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 08:06:13 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_twidsq_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 5 -dif -name q1fv_5 -include dft/simd/q1f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 100 FP additions, 95 FP multiplications, cannam@167: * (or, 55 additions, 50 multiplications, 45 fused multiply/add), cannam@167: * 44 stack variables, 4 constants, and 50 memory accesses cannam@167: */ cannam@167: #include "dft/simd/q1f.h" cannam@167: cannam@167: static void q1fv_5(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@167: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@167: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(10, rs), MAKE_VOLATILE_STRIDE(10, vs)) { cannam@167: V T1, Ta, Ti, Te, T8, T9, T1j, T1s, T1A, T1w, T1q, T1r, Tl, Tu, TC; cannam@167: V Ty, Ts, Tt, TF, TO, TW, TS, TM, TN, TZ, T18, T1g, T1c, T16, T17; cannam@167: { cannam@167: V T7, Td, T4, Tc; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: { cannam@167: V T5, T6, T2, T3; cannam@167: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T6 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: T7 = VADD(T5, T6); cannam@167: Td = VSUB(T5, T6); cannam@167: T2 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T3 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@167: T4 = VADD(T2, T3); cannam@167: Tc = VSUB(T2, T3); cannam@167: } cannam@167: Ta = VSUB(T4, T7); cannam@167: Ti = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tc, Td)); cannam@167: Te = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Td, Tc)); cannam@167: T8 = VADD(T4, T7); cannam@167: T9 = VFNMS(LDK(KP250000000), T8, T1); cannam@167: } cannam@167: { cannam@167: V T1p, T1v, T1m, T1u; cannam@167: T1j = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)])); cannam@167: { cannam@167: V T1n, T1o, T1k, T1l; cannam@167: T1n = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)])); cannam@167: T1o = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1p = VADD(T1n, T1o); cannam@167: T1v = VSUB(T1n, T1o); cannam@167: T1k = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1l = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)])); cannam@167: T1m = VADD(T1k, T1l); cannam@167: T1u = VSUB(T1k, T1l); cannam@167: } cannam@167: T1s = VSUB(T1m, T1p); cannam@167: T1A = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1u, T1v)); cannam@167: T1w = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1v, T1u)); cannam@167: T1q = VADD(T1m, T1p); cannam@167: T1r = VFNMS(LDK(KP250000000), T1q, T1j); cannam@167: } cannam@167: { cannam@167: V Tr, Tx, To, Tw; cannam@167: Tl = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@167: { cannam@167: V Tp, Tq, Tm, Tn; cannam@167: Tp = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@167: Tq = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tr = VADD(Tp, Tq); cannam@167: Tx = VSUB(Tp, Tq); cannam@167: Tm = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tn = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)])); cannam@167: To = VADD(Tm, Tn); cannam@167: Tw = VSUB(Tm, Tn); cannam@167: } cannam@167: Tu = VSUB(To, Tr); cannam@167: TC = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tw, Tx)); cannam@167: Ty = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tx, Tw)); cannam@167: Ts = VADD(To, Tr); cannam@167: Tt = VFNMS(LDK(KP250000000), Ts, Tl); cannam@167: } cannam@167: { cannam@167: V TL, TR, TI, TQ; cannam@167: TF = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: { cannam@167: V TJ, TK, TG, TH; cannam@167: TJ = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: TK = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: TL = VADD(TJ, TK); cannam@167: TR = VSUB(TJ, TK); cannam@167: TG = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: TH = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)])); cannam@167: TI = VADD(TG, TH); cannam@167: TQ = VSUB(TG, TH); cannam@167: } cannam@167: TO = VSUB(TI, TL); cannam@167: TW = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TQ, TR)); cannam@167: TS = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TR, TQ)); cannam@167: TM = VADD(TI, TL); cannam@167: TN = VFNMS(LDK(KP250000000), TM, TF); cannam@167: } cannam@167: { cannam@167: V T15, T1b, T12, T1a; cannam@167: TZ = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@167: { cannam@167: V T13, T14, T10, T11; cannam@167: T13 = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@167: T14 = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: T15 = VADD(T13, T14); cannam@167: T1b = VSUB(T13, T14); cannam@167: T10 = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: T11 = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)])); cannam@167: T12 = VADD(T10, T11); cannam@167: T1a = VSUB(T10, T11); cannam@167: } cannam@167: T18 = VSUB(T12, T15); cannam@167: T1g = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1a, T1b)); cannam@167: T1c = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1b, T1a)); cannam@167: T16 = VADD(T12, T15); cannam@167: T17 = VFNMS(LDK(KP250000000), T16, TZ); cannam@167: } cannam@167: ST(&(x[0]), VADD(T1, T8), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 4)]), VADD(T1j, T1q), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 2)]), VADD(TF, TM), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 3)]), VADD(TZ, T16), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 1)]), VADD(Tl, Ts), ms, &(x[WS(rs, 1)])); cannam@167: { cannam@167: V Tj, Tk, Th, T1B, T1C, T1z; cannam@167: Th = VFNMS(LDK(KP559016994), Ta, T9); cannam@167: Tj = BYTWJ(&(W[TWVL * 2]), VFMAI(Ti, Th)); cannam@167: Tk = BYTWJ(&(W[TWVL * 4]), VFNMSI(Ti, Th)); cannam@167: ST(&(x[WS(vs, 2)]), Tj, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3)]), Tk, ms, &(x[WS(vs, 3)])); cannam@167: T1z = VFNMS(LDK(KP559016994), T1s, T1r); cannam@167: T1B = BYTWJ(&(W[TWVL * 2]), VFMAI(T1A, T1z)); cannam@167: T1C = BYTWJ(&(W[TWVL * 4]), VFNMSI(T1A, T1z)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 4)]), T1B, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 4)]), T1C, ms, &(x[WS(vs, 3)])); cannam@167: } cannam@167: { cannam@167: V T1h, T1i, T1f, TD, TE, TB; cannam@167: T1f = VFNMS(LDK(KP559016994), T18, T17); cannam@167: T1h = BYTWJ(&(W[TWVL * 2]), VFMAI(T1g, T1f)); cannam@167: T1i = BYTWJ(&(W[TWVL * 4]), VFNMSI(T1g, T1f)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 3)]), T1h, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1i, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TB = VFNMS(LDK(KP559016994), Tu, Tt); cannam@167: TD = BYTWJ(&(W[TWVL * 2]), VFMAI(TC, TB)); cannam@167: TE = BYTWJ(&(W[TWVL * 4]), VFNMSI(TC, TB)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 1)]), TD, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 1)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V TX, TY, TV, TT, TU, TP; cannam@167: TV = VFNMS(LDK(KP559016994), TO, TN); cannam@167: TX = BYTWJ(&(W[TWVL * 2]), VFMAI(TW, TV)); cannam@167: TY = BYTWJ(&(W[TWVL * 4]), VFNMSI(TW, TV)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 2)]), TX, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 2)]), TY, ms, &(x[WS(vs, 3)])); cannam@167: TP = VFMA(LDK(KP559016994), TO, TN); cannam@167: TT = BYTWJ(&(W[0]), VFNMSI(TS, TP)); cannam@167: TU = BYTWJ(&(W[TWVL * 6]), VFMAI(TS, TP)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 2)]), TT, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 2)]), TU, ms, &(x[WS(vs, 4)])); cannam@167: } cannam@167: { cannam@167: V Tf, Tg, Tb, Tz, TA, Tv; cannam@167: Tb = VFMA(LDK(KP559016994), Ta, T9); cannam@167: Tf = BYTWJ(&(W[0]), VFNMSI(Te, Tb)); cannam@167: Tg = BYTWJ(&(W[TWVL * 6]), VFMAI(Te, Tb)); cannam@167: ST(&(x[WS(vs, 1)]), Tf, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4)]), Tg, ms, &(x[WS(vs, 4)])); cannam@167: Tv = VFMA(LDK(KP559016994), Tu, Tt); cannam@167: Tz = BYTWJ(&(W[0]), VFNMSI(Ty, Tv)); cannam@167: TA = BYTWJ(&(W[TWVL * 6]), VFMAI(Ty, Tv)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tz, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 1)]), TA, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V T1d, T1e, T19, T1x, T1y, T1t; cannam@167: T19 = VFMA(LDK(KP559016994), T18, T17); cannam@167: T1d = BYTWJ(&(W[0]), VFNMSI(T1c, T19)); cannam@167: T1e = BYTWJ(&(W[TWVL * 6]), VFMAI(T1c, T19)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1d, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 3)]), T1e, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1t = VFMA(LDK(KP559016994), T1s, T1r); cannam@167: T1x = BYTWJ(&(W[0]), VFNMSI(T1w, T1t)); cannam@167: T1y = BYTWJ(&(W[TWVL * 6]), VFMAI(T1w, T1t)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 4)]), T1x, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 4)]), T1y, ms, &(x[WS(vs, 4)])); 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, 2), cannam@167: VTW(0, 3), cannam@167: VTW(0, 4), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 5, XSIMD_STRING("q1fv_5"), twinstr, &GENUS, {55, 50, 45, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_q1fv_5) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1fv_5, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twidsq_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 5 -dif -name q1fv_5 -include dft/simd/q1f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 100 FP additions, 70 FP multiplications, cannam@167: * (or, 85 additions, 55 multiplications, 15 fused multiply/add), cannam@167: * 44 stack variables, 4 constants, and 50 memory accesses cannam@167: */ cannam@167: #include "dft/simd/q1f.h" cannam@167: cannam@167: static void q1fv_5(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@167: { cannam@167: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@167: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@167: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@167: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(10, rs), MAKE_VOLATILE_STRIDE(10, vs)) { cannam@167: V T8, T7, Th, Te, T9, Ta, T1q, T1p, T1z, T1w, T1r, T1s, Ts, Tr, TB; cannam@167: V Ty, Tt, Tu, TM, TL, TV, TS, TN, TO, T16, T15, T1f, T1c, T17, T18; cannam@167: { cannam@167: V T6, Td, T3, Tc; cannam@167: T8 = LD(&(x[0]), ms, &(x[0])); cannam@167: { cannam@167: V T4, T5, T1, T2; cannam@167: T4 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: T6 = VADD(T4, T5); cannam@167: Td = VSUB(T4, T5); cannam@167: T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@167: T3 = VADD(T1, T2); cannam@167: Tc = VSUB(T1, T2); cannam@167: } cannam@167: T7 = VMUL(LDK(KP559016994), VSUB(T3, T6)); cannam@167: Th = VBYI(VFNMS(LDK(KP587785252), Tc, VMUL(LDK(KP951056516), Td))); cannam@167: Te = VBYI(VFMA(LDK(KP951056516), Tc, VMUL(LDK(KP587785252), Td))); cannam@167: T9 = VADD(T3, T6); cannam@167: Ta = VFNMS(LDK(KP250000000), T9, T8); cannam@167: } cannam@167: { cannam@167: V T1o, T1v, T1l, T1u; cannam@167: T1q = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)])); cannam@167: { cannam@167: V T1m, T1n, T1j, T1k; cannam@167: T1m = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)])); cannam@167: T1n = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1o = VADD(T1m, T1n); cannam@167: T1v = VSUB(T1m, T1n); cannam@167: T1j = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1k = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)])); cannam@167: T1l = VADD(T1j, T1k); cannam@167: T1u = VSUB(T1j, T1k); cannam@167: } cannam@167: T1p = VMUL(LDK(KP559016994), VSUB(T1l, T1o)); cannam@167: T1z = VBYI(VFNMS(LDK(KP587785252), T1u, VMUL(LDK(KP951056516), T1v))); cannam@167: T1w = VBYI(VFMA(LDK(KP951056516), T1u, VMUL(LDK(KP587785252), T1v))); cannam@167: T1r = VADD(T1l, T1o); cannam@167: T1s = VFNMS(LDK(KP250000000), T1r, T1q); cannam@167: } cannam@167: { cannam@167: V Tq, Tx, Tn, Tw; cannam@167: Ts = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@167: { cannam@167: V To, Tp, Tl, Tm; cannam@167: To = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@167: Tp = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tq = VADD(To, Tp); cannam@167: Tx = VSUB(To, Tp); cannam@167: Tl = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: Tm = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)])); cannam@167: Tn = VADD(Tl, Tm); cannam@167: Tw = VSUB(Tl, Tm); cannam@167: } cannam@167: Tr = VMUL(LDK(KP559016994), VSUB(Tn, Tq)); cannam@167: TB = VBYI(VFNMS(LDK(KP587785252), Tw, VMUL(LDK(KP951056516), Tx))); cannam@167: Ty = VBYI(VFMA(LDK(KP951056516), Tw, VMUL(LDK(KP587785252), Tx))); cannam@167: Tt = VADD(Tn, Tq); cannam@167: Tu = VFNMS(LDK(KP250000000), Tt, Ts); cannam@167: } cannam@167: { cannam@167: V TK, TR, TH, TQ; cannam@167: TM = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: { cannam@167: V TI, TJ, TF, TG; cannam@167: TI = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@167: TJ = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: TK = VADD(TI, TJ); cannam@167: TR = VSUB(TI, TJ); cannam@167: TF = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: TG = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)])); cannam@167: TH = VADD(TF, TG); cannam@167: TQ = VSUB(TF, TG); cannam@167: } cannam@167: TL = VMUL(LDK(KP559016994), VSUB(TH, TK)); cannam@167: TV = VBYI(VFNMS(LDK(KP587785252), TQ, VMUL(LDK(KP951056516), TR))); cannam@167: TS = VBYI(VFMA(LDK(KP951056516), TQ, VMUL(LDK(KP587785252), TR))); cannam@167: TN = VADD(TH, TK); cannam@167: TO = VFNMS(LDK(KP250000000), TN, TM); cannam@167: } cannam@167: { cannam@167: V T14, T1b, T11, T1a; cannam@167: T16 = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@167: { cannam@167: V T12, T13, TZ, T10; cannam@167: T12 = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@167: T13 = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: T14 = VADD(T12, T13); cannam@167: T1b = VSUB(T12, T13); cannam@167: TZ = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: T10 = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)])); cannam@167: T11 = VADD(TZ, T10); cannam@167: T1a = VSUB(TZ, T10); cannam@167: } cannam@167: T15 = VMUL(LDK(KP559016994), VSUB(T11, T14)); cannam@167: T1f = VBYI(VFNMS(LDK(KP587785252), T1a, VMUL(LDK(KP951056516), T1b))); cannam@167: T1c = VBYI(VFMA(LDK(KP951056516), T1a, VMUL(LDK(KP587785252), T1b))); cannam@167: T17 = VADD(T11, T14); cannam@167: T18 = VFNMS(LDK(KP250000000), T17, T16); cannam@167: } cannam@167: ST(&(x[0]), VADD(T8, T9), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 4)]), VADD(T1q, T1r), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 2)]), VADD(TM, TN), ms, &(x[0])); cannam@167: ST(&(x[WS(rs, 3)]), VADD(T16, T17), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 1)]), VADD(Ts, Tt), ms, &(x[WS(rs, 1)])); cannam@167: { cannam@167: V Tj, Tk, Ti, T1B, T1C, T1A; cannam@167: Ti = VSUB(Ta, T7); cannam@167: Tj = BYTWJ(&(W[TWVL * 2]), VADD(Th, Ti)); cannam@167: Tk = BYTWJ(&(W[TWVL * 4]), VSUB(Ti, Th)); cannam@167: ST(&(x[WS(vs, 2)]), Tj, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3)]), Tk, ms, &(x[WS(vs, 3)])); cannam@167: T1A = VSUB(T1s, T1p); cannam@167: T1B = BYTWJ(&(W[TWVL * 2]), VADD(T1z, T1A)); cannam@167: T1C = BYTWJ(&(W[TWVL * 4]), VSUB(T1A, T1z)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 4)]), T1B, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 4)]), T1C, ms, &(x[WS(vs, 3)])); cannam@167: } cannam@167: { cannam@167: V T1h, T1i, T1g, TD, TE, TC; cannam@167: T1g = VSUB(T18, T15); cannam@167: T1h = BYTWJ(&(W[TWVL * 2]), VADD(T1f, T1g)); cannam@167: T1i = BYTWJ(&(W[TWVL * 4]), VSUB(T1g, T1f)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 3)]), T1h, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1i, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: TC = VSUB(Tu, Tr); cannam@167: TD = BYTWJ(&(W[TWVL * 2]), VADD(TB, TC)); cannam@167: TE = BYTWJ(&(W[TWVL * 4]), VSUB(TC, TB)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 1)]), TD, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 1)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V TX, TY, TW, TT, TU, TP; cannam@167: TW = VSUB(TO, TL); cannam@167: TX = BYTWJ(&(W[TWVL * 2]), VADD(TV, TW)); cannam@167: TY = BYTWJ(&(W[TWVL * 4]), VSUB(TW, TV)); cannam@167: ST(&(x[WS(vs, 2) + WS(rs, 2)]), TX, ms, &(x[WS(vs, 2)])); cannam@167: ST(&(x[WS(vs, 3) + WS(rs, 2)]), TY, ms, &(x[WS(vs, 3)])); cannam@167: TP = VADD(TL, TO); cannam@167: TT = BYTWJ(&(W[0]), VSUB(TP, TS)); cannam@167: TU = BYTWJ(&(W[TWVL * 6]), VADD(TS, TP)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 2)]), TT, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 2)]), TU, ms, &(x[WS(vs, 4)])); cannam@167: } cannam@167: { cannam@167: V Tf, Tg, Tb, Tz, TA, Tv; cannam@167: Tb = VADD(T7, Ta); cannam@167: Tf = BYTWJ(&(W[0]), VSUB(Tb, Te)); cannam@167: Tg = BYTWJ(&(W[TWVL * 6]), VADD(Te, Tb)); cannam@167: ST(&(x[WS(vs, 1)]), Tf, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4)]), Tg, ms, &(x[WS(vs, 4)])); cannam@167: Tv = VADD(Tr, Tu); cannam@167: Tz = BYTWJ(&(W[0]), VSUB(Tv, Ty)); cannam@167: TA = BYTWJ(&(W[TWVL * 6]), VADD(Ty, Tv)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tz, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 1)]), TA, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: } cannam@167: { cannam@167: V T1d, T1e, T19, T1x, T1y, T1t; cannam@167: T19 = VADD(T15, T18); cannam@167: T1d = BYTWJ(&(W[0]), VSUB(T19, T1c)); cannam@167: T1e = BYTWJ(&(W[TWVL * 6]), VADD(T1c, T19)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1d, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 3)]), T1e, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@167: T1t = VADD(T1p, T1s); cannam@167: T1x = BYTWJ(&(W[0]), VSUB(T1t, T1w)); cannam@167: T1y = BYTWJ(&(W[TWVL * 6]), VADD(T1w, T1t)); cannam@167: ST(&(x[WS(vs, 1) + WS(rs, 4)]), T1x, ms, &(x[WS(vs, 1)])); cannam@167: ST(&(x[WS(vs, 4) + WS(rs, 4)]), T1y, ms, &(x[WS(vs, 4)])); 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, 2), cannam@167: VTW(0, 3), cannam@167: VTW(0, 4), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 5, XSIMD_STRING("q1fv_5"), twinstr, &GENUS, {85, 55, 15, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_q1fv_5) (planner *p) { cannam@167: X(kdft_difsq_register) (p, q1fv_5, &desc); cannam@167: } cannam@167: #endif