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 07:37:28 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_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name n2fv_20 -with-ostride 2 -include n2f.h -store-multiple 2 */ Chris@10: Chris@10: /* Chris@10: * This function contains 104 FP additions, 50 FP multiplications, Chris@10: * (or, 58 additions, 4 multiplications, 46 fused multiply/add), Chris@10: * 79 stack variables, 4 constants, and 50 memory accesses Chris@10: */ Chris@10: #include "n2f.h" Chris@10: Chris@10: static void n2fv_20(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ri; Chris@10: xo = ro; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(40, is), MAKE_VOLATILE_STRIDE(40, os)) { Chris@10: V T1H, T1I, TU, TI, TP, TX, T1M, T1N, T1O, T1P, T1R, T1S, TM, TW, TT; Chris@10: V TF; Chris@10: { Chris@10: V T3, Tm, T1r, T13, Ta, TN, TH, TA, TG, Tt, Th, TO, T1u, T1C, T1n; Chris@10: V T1a, T1m, T1h, T1x, T1D, TE, Ti; Chris@10: { Chris@10: V T1, T2, Tk, Tl; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: Tk = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tl = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); Chris@10: { Chris@10: V T14, T6, T1c, Tw, Tn, T1f, Tz, T17, T9, To, Tq, T1b, Td, Tr, Te; Chris@10: V Tf, T15, Tp; Chris@10: { Chris@10: V Tx, Ty, T7, T8, Tb, Tc; Chris@10: { Chris@10: V T4, T5, Tu, Tv, T11, T12; Chris@10: T4 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: Tu = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tv = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tx = LD(&(xi[WS(is, 17)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = VSUB(T1, T2); Chris@10: T11 = VADD(T1, T2); Chris@10: Tm = VSUB(Tk, Tl); Chris@10: T12 = VADD(Tk, Tl); Chris@10: T14 = VADD(T4, T5); Chris@10: T6 = VSUB(T4, T5); Chris@10: T1c = VADD(Tu, Tv); Chris@10: Tw = VSUB(Tu, Tv); Chris@10: Ty = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 16)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T1r = VADD(T11, T12); Chris@10: T13 = VSUB(T11, T12); Chris@10: } Chris@10: Tb = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 18)]), ivs, &(xi[0])); Chris@10: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: T1f = VADD(Tx, Ty); Chris@10: Tz = VSUB(Tx, Ty); Chris@10: T17 = VADD(T7, T8); Chris@10: T9 = VSUB(T7, T8); Chris@10: To = LD(&(xi[WS(is, 19)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tq = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T1b = VADD(Tb, Tc); Chris@10: Td = VSUB(Tb, Tc); Chris@10: Tr = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Te = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: Tf = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: } Chris@10: Ta = VADD(T6, T9); Chris@10: TN = VSUB(T6, T9); Chris@10: T15 = VADD(Tn, To); Chris@10: Tp = VSUB(Tn, To); Chris@10: TH = VSUB(Tz, Tw); Chris@10: TA = VADD(Tw, Tz); Chris@10: { Chris@10: V T1d, T1v, T18, Ts, T1e, Tg, T16, T1s; Chris@10: T1d = VSUB(T1b, T1c); Chris@10: T1v = VADD(T1b, T1c); Chris@10: T18 = VADD(Tq, Tr); Chris@10: Ts = VSUB(Tq, Tr); Chris@10: T1e = VADD(Te, Tf); Chris@10: Tg = VSUB(Te, Tf); Chris@10: T16 = VSUB(T14, T15); Chris@10: T1s = VADD(T14, T15); Chris@10: { Chris@10: V T1t, T19, T1w, T1g; Chris@10: T1t = VADD(T17, T18); Chris@10: T19 = VSUB(T17, T18); Chris@10: TG = VSUB(Ts, Tp); Chris@10: Tt = VADD(Tp, Ts); Chris@10: T1w = VADD(T1e, T1f); Chris@10: T1g = VSUB(T1e, T1f); Chris@10: Th = VADD(Td, Tg); Chris@10: TO = VSUB(Td, Tg); Chris@10: T1u = VADD(T1s, T1t); Chris@10: T1C = VSUB(T1s, T1t); Chris@10: T1n = VSUB(T16, T19); Chris@10: T1a = VADD(T16, T19); Chris@10: T1m = VSUB(T1d, T1g); Chris@10: T1h = VADD(T1d, T1g); Chris@10: T1x = VADD(T1v, T1w); Chris@10: T1D = VSUB(T1v, T1w); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: TE = VSUB(Ta, Th); Chris@10: Ti = VADD(Ta, Th); Chris@10: { Chris@10: V TL, T1k, T1A, Tj, TD, T1E, T1G, TK, TC, T1j, T1z, T1i, T1y, TB; Chris@10: TL = VSUB(TA, Tt); Chris@10: TB = VADD(Tt, TA); Chris@10: T1i = VADD(T1a, T1h); Chris@10: T1k = VSUB(T1a, T1h); Chris@10: T1y = VADD(T1u, T1x); Chris@10: T1A = VSUB(T1u, T1x); Chris@10: Tj = VADD(T3, Ti); Chris@10: TD = VFNMS(LDK(KP250000000), Ti, T3); Chris@10: T1E = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1D, T1C)); Chris@10: T1G = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1C, T1D)); Chris@10: TK = VFNMS(LDK(KP250000000), TB, Tm); Chris@10: TC = VADD(Tm, TB); Chris@10: T1j = VFNMS(LDK(KP250000000), T1i, T13); Chris@10: T1H = VADD(T1r, T1y); Chris@10: STM2(&(xo[0]), T1H, ovs, &(xo[0])); Chris@10: T1z = VFNMS(LDK(KP250000000), T1y, T1r); Chris@10: T1I = VADD(T13, T1i); Chris@10: STM2(&(xo[20]), T1I, ovs, &(xo[0])); Chris@10: { Chris@10: V T1J, T1K, T1p, T1l, T1o, T1q, T1F, T1B, T1L, T1Q; Chris@10: TU = VFNMS(LDK(KP618033988), TG, TH); Chris@10: TI = VFMA(LDK(KP618033988), TH, TG); Chris@10: TP = VFMA(LDK(KP618033988), TO, TN); Chris@10: TX = VFNMS(LDK(KP618033988), TN, TO); Chris@10: T1J = VFMAI(TC, Tj); Chris@10: STM2(&(xo[30]), T1J, ovs, &(xo[2])); Chris@10: T1K = VFNMSI(TC, Tj); Chris@10: STM2(&(xo[10]), T1K, ovs, &(xo[2])); Chris@10: T1p = VFMA(LDK(KP559016994), T1k, T1j); Chris@10: T1l = VFNMS(LDK(KP559016994), T1k, T1j); Chris@10: T1o = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1n, T1m)); Chris@10: T1q = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1m, T1n)); Chris@10: T1F = VFNMS(LDK(KP559016994), T1A, T1z); Chris@10: T1B = VFMA(LDK(KP559016994), T1A, T1z); Chris@10: T1L = VFMAI(T1q, T1p); Chris@10: STM2(&(xo[28]), T1L, ovs, &(xo[0])); Chris@10: STN2(&(xo[28]), T1L, T1J, ovs); Chris@10: T1M = VFNMSI(T1q, T1p); Chris@10: STM2(&(xo[12]), T1M, ovs, &(xo[0])); Chris@10: T1N = VFNMSI(T1o, T1l); Chris@10: STM2(&(xo[36]), T1N, ovs, &(xo[0])); Chris@10: T1O = VFMAI(T1o, T1l); Chris@10: STM2(&(xo[4]), T1O, ovs, &(xo[0])); Chris@10: T1P = VFNMSI(T1E, T1B); Chris@10: STM2(&(xo[32]), T1P, ovs, &(xo[0])); Chris@10: T1Q = VFMAI(T1E, T1B); Chris@10: STM2(&(xo[8]), T1Q, ovs, &(xo[0])); Chris@10: STN2(&(xo[8]), T1Q, T1K, ovs); Chris@10: T1R = VFMAI(T1G, T1F); Chris@10: STM2(&(xo[24]), T1R, ovs, &(xo[0])); Chris@10: T1S = VFNMSI(T1G, T1F); Chris@10: STM2(&(xo[16]), T1S, ovs, &(xo[0])); Chris@10: TM = VFNMS(LDK(KP559016994), TL, TK); Chris@10: TW = VFMA(LDK(KP559016994), TL, TK); Chris@10: TT = VFNMS(LDK(KP559016994), TE, TD); Chris@10: TF = VFMA(LDK(KP559016994), TE, TD); Chris@10: } Chris@10: } Chris@10: } Chris@10: { Chris@10: V T10, TY, TQ, TS, TJ, TR, TZ, TV; Chris@10: T10 = VFMA(LDK(KP951056516), TX, TW); Chris@10: TY = VFNMS(LDK(KP951056516), TX, TW); Chris@10: TQ = VFMA(LDK(KP951056516), TP, TM); Chris@10: TS = VFNMS(LDK(KP951056516), TP, TM); Chris@10: TJ = VFMA(LDK(KP951056516), TI, TF); Chris@10: TR = VFNMS(LDK(KP951056516), TI, TF); Chris@10: TZ = VFMA(LDK(KP951056516), TU, TT); Chris@10: TV = VFNMS(LDK(KP951056516), TU, TT); Chris@10: { Chris@10: V T1T, T1U, T1V, T1W; Chris@10: T1T = VFMAI(TS, TR); Chris@10: STM2(&(xo[22]), T1T, ovs, &(xo[2])); Chris@10: STN2(&(xo[20]), T1I, T1T, ovs); Chris@10: T1U = VFNMSI(TS, TR); Chris@10: STM2(&(xo[18]), T1U, ovs, &(xo[2])); Chris@10: STN2(&(xo[16]), T1S, T1U, ovs); Chris@10: T1V = VFMAI(TQ, TJ); Chris@10: STM2(&(xo[38]), T1V, ovs, &(xo[2])); Chris@10: STN2(&(xo[36]), T1N, T1V, ovs); Chris@10: T1W = VFNMSI(TQ, TJ); Chris@10: STM2(&(xo[2]), T1W, ovs, &(xo[2])); Chris@10: STN2(&(xo[0]), T1H, T1W, ovs); Chris@10: { Chris@10: V T1X, T1Y, T1Z, T20; Chris@10: T1X = VFMAI(TY, TV); Chris@10: STM2(&(xo[6]), T1X, ovs, &(xo[2])); Chris@10: STN2(&(xo[4]), T1O, T1X, ovs); Chris@10: T1Y = VFNMSI(TY, TV); Chris@10: STM2(&(xo[34]), T1Y, ovs, &(xo[2])); Chris@10: STN2(&(xo[32]), T1P, T1Y, ovs); Chris@10: T1Z = VFMAI(T10, TZ); Chris@10: STM2(&(xo[14]), T1Z, ovs, &(xo[2])); Chris@10: STN2(&(xo[12]), T1M, T1Z, ovs); Chris@10: T20 = VFNMSI(T10, TZ); Chris@10: STM2(&(xo[26]), T20, ovs, &(xo[2])); Chris@10: STN2(&(xo[24]), T1R, T20, ovs); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 20, XSIMD_STRING("n2fv_20"), {58, 4, 46, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2fv_20) (planner *p) { Chris@10: X(kdft_register) (p, n2fv_20, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 20 -name n2fv_20 -with-ostride 2 -include n2f.h -store-multiple 2 */ Chris@10: Chris@10: /* Chris@10: * This function contains 104 FP additions, 24 FP multiplications, Chris@10: * (or, 92 additions, 12 multiplications, 12 fused multiply/add), Chris@10: * 57 stack variables, 4 constants, and 50 memory accesses Chris@10: */ Chris@10: #include "n2f.h" Chris@10: Chris@10: static void n2fv_20(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) 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 i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ri; Chris@10: xo = ro; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(40, is), MAKE_VOLATILE_STRIDE(40, os)) { Chris@10: V T3, T1B, Tm, T1i, TG, TN, TO, TH, T13, T16, T1k, T1u, T1v, T1z, T1r; Chris@10: V T1s, T1y, T1a, T1d, T1j, Ti, TD, TB, TL; Chris@10: { Chris@10: V T1, T2, T1g, Tk, Tl, T1h; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: T1g = VADD(T1, T2); Chris@10: Tk = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tl = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); Chris@10: T1h = VADD(Tk, Tl); Chris@10: T3 = VSUB(T1, T2); Chris@10: T1B = VADD(T1g, T1h); Chris@10: Tm = VSUB(Tk, Tl); Chris@10: T1i = VSUB(T1g, T1h); Chris@10: } Chris@10: { Chris@10: V T6, T18, Tw, T12, Tz, T15, T9, T1b, Td, T11, Tp, T19, Ts, T1c, Tg; Chris@10: V T14; Chris@10: { Chris@10: V T4, T5, Tu, Tv; Chris@10: T4 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: T6 = VSUB(T4, T5); Chris@10: T18 = VADD(T4, T5); Chris@10: Tu = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tv = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tw = VSUB(Tu, Tv); Chris@10: T12 = VADD(Tu, Tv); Chris@10: } Chris@10: { Chris@10: V Tx, Ty, T7, T8; Chris@10: Tx = LD(&(xi[WS(is, 17)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ty = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tz = VSUB(Tx, Ty); Chris@10: T15 = VADD(Tx, Ty); Chris@10: T7 = LD(&(xi[WS(is, 16)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T9 = VSUB(T7, T8); Chris@10: T1b = VADD(T7, T8); Chris@10: } Chris@10: { Chris@10: V Tb, Tc, Tn, To; Chris@10: Tb = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 18)]), ivs, &(xi[0])); Chris@10: Td = VSUB(Tb, Tc); Chris@10: T11 = VADD(Tb, Tc); Chris@10: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: To = LD(&(xi[WS(is, 19)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tp = VSUB(Tn, To); Chris@10: T19 = VADD(Tn, To); Chris@10: } Chris@10: { Chris@10: V Tq, Tr, Te, Tf; Chris@10: Tq = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tr = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ts = VSUB(Tq, Tr); Chris@10: T1c = VADD(Tq, Tr); Chris@10: Te = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: Tf = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Tg = VSUB(Te, Tf); Chris@10: T14 = VADD(Te, Tf); Chris@10: } Chris@10: TG = VSUB(Ts, Tp); Chris@10: TN = VSUB(T6, T9); Chris@10: TO = VSUB(Td, Tg); Chris@10: TH = VSUB(Tz, Tw); Chris@10: T13 = VSUB(T11, T12); Chris@10: T16 = VSUB(T14, T15); Chris@10: T1k = VADD(T13, T16); Chris@10: T1u = VADD(T11, T12); Chris@10: T1v = VADD(T14, T15); Chris@10: T1z = VADD(T1u, T1v); Chris@10: T1r = VADD(T18, T19); Chris@10: T1s = VADD(T1b, T1c); Chris@10: T1y = VADD(T1r, T1s); Chris@10: T1a = VSUB(T18, T19); Chris@10: T1d = VSUB(T1b, T1c); Chris@10: T1j = VADD(T1a, T1d); Chris@10: { Chris@10: V Ta, Th, Tt, TA; Chris@10: Ta = VADD(T6, T9); Chris@10: Th = VADD(Td, Tg); Chris@10: Ti = VADD(Ta, Th); Chris@10: TD = VMUL(LDK(KP559016994), VSUB(Ta, Th)); Chris@10: Tt = VADD(Tp, Ts); Chris@10: TA = VADD(Tw, Tz); Chris@10: TB = VADD(Tt, TA); Chris@10: TL = VMUL(LDK(KP559016994), VSUB(TA, Tt)); Chris@10: } Chris@10: } Chris@10: { Chris@10: V T1I, T1J, T1K, T1L, T1N, T1H, Tj, TC; Chris@10: Tj = VADD(T3, Ti); Chris@10: TC = VBYI(VADD(Tm, TB)); Chris@10: T1H = VSUB(Tj, TC); Chris@10: STM2(&(xo[10]), T1H, ovs, &(xo[2])); Chris@10: T1I = VADD(Tj, TC); Chris@10: STM2(&(xo[30]), T1I, ovs, &(xo[2])); Chris@10: { Chris@10: V T1A, T1C, T1D, T1x, T1G, T1t, T1w, T1F, T1E, T1M; Chris@10: T1A = VMUL(LDK(KP559016994), VSUB(T1y, T1z)); Chris@10: T1C = VADD(T1y, T1z); Chris@10: T1D = VFNMS(LDK(KP250000000), T1C, T1B); Chris@10: T1t = VSUB(T1r, T1s); Chris@10: T1w = VSUB(T1u, T1v); Chris@10: T1x = VBYI(VFMA(LDK(KP951056516), T1t, VMUL(LDK(KP587785252), T1w))); Chris@10: T1G = VBYI(VFNMS(LDK(KP587785252), T1t, VMUL(LDK(KP951056516), T1w))); Chris@10: T1J = VADD(T1B, T1C); Chris@10: STM2(&(xo[0]), T1J, ovs, &(xo[0])); Chris@10: T1F = VSUB(T1D, T1A); Chris@10: T1K = VSUB(T1F, T1G); Chris@10: STM2(&(xo[16]), T1K, ovs, &(xo[0])); Chris@10: T1L = VADD(T1G, T1F); Chris@10: STM2(&(xo[24]), T1L, ovs, &(xo[0])); Chris@10: T1E = VADD(T1A, T1D); Chris@10: T1M = VADD(T1x, T1E); Chris@10: STM2(&(xo[8]), T1M, ovs, &(xo[0])); Chris@10: STN2(&(xo[8]), T1M, T1H, ovs); Chris@10: T1N = VSUB(T1E, T1x); Chris@10: STM2(&(xo[32]), T1N, ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V T1O, T1P, T1R, T1S; Chris@10: { Chris@10: V T1n, T1l, T1m, T1f, T1q, T17, T1e, T1p, T1Q, T1o; Chris@10: T1n = VMUL(LDK(KP559016994), VSUB(T1j, T1k)); Chris@10: T1l = VADD(T1j, T1k); Chris@10: T1m = VFNMS(LDK(KP250000000), T1l, T1i); Chris@10: T17 = VSUB(T13, T16); Chris@10: T1e = VSUB(T1a, T1d); Chris@10: T1f = VBYI(VFNMS(LDK(KP587785252), T1e, VMUL(LDK(KP951056516), T17))); Chris@10: T1q = VBYI(VFMA(LDK(KP951056516), T1e, VMUL(LDK(KP587785252), T17))); Chris@10: T1O = VADD(T1i, T1l); Chris@10: STM2(&(xo[20]), T1O, ovs, &(xo[0])); Chris@10: T1p = VADD(T1n, T1m); Chris@10: T1P = VSUB(T1p, T1q); Chris@10: STM2(&(xo[12]), T1P, ovs, &(xo[0])); Chris@10: T1Q = VADD(T1q, T1p); Chris@10: STM2(&(xo[28]), T1Q, ovs, &(xo[0])); Chris@10: STN2(&(xo[28]), T1Q, T1I, ovs); Chris@10: T1o = VSUB(T1m, T1n); Chris@10: T1R = VADD(T1f, T1o); Chris@10: STM2(&(xo[4]), T1R, ovs, &(xo[0])); Chris@10: T1S = VSUB(T1o, T1f); Chris@10: STM2(&(xo[36]), T1S, ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V TI, TP, TX, TU, TM, TW, TF, TT, TK, TE; Chris@10: TI = VFMA(LDK(KP951056516), TG, VMUL(LDK(KP587785252), TH)); Chris@10: TP = VFMA(LDK(KP951056516), TN, VMUL(LDK(KP587785252), TO)); Chris@10: TX = VFNMS(LDK(KP587785252), TN, VMUL(LDK(KP951056516), TO)); Chris@10: TU = VFNMS(LDK(KP587785252), TG, VMUL(LDK(KP951056516), TH)); Chris@10: TK = VFMS(LDK(KP250000000), TB, Tm); Chris@10: TM = VADD(TK, TL); Chris@10: TW = VSUB(TL, TK); Chris@10: TE = VFNMS(LDK(KP250000000), Ti, T3); Chris@10: TF = VADD(TD, TE); Chris@10: TT = VSUB(TE, TD); Chris@10: { Chris@10: V TJ, TQ, T1T, T1U; Chris@10: TJ = VADD(TF, TI); Chris@10: TQ = VBYI(VSUB(TM, TP)); Chris@10: T1T = VSUB(TJ, TQ); Chris@10: STM2(&(xo[38]), T1T, ovs, &(xo[2])); Chris@10: STN2(&(xo[36]), T1S, T1T, ovs); Chris@10: T1U = VADD(TJ, TQ); Chris@10: STM2(&(xo[2]), T1U, ovs, &(xo[2])); Chris@10: STN2(&(xo[0]), T1J, T1U, ovs); Chris@10: } Chris@10: { Chris@10: V TZ, T10, T1V, T1W; Chris@10: TZ = VADD(TT, TU); Chris@10: T10 = VBYI(VADD(TX, TW)); Chris@10: T1V = VSUB(TZ, T10); Chris@10: STM2(&(xo[26]), T1V, ovs, &(xo[2])); Chris@10: STN2(&(xo[24]), T1L, T1V, ovs); Chris@10: T1W = VADD(TZ, T10); Chris@10: STM2(&(xo[14]), T1W, ovs, &(xo[2])); Chris@10: STN2(&(xo[12]), T1P, T1W, ovs); Chris@10: } Chris@10: { Chris@10: V TR, TS, T1X, T1Y; Chris@10: TR = VSUB(TF, TI); Chris@10: TS = VBYI(VADD(TP, TM)); Chris@10: T1X = VSUB(TR, TS); Chris@10: STM2(&(xo[22]), T1X, ovs, &(xo[2])); Chris@10: STN2(&(xo[20]), T1O, T1X, ovs); Chris@10: T1Y = VADD(TR, TS); Chris@10: STM2(&(xo[18]), T1Y, ovs, &(xo[2])); Chris@10: STN2(&(xo[16]), T1K, T1Y, ovs); Chris@10: } Chris@10: { Chris@10: V TV, TY, T1Z, T20; Chris@10: TV = VSUB(TT, TU); Chris@10: TY = VBYI(VSUB(TW, TX)); Chris@10: T1Z = VSUB(TV, TY); Chris@10: STM2(&(xo[34]), T1Z, ovs, &(xo[2])); Chris@10: STN2(&(xo[32]), T1N, T1Z, ovs); Chris@10: T20 = VADD(TV, TY); Chris@10: STM2(&(xo[6]), T20, ovs, &(xo[2])); Chris@10: STN2(&(xo[4]), T1R, T20, ovs); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 20, XSIMD_STRING("n2fv_20"), {92, 12, 12, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2fv_20) (planner *p) { Chris@10: X(kdft_register) (p, n2fv_20, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */