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 08:04:51 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_notw_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name n1fv_9 -include dft/simd/n1f.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 38 FP multiplications, Chris@82: * (or, 12 additions, 4 multiplications, 34 fused multiply/add), Chris@82: * 50 stack variables, 19 constants, and 18 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n1f.h" Chris@82: Chris@82: 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@82: { Chris@82: DVK(KP666666666, +0.666666666666666666666666666666666666666666667); Chris@82: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@82: DVK(KP673648177, +0.673648177666930348851716626769314796000375677); Chris@82: DVK(KP898197570, +0.898197570222573798468955502359086394667167570); Chris@82: DVK(KP879385241, +0.879385241571816768108218554649462939872416269); Chris@82: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@82: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@82: DVK(KP826351822, +0.826351822333069651148283373230685203999624323); Chris@82: DVK(KP420276625, +0.420276625461206169731530603237061658838781920); Chris@82: DVK(KP907603734, +0.907603734547952313649323976213898122064543220); Chris@82: DVK(KP347296355, +0.347296355333860697703433253538629592000751354); Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DVK(KP726681596, +0.726681596905677465811651808188092531873167623); Chris@82: DVK(KP968908795, +0.968908795874236621082202410917456709164223497); Chris@82: DVK(KP586256827, +0.586256827714544512072145703099641959914944179); Chris@82: DVK(KP203604859, +0.203604859554852403062088995281827210665664861); Chris@82: DVK(KP152703644, +0.152703644666139302296566746461370407999248646); Chris@82: DVK(KP439692620, +0.439692620785908384054109277324731469936208134); Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: 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@82: V T5, Tv, Tj, Tl, Tm, Ta, Tf, Tk, Ts, TB, Tx, Tn, To, TC, Ty; Chris@82: V Ti, Tg, Th; Chris@82: { Chris@82: V T1, T2, T3, T4; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T2 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: T3 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: T4 = VADD(T2, T3); Chris@82: T5 = VADD(T1, T4); Chris@82: Tv = VSUB(T3, T2); Chris@82: Tj = VFNMS(LDK(KP500000000), T4, T1); Chris@82: } Chris@82: { Chris@82: V T6, Tb, T9, Te; Chris@82: T6 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tb = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: { Chris@82: V T7, T8, Tc, Td; Chris@82: T7 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T8 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: T9 = VADD(T7, T8); Chris@82: Tl = VSUB(T7, T8); Chris@82: Tc = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: Td = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@82: Te = VADD(Tc, Td); Chris@82: Tm = VSUB(Td, Tc); Chris@82: } Chris@82: Ta = VADD(T6, T9); Chris@82: Tf = VADD(Tb, Te); Chris@82: Tk = VFNMS(LDK(KP500000000), Te, Tb); Chris@82: Ts = VFNMS(LDK(KP439692620), Tl, Tk); Chris@82: TB = VFNMS(LDK(KP152703644), Tm, Tk); Chris@82: Tx = VFMA(LDK(KP203604859), Tk, Tm); Chris@82: Tn = VFNMS(LDK(KP500000000), T9, T6); Chris@82: To = VFNMS(LDK(KP586256827), Tn, Tm); Chris@82: TC = VFMA(LDK(KP968908795), Tn, Tl); Chris@82: Ty = VFNMS(LDK(KP726681596), Tl, Tn); Chris@82: } Chris@82: Ti = VMUL(LDK(KP866025403), VSUB(Tf, Ta)); Chris@82: Tg = VADD(Ta, Tf); Chris@82: Th = VFNMS(LDK(KP500000000), Tg, T5); Chris@82: ST(&(xo[0]), VADD(T5, Tg), ovs, &(xo[0])); Chris@82: ST(&(xo[WS(os, 3)]), VFMAI(Ti, Th), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 6)]), VFNMSI(Ti, Th), ovs, &(xo[0])); Chris@82: { Chris@82: V Tq, Tu, Tp, Tt, Tr, Tw; Chris@82: Tp = VFNMS(LDK(KP347296355), To, Tl); Chris@82: Tq = VFNMS(LDK(KP907603734), Tp, Tk); Chris@82: Tt = VFNMS(LDK(KP420276625), Ts, Tm); Chris@82: Tu = VFNMS(LDK(KP826351822), Tt, Tn); Chris@82: Tr = VFNMS(LDK(KP939692620), Tq, Tj); Chris@82: Tw = VMUL(LDK(KP984807753), VFMA(LDK(KP879385241), Tv, Tu)); Chris@82: ST(&(xo[WS(os, 2)]), VFNMSI(Tw, Tr), ovs, &(xo[0])); Chris@82: ST(&(xo[WS(os, 7)]), VFMAI(Tw, Tr), ovs, &(xo[WS(os, 1)])); Chris@82: } Chris@82: { Chris@82: V TA, TG, TE, TJ, TH, TK; Chris@82: { Chris@82: V Tz, TF, TD, TI; Chris@82: Tz = VFMA(LDK(KP898197570), Ty, Tx); Chris@82: TF = VFNMS(LDK(KP673648177), TC, TB); Chris@82: TA = VFMA(LDK(KP852868531), Tz, Tj); Chris@82: TG = VFNMS(LDK(KP500000000), Tz, TF); Chris@82: TD = VFMA(LDK(KP673648177), TC, TB); Chris@82: TI = VFNMS(LDK(KP898197570), Ty, Tx); Chris@82: TE = VMUL(LDK(KP984807753), VFNMS(LDK(KP879385241), Tv, TD)); Chris@82: TJ = VFMA(LDK(KP666666666), TD, TI); Chris@82: } Chris@82: ST(&(xo[WS(os, 1)]), VFNMSI(TE, TA), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 8)]), VFMAI(TE, TA), ovs, &(xo[0])); Chris@82: TH = VFMA(LDK(KP852868531), TG, Tj); Chris@82: TK = VMUL(LDK(KP866025403), VFMA(LDK(KP852868531), TJ, Tv)); Chris@82: ST(&(xo[WS(os, 5)]), VFNMSI(TK, TH), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 4)]), VFMAI(TK, TH), ovs, &(xo[0])); Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 9, XSIMD_STRING("n1fv_9"), {12, 4, 34, 0}, &GENUS, 0, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n1fv_9) (planner *p) { Chris@82: X(kdft_register) (p, n1fv_9, &desc); Chris@82: } Chris@82: Chris@82: #else Chris@82: Chris@82: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name n1fv_9 -include dft/simd/n1f.h */ Chris@82: Chris@82: /* Chris@82: * This function contains 46 FP additions, 26 FP multiplications, Chris@82: * (or, 30 additions, 10 multiplications, 16 fused multiply/add), Chris@82: * 41 stack variables, 14 constants, and 18 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n1f.h" Chris@82: Chris@82: 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@82: { Chris@82: DVK(KP342020143, +0.342020143325668733044099614682259580763083368); Chris@82: DVK(KP813797681, +0.813797681349373692844693217248393223289101568); Chris@82: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@82: DVK(KP296198132, +0.296198132726023843175338011893050938967728390); Chris@82: DVK(KP642787609, +0.642787609686539326322643409907263432907559884); Chris@82: DVK(KP663413948, +0.663413948168938396205421319635891297216863310); Chris@82: DVK(KP556670399, +0.556670399226419366452912952047023132968291906); Chris@82: DVK(KP766044443, +0.766044443118978035202392650555416673935832457); Chris@82: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@82: DVK(KP150383733, +0.150383733180435296639271897612501926072238258); Chris@82: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@82: DVK(KP173648177, +0.173648177666930348851716626769314796000375677); Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: 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@82: V T5, Ts, Tj, To, Tf, Tn, Tp, Tu, Tl, Ta, Tk, Tm, Tt; Chris@82: { Chris@82: V T1, T2, T3, T4; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T2 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: T3 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: T4 = VADD(T2, T3); Chris@82: T5 = VADD(T1, T4); Chris@82: Ts = VMUL(LDK(KP866025403), VSUB(T3, T2)); Chris@82: Tj = VFNMS(LDK(KP500000000), T4, T1); Chris@82: } Chris@82: { Chris@82: V Tb, Te, Tc, Td; Chris@82: Tb = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: Tc = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: Td = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@82: Te = VADD(Tc, Td); Chris@82: To = VSUB(Td, Tc); Chris@82: Tf = VADD(Tb, Te); Chris@82: Tn = VFNMS(LDK(KP500000000), Te, Tb); Chris@82: Tp = VFMA(LDK(KP173648177), Tn, VMUL(LDK(KP852868531), To)); Chris@82: Tu = VFNMS(LDK(KP984807753), Tn, VMUL(LDK(KP150383733), To)); Chris@82: } Chris@82: { Chris@82: V T6, T9, T7, T8; Chris@82: T6 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: T7 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T8 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: T9 = VADD(T7, T8); Chris@82: Tl = VSUB(T8, T7); Chris@82: Ta = VADD(T6, T9); Chris@82: Tk = VFNMS(LDK(KP500000000), T9, T6); Chris@82: Tm = VFMA(LDK(KP766044443), Tk, VMUL(LDK(KP556670399), Tl)); Chris@82: Tt = VFNMS(LDK(KP642787609), Tk, VMUL(LDK(KP663413948), Tl)); Chris@82: } Chris@82: { Chris@82: V Ti, Tg, Th, Tz, TA; Chris@82: Ti = VBYI(VMUL(LDK(KP866025403), VSUB(Tf, Ta))); Chris@82: Tg = VADD(Ta, Tf); Chris@82: Th = VFNMS(LDK(KP500000000), Tg, T5); Chris@82: ST(&(xo[0]), VADD(T5, Tg), ovs, &(xo[0])); Chris@82: ST(&(xo[WS(os, 3)]), VADD(Th, Ti), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 6)]), VSUB(Th, Ti), ovs, &(xo[0])); Chris@82: Tz = VFMA(LDK(KP173648177), Tk, VFNMS(LDK(KP296198132), To, VFNMS(LDK(KP939692620), Tn, VFNMS(LDK(KP852868531), Tl, Tj)))); Chris@82: TA = VBYI(VSUB(VFNMS(LDK(KP342020143), Tn, VFNMS(LDK(KP150383733), Tl, VFNMS(LDK(KP984807753), Tk, VMUL(LDK(KP813797681), To)))), Ts)); Chris@82: ST(&(xo[WS(os, 7)]), VSUB(Tz, TA), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 2)]), VADD(Tz, TA), ovs, &(xo[0])); Chris@82: { Chris@82: V Tr, Tx, Tw, Ty, Tq, Tv; Chris@82: Tq = VADD(Tm, Tp); Chris@82: Tr = VADD(Tj, Tq); Chris@82: Tx = VFMA(LDK(KP866025403), VSUB(Tt, Tu), VFNMS(LDK(KP500000000), Tq, Tj)); Chris@82: Tv = VADD(Tt, Tu); Chris@82: Tw = VBYI(VADD(Ts, Tv)); Chris@82: Ty = VBYI(VADD(Ts, VFNMS(LDK(KP500000000), Tv, VMUL(LDK(KP866025403), VSUB(Tp, Tm))))); Chris@82: ST(&(xo[WS(os, 8)]), VSUB(Tr, Tw), ovs, &(xo[0])); Chris@82: ST(&(xo[WS(os, 4)]), VADD(Tx, Ty), ovs, &(xo[0])); Chris@82: ST(&(xo[WS(os, 1)]), VADD(Tw, Tr), ovs, &(xo[WS(os, 1)])); Chris@82: ST(&(xo[WS(os, 5)]), VSUB(Tx, Ty), ovs, &(xo[WS(os, 1)])); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 9, XSIMD_STRING("n1fv_9"), {30, 10, 16, 0}, &GENUS, 0, 0, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n1fv_9) (planner *p) { Chris@82: X(kdft_register) (p, n1fv_9, &desc); Chris@82: } Chris@82: Chris@82: #endif