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 07:38:00 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_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name t1fuv_9 -include t1fu.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 54 FP additions, 54 FP multiplications, Chris@10: * (or, 20 additions, 20 multiplications, 34 fused multiply/add), Chris@10: * 67 stack variables, 19 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "t1fu.h" Chris@10: Chris@10: static void t1fuv_9(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) 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 m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 16)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 16), MAKE_VOLATILE_STRIDE(9, rs)) { Chris@10: V T1, T3, T5, T9, Th, Tb, Td, Tj, Tl, TD, T6; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: { Chris@10: V T2, T4, T8, Tg; Chris@10: T2 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: T4 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: Tg = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: { Chris@10: V Ta, Tc, Ti, Tk; Chris@10: Ta = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: Tc = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: Ti = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tk = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: T3 = BYTWJ(&(W[TWVL * 4]), T2); Chris@10: T5 = BYTWJ(&(W[TWVL * 10]), T4); Chris@10: T9 = BYTWJ(&(W[0]), T8); Chris@10: Th = BYTWJ(&(W[TWVL * 2]), Tg); Chris@10: Tb = BYTWJ(&(W[TWVL * 6]), Ta); Chris@10: Td = BYTWJ(&(W[TWVL * 12]), Tc); Chris@10: Tj = BYTWJ(&(W[TWVL * 8]), Ti); Chris@10: Tl = BYTWJ(&(W[TWVL * 14]), Tk); Chris@10: } Chris@10: } Chris@10: TD = VSUB(T5, T3); Chris@10: T6 = VADD(T3, T5); Chris@10: { Chris@10: V Tt, Te, Tu, Tm, Tr, T7; Chris@10: Tt = VSUB(Tb, Td); Chris@10: Te = VADD(Tb, Td); Chris@10: Tu = VSUB(Tl, Tj); Chris@10: Tm = VADD(Tj, Tl); Chris@10: Tr = VFNMS(LDK(KP500000000), T6, T1); Chris@10: T7 = VADD(T1, T6); Chris@10: { Chris@10: V Tv, Tf, Ts, Tn; Chris@10: Tv = VFNMS(LDK(KP500000000), Te, T9); Chris@10: Tf = VADD(T9, Te); Chris@10: Ts = VFNMS(LDK(KP500000000), Tm, Th); Chris@10: Tn = VADD(Th, Tm); Chris@10: { Chris@10: V TG, TK, Tw, TJ, TF, TA, To, Tq; Chris@10: TG = VFNMS(LDK(KP726681596), Tt, Tv); Chris@10: TK = VFMA(LDK(KP968908795), Tv, Tt); Chris@10: Tw = VFNMS(LDK(KP586256827), Tv, Tu); Chris@10: TJ = VFNMS(LDK(KP152703644), Tu, Ts); Chris@10: TF = VFMA(LDK(KP203604859), Ts, Tu); Chris@10: TA = VFNMS(LDK(KP439692620), Tt, Ts); Chris@10: To = VADD(Tf, Tn); Chris@10: Tq = VMUL(LDK(KP866025403), VSUB(Tn, Tf)); Chris@10: { Chris@10: V TQ, TH, TL, TN, TB, Tp, Ty, TI, Tx; Chris@10: Tx = VFNMS(LDK(KP347296355), Tw, Tt); Chris@10: TQ = VFNMS(LDK(KP898197570), TG, TF); Chris@10: TH = VFMA(LDK(KP898197570), TG, TF); Chris@10: TL = VFMA(LDK(KP673648177), TK, TJ); Chris@10: TN = VFNMS(LDK(KP673648177), TK, TJ); Chris@10: TB = VFNMS(LDK(KP420276625), TA, Tu); Chris@10: ST(&(x[0]), VADD(T7, To), ms, &(x[0])); Chris@10: Tp = VFNMS(LDK(KP500000000), To, T7); Chris@10: Ty = VFNMS(LDK(KP907603734), Tx, Ts); Chris@10: TI = VFMA(LDK(KP852868531), TH, Tr); Chris@10: { Chris@10: V TO, TR, TM, TC, Tz, TP, TS, TE; Chris@10: TO = VFNMS(LDK(KP500000000), TH, TN); Chris@10: TR = VFMA(LDK(KP666666666), TL, TQ); Chris@10: TM = VMUL(LDK(KP984807753), VFNMS(LDK(KP879385241), TD, TL)); Chris@10: TC = VFNMS(LDK(KP826351822), TB, Tv); Chris@10: ST(&(x[WS(rs, 6)]), VFNMSI(Tq, Tp), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 3)]), VFMAI(Tq, Tp), ms, &(x[WS(rs, 1)])); Chris@10: Tz = VFNMS(LDK(KP939692620), Ty, Tr); Chris@10: TP = VFMA(LDK(KP852868531), TO, Tr); Chris@10: TS = VMUL(LDK(KP866025403), VFMA(LDK(KP852868531), TR, TD)); Chris@10: ST(&(x[WS(rs, 8)]), VFMAI(TM, TI), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 1)]), VFNMSI(TM, TI), ms, &(x[WS(rs, 1)])); Chris@10: TE = VMUL(LDK(KP984807753), VFMA(LDK(KP879385241), TD, TC)); Chris@10: ST(&(x[WS(rs, 4)]), VFMAI(TS, TP), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 5)]), VFNMSI(TS, TP), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VFMAI(TE, Tz), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 2)]), VFNMSI(TE, Tz), ms, &(x[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 tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 9, XSIMD_STRING("t1fuv_9"), twinstr, &GENUS, {20, 20, 34, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1fuv_9) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1fuv_9, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name t1fuv_9 -include t1fu.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 54 FP additions, 42 FP multiplications, Chris@10: * (or, 38 additions, 26 multiplications, 16 fused multiply/add), Chris@10: * 38 stack variables, 14 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "t1fu.h" Chris@10: Chris@10: static void t1fuv_9(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); Chris@10: DVK(KP296198132, +0.296198132726023843175338011893050938967728390); Chris@10: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); Chris@10: DVK(KP173648177, +0.173648177666930348851716626769314796000375677); Chris@10: DVK(KP556670399, +0.556670399226419366452912952047023132968291906); Chris@10: DVK(KP766044443, +0.766044443118978035202392650555416673935832457); Chris@10: DVK(KP642787609, +0.642787609686539326322643409907263432907559884); Chris@10: DVK(KP663413948, +0.663413948168938396205421319635891297216863310); Chris@10: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); Chris@10: DVK(KP150383733, +0.150383733180435296639271897612501926072238258); Chris@10: DVK(KP342020143, +0.342020143325668733044099614682259580763083368); Chris@10: DVK(KP813797681, +0.813797681349373692844693217248393223289101568); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 16)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 16), MAKE_VOLATILE_STRIDE(9, rs)) { Chris@10: V T1, T6, TA, Tt, Tf, Ts, Tw, Tn, Tv; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: { Chris@10: V T3, T5, T2, T4; Chris@10: T2 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: T3 = BYTWJ(&(W[TWVL * 4]), T2); Chris@10: T4 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: T5 = BYTWJ(&(W[TWVL * 10]), T4); Chris@10: T6 = VADD(T3, T5); Chris@10: TA = VMUL(LDK(KP866025403), VSUB(T5, T3)); Chris@10: } Chris@10: { Chris@10: V T9, Td, Tb, T8, Tc, Ta, Te; Chris@10: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: T9 = BYTWJ(&(W[0]), T8); Chris@10: Tc = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: Td = BYTWJ(&(W[TWVL * 12]), Tc); Chris@10: Ta = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: Tb = BYTWJ(&(W[TWVL * 6]), Ta); Chris@10: Tt = VSUB(Td, Tb); Chris@10: Te = VADD(Tb, Td); Chris@10: Tf = VADD(T9, Te); Chris@10: Ts = VFNMS(LDK(KP500000000), Te, T9); Chris@10: } Chris@10: { Chris@10: V Th, Tl, Tj, Tg, Tk, Ti, Tm; Chris@10: Tg = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: Th = BYTWJ(&(W[TWVL * 2]), Tg); Chris@10: Tk = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: Tl = BYTWJ(&(W[TWVL * 14]), Tk); Chris@10: Ti = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tj = BYTWJ(&(W[TWVL * 8]), Ti); Chris@10: Tw = VSUB(Tl, Tj); Chris@10: Tm = VADD(Tj, Tl); Chris@10: Tn = VADD(Th, Tm); Chris@10: Tv = VFNMS(LDK(KP500000000), Tm, Th); Chris@10: } Chris@10: { Chris@10: V Tq, T7, To, Tp; Chris@10: Tq = VBYI(VMUL(LDK(KP866025403), VSUB(Tn, Tf))); Chris@10: T7 = VADD(T1, T6); Chris@10: To = VADD(Tf, Tn); Chris@10: Tp = VFNMS(LDK(KP500000000), To, T7); Chris@10: ST(&(x[0]), VADD(T7, To), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 3)]), VADD(Tp, Tq), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 6)]), VSUB(Tp, Tq), ms, &(x[0])); Chris@10: } Chris@10: { Chris@10: V TI, TB, TC, TD, Tu, Tx, Ty, Tr, TH; Chris@10: TI = VBYI(VSUB(VFNMS(LDK(KP342020143), Tv, VFNMS(LDK(KP150383733), Tt, VFNMS(LDK(KP984807753), Ts, VMUL(LDK(KP813797681), Tw)))), TA)); Chris@10: TB = VFNMS(LDK(KP642787609), Ts, VMUL(LDK(KP663413948), Tt)); Chris@10: TC = VFNMS(LDK(KP984807753), Tv, VMUL(LDK(KP150383733), Tw)); Chris@10: TD = VADD(TB, TC); Chris@10: Tu = VFMA(LDK(KP766044443), Ts, VMUL(LDK(KP556670399), Tt)); Chris@10: Tx = VFMA(LDK(KP173648177), Tv, VMUL(LDK(KP852868531), Tw)); Chris@10: Ty = VADD(Tu, Tx); Chris@10: Tr = VFNMS(LDK(KP500000000), T6, T1); Chris@10: TH = VFMA(LDK(KP173648177), Ts, VFNMS(LDK(KP296198132), Tw, VFNMS(LDK(KP939692620), Tv, VFNMS(LDK(KP852868531), Tt, Tr)))); Chris@10: ST(&(x[WS(rs, 7)]), VSUB(TH, TI), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 2)]), VADD(TH, TI), ms, &(x[0])); Chris@10: { Chris@10: V Tz, TE, TF, TG; Chris@10: Tz = VADD(Tr, Ty); Chris@10: TE = VBYI(VADD(TA, TD)); Chris@10: ST(&(x[WS(rs, 8)]), VSUB(Tz, TE), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 1)]), VADD(TE, Tz), ms, &(x[WS(rs, 1)])); Chris@10: TF = VFMA(LDK(KP866025403), VSUB(TB, TC), VFNMS(LDK(KP500000000), Ty, Tr)); Chris@10: TG = VBYI(VADD(TA, VFNMS(LDK(KP500000000), TD, VMUL(LDK(KP866025403), VSUB(Tx, Tu))))); Chris@10: ST(&(x[WS(rs, 5)]), VSUB(TF, TG), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 4)]), VADD(TF, TG), ms, &(x[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), Chris@10: VTW(0, 3), Chris@10: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 9, XSIMD_STRING("t1fuv_9"), twinstr, &GENUS, {38, 26, 16, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1fuv_9) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1fuv_9, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */