T9 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: Tb = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: T8 = LDW(&(W[TWVL * 2])); cannam@95: Ta = LDW(&(W[TWVL * 3])); cannam@95: Tc = VFMA(T8, T9, VMUL(Ta, Tb)); cannam@95: TI = VFNMS(Ta, T9, VMUL(T8, Tb)); cannam@95: } cannam@95: { cannam@95: V Te, Tg, Td, Tf; cannam@95: Te = LD(&(ri[WS(rs, 6)]), ms, &(ri[0])); cannam@95: Tg = LD(&(ii[WS(rs, 6)]), ms, &(ii[0])); cannam@95: Td = LDW(&(W[TWVL * 10])); cannam@95: Tf = LDW(&(W[TWVL * 11])); cannam@95: Th = VFMA(Td, Te, VMUL(Tf, Tg)); cannam@95: TJ = VFNMS(Tf, Te, VMUL(Td, Tg)); cannam@95: } cannam@95: Ti = VADD(Tc, Th); cannam@95: T1f = VSUB(Tc, Th); cannam@95: TK = VSUB(TI, TJ); cannam@95: T16 = VADD(TI, TJ); cannam@95: } cannam@95: { cannam@95: V To, TN, Tt, TO; cannam@95: { cannam@95: V Tl, Tn, Tk, Tm; cannam@95: Tl = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: Tn = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tk = LDW(&(W[0])); cannam@95: Tm = LDW(&(W[TWVL * 1])); cannam@95: To = VFMA(Tk, Tl, VMUL(Tm, Tn)); cannam@95: TN = VFNMS(Tm, Tl, VMUL(Tk, Tn)); cannam@95: } cannam@95: { cannam@95: V Tq, Ts, Tp, Tr; cannam@95: Tq = LD(&(ri[WS(rs, 5)]), ms, &(ri[WS(rs, 1)])); cannam@95: Ts = LD(&(ii[WS(rs, 5)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tp = LDW(&(W[TWVL * 8])); cannam@95: Tr = LDW(&(W[TWVL * 9])); cannam@95: Tt = VFMA(Tp, Tq, VMUL(Tr, Ts)); cannam@95: TO = VFNMS(Tr, Tq, VMUL(Tp, Ts)); cannam@95: } cannam@95: Tu = VADD(To, Tt); cannam@95: T12 = VADD(TN, TO); cannam@95: TM = VSUB(To, Tt); cannam@95: TP = VSUB(TN, TO); cannam@95: } cannam@95: { cannam@95: V Tj, TG, T1b, T1c; cannam@95: Tj = VADD(T7, Ti); cannam@95: TG = VADD(Tu, TF); cannam@95: ST(&(ri[WS(rs, 4)]), VSUB(Tj, TG), ms, &(ri[0])); cannam@95: ST(&(ri[0]), VADD(Tj, TG), ms, &(ri[0])); cannam@95: { cannam@95: V T15, T1a, T11, T14; cannam@95: T15 = VADD(T12, T13); cannam@95: T1a = VADD(T16, T19); cannam@95: ST(&(ii[0]), VADD(T15, T1a), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 4)]), VSUB(T1a, T15), ms, &(ii[0])); cannam@95: T11 = VSUB(T7, Ti); cannam@95: T14 = VSUB(T12, T13); cannam@95: ST(&(ri[WS(rs, 6)]), VSUB(T11, T14), ms, &(ri[0])); cannam@95: ST(&(ri[WS(rs, 2)]), VADD(T11, T14), ms, &(ri[0])); cannam@95: } cannam@95: T1b = VSUB(TF, Tu); cannam@95: T1c = VSUB(T19, T16); cannam@95: ST(&(ii[WS(rs, 2)]), VADD(T1b, T1c), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 6)]), VSUB(T1c, T1b), ms, &(ii[0])); cannam@95: { cannam@95: V TX, T1g, T10, T1d, TY, TZ; cannam@95: TX = VSUB(TH, TK); cannam@95: T1g = VSUB(T1e, T1f); cannam@95: TY = VSUB(TP, TM); cannam@95: TZ = VADD(TR, TU); cannam@95: T10 = VMUL(LDK(KP707106781), VSUB(TY, TZ)); cannam@95: T1d = VMUL(LDK(KP707106781), VADD(TY, TZ)); cannam@95: ST(&(ri[WS(rs, 7)]), VSUB(TX, T10), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 5)]), VSUB(T1g, T1d), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 3)]), VADD(TX, T10), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 1)]), VADD(T1d, T1g), ms, &(ii[WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V TL, T1i, TW, T1h, TQ, TV; cannam@95: TL = VADD(TH, TK); cannam@95: T1i = VADD(T1f, T1e); cannam@95: TQ = VADD(TM, TP); cannam@95: TV = VSUB(TR, TU); cannam@95: TW = VMUL(LDK(KP707106781), VADD(TQ, TV)); cannam@95: T1h = VMUL(LDK(KP707106781), VSUB(TV, TQ)); cannam@95: ST(&(ri[WS(rs, 5)]), VSUB(TL, TW), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 7)]), VSUB(T1i, T1h), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 1)]), VADD(TL, TW), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 3)]), VADD(T1h, T1i), ms, &(ii[WS(rs, 1)])); 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, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), 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("t1sv_8"), twinstr, &GENUS, {52, 18, 14, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1sv_8) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1sv_8, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */