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 07:36:52 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 15 -name n1fv_15 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 78 FP additions, 49 FP multiplications, Chris@10: * (or, 36 additions, 7 multiplications, 42 fused multiply/add), Chris@10: * 78 stack variables, 8 constants, and 30 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_15(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(KP823639103, +0.823639103546331925877420039278190003029660514); Chris@10: DVK(KP910592997, +0.910592997310029334643087372129977886038870291); Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); 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(30, is), MAKE_VOLATILE_STRIDE(30, os)) { Chris@10: V Tb, TX, TM, TQ, Th, TB, T5, Ti, Ta, TC, TN, Te, TG, Tq, Tj; Chris@10: V T1, T2, T3; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: { Chris@10: V T6, T7, T8, Tm, Tn, To; Chris@10: T6 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tm = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tn = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: To = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: { Chris@10: V T4, Tc, T9, Td, Tp; Chris@10: Tb = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: T4 = VADD(T2, T3); Chris@10: TX = VSUB(T3, T2); Chris@10: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: TM = VSUB(T8, T7); Chris@10: T9 = VADD(T7, T8); Chris@10: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tp = VADD(Tn, To); Chris@10: TQ = VSUB(To, Tn); Chris@10: Th = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: TB = VFNMS(LDK(KP500000000), T4, T1); Chris@10: T5 = VADD(T1, T4); Chris@10: Ti = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ta = VADD(T6, T9); Chris@10: TC = VFNMS(LDK(KP500000000), T9, T6); Chris@10: TN = VSUB(Td, Tc); Chris@10: Te = VADD(Tc, Td); Chris@10: TG = VFNMS(LDK(KP500000000), Tp, Tm); Chris@10: Tq = VADD(Tm, Tp); Chris@10: Tj = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: } Chris@10: } Chris@10: { Chris@10: V TY, TO, Tf, TD, TP, Tk; Chris@10: TY = VADD(TM, TN); Chris@10: TO = VSUB(TM, TN); Chris@10: Tf = VADD(Tb, Te); Chris@10: TD = VFNMS(LDK(KP500000000), Te, Tb); Chris@10: TP = VSUB(Tj, Ti); Chris@10: Tk = VADD(Ti, Tj); Chris@10: { Chris@10: V Tx, Tg, TE, TU, TZ, TR, Tl, TF; Chris@10: Tx = VSUB(Ta, Tf); Chris@10: Tg = VADD(Ta, Tf); Chris@10: TE = VADD(TC, TD); Chris@10: TU = VSUB(TC, TD); Chris@10: TZ = VADD(TP, TQ); Chris@10: TR = VSUB(TP, TQ); Chris@10: Tl = VADD(Th, Tk); Chris@10: TF = VFNMS(LDK(KP500000000), Tk, Th); Chris@10: { Chris@10: V T12, T10, T18, TS, Tw, Tr, TH, TV, T11, T1g; Chris@10: T12 = VSUB(TY, TZ); Chris@10: T10 = VADD(TY, TZ); Chris@10: T18 = VFNMS(LDK(KP618033988), TO, TR); Chris@10: TS = VFMA(LDK(KP618033988), TR, TO); Chris@10: Tw = VSUB(Tl, Tq); Chris@10: Tr = VADD(Tl, Tq); Chris@10: TH = VADD(TF, TG); Chris@10: TV = VSUB(TF, TG); Chris@10: T11 = VFNMS(LDK(KP250000000), T10, TX); Chris@10: T1g = VMUL(LDK(KP866025403), VADD(TX, T10)); Chris@10: { Chris@10: V TA, Ty, Tu, TK, TI, T1a, TW, T1b, T13, Tt, Ts, TJ, T1f; Chris@10: TA = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tw, Tx)); Chris@10: Ty = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tx, Tw)); Chris@10: Ts = VADD(Tg, Tr); Chris@10: Tu = VSUB(Tg, Tr); Chris@10: TK = VSUB(TE, TH); Chris@10: TI = VADD(TE, TH); Chris@10: T1a = VFNMS(LDK(KP618033988), TU, TV); Chris@10: TW = VFMA(LDK(KP618033988), TV, TU); Chris@10: T1b = VFNMS(LDK(KP559016994), T12, T11); Chris@10: T13 = VFMA(LDK(KP559016994), T12, T11); Chris@10: ST(&(xo[0]), VADD(T5, Ts), ovs, &(xo[0])); Chris@10: Tt = VFNMS(LDK(KP250000000), Ts, T5); Chris@10: TJ = VFNMS(LDK(KP250000000), TI, TB); Chris@10: T1f = VADD(TB, TI); Chris@10: { Chris@10: V T1c, T1e, T16, T14, Tv, Tz, T17, TL; Chris@10: T1c = VMUL(LDK(KP951056516), VFNMS(LDK(KP910592997), T1b, T1a)); Chris@10: T1e = VMUL(LDK(KP951056516), VFMA(LDK(KP910592997), T1b, T1a)); Chris@10: T16 = VMUL(LDK(KP951056516), VFMA(LDK(KP910592997), T13, TW)); Chris@10: T14 = VMUL(LDK(KP951056516), VFNMS(LDK(KP910592997), T13, TW)); Chris@10: Tv = VFNMS(LDK(KP559016994), Tu, Tt); Chris@10: Tz = VFMA(LDK(KP559016994), Tu, Tt); Chris@10: T17 = VFNMS(LDK(KP559016994), TK, TJ); Chris@10: TL = VFMA(LDK(KP559016994), TK, TJ); Chris@10: ST(&(xo[WS(os, 10)]), VFMAI(T1g, T1f), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 5)]), VFNMSI(T1g, T1f), ovs, &(xo[WS(os, 1)])); Chris@10: { Chris@10: V T19, T1d, T15, TT; Chris@10: ST(&(xo[WS(os, 12)]), VFMAI(Ty, Tv), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 3)]), VFNMSI(Ty, Tv), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VFMAI(TA, Tz), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 6)]), VFNMSI(TA, Tz), ovs, &(xo[0])); Chris@10: T19 = VFMA(LDK(KP823639103), T18, T17); Chris@10: T1d = VFNMS(LDK(KP823639103), T18, T17); Chris@10: T15 = VFNMS(LDK(KP823639103), TS, TL); Chris@10: TT = VFMA(LDK(KP823639103), TS, TL); Chris@10: ST(&(xo[WS(os, 2)]), VFMAI(T1c, T19), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 13)]), VFNMSI(T1c, T19), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VFMAI(T1e, T1d), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 8)]), VFNMSI(T1e, T1d), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VFMAI(T16, T15), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 11)]), VFNMSI(T16, T15), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 14)]), VFMAI(T14, TT), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 1)]), VFNMSI(T14, TT), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: } 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 = { 15, XSIMD_STRING("n1fv_15"), {36, 7, 42, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_15) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_15, &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 15 -name n1fv_15 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 78 FP additions, 25 FP multiplications, Chris@10: * (or, 64 additions, 11 multiplications, 14 fused multiply/add), Chris@10: * 55 stack variables, 10 constants, and 30 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_15(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(KP216506350, +0.216506350946109661690930792688234045867850657); Chris@10: DVK(KP509036960, +0.509036960455127183450980863393907648510733164); Chris@10: DVK(KP823639103, +0.823639103546331925877420039278190003029660514); 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: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP484122918, +0.484122918275927110647408174972799951354115213); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); 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(30, is), MAKE_VOLATILE_STRIDE(30, os)) { Chris@10: V T5, T10, TB, TO, TU, TV, TR, Ta, Tf, Tg, Tl, Tq, Tr, TE, TH; Chris@10: V TI, TZ, T11, T1f, T1g; Chris@10: { Chris@10: V T1, T2, T3, T4; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: T4 = VADD(T2, T3); Chris@10: T5 = VADD(T1, T4); Chris@10: T10 = VSUB(T3, T2); Chris@10: TB = VFNMS(LDK(KP500000000), T4, T1); Chris@10: } Chris@10: { Chris@10: V T6, T9, TC, TP, Tm, Tp, TG, TN, Tb, Te, TD, TQ, Th, Tk, TF; Chris@10: V TM, TX, TY; Chris@10: { Chris@10: V T7, T8, Tn, To; Chris@10: T6 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: T9 = VADD(T7, T8); Chris@10: TC = VFNMS(LDK(KP500000000), T9, T6); Chris@10: TP = VSUB(T8, T7); Chris@10: Tm = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tn = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: To = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: Tp = VADD(Tn, To); Chris@10: TG = VFNMS(LDK(KP500000000), Tp, Tm); Chris@10: TN = VSUB(To, Tn); Chris@10: } Chris@10: { Chris@10: V Tc, Td, Ti, Tj; Chris@10: Tb = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Te = VADD(Tc, Td); Chris@10: TD = VFNMS(LDK(KP500000000), Te, Tb); Chris@10: TQ = VSUB(Td, Tc); Chris@10: Th = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Ti = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tj = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tk = VADD(Ti, Tj); Chris@10: TF = VFNMS(LDK(KP500000000), Tk, Th); Chris@10: TM = VSUB(Tj, Ti); Chris@10: } Chris@10: TO = VSUB(TM, TN); Chris@10: TU = VSUB(TF, TG); Chris@10: TV = VSUB(TC, TD); Chris@10: TR = VSUB(TP, TQ); Chris@10: Ta = VADD(T6, T9); Chris@10: Tf = VADD(Tb, Te); Chris@10: Tg = VADD(Ta, Tf); Chris@10: Tl = VADD(Th, Tk); Chris@10: Tq = VADD(Tm, Tp); Chris@10: Tr = VADD(Tl, Tq); Chris@10: TE = VADD(TC, TD); Chris@10: TH = VADD(TF, TG); Chris@10: TI = VADD(TE, TH); Chris@10: TX = VADD(TP, TQ); Chris@10: TY = VADD(TM, TN); Chris@10: TZ = VMUL(LDK(KP484122918), VSUB(TX, TY)); Chris@10: T11 = VADD(TX, TY); Chris@10: } Chris@10: T1f = VADD(TB, TI); Chris@10: T1g = VBYI(VMUL(LDK(KP866025403), VADD(T10, T11))); Chris@10: ST(&(xo[WS(os, 5)]), VSUB(T1f, T1g), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 10)]), VADD(T1f, T1g), ovs, &(xo[0])); Chris@10: { Chris@10: V Tu, Ts, Tt, Ty, TA, Tw, Tx, Tz, Tv; Chris@10: Tu = VMUL(LDK(KP559016994), VSUB(Tg, Tr)); Chris@10: Ts = VADD(Tg, Tr); Chris@10: Tt = VFNMS(LDK(KP250000000), Ts, T5); Chris@10: Tw = VSUB(Tl, Tq); Chris@10: Tx = VSUB(Ta, Tf); Chris@10: Ty = VBYI(VFNMS(LDK(KP587785252), Tx, VMUL(LDK(KP951056516), Tw))); Chris@10: TA = VBYI(VFMA(LDK(KP951056516), Tx, VMUL(LDK(KP587785252), Tw))); Chris@10: ST(&(xo[0]), VADD(T5, Ts), ovs, &(xo[0])); Chris@10: Tz = VADD(Tu, Tt); Chris@10: ST(&(xo[WS(os, 6)]), VSUB(Tz, TA), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 9)]), VADD(TA, Tz), ovs, &(xo[WS(os, 1)])); Chris@10: Tv = VSUB(Tt, Tu); Chris@10: ST(&(xo[WS(os, 3)]), VSUB(Tv, Ty), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 12)]), VADD(Ty, Tv), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V TS, TW, T1b, T18, T13, T1a, TL, T17, T12, TJ, TK; Chris@10: TS = VFNMS(LDK(KP509036960), TR, VMUL(LDK(KP823639103), TO)); Chris@10: TW = VFNMS(LDK(KP587785252), TV, VMUL(LDK(KP951056516), TU)); Chris@10: T1b = VFMA(LDK(KP951056516), TV, VMUL(LDK(KP587785252), TU)); Chris@10: T18 = VFMA(LDK(KP823639103), TR, VMUL(LDK(KP509036960), TO)); Chris@10: T12 = VFNMS(LDK(KP216506350), T11, VMUL(LDK(KP866025403), T10)); Chris@10: T13 = VSUB(TZ, T12); Chris@10: T1a = VADD(TZ, T12); Chris@10: TJ = VFNMS(LDK(KP250000000), TI, TB); Chris@10: TK = VMUL(LDK(KP559016994), VSUB(TE, TH)); Chris@10: TL = VSUB(TJ, TK); Chris@10: T17 = VADD(TK, TJ); Chris@10: { Chris@10: V TT, T14, T1d, T1e; Chris@10: TT = VSUB(TL, TS); Chris@10: T14 = VBYI(VSUB(TW, T13)); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(TT, T14), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 7)]), VADD(TT, T14), ovs, &(xo[WS(os, 1)])); Chris@10: T1d = VSUB(T17, T18); Chris@10: T1e = VBYI(VADD(T1b, T1a)); Chris@10: ST(&(xo[WS(os, 11)]), VSUB(T1d, T1e), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 4)]), VADD(T1d, T1e), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V T15, T16, T19, T1c; Chris@10: T15 = VADD(TL, TS); Chris@10: T16 = VBYI(VADD(TW, T13)); Chris@10: ST(&(xo[WS(os, 13)]), VSUB(T15, T16), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 2)]), VADD(T15, T16), ovs, &(xo[0])); Chris@10: T19 = VADD(T17, T18); Chris@10: T1c = VBYI(VSUB(T1a, T1b)); Chris@10: ST(&(xo[WS(os, 14)]), VSUB(T19, T1c), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 1)]), VADD(T19, T1c), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 15, XSIMD_STRING("n1fv_15"), {64, 11, 14, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_15) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_15, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */