Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:38:41 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name t2fv_20 -include t2f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 123 FP additions, 88 FP multiplications, Chris@10: * (or, 77 additions, 42 multiplications, 46 fused multiply/add), Chris@10: * 68 stack variables, 4 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "t2f.h" Chris@10: Chris@10: static void t2fv_20(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 38)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 38), MAKE_VOLATILE_STRIDE(20, rs)) { Chris@10: V T4, Tx, T1m, T1K, T1y, Tk, Tf, T16, T10, TT, T1O, T1w, T1L, T1p, T1M; Chris@10: V T1s, TZ, TI, T1x, Tp; Chris@10: { Chris@10: V T1, Tv, T2, Tt; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: Tv = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)])); Chris@10: T2 = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@10: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V T9, T1n, TN, T1v, TS, Te, T1q, T1u, TE, TG, Tm, T1o, TC, Tn, T1r; Chris@10: V TH, To; Chris@10: { Chris@10: V TP, TR, Ta, Tc; Chris@10: { Chris@10: V T5, T7, TJ, TL, T1k, T1l; Chris@10: T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: T7 = LD(&(x[WS(rs, 14)]), ms, &(x[0])); Chris@10: TJ = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)])); Chris@10: TL = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V Tw, T3, Tu, T6, T8, TK, TM, TO, TQ; Chris@10: TO = LD(&(x[WS(rs, 17)]), ms, &(x[WS(rs, 1)])); Chris@10: Tw = BYTWJ(&(W[TWVL * 28]), Tv); Chris@10: T3 = BYTWJ(&(W[TWVL * 18]), T2); Chris@10: Tu = BYTWJ(&(W[TWVL * 8]), Tt); Chris@10: T6 = BYTWJ(&(W[TWVL * 6]), T5); Chris@10: T8 = BYTWJ(&(W[TWVL * 26]), T7); Chris@10: TK = BYTWJ(&(W[TWVL * 24]), TJ); Chris@10: TM = BYTWJ(&(W[TWVL * 4]), TL); Chris@10: TP = BYTWJ(&(W[TWVL * 32]), TO); Chris@10: TQ = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: T4 = VSUB(T1, T3); Chris@10: T1k = VADD(T1, T3); Chris@10: Tx = VSUB(Tu, Tw); Chris@10: T1l = VADD(Tu, Tw); Chris@10: T9 = VSUB(T6, T8); Chris@10: T1n = VADD(T6, T8); Chris@10: TN = VSUB(TK, TM); Chris@10: T1v = VADD(TK, TM); Chris@10: TR = BYTWJ(&(W[TWVL * 12]), TQ); Chris@10: } Chris@10: Ta = LD(&(x[WS(rs, 16)]), ms, &(x[0])); Chris@10: T1m = VSUB(T1k, T1l); Chris@10: T1K = VADD(T1k, T1l); Chris@10: Tc = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: } Chris@10: { Chris@10: V Tb, TA, Td, Th, Tj, Tz, Tg, Ti, Ty; Chris@10: Tg = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: Ti = LD(&(x[WS(rs, 18)]), ms, &(x[0])); Chris@10: Ty = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: TS = VSUB(TP, TR); Chris@10: T1y = VADD(TP, TR); Chris@10: Tb = BYTWJ(&(W[TWVL * 30]), Ta); Chris@10: TA = LD(&(x[WS(rs, 19)]), ms, &(x[WS(rs, 1)])); Chris@10: Td = BYTWJ(&(W[TWVL * 10]), Tc); Chris@10: Th = BYTWJ(&(W[TWVL * 14]), Tg); Chris@10: Tj = BYTWJ(&(W[TWVL * 34]), Ti); Chris@10: Tz = BYTWJ(&(W[TWVL * 16]), Ty); Chris@10: { Chris@10: V TD, TF, TB, Tl; Chris@10: TD = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: TF = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@10: Tl = LD(&(x[WS(rs, 12)]), ms, &(x[0])); Chris@10: TB = BYTWJ(&(W[TWVL * 36]), TA); Chris@10: Te = VSUB(Tb, Td); Chris@10: T1q = VADD(Tb, Td); Chris@10: Tk = VSUB(Th, Tj); Chris@10: T1u = VADD(Th, Tj); Chris@10: TE = BYTWJ(&(W[0]), TD); Chris@10: TG = BYTWJ(&(W[TWVL * 20]), TF); Chris@10: Tm = BYTWJ(&(W[TWVL * 22]), Tl); Chris@10: T1o = VADD(Tz, TB); Chris@10: TC = VSUB(Tz, TB); Chris@10: Tn = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: Tf = VADD(T9, Te); Chris@10: T16 = VSUB(T9, Te); Chris@10: T10 = VSUB(TS, TN); Chris@10: TT = VADD(TN, TS); Chris@10: T1r = VADD(TE, TG); Chris@10: TH = VSUB(TE, TG); Chris@10: T1O = VADD(T1u, T1v); Chris@10: T1w = VSUB(T1u, T1v); Chris@10: To = BYTWJ(&(W[TWVL * 2]), Tn); Chris@10: T1L = VADD(T1n, T1o); Chris@10: T1p = VSUB(T1n, T1o); Chris@10: T1M = VADD(T1q, T1r); Chris@10: T1s = VSUB(T1q, T1r); Chris@10: TZ = VSUB(TH, TC); Chris@10: TI = VADD(TC, TH); Chris@10: T1x = VADD(Tm, To); Chris@10: Tp = VSUB(Tm, To); Chris@10: } Chris@10: } Chris@10: { Chris@10: V T1V, T1N, T14, T1d, T11, T1G, T1t, T1z, T1P, Tq, T17, T13, TV, TU; Chris@10: T1V = VSUB(T1L, T1M); Chris@10: T1N = VADD(T1L, T1M); Chris@10: T14 = VSUB(TT, TI); Chris@10: TU = VADD(TI, TT); Chris@10: T1d = VFNMS(LDK(KP618033988), TZ, T10); Chris@10: T11 = VFMA(LDK(KP618033988), T10, TZ); Chris@10: T1G = VSUB(T1p, T1s); Chris@10: T1t = VADD(T1p, T1s); Chris@10: T1z = VSUB(T1x, T1y); Chris@10: T1P = VADD(T1x, T1y); Chris@10: Tq = VADD(Tk, Tp); Chris@10: T17 = VSUB(Tk, Tp); Chris@10: T13 = VFNMS(LDK(KP250000000), TU, Tx); Chris@10: TV = VADD(Tx, TU); Chris@10: { Chris@10: V T1J, T1H, T1D, T1Z, T1X, T1T, T1h, T1j, T1b, T19, T1C, T1S, T1c, TY, T1F; Chris@10: V T1A; Chris@10: T1F = VSUB(T1w, T1z); Chris@10: T1A = VADD(T1w, T1z); Chris@10: { Chris@10: V T1W, T1Q, TX, Tr; Chris@10: T1W = VSUB(T1O, T1P); Chris@10: T1Q = VADD(T1O, T1P); Chris@10: TX = VSUB(Tf, Tq); Chris@10: Tr = VADD(Tf, Tq); Chris@10: { Chris@10: V T1g, T18, T1f, T15; Chris@10: T1g = VFNMS(LDK(KP618033988), T16, T17); Chris@10: T18 = VFMA(LDK(KP618033988), T17, T16); Chris@10: T1f = VFMA(LDK(KP559016994), T14, T13); Chris@10: T15 = VFNMS(LDK(KP559016994), T14, T13); Chris@10: T1J = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1F, T1G)); Chris@10: T1H = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1G, T1F)); Chris@10: { Chris@10: V T1B, T1R, TW, Ts; Chris@10: T1B = VADD(T1t, T1A); Chris@10: T1D = VSUB(T1t, T1A); Chris@10: T1Z = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1V, T1W)); Chris@10: T1X = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1W, T1V)); Chris@10: T1R = VADD(T1N, T1Q); Chris@10: T1T = VSUB(T1N, T1Q); Chris@10: TW = VFNMS(LDK(KP250000000), Tr, T4); Chris@10: Ts = VADD(T4, Tr); Chris@10: T1h = VFNMS(LDK(KP951056516), T1g, T1f); Chris@10: T1j = VFMA(LDK(KP951056516), T1g, T1f); Chris@10: T1b = VFNMS(LDK(KP951056516), T18, T15); Chris@10: T19 = VFMA(LDK(KP951056516), T18, T15); Chris@10: ST(&(x[WS(rs, 10)]), VADD(T1m, T1B), ms, &(x[0])); Chris@10: T1C = VFNMS(LDK(KP250000000), T1B, T1m); Chris@10: ST(&(x[0]), VADD(T1K, T1R), ms, &(x[0])); Chris@10: T1S = VFNMS(LDK(KP250000000), T1R, T1K); Chris@10: T1c = VFNMS(LDK(KP559016994), TX, TW); Chris@10: TY = VFMA(LDK(KP559016994), TX, TW); Chris@10: ST(&(x[WS(rs, 15)]), VFMAI(TV, Ts), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 5)]), VFNMSI(TV, Ts), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: V T1E, T1I, T1U, T1Y; Chris@10: T1E = VFNMS(LDK(KP559016994), T1D, T1C); Chris@10: T1I = VFMA(LDK(KP559016994), T1D, T1C); Chris@10: T1U = VFMA(LDK(KP559016994), T1T, T1S); Chris@10: T1Y = VFNMS(LDK(KP559016994), T1T, T1S); Chris@10: { Chris@10: V T1e, T1i, T1a, T12; Chris@10: T1e = VFNMS(LDK(KP951056516), T1d, T1c); Chris@10: T1i = VFMA(LDK(KP951056516), T1d, T1c); Chris@10: T1a = VFNMS(LDK(KP951056516), T11, TY); Chris@10: T12 = VFMA(LDK(KP951056516), T11, TY); Chris@10: ST(&(x[WS(rs, 18)]), VFNMSI(T1H, T1E), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 2)]), VFMAI(T1H, T1E), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 14)]), VFMAI(T1J, T1I), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 6)]), VFNMSI(T1J, T1I), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 16)]), VFNMSI(T1X, T1U), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 4)]), VFMAI(T1X, T1U), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 12)]), VFMAI(T1Z, T1Y), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 8)]), VFNMSI(T1Z, T1Y), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 3)]), VFMAI(T1h, T1e), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 17)]), VFNMSI(T1h, T1e), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VFMAI(T1j, T1i), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 13)]), VFNMSI(T1j, T1i), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 11)]), VFMAI(T1b, T1a), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 9)]), VFNMSI(T1b, T1a), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 19)]), VFMAI(T19, T12), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 1)]), VFNMSI(T19, T12), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: VTW(0, 9), Chris@10: VTW(0, 10), Chris@10: VTW(0, 11), Chris@10: VTW(0, 12), Chris@10: VTW(0, 13), Chris@10: VTW(0, 14), Chris@10: VTW(0, 15), Chris@10: VTW(0, 16), Chris@10: VTW(0, 17), Chris@10: VTW(0, 18), Chris@10: VTW(0, 19), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, XSIMD_STRING("t2fv_20"), twinstr, &GENUS, {77, 42, 46, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t2fv_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2fv_20, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name t2fv_20 -include t2f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 123 FP additions, 62 FP multiplications, Chris@10: * (or, 111 additions, 50 multiplications, 12 fused multiply/add), Chris@10: * 54 stack variables, 4 constants, and 40 memory accesses Chris@10: */ Chris@10: #include "t2f.h" Chris@10: Chris@10: static void t2fv_20(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 38)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 38), MAKE_VOLATILE_STRIDE(20, rs)) { Chris@10: V T4, Tx, T1B, T1U, TZ, T16, T17, T10, Tf, Tq, Tr, T1N, T1O, T1S, T1t; Chris@10: V T1w, T1C, TI, TT, TU, T1K, T1L, T1R, T1m, T1p, T1D, Ts, TV; Chris@10: { Chris@10: V T1, Tw, T3, Tu, Tv, T2, Tt, T1z, T1A; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: Tv = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)])); Chris@10: Tw = BYTWJ(&(W[TWVL * 28]), Tv); Chris@10: T2 = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@10: T3 = BYTWJ(&(W[TWVL * 18]), T2); Chris@10: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tu = BYTWJ(&(W[TWVL * 8]), Tt); Chris@10: T4 = VSUB(T1, T3); Chris@10: Tx = VSUB(Tu, Tw); Chris@10: T1z = VADD(T1, T3); Chris@10: T1A = VADD(Tu, Tw); Chris@10: T1B = VSUB(T1z, T1A); Chris@10: T1U = VADD(T1z, T1A); Chris@10: } Chris@10: { Chris@10: V T9, T1r, TN, T1l, TS, T1o, Te, T1u, Tk, T1k, TC, T1s, TH, T1v, Tp; Chris@10: V T1n; Chris@10: { Chris@10: V T6, T8, T5, T7; Chris@10: T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: T6 = BYTWJ(&(W[TWVL * 6]), T5); Chris@10: T7 = LD(&(x[WS(rs, 14)]), ms, &(x[0])); Chris@10: T8 = BYTWJ(&(W[TWVL * 26]), T7); Chris@10: T9 = VSUB(T6, T8); Chris@10: T1r = VADD(T6, T8); Chris@10: } Chris@10: { Chris@10: V TK, TM, TJ, TL; Chris@10: TJ = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)])); Chris@10: TK = BYTWJ(&(W[TWVL * 24]), TJ); Chris@10: TL = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: TM = BYTWJ(&(W[TWVL * 4]), TL); Chris@10: TN = VSUB(TK, TM); Chris@10: T1l = VADD(TK, TM); Chris@10: } Chris@10: { Chris@10: V TP, TR, TO, TQ; Chris@10: TO = LD(&(x[WS(rs, 17)]), ms, &(x[WS(rs, 1)])); Chris@10: TP = BYTWJ(&(W[TWVL * 32]), TO); Chris@10: TQ = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: TR = BYTWJ(&(W[TWVL * 12]), TQ); Chris@10: TS = VSUB(TP, TR); Chris@10: T1o = VADD(TP, TR); Chris@10: } Chris@10: { Chris@10: V Tb, Td, Ta, Tc; Chris@10: Ta = LD(&(x[WS(rs, 16)]), ms, &(x[0])); Chris@10: Tb = BYTWJ(&(W[TWVL * 30]), Ta); Chris@10: Tc = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: Td = BYTWJ(&(W[TWVL * 10]), Tc); Chris@10: Te = VSUB(Tb, Td); Chris@10: T1u = VADD(Tb, Td); Chris@10: } Chris@10: { Chris@10: V Th, Tj, Tg, Ti; Chris@10: Tg = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: Th = BYTWJ(&(W[TWVL * 14]), Tg); Chris@10: Ti = LD(&(x[WS(rs, 18)]), ms, &(x[0])); Chris@10: Tj = BYTWJ(&(W[TWVL * 34]), Ti); Chris@10: Tk = VSUB(Th, Tj); Chris@10: T1k = VADD(Th, Tj); Chris@10: } Chris@10: { Chris@10: V Tz, TB, Ty, TA; Chris@10: Ty = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: Tz = BYTWJ(&(W[TWVL * 16]), Ty); Chris@10: TA = LD(&(x[WS(rs, 19)]), ms, &(x[WS(rs, 1)])); Chris@10: TB = BYTWJ(&(W[TWVL * 36]), TA); Chris@10: TC = VSUB(Tz, TB); Chris@10: T1s = VADD(Tz, TB); Chris@10: } Chris@10: { Chris@10: V TE, TG, TD, TF; Chris@10: TD = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: TE = BYTWJ(&(W[0]), TD); Chris@10: TF = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@10: TG = BYTWJ(&(W[TWVL * 20]), TF); Chris@10: TH = VSUB(TE, TG); Chris@10: T1v = VADD(TE, TG); Chris@10: } Chris@10: { Chris@10: V Tm, To, Tl, Tn; Chris@10: Tl = LD(&(x[WS(rs, 12)]), ms, &(x[0])); Chris@10: Tm = BYTWJ(&(W[TWVL * 22]), Tl); Chris@10: Tn = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: To = BYTWJ(&(W[TWVL * 2]), Tn); Chris@10: Tp = VSUB(Tm, To); Chris@10: T1n = VADD(Tm, To); Chris@10: } Chris@10: TZ = VSUB(TH, TC); Chris@10: T16 = VSUB(T9, Te); Chris@10: T17 = VSUB(Tk, Tp); Chris@10: T10 = VSUB(TS, TN); Chris@10: Tf = VADD(T9, Te); Chris@10: Tq = VADD(Tk, Tp); Chris@10: Tr = VADD(Tf, Tq); Chris@10: T1N = VADD(T1k, T1l); Chris@10: T1O = VADD(T1n, T1o); Chris@10: T1S = VADD(T1N, T1O); Chris@10: T1t = VSUB(T1r, T1s); Chris@10: T1w = VSUB(T1u, T1v); Chris@10: T1C = VADD(T1t, T1w); Chris@10: TI = VADD(TC, TH); Chris@10: TT = VADD(TN, TS); Chris@10: TU = VADD(TI, TT); Chris@10: T1K = VADD(T1r, T1s); Chris@10: T1L = VADD(T1u, T1v); Chris@10: T1R = VADD(T1K, T1L); Chris@10: T1m = VSUB(T1k, T1l); Chris@10: T1p = VSUB(T1n, T1o); Chris@10: T1D = VADD(T1m, T1p); Chris@10: } Chris@10: Ts = VADD(T4, Tr); Chris@10: TV = VBYI(VADD(Tx, TU)); Chris@10: ST(&(x[WS(rs, 5)]), VSUB(Ts, TV), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 15)]), VADD(Ts, TV), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V T1T, T1V, T1W, T1Q, T1Z, T1M, T1P, T1Y, T1X; Chris@10: T1T = VMUL(LDK(KP559016994), VSUB(T1R, T1S)); Chris@10: T1V = VADD(T1R, T1S); Chris@10: T1W = VFNMS(LDK(KP250000000), T1V, T1U); Chris@10: T1M = VSUB(T1K, T1L); Chris@10: T1P = VSUB(T1N, T1O); Chris@10: T1Q = VBYI(VFMA(LDK(KP951056516), T1M, VMUL(LDK(KP587785252), T1P))); Chris@10: T1Z = VBYI(VFNMS(LDK(KP587785252), T1M, VMUL(LDK(KP951056516), T1P))); Chris@10: ST(&(x[0]), VADD(T1U, T1V), ms, &(x[0])); Chris@10: T1Y = VSUB(T1W, T1T); Chris@10: ST(&(x[WS(rs, 8)]), VSUB(T1Y, T1Z), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 12)]), VADD(T1Z, T1Y), ms, &(x[0])); Chris@10: T1X = VADD(T1T, T1W); Chris@10: ST(&(x[WS(rs, 4)]), VADD(T1Q, T1X), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 16)]), VSUB(T1X, T1Q), ms, &(x[0])); Chris@10: } Chris@10: { Chris@10: V T1G, T1E, T1F, T1y, T1J, T1q, T1x, T1I, T1H; Chris@10: T1G = VMUL(LDK(KP559016994), VSUB(T1C, T1D)); Chris@10: T1E = VADD(T1C, T1D); Chris@10: T1F = VFNMS(LDK(KP250000000), T1E, T1B); Chris@10: T1q = VSUB(T1m, T1p); Chris@10: T1x = VSUB(T1t, T1w); Chris@10: T1y = VBYI(VFNMS(LDK(KP587785252), T1x, VMUL(LDK(KP951056516), T1q))); Chris@10: T1J = VBYI(VFMA(LDK(KP951056516), T1x, VMUL(LDK(KP587785252), T1q))); Chris@10: ST(&(x[WS(rs, 10)]), VADD(T1B, T1E), ms, &(x[0])); Chris@10: T1I = VADD(T1G, T1F); Chris@10: ST(&(x[WS(rs, 6)]), VSUB(T1I, T1J), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 14)]), VADD(T1J, T1I), ms, &(x[0])); Chris@10: T1H = VSUB(T1F, T1G); Chris@10: ST(&(x[WS(rs, 2)]), VADD(T1y, T1H), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 18)]), VSUB(T1H, T1y), ms, &(x[0])); Chris@10: } Chris@10: { Chris@10: V T11, T18, T1g, T1d, T15, T1f, TY, T1c; Chris@10: T11 = VFMA(LDK(KP951056516), TZ, VMUL(LDK(KP587785252), T10)); Chris@10: T18 = VFMA(LDK(KP951056516), T16, VMUL(LDK(KP587785252), T17)); Chris@10: T1g = VFNMS(LDK(KP587785252), T16, VMUL(LDK(KP951056516), T17)); Chris@10: T1d = VFNMS(LDK(KP587785252), TZ, VMUL(LDK(KP951056516), T10)); Chris@10: { Chris@10: V T13, T14, TW, TX; Chris@10: T13 = VFMS(LDK(KP250000000), TU, Tx); Chris@10: T14 = VMUL(LDK(KP559016994), VSUB(TT, TI)); Chris@10: T15 = VADD(T13, T14); Chris@10: T1f = VSUB(T14, T13); Chris@10: TW = VMUL(LDK(KP559016994), VSUB(Tf, Tq)); Chris@10: TX = VFNMS(LDK(KP250000000), Tr, T4); Chris@10: TY = VADD(TW, TX); Chris@10: T1c = VSUB(TX, TW); Chris@10: } Chris@10: { Chris@10: V T12, T19, T1i, T1j; Chris@10: T12 = VADD(TY, T11); Chris@10: T19 = VBYI(VSUB(T15, T18)); Chris@10: ST(&(x[WS(rs, 19)]), VSUB(T12, T19), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 1)]), VADD(T12, T19), ms, &(x[WS(rs, 1)])); Chris@10: T1i = VADD(T1c, T1d); Chris@10: T1j = VBYI(VADD(T1g, T1f)); Chris@10: ST(&(x[WS(rs, 13)]), VSUB(T1i, T1j), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VADD(T1i, T1j), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V T1a, T1b, T1e, T1h; Chris@10: T1a = VSUB(TY, T11); Chris@10: T1b = VBYI(VADD(T18, T15)); Chris@10: ST(&(x[WS(rs, 11)]), VSUB(T1a, T1b), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 9)]), VADD(T1a, T1b), ms, &(x[WS(rs, 1)])); Chris@10: T1e = VSUB(T1c, T1d); Chris@10: T1h = VBYI(VSUB(T1f, T1g)); Chris@10: ST(&(x[WS(rs, 17)]), VSUB(T1e, T1h), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 3)]), VADD(T1e, T1h), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: VTW(0, 9), Chris@10: VTW(0, 10), Chris@10: VTW(0, 11), Chris@10: VTW(0, 12), Chris@10: VTW(0, 13), Chris@10: VTW(0, 14), Chris@10: VTW(0, 15), Chris@10: VTW(0, 16), Chris@10: VTW(0, 17), Chris@10: VTW(0, 18), Chris@10: VTW(0, 19), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 20, XSIMD_STRING("t2fv_20"), twinstr, &GENUS, {111, 50, 12, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t2fv_20) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2fv_20, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */