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 07:37:04 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 -sign 1 -n 15 -name n1bv_15 -include n1b.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 "n1b.h" Chris@10: Chris@10: static void n1bv_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 = ii; Chris@10: xo = io; 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, TH, Tw, TA, Th, T11, T5, Ti, T12, Ta, Tx, Te, Tq, T16, 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: TH = VSUB(T2, T3); Chris@10: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Tw = VSUB(T7, T8); 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: TA = VSUB(Tn, To); Chris@10: Th = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T11 = VADD(T1, T4); Chris@10: T5 = VFNMS(LDK(KP500000000), T4, T1); Chris@10: Ti = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: T12 = VADD(T6, T9); Chris@10: Ta = VFNMS(LDK(KP500000000), T9, T6); Chris@10: Tx = VSUB(Tc, Td); Chris@10: Te = VADD(Tc, Td); Chris@10: Tq = VFNMS(LDK(KP500000000), Tp, Tm); Chris@10: T16 = 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 TI, Ty, T13, Tf, Tz, Tk; Chris@10: TI = VADD(Tw, Tx); Chris@10: Ty = VSUB(Tw, Tx); Chris@10: T13 = VADD(Tb, Te); Chris@10: Tf = VFNMS(LDK(KP500000000), Te, Tb); Chris@10: Tz = VSUB(Ti, Tj); Chris@10: Tk = VADD(Ti, Tj); Chris@10: { Chris@10: V T1d, T14, Tg, TE, TJ, TB, T15, Tl; Chris@10: T1d = VSUB(T12, T13); Chris@10: T14 = VADD(T12, T13); Chris@10: Tg = VADD(Ta, Tf); Chris@10: TE = VSUB(Ta, Tf); Chris@10: TJ = VADD(Tz, TA); Chris@10: TB = VSUB(Tz, TA); Chris@10: T15 = VADD(Th, Tk); Chris@10: Tl = VFNMS(LDK(KP500000000), Tk, Th); Chris@10: { Chris@10: V TM, TK, TS, TC, T1c, T17, Tr, TF, TL, T10; Chris@10: TM = VSUB(TI, TJ); Chris@10: TK = VADD(TI, TJ); Chris@10: TS = VFNMS(LDK(KP618033988), Ty, TB); Chris@10: TC = VFMA(LDK(KP618033988), TB, Ty); Chris@10: T1c = VSUB(T15, T16); Chris@10: T17 = VADD(T15, T16); Chris@10: Tr = VADD(Tl, Tq); Chris@10: TF = VSUB(Tl, Tq); Chris@10: TL = VFNMS(LDK(KP250000000), TK, TH); Chris@10: T10 = VMUL(LDK(KP866025403), VADD(TH, TK)); Chris@10: { Chris@10: V T1g, T1e, T1a, Tu, Ts, TU, TG, TV, TN, T19, T18, Tt, TZ; Chris@10: T1g = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), T1c, T1d)); Chris@10: T1e = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), T1d, T1c)); Chris@10: T18 = VADD(T14, T17); Chris@10: T1a = VSUB(T14, T17); Chris@10: Tu = VSUB(Tg, Tr); Chris@10: Ts = VADD(Tg, Tr); Chris@10: TU = VFNMS(LDK(KP618033988), TE, TF); Chris@10: TG = VFMA(LDK(KP618033988), TF, TE); Chris@10: TV = VFNMS(LDK(KP559016994), TM, TL); Chris@10: TN = VFMA(LDK(KP559016994), TM, TL); Chris@10: ST(&(xo[0]), VADD(T11, T18), ovs, &(xo[0])); Chris@10: T19 = VFNMS(LDK(KP250000000), T18, T11); Chris@10: Tt = VFNMS(LDK(KP250000000), Ts, T5); Chris@10: TZ = VADD(T5, Ts); Chris@10: { Chris@10: V TW, TY, TQ, TO, T1b, T1f, TR, Tv; Chris@10: TW = VMUL(LDK(KP951056516), VFMA(LDK(KP910592997), TV, TU)); Chris@10: TY = VMUL(LDK(KP951056516), VFNMS(LDK(KP910592997), TV, TU)); Chris@10: TQ = VMUL(LDK(KP951056516), VFNMS(LDK(KP910592997), TN, TG)); Chris@10: TO = VMUL(LDK(KP951056516), VFMA(LDK(KP910592997), TN, TG)); Chris@10: T1b = VFNMS(LDK(KP559016994), T1a, T19); Chris@10: T1f = VFMA(LDK(KP559016994), T1a, T19); Chris@10: TR = VFNMS(LDK(KP559016994), Tu, Tt); Chris@10: Tv = VFMA(LDK(KP559016994), Tu, Tt); Chris@10: ST(&(xo[WS(os, 10)]), VFMAI(T10, TZ), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 5)]), VFNMSI(T10, TZ), ovs, &(xo[WS(os, 1)])); Chris@10: { Chris@10: V TT, TX, TP, TD; Chris@10: ST(&(xo[WS(os, 12)]), VFNMSI(T1e, T1b), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 3)]), VFMAI(T1e, T1b), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VFNMSI(T1g, T1f), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 6)]), VFMAI(T1g, T1f), ovs, &(xo[0])); Chris@10: TT = VFNMS(LDK(KP823639103), TS, TR); Chris@10: TX = VFMA(LDK(KP823639103), TS, TR); Chris@10: TP = VFMA(LDK(KP823639103), TC, Tv); Chris@10: TD = VFNMS(LDK(KP823639103), TC, Tv); Chris@10: ST(&(xo[WS(os, 13)]), VFMAI(TW, TT), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 2)]), VFNMSI(TW, TT), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 8)]), VFMAI(TY, TX), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 7)]), VFNMSI(TY, TX), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 11)]), VFMAI(TQ, TP), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 4)]), VFNMSI(TQ, TP), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 14)]), VFNMSI(TO, TD), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 1)]), VFMAI(TO, TD), 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("n1bv_15"), {36, 7, 42, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_15) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_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 -sign 1 -n 15 -name n1bv_15 -include n1b.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 "n1b.h" Chris@10: Chris@10: static void n1bv_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(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); 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 = ii; Chris@10: xo = io; 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 Ti, T11, TH, Ts, TL, TM, Tz, TC, TD, TI, T12, T13, T14, T15, T16; Chris@10: V T17, Tf, Tj, TZ, T10; Chris@10: { Chris@10: V TF, Tg, Th, TG; Chris@10: TF = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Th = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: TG = VADD(Tg, Th); Chris@10: Ti = VSUB(Tg, Th); Chris@10: T11 = VADD(TF, TG); Chris@10: TH = VFNMS(LDK(KP500000000), TG, TF); Chris@10: } Chris@10: { Chris@10: V Tm, Tn, T3, To, Tw, Tx, Td, Ty, Tp, Tq, T6, Tr, Tt, Tu, Ta; Chris@10: V Tv, T7, Te; Chris@10: { Chris@10: V T1, T2, Tb, Tc; Chris@10: Tm = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T1 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tn = VADD(T1, T2); Chris@10: T3 = VSUB(T1, T2); Chris@10: To = VFNMS(LDK(KP500000000), Tn, Tm); Chris@10: Tw = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tb = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: Tx = VADD(Tb, Tc); Chris@10: Td = VSUB(Tb, Tc); Chris@10: Ty = VFNMS(LDK(KP500000000), Tx, Tw); Chris@10: } Chris@10: { Chris@10: V T4, T5, T8, T9; Chris@10: Tp = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: T4 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tq = VADD(T4, T5); Chris@10: T6 = VSUB(T4, T5); Chris@10: Tr = VFNMS(LDK(KP500000000), Tq, Tp); Chris@10: Tt = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: T9 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tu = VADD(T8, T9); Chris@10: Ta = VSUB(T8, T9); Chris@10: Tv = VFNMS(LDK(KP500000000), Tu, Tt); Chris@10: } Chris@10: Ts = VSUB(To, Tr); Chris@10: TL = VSUB(T3, T6); Chris@10: TM = VSUB(Ta, Td); Chris@10: Tz = VSUB(Tv, Ty); Chris@10: TC = VADD(To, Tr); Chris@10: TD = VADD(Tv, Ty); Chris@10: TI = VADD(TC, TD); Chris@10: T12 = VADD(Tm, Tn); Chris@10: T13 = VADD(Tp, Tq); Chris@10: T14 = VADD(T12, T13); Chris@10: T15 = VADD(Tt, Tu); Chris@10: T16 = VADD(Tw, Tx); Chris@10: T17 = VADD(T15, T16); Chris@10: T7 = VADD(T3, T6); Chris@10: Te = VADD(Ta, Td); Chris@10: Tf = VMUL(LDK(KP484122918), VSUB(T7, Te)); Chris@10: Tj = VADD(T7, Te); Chris@10: } Chris@10: TZ = VADD(TH, TI); Chris@10: T10 = VBYI(VMUL(LDK(KP866025403), VADD(Ti, Tj))); Chris@10: ST(&(xo[WS(os, 5)]), VSUB(TZ, T10), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 10)]), VADD(T10, TZ), ovs, &(xo[0])); Chris@10: { Chris@10: V T1a, T18, T19, T1e, T1f, T1c, T1d, T1g, T1b; Chris@10: T1a = VMUL(LDK(KP559016994), VSUB(T14, T17)); Chris@10: T18 = VADD(T14, T17); Chris@10: T19 = VFNMS(LDK(KP250000000), T18, T11); Chris@10: T1c = VSUB(T12, T13); Chris@10: T1d = VSUB(T15, T16); Chris@10: T1e = VBYI(VFNMS(LDK(KP951056516), T1d, VMUL(LDK(KP587785252), T1c))); Chris@10: T1f = VBYI(VFMA(LDK(KP951056516), T1c, VMUL(LDK(KP587785252), T1d))); Chris@10: ST(&(xo[0]), VADD(T11, T18), ovs, &(xo[0])); Chris@10: T1g = VADD(T1a, T19); Chris@10: ST(&(xo[WS(os, 6)]), VADD(T1f, T1g), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 9)]), VSUB(T1g, T1f), ovs, &(xo[WS(os, 1)])); Chris@10: T1b = VSUB(T19, T1a); Chris@10: ST(&(xo[WS(os, 3)]), VSUB(T1b, T1e), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 12)]), VADD(T1e, T1b), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V TA, TN, TU, TS, Tl, TR, TK, TV, Tk, TE, TJ; Chris@10: TA = VFMA(LDK(KP951056516), Ts, VMUL(LDK(KP587785252), Tz)); Chris@10: TN = VFMA(LDK(KP823639103), TL, VMUL(LDK(KP509036960), TM)); Chris@10: TU = VFNMS(LDK(KP823639103), TM, VMUL(LDK(KP509036960), TL)); Chris@10: TS = VFNMS(LDK(KP951056516), Tz, VMUL(LDK(KP587785252), Ts)); Chris@10: Tk = VFNMS(LDK(KP216506350), Tj, VMUL(LDK(KP866025403), Ti)); Chris@10: Tl = VADD(Tf, Tk); Chris@10: TR = VSUB(Tf, Tk); Chris@10: TE = VMUL(LDK(KP559016994), VSUB(TC, TD)); Chris@10: TJ = VFNMS(LDK(KP250000000), TI, TH); Chris@10: TK = VADD(TE, TJ); Chris@10: TV = VSUB(TJ, TE); Chris@10: { Chris@10: V TB, TO, TX, TY; Chris@10: TB = VBYI(VADD(Tl, TA)); Chris@10: TO = VSUB(TK, TN); Chris@10: ST(&(xo[WS(os, 1)]), VADD(TB, TO), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 14)]), VSUB(TO, TB), ovs, &(xo[0])); Chris@10: TX = VBYI(VSUB(TS, TR)); Chris@10: TY = VSUB(TV, TU); Chris@10: ST(&(xo[WS(os, 7)]), VADD(TX, TY), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(TY, TX), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V TP, TQ, TT, TW; Chris@10: TP = VBYI(VSUB(Tl, TA)); Chris@10: TQ = VADD(TN, TK); Chris@10: ST(&(xo[WS(os, 4)]), VADD(TP, TQ), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 11)]), VSUB(TQ, TP), ovs, &(xo[WS(os, 1)])); Chris@10: TT = VBYI(VADD(TR, TS)); Chris@10: TW = VADD(TU, TV); Chris@10: ST(&(xo[WS(os, 2)]), VADD(TT, TW), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 13)]), VSUB(TW, TT), 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("n1bv_15"), {64, 11, 14, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_15) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_15, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */