cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:38:02 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name t1fv_9 -include t1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 54 FP additions, 54 FP multiplications, cannam@95: * (or, 20 additions, 20 multiplications, 34 fused multiply/add), cannam@95: * 67 stack variables, 19 constants, and 18 memory accesses cannam@95: */ cannam@95: #include "t1f.h" cannam@95: cannam@95: static void t1fv_9(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); cannam@95: DVK(KP826351822, +0.826351822333069651148283373230685203999624323); cannam@95: DVK(KP879385241, +0.879385241571816768108218554649462939872416269); cannam@95: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); cannam@95: DVK(KP666666666, +0.666666666666666666666666666666666666666666667); cannam@95: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); cannam@95: DVK(KP907603734, +0.907603734547952313649323976213898122064543220); cannam@95: DVK(KP420276625, +0.420276625461206169731530603237061658838781920); cannam@95: DVK(KP673648177, +0.673648177666930348851716626769314796000375677); cannam@95: DVK(KP898197570, +0.898197570222573798468955502359086394667167570); cannam@95: DVK(KP347296355, +0.347296355333860697703433253538629592000751354); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP439692620, +0.439692620785908384054109277324731469936208134); cannam@95: DVK(KP203604859, +0.203604859554852403062088995281827210665664861); cannam@95: DVK(KP152703644, +0.152703644666139302296566746461370407999248646); cannam@95: DVK(KP586256827, +0.586256827714544512072145703099641959914944179); cannam@95: DVK(KP968908795, +0.968908795874236621082202410917456709164223497); cannam@95: DVK(KP726681596, +0.726681596905677465811651808188092531873167623); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: 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)) { cannam@95: V T1, T3, T5, T9, Th, Tb, Td, Tj, Tl, TD, T6; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: { cannam@95: V T2, T4, T8, Tg; cannam@95: T2 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: T4 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Tg = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: { cannam@95: V Ta, Tc, Ti, Tk; cannam@95: Ta = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Tc = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Ti = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tk = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: T3 = BYTWJ(&(W[TWVL * 4]), T2); cannam@95: T5 = BYTWJ(&(W[TWVL * 10]), T4); cannam@95: T9 = BYTWJ(&(W[0]), T8); cannam@95: Th = BYTWJ(&(W[TWVL * 2]), Tg); cannam@95: Tb = BYTWJ(&(W[TWVL * 6]), Ta); cannam@95: Td = BYTWJ(&(W[TWVL * 12]), Tc); cannam@95: Tj = BYTWJ(&(W[TWVL * 8]), Ti); cannam@95: Tl = BYTWJ(&(W[TWVL * 14]), Tk); cannam@95: } cannam@95: } cannam@95: TD = VSUB(T5, T3); cannam@95: T6 = VADD(T3, T5); cannam@95: { cannam@95: V Tt, Te, Tu, Tm, Tr, T7; cannam@95: Tt = VSUB(Tb, Td); cannam@95: Te = VADD(Tb, Td); cannam@95: Tu = VSUB(Tl, Tj); cannam@95: Tm = VADD(Tj, Tl); cannam@95: Tr = VFNMS(LDK(KP500000000), T6, T1); cannam@95: T7 = VADD(T1, T6); cannam@95: { cannam@95: V Tv, Tf, Ts, Tn; cannam@95: Tv = VFNMS(LDK(KP500000000), Te, T9); cannam@95: Tf = VADD(T9, Te); cannam@95: Ts = VFNMS(LDK(KP500000000), Tm, Th); cannam@95: Tn = VADD(Th, Tm); cannam@95: { cannam@95: V TG, TK, Tw, TJ, TF, TA, To, Tq; cannam@95: TG = VFNMS(LDK(KP726681596), Tt, Tv); cannam@95: TK = VFMA(LDK(KP968908795), Tv, Tt); cannam@95: Tw = VFNMS(LDK(KP586256827), Tv, Tu); cannam@95: TJ = VFNMS(LDK(KP152703644), Tu, Ts); cannam@95: TF = VFMA(LDK(KP203604859), Ts, Tu); cannam@95: TA = VFNMS(LDK(KP439692620), Tt, Ts); cannam@95: To = VADD(Tf, Tn); cannam@95: Tq = VMUL(LDK(KP866025403), VSUB(Tn, Tf)); cannam@95: { cannam@95: V TQ, TH, TL, TN, TB, Tp, Ty, TI, Tx; cannam@95: Tx = VFNMS(LDK(KP347296355), Tw, Tt); cannam@95: TQ = VFNMS(LDK(KP898197570), TG, TF); cannam@95: TH = VFMA(LDK(KP898197570), TG, TF); cannam@95: TL = VFMA(LDK(KP673648177), TK, TJ); cannam@95: TN = VFNMS(LDK(KP673648177), TK, TJ); cannam@95: TB = VFNMS(LDK(KP420276625), TA, Tu); cannam@95: ST(&(x[0]), VADD(T7, To), ms, &(x[0])); cannam@95: Tp = VFNMS(LDK(KP500000000), To, T7); cannam@95: Ty = VFNMS(LDK(KP907603734), Tx, Ts); cannam@95: TI = VFMA(LDK(KP852868531), TH, Tr); cannam@95: { cannam@95: V TO, TR, TM, TC, Tz, TP, TS, TE; cannam@95: TO = VFNMS(LDK(KP500000000), TH, TN); cannam@95: TR = VFMA(LDK(KP666666666), TL, TQ); cannam@95: TM = VMUL(LDK(KP984807753), VFNMS(LDK(KP879385241), TD, TL)); cannam@95: TC = VFNMS(LDK(KP826351822), TB, Tv); cannam@95: ST(&(x[WS(rs, 6)]), VFNMSI(Tq, Tp), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 3)]), VFMAI(Tq, Tp), ms, &(x[WS(rs, 1)])); cannam@95: Tz = VFNMS(LDK(KP939692620), Ty, Tr); cannam@95: TP = VFMA(LDK(KP852868531), TO, Tr); cannam@95: TS = VMUL(LDK(KP866025403), VFMA(LDK(KP852868531), TR, TD)); cannam@95: ST(&(x[WS(rs, 8)]), VFMAI(TM, TI), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 1)]), VFNMSI(TM, TI), ms, &(x[WS(rs, 1)])); cannam@95: TE = VMUL(LDK(KP984807753), VFMA(LDK(KP879385241), TD, TC)); cannam@95: ST(&(x[WS(rs, 4)]), VFMAI(TS, TP), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 5)]), VFNMSI(TS, TP), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VFMAI(TE, Tz), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 2)]), VFNMSI(TE, Tz), ms, &(x[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 9, XSIMD_STRING("t1fv_9"), twinstr, &GENUS, {20, 20, 34, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fv_9) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fv_9, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 9 -name t1fv_9 -include t1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 54 FP additions, 42 FP multiplications, cannam@95: * (or, 38 additions, 26 multiplications, 16 fused multiply/add), cannam@95: * 38 stack variables, 14 constants, and 18 memory accesses cannam@95: */ cannam@95: #include "t1f.h" cannam@95: cannam@95: static void t1fv_9(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP939692620, +0.939692620785908384054109277324731469936208134); cannam@95: DVK(KP296198132, +0.296198132726023843175338011893050938967728390); cannam@95: DVK(KP852868531, +0.852868531952443209628250963940074071936020296); cannam@95: DVK(KP173648177, +0.173648177666930348851716626769314796000375677); cannam@95: DVK(KP556670399, +0.556670399226419366452912952047023132968291906); cannam@95: DVK(KP766044443, +0.766044443118978035202392650555416673935832457); cannam@95: DVK(KP642787609, +0.642787609686539326322643409907263432907559884); cannam@95: DVK(KP663413948, +0.663413948168938396205421319635891297216863310); cannam@95: DVK(KP984807753, +0.984807753012208059366743024589523013670643252); cannam@95: DVK(KP150383733, +0.150383733180435296639271897612501926072238258); cannam@95: DVK(KP342020143, +0.342020143325668733044099614682259580763083368); cannam@95: DVK(KP813797681, +0.813797681349373692844693217248393223289101568); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: 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)) { cannam@95: V T1, T6, TA, Tt, Tf, Ts, Tw, Tn, Tv; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: { cannam@95: V T3, T5, T2, T4; cannam@95: T2 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: T3 = BYTWJ(&(W[TWVL * 4]), T2); cannam@95: T4 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: T5 = BYTWJ(&(W[TWVL * 10]), T4); cannam@95: T6 = VADD(T3, T5); cannam@95: TA = VMUL(LDK(KP866025403), VSUB(T5, T3)); cannam@95: } cannam@95: { cannam@95: V T9, Td, Tb, T8, Tc, Ta, Te; cannam@95: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: T9 = BYTWJ(&(W[0]), T8); cannam@95: Tc = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Td = BYTWJ(&(W[TWVL * 12]), Tc); cannam@95: Ta = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Tb = BYTWJ(&(W[TWVL * 6]), Ta); cannam@95: Tt = VSUB(Td, Tb); cannam@95: Te = VADD(Tb, Td); cannam@95: Tf = VADD(T9, Te); cannam@95: Ts = VFNMS(LDK(KP500000000), Te, T9); cannam@95: } cannam@95: { cannam@95: V Th, Tl, Tj, Tg, Tk, Ti, Tm; cannam@95: Tg = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Th = BYTWJ(&(W[TWVL * 2]), Tg); cannam@95: Tk = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: Tl = BYTWJ(&(W[TWVL * 14]), Tk); cannam@95: Ti = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tj = BYTWJ(&(W[TWVL * 8]), Ti); cannam@95: Tw = VSUB(Tl, Tj); cannam@95: Tm = VADD(Tj, Tl); cannam@95: Tn = VADD(Th, Tm); cannam@95: Tv = VFNMS(LDK(KP500000000), Tm, Th); cannam@95: } cannam@95: { cannam@95: V Tq, T7, To, Tp; cannam@95: Tq = VBYI(VMUL(LDK(KP866025403), VSUB(Tn, Tf))); cannam@95: T7 = VADD(T1, T6); cannam@95: To = VADD(Tf, Tn); cannam@95: Tp = VFNMS(LDK(KP500000000), To, T7); cannam@95: ST(&(x[0]), VADD(T7, To), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 3)]), VADD(Tp, Tq), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(Tp, Tq), ms, &(x[0])); cannam@95: } cannam@95: { cannam@95: V TI, TB, TC, TD, Tu, Tx, Ty, Tr, TH; cannam@95: TI = VBYI(VSUB(VFNMS(LDK(KP342020143), Tv, VFNMS(LDK(KP150383733), Tt, VFNMS(LDK(KP984807753), Ts, VMUL(LDK(KP813797681), Tw)))), TA)); cannam@95: TB = VFNMS(LDK(KP642787609), Ts, VMUL(LDK(KP663413948), Tt)); cannam@95: TC = VFNMS(LDK(KP984807753), Tv, VMUL(LDK(KP150383733), Tw)); cannam@95: TD = VADD(TB, TC); cannam@95: Tu = VFMA(LDK(KP766044443), Ts, VMUL(LDK(KP556670399), Tt)); cannam@95: Tx = VFMA(LDK(KP173648177), Tv, VMUL(LDK(KP852868531), Tw)); cannam@95: Ty = VADD(Tu, Tx); cannam@95: Tr = VFNMS(LDK(KP500000000), T6, T1); cannam@95: TH = VFMA(LDK(KP173648177), Ts, VFNMS(LDK(KP296198132), Tw, VFNMS(LDK(KP939692620), Tv, VFNMS(LDK(KP852868531), Tt, Tr)))); cannam@95: ST(&(x[WS(rs, 7)]), VSUB(TH, TI), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 2)]), VADD(TH, TI), ms, &(x[0])); cannam@95: { cannam@95: V Tz, TE, TF, TG; cannam@95: Tz = VADD(Tr, Ty); cannam@95: TE = VBYI(VADD(TA, TD)); cannam@95: ST(&(x[WS(rs, 8)]), VSUB(Tz, TE), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 1)]), VADD(TE, Tz), ms, &(x[WS(rs, 1)])); cannam@95: TF = VFMA(LDK(KP866025403), VSUB(TB, TC), VFNMS(LDK(KP500000000), Ty, Tr)); cannam@95: TG = VBYI(VADD(TA, VFNMS(LDK(KP500000000), TD, VMUL(LDK(KP866025403), VSUB(Tx, Tu))))); cannam@95: ST(&(x[WS(rs, 5)]), VSUB(TF, TG), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 4)]), VADD(TF, TG), ms, &(x[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 9, XSIMD_STRING("t1fv_9"), twinstr, &GENUS, {38, 26, 16, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fv_9) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fv_9, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */