cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:45:28 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twidsq_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 5 -dif -name q1bv_5 -include q1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 100 FP additions, 95 FP multiplications, cannam@127: * (or, 55 additions, 50 multiplications, 45 fused multiply/add), cannam@127: * 69 stack variables, 4 constants, and 50 memory accesses cannam@127: */ cannam@127: #include "q1b.h" cannam@127: cannam@127: static void q1bv_5(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: 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@127: V Te, T1w, Ty, TS, TW, Tb, T1t, Tv, T1g, T1c, TP, TV, T1f, T19, TY; cannam@127: V TX; cannam@127: { cannam@127: V T1, T1j, Tl, Ti, Ta, T8, T1A, T1q, T1s, T9, TF, T1r, TZ, TR, TL; cannam@127: V TC, Ts, Tu, TQ, TI, T15, T1b, T10, T11, Tt; cannam@127: { cannam@127: V T1n, T1o, T1k, T1l, T7, Td, T4, Tc; cannam@127: { cannam@127: V T5, T6, T2, T3; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: T6 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T2 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T3 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: T1j = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)])); cannam@127: T1n = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)])); cannam@127: T1o = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: T1k = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: T1l = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)])); cannam@127: T7 = VADD(T5, T6); cannam@127: Td = VSUB(T5, T6); cannam@127: T4 = VADD(T2, T3); cannam@127: Tc = VSUB(T2, T3); cannam@127: } cannam@127: { cannam@127: V Tm, Tn, Tr, Tx, T1v, T1p; cannam@127: Tl = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@127: T1v = VSUB(T1n, T1o); cannam@127: T1p = VADD(T1n, T1o); cannam@127: { cannam@127: V T1u, T1m, Tp, Tq; cannam@127: T1u = VSUB(T1k, T1l); cannam@127: T1m = VADD(T1k, T1l); cannam@127: Tp = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@127: Ti = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tc, Td)); cannam@127: Te = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Td, Tc)); cannam@127: Ta = VSUB(T4, T7); cannam@127: T8 = VADD(T4, T7); cannam@127: Tq = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: T1w = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1v, T1u)); cannam@127: T1A = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1u, T1v)); cannam@127: T1q = VADD(T1m, T1p); cannam@127: T1s = VSUB(T1m, T1p); cannam@127: Tm = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: T9 = VFNMS(LDK(KP250000000), T8, T1); cannam@127: Tn = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)])); cannam@127: Tr = VADD(Tp, Tq); cannam@127: Tx = VSUB(Tp, Tq); cannam@127: } cannam@127: { cannam@127: V TJ, TK, TG, Tw, To, TH, T13, T14; cannam@127: TF = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: T1r = VFNMS(LDK(KP250000000), T1q, T1j); cannam@127: TJ = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: TK = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: TG = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: Tw = VSUB(Tm, Tn); cannam@127: To = VADD(Tm, Tn); cannam@127: TH = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)])); cannam@127: TZ = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@127: T13 = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@127: T14 = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: TR = VSUB(TJ, TK); cannam@127: TL = VADD(TJ, TK); cannam@127: Ty = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tx, Tw)); cannam@127: TC = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tw, Tx)); cannam@127: Ts = VADD(To, Tr); cannam@127: Tu = VSUB(To, Tr); cannam@127: TQ = VSUB(TG, TH); cannam@127: TI = VADD(TG, TH); cannam@127: T15 = VADD(T13, T14); cannam@127: T1b = VSUB(T13, T14); cannam@127: T10 = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: T11 = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)])); cannam@127: Tt = VFNMS(LDK(KP250000000), Ts, Tl); cannam@127: } cannam@127: } cannam@127: } cannam@127: { cannam@127: V TO, T12, T1a, Th, T1z, TN, TM, T18, T17; cannam@127: ST(&(x[0]), VADD(T1, T8), ms, &(x[0])); cannam@127: TS = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TR, TQ)); cannam@127: TW = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TQ, TR)); cannam@127: TM = VADD(TI, TL); cannam@127: TO = VSUB(TI, TL); cannam@127: ST(&(x[WS(rs, 4)]), VADD(T1j, T1q), ms, &(x[0])); cannam@127: T12 = VADD(T10, T11); cannam@127: T1a = VSUB(T10, T11); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tl, Ts), ms, &(x[WS(rs, 1)])); cannam@127: Th = VFNMS(LDK(KP559016994), Ta, T9); cannam@127: Tb = VFMA(LDK(KP559016994), Ta, T9); cannam@127: T1t = VFMA(LDK(KP559016994), T1s, T1r); cannam@127: T1z = VFNMS(LDK(KP559016994), T1s, T1r); cannam@127: ST(&(x[WS(rs, 2)]), VADD(TF, TM), ms, &(x[0])); cannam@127: TN = VFNMS(LDK(KP250000000), TM, TF); cannam@127: { cannam@127: V T16, Tk, Tj, T1C, T1B, TD, TE, TB; cannam@127: TB = VFNMS(LDK(KP559016994), Tu, Tt); cannam@127: Tv = VFMA(LDK(KP559016994), Tu, Tt); cannam@127: T1g = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1a, T1b)); cannam@127: T1c = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1b, T1a)); cannam@127: T18 = VSUB(T12, T15); cannam@127: T16 = VADD(T12, T15); cannam@127: Tk = BYTW(&(W[TWVL * 4]), VFMAI(Ti, Th)); cannam@127: Tj = BYTW(&(W[TWVL * 2]), VFNMSI(Ti, Th)); cannam@127: T1C = BYTW(&(W[TWVL * 4]), VFMAI(T1A, T1z)); cannam@127: T1B = BYTW(&(W[TWVL * 2]), VFNMSI(T1A, T1z)); cannam@127: TD = BYTW(&(W[TWVL * 2]), VFNMSI(TC, TB)); cannam@127: TE = BYTW(&(W[TWVL * 4]), VFMAI(TC, TB)); cannam@127: ST(&(x[WS(rs, 3)]), VADD(TZ, T16), ms, &(x[WS(rs, 1)])); cannam@127: T17 = VFNMS(LDK(KP250000000), T16, TZ); cannam@127: ST(&(x[WS(vs, 3)]), Tk, ms, &(x[WS(vs, 3)])); cannam@127: ST(&(x[WS(vs, 2)]), Tj, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 4)]), T1C, ms, &(x[WS(vs, 3)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 4)]), T1B, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 1)]), TD, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 1)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: } cannam@127: TP = VFMA(LDK(KP559016994), TO, TN); cannam@127: TV = VFNMS(LDK(KP559016994), TO, TN); cannam@127: T1f = VFNMS(LDK(KP559016994), T18, T17); cannam@127: T19 = VFMA(LDK(KP559016994), T18, T17); cannam@127: } cannam@127: } cannam@127: TY = BYTW(&(W[TWVL * 4]), VFMAI(TW, TV)); cannam@127: TX = BYTW(&(W[TWVL * 2]), VFNMSI(TW, TV)); cannam@127: { cannam@127: V T1i, T1h, TU, TT; cannam@127: T1i = BYTW(&(W[TWVL * 4]), VFMAI(T1g, T1f)); cannam@127: T1h = BYTW(&(W[TWVL * 2]), VFNMSI(T1g, T1f)); cannam@127: TU = BYTW(&(W[TWVL * 6]), VFNMSI(TS, TP)); cannam@127: TT = BYTW(&(W[0]), VFMAI(TS, TP)); cannam@127: { cannam@127: V Tg, Tf, TA, Tz; cannam@127: Tg = BYTW(&(W[TWVL * 6]), VFNMSI(Te, Tb)); cannam@127: Tf = BYTW(&(W[0]), VFMAI(Te, Tb)); cannam@127: TA = BYTW(&(W[TWVL * 6]), VFNMSI(Ty, Tv)); cannam@127: Tz = BYTW(&(W[0]), VFMAI(Ty, Tv)); cannam@127: { cannam@127: V T1e, T1d, T1y, T1x; cannam@127: T1e = BYTW(&(W[TWVL * 6]), VFNMSI(T1c, T19)); cannam@127: T1d = BYTW(&(W[0]), VFMAI(T1c, T19)); cannam@127: T1y = BYTW(&(W[TWVL * 6]), VFNMSI(T1w, T1t)); cannam@127: T1x = BYTW(&(W[0]), VFMAI(T1w, T1t)); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 2)]), TY, ms, &(x[WS(vs, 3)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 2)]), TX, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1i, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 3)]), T1h, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 2)]), TU, ms, &(x[WS(vs, 4)])); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 2)]), TT, ms, &(x[WS(vs, 1)])); cannam@127: ST(&(x[WS(vs, 4)]), Tg, ms, &(x[WS(vs, 4)])); cannam@127: ST(&(x[WS(vs, 1)]), Tf, ms, &(x[WS(vs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 1)]), TA, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tz, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 3)]), T1e, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1d, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 4)]), T1y, ms, &(x[WS(vs, 4)])); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 4)]), T1x, ms, &(x[WS(vs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, XSIMD_STRING("q1bv_5"), twinstr, &GENUS, {55, 50, 45, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_q1bv_5) (planner *p) { cannam@127: X(kdft_difsq_register) (p, q1bv_5, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twidsq_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 5 -dif -name q1bv_5 -include q1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 100 FP additions, 70 FP multiplications, cannam@127: * (or, 85 additions, 55 multiplications, 15 fused multiply/add), cannam@127: * 44 stack variables, 4 constants, and 50 memory accesses cannam@127: */ cannam@127: #include "q1b.h" cannam@127: cannam@127: static void q1bv_5(R *ri, R *ii, const R *W, stride rs, stride vs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: 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@127: V Tb, T7, Th, Ta, Tc, Td, T1t, T1p, T1z, T1s, T1u, T1v, Tv, Tr, TB; cannam@127: V Tu, Tw, Tx, TP, TL, TV, TO, TQ, TR, T19, T15, T1f, T18, T1a, T1b; cannam@127: { cannam@127: V T6, T9, T3, T8; cannam@127: Tb = LD(&(x[0]), ms, &(x[0])); cannam@127: { cannam@127: V T4, T5, T1, T2; cannam@127: T4 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: T5 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T6 = VSUB(T4, T5); cannam@127: T9 = VADD(T4, T5); cannam@127: T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: T3 = VSUB(T1, T2); cannam@127: T8 = VADD(T1, T2); cannam@127: } cannam@127: T7 = VBYI(VFMA(LDK(KP951056516), T3, VMUL(LDK(KP587785252), T6))); cannam@127: Th = VBYI(VFNMS(LDK(KP951056516), T6, VMUL(LDK(KP587785252), T3))); cannam@127: Ta = VMUL(LDK(KP559016994), VSUB(T8, T9)); cannam@127: Tc = VADD(T8, T9); cannam@127: Td = VFNMS(LDK(KP250000000), Tc, Tb); cannam@127: } cannam@127: { cannam@127: V T1o, T1r, T1l, T1q; cannam@127: T1t = LD(&(x[WS(vs, 4)]), ms, &(x[WS(vs, 4)])); cannam@127: { cannam@127: V T1m, T1n, T1j, T1k; cannam@127: T1m = LD(&(x[WS(vs, 4) + WS(rs, 2)]), ms, &(x[WS(vs, 4)])); cannam@127: T1n = LD(&(x[WS(vs, 4) + WS(rs, 3)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: T1o = VSUB(T1m, T1n); cannam@127: T1r = VADD(T1m, T1n); cannam@127: T1j = LD(&(x[WS(vs, 4) + WS(rs, 1)]), ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: T1k = LD(&(x[WS(vs, 4) + WS(rs, 4)]), ms, &(x[WS(vs, 4)])); cannam@127: T1l = VSUB(T1j, T1k); cannam@127: T1q = VADD(T1j, T1k); cannam@127: } cannam@127: T1p = VBYI(VFMA(LDK(KP951056516), T1l, VMUL(LDK(KP587785252), T1o))); cannam@127: T1z = VBYI(VFNMS(LDK(KP951056516), T1o, VMUL(LDK(KP587785252), T1l))); cannam@127: T1s = VMUL(LDK(KP559016994), VSUB(T1q, T1r)); cannam@127: T1u = VADD(T1q, T1r); cannam@127: T1v = VFNMS(LDK(KP250000000), T1u, T1t); cannam@127: } cannam@127: { cannam@127: V Tq, Tt, Tn, Ts; cannam@127: Tv = LD(&(x[WS(vs, 1)]), ms, &(x[WS(vs, 1)])); cannam@127: { cannam@127: V To, Tp, Tl, Tm; cannam@127: To = LD(&(x[WS(vs, 1) + WS(rs, 2)]), ms, &(x[WS(vs, 1)])); cannam@127: Tp = LD(&(x[WS(vs, 1) + WS(rs, 3)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Tq = VSUB(To, Tp); cannam@127: Tt = VADD(To, Tp); cannam@127: Tl = LD(&(x[WS(vs, 1) + WS(rs, 1)]), ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: Tm = LD(&(x[WS(vs, 1) + WS(rs, 4)]), ms, &(x[WS(vs, 1)])); cannam@127: Tn = VSUB(Tl, Tm); cannam@127: Ts = VADD(Tl, Tm); cannam@127: } cannam@127: Tr = VBYI(VFMA(LDK(KP951056516), Tn, VMUL(LDK(KP587785252), Tq))); cannam@127: TB = VBYI(VFNMS(LDK(KP951056516), Tq, VMUL(LDK(KP587785252), Tn))); cannam@127: Tu = VMUL(LDK(KP559016994), VSUB(Ts, Tt)); cannam@127: Tw = VADD(Ts, Tt); cannam@127: Tx = VFNMS(LDK(KP250000000), Tw, Tv); cannam@127: } cannam@127: { cannam@127: V TK, TN, TH, TM; cannam@127: TP = LD(&(x[WS(vs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: { cannam@127: V TI, TJ, TF, TG; cannam@127: TI = LD(&(x[WS(vs, 2) + WS(rs, 2)]), ms, &(x[WS(vs, 2)])); cannam@127: TJ = LD(&(x[WS(vs, 2) + WS(rs, 3)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: TK = VSUB(TI, TJ); cannam@127: TN = VADD(TI, TJ); cannam@127: TF = LD(&(x[WS(vs, 2) + WS(rs, 1)]), ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: TG = LD(&(x[WS(vs, 2) + WS(rs, 4)]), ms, &(x[WS(vs, 2)])); cannam@127: TH = VSUB(TF, TG); cannam@127: TM = VADD(TF, TG); cannam@127: } cannam@127: TL = VBYI(VFMA(LDK(KP951056516), TH, VMUL(LDK(KP587785252), TK))); cannam@127: TV = VBYI(VFNMS(LDK(KP951056516), TK, VMUL(LDK(KP587785252), TH))); cannam@127: TO = VMUL(LDK(KP559016994), VSUB(TM, TN)); cannam@127: TQ = VADD(TM, TN); cannam@127: TR = VFNMS(LDK(KP250000000), TQ, TP); cannam@127: } cannam@127: { cannam@127: V T14, T17, T11, T16; cannam@127: T19 = LD(&(x[WS(vs, 3)]), ms, &(x[WS(vs, 3)])); cannam@127: { cannam@127: V T12, T13, TZ, T10; cannam@127: T12 = LD(&(x[WS(vs, 3) + WS(rs, 2)]), ms, &(x[WS(vs, 3)])); cannam@127: T13 = LD(&(x[WS(vs, 3) + WS(rs, 3)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: T14 = VSUB(T12, T13); cannam@127: T17 = VADD(T12, T13); cannam@127: TZ = LD(&(x[WS(vs, 3) + WS(rs, 1)]), ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: T10 = LD(&(x[WS(vs, 3) + WS(rs, 4)]), ms, &(x[WS(vs, 3)])); cannam@127: T11 = VSUB(TZ, T10); cannam@127: T16 = VADD(TZ, T10); cannam@127: } cannam@127: T15 = VBYI(VFMA(LDK(KP951056516), T11, VMUL(LDK(KP587785252), T14))); cannam@127: T1f = VBYI(VFNMS(LDK(KP951056516), T14, VMUL(LDK(KP587785252), T11))); cannam@127: T18 = VMUL(LDK(KP559016994), VSUB(T16, T17)); cannam@127: T1a = VADD(T16, T17); cannam@127: T1b = VFNMS(LDK(KP250000000), T1a, T19); cannam@127: } cannam@127: ST(&(x[0]), VADD(Tb, Tc), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 4)]), VADD(T1t, T1u), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 2)]), VADD(TP, TQ), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 3)]), VADD(T19, T1a), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tv, Tw), ms, &(x[WS(rs, 1)])); cannam@127: { cannam@127: V Tj, Tk, Ti, T1B, T1C, T1A; cannam@127: Ti = VSUB(Td, Ta); cannam@127: Tj = BYTW(&(W[TWVL * 2]), VADD(Th, Ti)); cannam@127: Tk = BYTW(&(W[TWVL * 4]), VSUB(Ti, Th)); cannam@127: ST(&(x[WS(vs, 2)]), Tj, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 3)]), Tk, ms, &(x[WS(vs, 3)])); cannam@127: T1A = VSUB(T1v, T1s); cannam@127: T1B = BYTW(&(W[TWVL * 2]), VADD(T1z, T1A)); cannam@127: T1C = BYTW(&(W[TWVL * 4]), VSUB(T1A, T1z)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 4)]), T1B, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 4)]), T1C, ms, &(x[WS(vs, 3)])); cannam@127: } cannam@127: { cannam@127: V T1h, T1i, T1g, TD, TE, TC; cannam@127: T1g = VSUB(T1b, T18); cannam@127: T1h = BYTW(&(W[TWVL * 2]), VADD(T1f, T1g)); cannam@127: T1i = BYTW(&(W[TWVL * 4]), VSUB(T1g, T1f)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 3)]), T1h, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 3)]), T1i, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: TC = VSUB(Tx, Tu); cannam@127: TD = BYTW(&(W[TWVL * 2]), VADD(TB, TC)); cannam@127: TE = BYTW(&(W[TWVL * 4]), VSUB(TC, TB)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 1)]), TD, ms, &(x[WS(vs, 2) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 1)]), TE, ms, &(x[WS(vs, 3) + WS(rs, 1)])); cannam@127: } cannam@127: { cannam@127: V TX, TY, TW, TT, TU, TS; cannam@127: TW = VSUB(TR, TO); cannam@127: TX = BYTW(&(W[TWVL * 2]), VADD(TV, TW)); cannam@127: TY = BYTW(&(W[TWVL * 4]), VSUB(TW, TV)); cannam@127: ST(&(x[WS(vs, 2) + WS(rs, 2)]), TX, ms, &(x[WS(vs, 2)])); cannam@127: ST(&(x[WS(vs, 3) + WS(rs, 2)]), TY, ms, &(x[WS(vs, 3)])); cannam@127: TS = VADD(TO, TR); cannam@127: TT = BYTW(&(W[0]), VADD(TL, TS)); cannam@127: TU = BYTW(&(W[TWVL * 6]), VSUB(TS, TL)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 2)]), TT, ms, &(x[WS(vs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 2)]), TU, ms, &(x[WS(vs, 4)])); cannam@127: } cannam@127: { cannam@127: V Tf, Tg, Te, Tz, TA, Ty; cannam@127: Te = VADD(Ta, Td); cannam@127: Tf = BYTW(&(W[0]), VADD(T7, Te)); cannam@127: Tg = BYTW(&(W[TWVL * 6]), VSUB(Te, T7)); cannam@127: ST(&(x[WS(vs, 1)]), Tf, ms, &(x[WS(vs, 1)])); cannam@127: ST(&(x[WS(vs, 4)]), Tg, ms, &(x[WS(vs, 4)])); cannam@127: Ty = VADD(Tu, Tx); cannam@127: Tz = BYTW(&(W[0]), VADD(Tr, Ty)); cannam@127: TA = BYTW(&(W[TWVL * 6]), VSUB(Ty, Tr)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 1)]), Tz, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 1)]), TA, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: } cannam@127: { cannam@127: V T1d, T1e, T1c, T1x, T1y, T1w; cannam@127: T1c = VADD(T18, T1b); cannam@127: T1d = BYTW(&(W[0]), VADD(T15, T1c)); cannam@127: T1e = BYTW(&(W[TWVL * 6]), VSUB(T1c, T15)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 3)]), T1d, ms, &(x[WS(vs, 1) + WS(rs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 3)]), T1e, ms, &(x[WS(vs, 4) + WS(rs, 1)])); cannam@127: T1w = VADD(T1s, T1v); cannam@127: T1x = BYTW(&(W[0]), VADD(T1p, T1w)); cannam@127: T1y = BYTW(&(W[TWVL * 6]), VSUB(T1w, T1p)); cannam@127: ST(&(x[WS(vs, 1) + WS(rs, 4)]), T1x, ms, &(x[WS(vs, 1)])); cannam@127: ST(&(x[WS(vs, 4) + WS(rs, 4)]), T1y, ms, &(x[WS(vs, 4)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, XSIMD_STRING("q1bv_5"), twinstr, &GENUS, {85, 55, 15, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_q1bv_5) (planner *p) { cannam@127: X(kdft_difsq_register) (p, q1bv_5, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */