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 9 -name n1fv_9 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 38 FP multiplications, Chris@10: * (or, 12 additions, 4 multiplications, 34 fused multiply/add), Chris@10: * 68 stack variables, 19 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_9(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(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@10: DVK(KP826351822, +0.826351822333069651148283373230685203999624323); Chris@10: DVK(KP879385241, +0.879385241571816768108218554649462939872416269); Chris@10: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@10: DVK(KP666666666, +0.666666666666666666666666666666666666666666667); Chris@10: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@10: DVK(KP907603734, +0.907603734547952313649323976213898122064543220); Chris@10: DVK(KP420276625, +0.420276625461206169731530603237061658838781920); Chris@10: DVK(KP673648177, +0.673648177666930348851716626769314796000375677); Chris@10: DVK(KP898197570, +0.898197570222573798468955502359086394667167570); Chris@10: DVK(KP347296355, +0.347296355333860697703433253538629592000751354); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP439692620, +0.439692620785908384054109277324731469936208134); Chris@10: DVK(KP203604859, +0.203604859554852403062088995281827210665664861); Chris@10: DVK(KP152703644, +0.152703644666139302296566746461370407999248646); Chris@10: DVK(KP586256827, +0.586256827714544512072145703099641959914944179); Chris@10: DVK(KP968908795, +0.968908795874236621082202410917456709164223497); Chris@10: DVK(KP726681596, +0.726681596905677465811651808188092531873167623); 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(18, is), MAKE_VOLATILE_STRIDE(18, os)) { Chris@10: V T1, T2, T3, T6, Tb, T7, T8, Tc, Td, Tv, T4; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T6 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tb = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T7 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tc = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: Tv = VSUB(T3, T2); Chris@10: T4 = VADD(T2, T3); Chris@10: { Chris@10: V Tl, T9, Tm, Te, Tj, T5; Chris@10: Tl = VSUB(T7, T8); Chris@10: T9 = VADD(T7, T8); Chris@10: Tm = VSUB(Td, Tc); Chris@10: Te = VADD(Tc, Td); Chris@10: Tj = VFNMS(LDK(KP500000000), T4, T1); Chris@10: T5 = VADD(T1, T4); Chris@10: { Chris@10: V Tn, Ta, Tk, Tf; Chris@10: Tn = VFNMS(LDK(KP500000000), T9, T6); Chris@10: Ta = VADD(T6, T9); Chris@10: Tk = VFNMS(LDK(KP500000000), Te, Tb); Chris@10: Tf = VADD(Tb, Te); Chris@10: { Chris@10: V Ty, TC, To, TB, Tx, Ts, Tg, Ti; Chris@10: Ty = VFNMS(LDK(KP726681596), Tl, Tn); Chris@10: TC = VFMA(LDK(KP968908795), Tn, Tl); Chris@10: To = VFNMS(LDK(KP586256827), Tn, Tm); Chris@10: TB = VFNMS(LDK(KP152703644), Tm, Tk); Chris@10: Tx = VFMA(LDK(KP203604859), Tk, Tm); Chris@10: Ts = VFNMS(LDK(KP439692620), Tl, Tk); Chris@10: Tg = VADD(Ta, Tf); Chris@10: Ti = VMUL(LDK(KP866025403), VSUB(Tf, Ta)); Chris@10: { Chris@10: V Tz, TI, TF, TD, Tt, Th, Tq, Tp; Chris@10: Tp = VFNMS(LDK(KP347296355), To, Tl); Chris@10: Tz = VFMA(LDK(KP898197570), Ty, Tx); Chris@10: TI = VFNMS(LDK(KP898197570), Ty, Tx); Chris@10: TF = VFNMS(LDK(KP673648177), TC, TB); Chris@10: TD = VFMA(LDK(KP673648177), TC, TB); Chris@10: Tt = VFNMS(LDK(KP420276625), Ts, Tm); Chris@10: ST(&(xo[0]), VADD(T5, Tg), ovs, &(xo[0])); Chris@10: Th = VFNMS(LDK(KP500000000), Tg, T5); Chris@10: Tq = VFNMS(LDK(KP907603734), Tp, Tk); Chris@10: { Chris@10: V TA, TJ, TE, TG, Tu, Tr, TK, TH, Tw; Chris@10: TA = VFMA(LDK(KP852868531), Tz, Tj); Chris@10: TJ = VFMA(LDK(KP666666666), TD, TI); Chris@10: TE = VMUL(LDK(KP984807753), VFNMS(LDK(KP879385241), Tv, TD)); Chris@10: TG = VFNMS(LDK(KP500000000), Tz, TF); Chris@10: Tu = VFNMS(LDK(KP826351822), Tt, Tn); Chris@10: ST(&(xo[WS(os, 6)]), VFNMSI(Ti, Th), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 3)]), VFMAI(Ti, Th), ovs, &(xo[WS(os, 1)])); Chris@10: Tr = VFNMS(LDK(KP939692620), Tq, Tj); Chris@10: TK = VMUL(LDK(KP866025403), VFMA(LDK(KP852868531), TJ, Tv)); Chris@10: ST(&(xo[WS(os, 8)]), VFMAI(TE, TA), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 1)]), VFNMSI(TE, TA), ovs, &(xo[WS(os, 1)])); Chris@10: TH = VFMA(LDK(KP852868531), TG, Tj); Chris@10: Tw = VMUL(LDK(KP984807753), VFMA(LDK(KP879385241), Tv, Tu)); Chris@10: ST(&(xo[WS(os, 4)]), VFMAI(TK, TH), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 5)]), VFNMSI(TK, TH), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VFMAI(Tw, Tr), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 2)]), VFNMSI(Tw, Tr), ovs, &(xo[0])); 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 = { 9, XSIMD_STRING("n1fv_9"), {12, 4, 34, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_9) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_9, &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 9 -name n1fv_9 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 46 FP additions, 26 FP multiplications, Chris@10: * (or, 30 additions, 10 multiplications, 16 fused multiply/add), Chris@10: * 41 stack variables, 14 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_9(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(KP342020143, +0.342020143325668733044099614682259580763083368); Chris@10: DVK(KP813797681, +0.813797681349373692844693217248393223289101568); Chris@10: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@10: DVK(KP296198132, +0.296198132726023843175338011893050938967728390); Chris@10: DVK(KP642787609, +0.642787609686539326322643409907263432907559884); Chris@10: DVK(KP663413948, +0.663413948168938396205421319635891297216863310); Chris@10: DVK(KP556670399, +0.556670399226419366452912952047023132968291906); Chris@10: DVK(KP766044443, +0.766044443118978035202392650555416673935832457); Chris@10: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@10: DVK(KP150383733, +0.150383733180435296639271897612501926072238258); Chris@10: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@10: DVK(KP173648177, +0.173648177666930348851716626769314796000375677); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); 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(18, is), MAKE_VOLATILE_STRIDE(18, os)) { Chris@10: V T5, Ts, Tj, To, Tf, Tn, Tp, Tu, Tl, Ta, Tk, Tm, Tt; 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, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T4 = VADD(T2, T3); Chris@10: T5 = VADD(T1, T4); Chris@10: Ts = VMUL(LDK(KP866025403), VSUB(T3, T2)); Chris@10: Tj = VFNMS(LDK(KP500000000), T4, T1); Chris@10: } Chris@10: { Chris@10: V Tb, Te, Tc, Td; Chris@10: Tb = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: Te = VADD(Tc, Td); Chris@10: To = VSUB(Td, Tc); Chris@10: Tf = VADD(Tb, Te); Chris@10: Tn = VFNMS(LDK(KP500000000), Te, Tb); Chris@10: Tp = VFMA(LDK(KP173648177), Tn, VMUL(LDK(KP852868531), To)); Chris@10: Tu = VFNMS(LDK(KP984807753), Tn, VMUL(LDK(KP150383733), To)); Chris@10: } Chris@10: { Chris@10: V T6, T9, T7, T8; Chris@10: T6 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T9 = VADD(T7, T8); Chris@10: Tl = VSUB(T8, T7); Chris@10: Ta = VADD(T6, T9); Chris@10: Tk = VFNMS(LDK(KP500000000), T9, T6); Chris@10: Tm = VFMA(LDK(KP766044443), Tk, VMUL(LDK(KP556670399), Tl)); Chris@10: Tt = VFNMS(LDK(KP642787609), Tk, VMUL(LDK(KP663413948), Tl)); Chris@10: } Chris@10: { Chris@10: V Ti, Tg, Th, Tz, TA; Chris@10: Ti = VBYI(VMUL(LDK(KP866025403), VSUB(Tf, Ta))); Chris@10: Tg = VADD(Ta, Tf); Chris@10: Th = VFNMS(LDK(KP500000000), Tg, T5); Chris@10: ST(&(xo[0]), VADD(T5, Tg), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 3)]), VADD(Th, Ti), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 6)]), VSUB(Th, Ti), ovs, &(xo[0])); Chris@10: Tz = VFMA(LDK(KP173648177), Tk, VFNMS(LDK(KP296198132), To, VFNMS(LDK(KP939692620), Tn, VFNMS(LDK(KP852868531), Tl, Tj)))); Chris@10: TA = VBYI(VSUB(VFNMS(LDK(KP342020143), Tn, VFNMS(LDK(KP150383733), Tl, VFNMS(LDK(KP984807753), Tk, VMUL(LDK(KP813797681), To)))), Ts)); Chris@10: ST(&(xo[WS(os, 7)]), VSUB(Tz, TA), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 2)]), VADD(Tz, TA), ovs, &(xo[0])); Chris@10: { Chris@10: V Tr, Tx, Tw, Ty, Tq, Tv; Chris@10: Tq = VADD(Tm, Tp); Chris@10: Tr = VADD(Tj, Tq); Chris@10: Tx = VFMA(LDK(KP866025403), VSUB(Tt, Tu), VFNMS(LDK(KP500000000), Tq, Tj)); Chris@10: Tv = VADD(Tt, Tu); Chris@10: Tw = VBYI(VADD(Ts, Tv)); Chris@10: Ty = VBYI(VADD(Ts, VFNMS(LDK(KP500000000), Tv, VMUL(LDK(KP866025403), VSUB(Tp, Tm))))); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(Tr, Tw), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VADD(Tx, Ty), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 1)]), VADD(Tw, Tr), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VSUB(Tx, Ty), 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 = { 9, XSIMD_STRING("n1fv_9"), {30, 10, 16, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_9) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_9, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */