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 08:06:05 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_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name t2bv_20 -include dft/simd/t2b.h -sign 1 */ Chris@82: Chris@82: /* Chris@82: * This function contains 123 FP additions, 88 FP multiplications, Chris@82: * (or, 77 additions, 42 multiplications, 46 fused multiply/add), Chris@82: * 54 stack variables, 4 constants, and 40 memory accesses Chris@82: */ Chris@82: #include "dft/simd/t2b.h" Chris@82: Chris@82: static void t2bv_20(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@82: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@82: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@82: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT m; Chris@82: R *x; Chris@82: x = ii; Chris@82: 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@82: V T4, TX, T1m, T1K, TF, T14, T15, TQ, Tf, Tq, Tr, T1O, T1P, T1Q, T1w; Chris@82: V T1z, T1A, TY, TZ, T10, T1L, T1M, T1N, T1p, T1s, T1t, T1i, T1j; Chris@82: { Chris@82: V T1, TW, T3, TU, TV, T2, TT, T1k, T1l; Chris@82: T1 = LD(&(x[0]), ms, &(x[0])); Chris@82: TV = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)])); Chris@82: TW = BYTW(&(W[TWVL * 28]), TV); Chris@82: T2 = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@82: T3 = BYTW(&(W[TWVL * 18]), T2); Chris@82: TT = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@82: TU = BYTW(&(W[TWVL * 8]), TT); Chris@82: T4 = VSUB(T1, T3); Chris@82: TX = VSUB(TU, TW); Chris@82: T1k = VADD(T1, T3); Chris@82: T1l = VADD(TU, TW); Chris@82: T1m = VSUB(T1k, T1l); Chris@82: T1K = VADD(T1k, T1l); Chris@82: } Chris@82: { Chris@82: V T9, T1n, TK, T1v, TP, T1y, Te, T1q, Tk, T1u, Tz, T1o, TE, T1r, Tp; Chris@82: V T1x; Chris@82: { Chris@82: V T6, T8, T5, T7; Chris@82: T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@82: T6 = BYTW(&(W[TWVL * 6]), T5); Chris@82: T7 = LD(&(x[WS(rs, 14)]), ms, &(x[0])); Chris@82: T8 = BYTW(&(W[TWVL * 26]), T7); Chris@82: T9 = VSUB(T6, T8); Chris@82: T1n = VADD(T6, T8); Chris@82: } Chris@82: { Chris@82: V TH, TJ, TG, TI; Chris@82: TG = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)])); Chris@82: TH = BYTW(&(W[TWVL * 24]), TG); Chris@82: TI = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@82: TJ = BYTW(&(W[TWVL * 4]), TI); Chris@82: TK = VSUB(TH, TJ); Chris@82: T1v = VADD(TH, TJ); Chris@82: } Chris@82: { Chris@82: V TM, TO, TL, TN; Chris@82: TL = LD(&(x[WS(rs, 17)]), ms, &(x[WS(rs, 1)])); Chris@82: TM = BYTW(&(W[TWVL * 32]), TL); Chris@82: TN = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@82: TO = BYTW(&(W[TWVL * 12]), TN); Chris@82: TP = VSUB(TM, TO); Chris@82: T1y = VADD(TM, TO); Chris@82: } Chris@82: { Chris@82: V Tb, Td, Ta, Tc; Chris@82: Ta = LD(&(x[WS(rs, 16)]), ms, &(x[0])); Chris@82: Tb = BYTW(&(W[TWVL * 30]), Ta); Chris@82: Tc = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@82: Td = BYTW(&(W[TWVL * 10]), Tc); Chris@82: Te = VSUB(Tb, Td); Chris@82: T1q = VADD(Tb, Td); Chris@82: } Chris@82: { Chris@82: V Th, Tj, Tg, Ti; Chris@82: Tg = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@82: Th = BYTW(&(W[TWVL * 14]), Tg); Chris@82: Ti = LD(&(x[WS(rs, 18)]), ms, &(x[0])); Chris@82: Tj = BYTW(&(W[TWVL * 34]), Ti); Chris@82: Tk = VSUB(Th, Tj); Chris@82: T1u = VADD(Th, Tj); Chris@82: } Chris@82: { Chris@82: V Tw, Ty, Tv, Tx; Chris@82: Tv = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@82: Tw = BYTW(&(W[TWVL * 16]), Tv); Chris@82: Tx = LD(&(x[WS(rs, 19)]), ms, &(x[WS(rs, 1)])); Chris@82: Ty = BYTW(&(W[TWVL * 36]), Tx); Chris@82: Tz = VSUB(Tw, Ty); Chris@82: T1o = VADD(Tw, Ty); Chris@82: } Chris@82: { Chris@82: V TB, TD, TA, TC; Chris@82: TA = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@82: TB = BYTW(&(W[0]), TA); Chris@82: TC = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@82: TD = BYTW(&(W[TWVL * 20]), TC); Chris@82: TE = VSUB(TB, TD); Chris@82: T1r = VADD(TB, TD); Chris@82: } Chris@82: { Chris@82: V Tm, To, Tl, Tn; Chris@82: Tl = LD(&(x[WS(rs, 12)]), ms, &(x[0])); Chris@82: Tm = BYTW(&(W[TWVL * 22]), Tl); Chris@82: Tn = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@82: To = BYTW(&(W[TWVL * 2]), Tn); Chris@82: Tp = VSUB(Tm, To); Chris@82: T1x = VADD(Tm, To); Chris@82: } Chris@82: TF = VSUB(Tz, TE); Chris@82: T14 = VSUB(T9, Te); Chris@82: T15 = VSUB(Tk, Tp); Chris@82: TQ = VSUB(TK, TP); Chris@82: Tf = VADD(T9, Te); Chris@82: Tq = VADD(Tk, Tp); Chris@82: Tr = VADD(Tf, Tq); Chris@82: T1O = VADD(T1u, T1v); Chris@82: T1P = VADD(T1x, T1y); Chris@82: T1Q = VADD(T1O, T1P); Chris@82: T1w = VSUB(T1u, T1v); Chris@82: T1z = VSUB(T1x, T1y); Chris@82: T1A = VADD(T1w, T1z); Chris@82: TY = VADD(Tz, TE); Chris@82: TZ = VADD(TK, TP); Chris@82: T10 = VADD(TY, TZ); Chris@82: T1L = VADD(T1n, T1o); Chris@82: T1M = VADD(T1q, T1r); Chris@82: T1N = VADD(T1L, T1M); Chris@82: T1p = VSUB(T1n, T1o); Chris@82: T1s = VSUB(T1q, T1r); Chris@82: T1t = VADD(T1p, T1s); Chris@82: } Chris@82: T1i = VADD(T4, Tr); Chris@82: T1j = VADD(TX, T10); Chris@82: ST(&(x[WS(rs, 15)]), VFNMSI(T1j, T1i), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 5)]), VFMAI(T1j, T1i), ms, &(x[WS(rs, 1)])); Chris@82: { Chris@82: V T1T, T1R, T1S, T1X, T1Z, T1V, T1W, T1Y, T1U; Chris@82: T1T = VSUB(T1N, T1Q); Chris@82: T1R = VADD(T1N, T1Q); Chris@82: T1S = VFNMS(LDK(KP250000000), T1R, T1K); Chris@82: T1V = VSUB(T1L, T1M); Chris@82: T1W = VSUB(T1O, T1P); Chris@82: T1X = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1W, T1V)); Chris@82: T1Z = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1V, T1W)); Chris@82: ST(&(x[0]), VADD(T1K, T1R), ms, &(x[0])); Chris@82: T1Y = VFNMS(LDK(KP559016994), T1T, T1S); Chris@82: ST(&(x[WS(rs, 8)]), VFMAI(T1Z, T1Y), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 12)]), VFNMSI(T1Z, T1Y), ms, &(x[0])); Chris@82: T1U = VFMA(LDK(KP559016994), T1T, T1S); Chris@82: ST(&(x[WS(rs, 4)]), VFNMSI(T1X, T1U), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 16)]), VFMAI(T1X, T1U), ms, &(x[0])); Chris@82: } Chris@82: { Chris@82: V T1D, T1B, T1C, T1H, T1J, T1F, T1G, T1I, T1E; Chris@82: T1D = VSUB(T1t, T1A); Chris@82: T1B = VADD(T1t, T1A); Chris@82: T1C = VFNMS(LDK(KP250000000), T1B, T1m); Chris@82: T1F = VSUB(T1w, T1z); Chris@82: T1G = VSUB(T1p, T1s); Chris@82: T1H = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1G, T1F)); Chris@82: T1J = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1F, T1G)); Chris@82: ST(&(x[WS(rs, 10)]), VADD(T1m, T1B), ms, &(x[0])); Chris@82: T1I = VFMA(LDK(KP559016994), T1D, T1C); Chris@82: ST(&(x[WS(rs, 6)]), VFMAI(T1J, T1I), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 14)]), VFNMSI(T1J, T1I), ms, &(x[0])); Chris@82: T1E = VFNMS(LDK(KP559016994), T1D, T1C); Chris@82: ST(&(x[WS(rs, 2)]), VFNMSI(T1H, T1E), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 18)]), VFMAI(T1H, T1E), ms, &(x[0])); Chris@82: } Chris@82: { Chris@82: V TR, T16, T1e, T1b, T13, T1d, Tu, T1a; Chris@82: TR = VFMA(LDK(KP618033988), TQ, TF); Chris@82: T16 = VFMA(LDK(KP618033988), T15, T14); Chris@82: T1e = VFNMS(LDK(KP618033988), T14, T15); Chris@82: T1b = VFNMS(LDK(KP618033988), TF, TQ); Chris@82: { Chris@82: V T11, T12, Ts, Tt; Chris@82: T11 = VFNMS(LDK(KP250000000), T10, TX); Chris@82: T12 = VSUB(TY, TZ); Chris@82: T13 = VFMA(LDK(KP559016994), T12, T11); Chris@82: T1d = VFNMS(LDK(KP559016994), T12, T11); Chris@82: Ts = VFNMS(LDK(KP250000000), Tr, T4); Chris@82: Tt = VSUB(Tf, Tq); Chris@82: Tu = VFMA(LDK(KP559016994), Tt, Ts); Chris@82: T1a = VFNMS(LDK(KP559016994), Tt, Ts); Chris@82: } Chris@82: { Chris@82: V TS, T17, T1g, T1h; Chris@82: TS = VFNMS(LDK(KP951056516), TR, Tu); Chris@82: T17 = VFMA(LDK(KP951056516), T16, T13); Chris@82: ST(&(x[WS(rs, 19)]), VFNMSI(T17, TS), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 1)]), VFMAI(T17, TS), ms, &(x[WS(rs, 1)])); Chris@82: T1g = VFNMS(LDK(KP951056516), T1b, T1a); Chris@82: T1h = VFMA(LDK(KP951056516), T1e, T1d); Chris@82: ST(&(x[WS(rs, 7)]), VFNMSI(T1h, T1g), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 13)]), VFMAI(T1h, T1g), ms, &(x[WS(rs, 1)])); Chris@82: } Chris@82: { Chris@82: V T18, T19, T1c, T1f; Chris@82: T18 = VFMA(LDK(KP951056516), TR, Tu); Chris@82: T19 = VFNMS(LDK(KP951056516), T16, T13); Chris@82: ST(&(x[WS(rs, 11)]), VFNMSI(T19, T18), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 9)]), VFMAI(T19, T18), ms, &(x[WS(rs, 1)])); Chris@82: T1c = VFMA(LDK(KP951056516), T1b, T1a); Chris@82: T1f = VFNMS(LDK(KP951056516), T1e, T1d); Chris@82: ST(&(x[WS(rs, 3)]), VFNMSI(T1f, T1c), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 17)]), VFMAI(T1f, T1c), ms, &(x[WS(rs, 1)])); 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, 2), Chris@82: VTW(0, 3), Chris@82: VTW(0, 4), Chris@82: VTW(0, 5), Chris@82: VTW(0, 6), Chris@82: VTW(0, 7), Chris@82: VTW(0, 8), Chris@82: VTW(0, 9), Chris@82: VTW(0, 10), Chris@82: VTW(0, 11), Chris@82: VTW(0, 12), Chris@82: VTW(0, 13), Chris@82: VTW(0, 14), Chris@82: VTW(0, 15), Chris@82: VTW(0, 16), Chris@82: VTW(0, 17), Chris@82: VTW(0, 18), Chris@82: VTW(0, 19), Chris@82: {TW_NEXT, VL, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 20, XSIMD_STRING("t2bv_20"), twinstr, &GENUS, {77, 42, 46, 0}, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_t2bv_20) (planner *p) { Chris@82: X(kdft_dit_register) (p, t2bv_20, &desc); Chris@82: } Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name t2bv_20 -include dft/simd/t2b.h -sign 1 */ Chris@82: Chris@82: /* Chris@82: * This function contains 123 FP additions, 62 FP multiplications, Chris@82: * (or, 111 additions, 50 multiplications, 12 fused multiply/add), Chris@82: * 54 stack variables, 4 constants, and 40 memory accesses Chris@82: */ Chris@82: #include "dft/simd/t2b.h" Chris@82: Chris@82: static void t2bv_20(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@82: { Chris@82: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@82: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@82: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@82: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@82: { Chris@82: INT m; Chris@82: R *x; Chris@82: x = ii; Chris@82: 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@82: V T4, T10, T1B, T1R, TF, T14, T15, TQ, Tf, Tq, Tr, T1N, T1O, T1P, T1t; Chris@82: V T1w, T1D, TT, TU, T11, T1K, T1L, T1M, T1m, T1p, T1C, T1i, T1j; Chris@82: { Chris@82: V T1, TZ, T3, TX, TY, T2, TW, T1z, T1A; Chris@82: T1 = LD(&(x[0]), ms, &(x[0])); Chris@82: TY = LD(&(x[WS(rs, 15)]), ms, &(x[WS(rs, 1)])); Chris@82: TZ = BYTW(&(W[TWVL * 28]), TY); Chris@82: T2 = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@82: T3 = BYTW(&(W[TWVL * 18]), T2); Chris@82: TW = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@82: TX = BYTW(&(W[TWVL * 8]), TW); Chris@82: T4 = VSUB(T1, T3); Chris@82: T10 = VSUB(TX, TZ); Chris@82: T1z = VADD(T1, T3); Chris@82: T1A = VADD(TX, TZ); Chris@82: T1B = VSUB(T1z, T1A); Chris@82: T1R = VADD(T1z, T1A); Chris@82: } Chris@82: { Chris@82: V T9, T1k, TK, T1s, TP, T1v, Te, T1n, Tk, T1r, Tz, T1l, TE, T1o, Tp; Chris@82: V T1u; Chris@82: { Chris@82: V T6, T8, T5, T7; Chris@82: T5 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@82: T6 = BYTW(&(W[TWVL * 6]), T5); Chris@82: T7 = LD(&(x[WS(rs, 14)]), ms, &(x[0])); Chris@82: T8 = BYTW(&(W[TWVL * 26]), T7); Chris@82: T9 = VSUB(T6, T8); Chris@82: T1k = VADD(T6, T8); Chris@82: } Chris@82: { Chris@82: V TH, TJ, TG, TI; Chris@82: TG = LD(&(x[WS(rs, 13)]), ms, &(x[WS(rs, 1)])); Chris@82: TH = BYTW(&(W[TWVL * 24]), TG); Chris@82: TI = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@82: TJ = BYTW(&(W[TWVL * 4]), TI); Chris@82: TK = VSUB(TH, TJ); Chris@82: T1s = VADD(TH, TJ); Chris@82: } Chris@82: { Chris@82: V TM, TO, TL, TN; Chris@82: TL = LD(&(x[WS(rs, 17)]), ms, &(x[WS(rs, 1)])); Chris@82: TM = BYTW(&(W[TWVL * 32]), TL); Chris@82: TN = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@82: TO = BYTW(&(W[TWVL * 12]), TN); Chris@82: TP = VSUB(TM, TO); Chris@82: T1v = VADD(TM, TO); Chris@82: } Chris@82: { Chris@82: V Tb, Td, Ta, Tc; Chris@82: Ta = LD(&(x[WS(rs, 16)]), ms, &(x[0])); Chris@82: Tb = BYTW(&(W[TWVL * 30]), Ta); Chris@82: Tc = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@82: Td = BYTW(&(W[TWVL * 10]), Tc); Chris@82: Te = VSUB(Tb, Td); Chris@82: T1n = VADD(Tb, Td); Chris@82: } Chris@82: { Chris@82: V Th, Tj, Tg, Ti; Chris@82: Tg = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@82: Th = BYTW(&(W[TWVL * 14]), Tg); Chris@82: Ti = LD(&(x[WS(rs, 18)]), ms, &(x[0])); Chris@82: Tj = BYTW(&(W[TWVL * 34]), Ti); Chris@82: Tk = VSUB(Th, Tj); Chris@82: T1r = VADD(Th, Tj); Chris@82: } Chris@82: { Chris@82: V Tw, Ty, Tv, Tx; Chris@82: Tv = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@82: Tw = BYTW(&(W[TWVL * 16]), Tv); Chris@82: Tx = LD(&(x[WS(rs, 19)]), ms, &(x[WS(rs, 1)])); Chris@82: Ty = BYTW(&(W[TWVL * 36]), Tx); Chris@82: Tz = VSUB(Tw, Ty); Chris@82: T1l = VADD(Tw, Ty); Chris@82: } Chris@82: { Chris@82: V TB, TD, TA, TC; Chris@82: TA = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@82: TB = BYTW(&(W[0]), TA); Chris@82: TC = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@82: TD = BYTW(&(W[TWVL * 20]), TC); Chris@82: TE = VSUB(TB, TD); Chris@82: T1o = VADD(TB, TD); Chris@82: } Chris@82: { Chris@82: V Tm, To, Tl, Tn; Chris@82: Tl = LD(&(x[WS(rs, 12)]), ms, &(x[0])); Chris@82: Tm = BYTW(&(W[TWVL * 22]), Tl); Chris@82: Tn = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@82: To = BYTW(&(W[TWVL * 2]), Tn); Chris@82: Tp = VSUB(Tm, To); Chris@82: T1u = VADD(Tm, To); Chris@82: } Chris@82: TF = VSUB(Tz, TE); Chris@82: T14 = VSUB(T9, Te); Chris@82: T15 = VSUB(Tk, Tp); Chris@82: TQ = VSUB(TK, TP); Chris@82: Tf = VADD(T9, Te); Chris@82: Tq = VADD(Tk, Tp); Chris@82: Tr = VADD(Tf, Tq); Chris@82: T1N = VADD(T1r, T1s); Chris@82: T1O = VADD(T1u, T1v); Chris@82: T1P = VADD(T1N, T1O); Chris@82: T1t = VSUB(T1r, T1s); Chris@82: T1w = VSUB(T1u, T1v); Chris@82: T1D = VADD(T1t, T1w); Chris@82: TT = VADD(Tz, TE); Chris@82: TU = VADD(TK, TP); Chris@82: T11 = VADD(TT, TU); Chris@82: T1K = VADD(T1k, T1l); Chris@82: T1L = VADD(T1n, T1o); Chris@82: T1M = VADD(T1K, T1L); Chris@82: T1m = VSUB(T1k, T1l); Chris@82: T1p = VSUB(T1n, T1o); Chris@82: T1C = VADD(T1m, T1p); Chris@82: } Chris@82: T1i = VADD(T4, Tr); Chris@82: T1j = VBYI(VADD(T10, T11)); Chris@82: ST(&(x[WS(rs, 15)]), VSUB(T1i, T1j), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 5)]), VADD(T1i, T1j), ms, &(x[WS(rs, 1)])); Chris@82: { Chris@82: V T1Q, T1S, T1T, T1X, T1Z, T1V, T1W, T1Y, T1U; Chris@82: T1Q = VMUL(LDK(KP559016994), VSUB(T1M, T1P)); Chris@82: T1S = VADD(T1M, T1P); Chris@82: T1T = VFNMS(LDK(KP250000000), T1S, T1R); Chris@82: T1V = VSUB(T1K, T1L); Chris@82: T1W = VSUB(T1N, T1O); Chris@82: T1X = VBYI(VFMA(LDK(KP951056516), T1V, VMUL(LDK(KP587785252), T1W))); Chris@82: T1Z = VBYI(VFNMS(LDK(KP951056516), T1W, VMUL(LDK(KP587785252), T1V))); Chris@82: ST(&(x[0]), VADD(T1R, T1S), ms, &(x[0])); Chris@82: T1Y = VSUB(T1T, T1Q); Chris@82: ST(&(x[WS(rs, 8)]), VSUB(T1Y, T1Z), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 12)]), VADD(T1Z, T1Y), ms, &(x[0])); Chris@82: T1U = VADD(T1Q, T1T); Chris@82: ST(&(x[WS(rs, 4)]), VSUB(T1U, T1X), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 16)]), VADD(T1X, T1U), ms, &(x[0])); Chris@82: } Chris@82: { Chris@82: V T1G, T1E, T1F, T1y, T1I, T1q, T1x, T1J, T1H; Chris@82: T1G = VMUL(LDK(KP559016994), VSUB(T1C, T1D)); Chris@82: T1E = VADD(T1C, T1D); Chris@82: T1F = VFNMS(LDK(KP250000000), T1E, T1B); Chris@82: T1q = VSUB(T1m, T1p); Chris@82: T1x = VSUB(T1t, T1w); Chris@82: T1y = VBYI(VFNMS(LDK(KP951056516), T1x, VMUL(LDK(KP587785252), T1q))); Chris@82: T1I = VBYI(VFMA(LDK(KP951056516), T1q, VMUL(LDK(KP587785252), T1x))); Chris@82: ST(&(x[WS(rs, 10)]), VADD(T1B, T1E), ms, &(x[0])); Chris@82: T1J = VADD(T1G, T1F); Chris@82: ST(&(x[WS(rs, 6)]), VADD(T1I, T1J), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 14)]), VSUB(T1J, T1I), ms, &(x[0])); Chris@82: T1H = VSUB(T1F, T1G); Chris@82: ST(&(x[WS(rs, 2)]), VADD(T1y, T1H), ms, &(x[0])); Chris@82: ST(&(x[WS(rs, 18)]), VSUB(T1H, T1y), ms, &(x[0])); Chris@82: } Chris@82: { Chris@82: V TR, T16, T1d, T1b, T13, T1e, Tu, T1a; Chris@82: TR = VFNMS(LDK(KP951056516), TQ, VMUL(LDK(KP587785252), TF)); Chris@82: T16 = VFNMS(LDK(KP951056516), T15, VMUL(LDK(KP587785252), T14)); Chris@82: T1d = VFMA(LDK(KP951056516), T14, VMUL(LDK(KP587785252), T15)); Chris@82: T1b = VFMA(LDK(KP951056516), TF, VMUL(LDK(KP587785252), TQ)); Chris@82: { Chris@82: V TV, T12, Ts, Tt; Chris@82: TV = VMUL(LDK(KP559016994), VSUB(TT, TU)); Chris@82: T12 = VFNMS(LDK(KP250000000), T11, T10); Chris@82: T13 = VSUB(TV, T12); Chris@82: T1e = VADD(TV, T12); Chris@82: Ts = VFNMS(LDK(KP250000000), Tr, T4); Chris@82: Tt = VMUL(LDK(KP559016994), VSUB(Tf, Tq)); Chris@82: Tu = VSUB(Ts, Tt); Chris@82: T1a = VADD(Tt, Ts); Chris@82: } Chris@82: { Chris@82: V TS, T17, T1g, T1h; Chris@82: TS = VSUB(Tu, TR); Chris@82: T17 = VBYI(VSUB(T13, T16)); Chris@82: ST(&(x[WS(rs, 17)]), VSUB(TS, T17), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 3)]), VADD(TS, T17), ms, &(x[WS(rs, 1)])); Chris@82: T1g = VADD(T1a, T1b); Chris@82: T1h = VBYI(VSUB(T1e, T1d)); Chris@82: ST(&(x[WS(rs, 11)]), VSUB(T1g, T1h), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 9)]), VADD(T1g, T1h), ms, &(x[WS(rs, 1)])); Chris@82: } Chris@82: { Chris@82: V T18, T19, T1c, T1f; Chris@82: T18 = VADD(Tu, TR); Chris@82: T19 = VBYI(VADD(T16, T13)); Chris@82: ST(&(x[WS(rs, 13)]), VSUB(T18, T19), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 7)]), VADD(T18, T19), ms, &(x[WS(rs, 1)])); Chris@82: T1c = VSUB(T1a, T1b); Chris@82: T1f = VBYI(VADD(T1d, T1e)); Chris@82: ST(&(x[WS(rs, 19)]), VSUB(T1c, T1f), ms, &(x[WS(rs, 1)])); Chris@82: ST(&(x[WS(rs, 1)]), VADD(T1c, T1f), ms, &(x[WS(rs, 1)])); 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, 2), Chris@82: VTW(0, 3), Chris@82: VTW(0, 4), Chris@82: VTW(0, 5), Chris@82: VTW(0, 6), Chris@82: VTW(0, 7), Chris@82: VTW(0, 8), Chris@82: VTW(0, 9), Chris@82: VTW(0, 10), Chris@82: VTW(0, 11), Chris@82: VTW(0, 12), Chris@82: VTW(0, 13), Chris@82: VTW(0, 14), Chris@82: VTW(0, 15), Chris@82: VTW(0, 16), Chris@82: VTW(0, 17), Chris@82: VTW(0, 18), Chris@82: VTW(0, 19), Chris@82: {TW_NEXT, VL, 0} Chris@82: }; Chris@82: Chris@82: static const ct_desc desc = { 20, XSIMD_STRING("t2bv_20"), twinstr, &GENUS, {111, 50, 12, 0}, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_t2bv_20) (planner *p) { Chris@82: X(kdft_dit_register) (p, t2bv_20, &desc); Chris@82: } Chris@82: #endif