cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:39:20 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 16 -name n1bv_16 -include n1b.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 72 FP additions, 34 FP multiplications, cannam@127: * (or, 38 additions, 0 multiplications, 34 fused multiply/add), cannam@127: * 54 stack variables, 3 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "n1b.h" cannam@127: cannam@127: static void n1bv_16(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@127: DVK(KP414213562, +0.414213562373095048801688724209698078569671875); cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT i; cannam@127: const R *xi; cannam@127: R *xo; cannam@127: xi = ii; cannam@127: xo = io; cannam@127: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { cannam@127: V T7, Tu, TF, TB, T13, TL, TO, TX, TC, Te, TP, Th, TQ, Tk, TW; cannam@127: V T16; cannam@127: { cannam@127: V TH, TU, Tz, Tf, TK, TV, TA, TM, Ta, TN, Td, Tg, Ti, Tj; cannam@127: { cannam@127: V T1, T2, T4, T5, To, Tp, Tr, Ts; cannam@127: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@127: T2 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@127: T4 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@127: T5 = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); cannam@127: To = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@127: Tp = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@127: Tr = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); cannam@127: Ts = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@127: { cannam@127: V T8, TI, Tq, TJ, Tt, T9, Tb, Tc, T3, T6; cannam@127: T8 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@127: TH = VSUB(T1, T2); cannam@127: T3 = VADD(T1, T2); cannam@127: TU = VSUB(T4, T5); cannam@127: T6 = VADD(T4, T5); cannam@127: TI = VSUB(To, Tp); cannam@127: Tq = VADD(To, Tp); cannam@127: TJ = VSUB(Tr, Ts); cannam@127: Tt = VADD(Tr, Ts); cannam@127: T9 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tb = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tc = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); cannam@127: T7 = VSUB(T3, T6); cannam@127: Tz = VADD(T3, T6); cannam@127: Tf = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); cannam@127: TK = VADD(TI, TJ); cannam@127: TV = VSUB(TI, TJ); cannam@127: TA = VADD(Tq, Tt); cannam@127: Tu = VSUB(Tq, Tt); cannam@127: TM = VSUB(T8, T9); cannam@127: Ta = VADD(T8, T9); cannam@127: TN = VSUB(Tb, Tc); cannam@127: Td = VADD(Tb, Tc); cannam@127: Tg = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@127: Ti = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tj = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@127: } cannam@127: } cannam@127: TF = VADD(Tz, TA); cannam@127: TB = VSUB(Tz, TA); cannam@127: T13 = VFNMS(LDK(KP707106781), TK, TH); cannam@127: TL = VFMA(LDK(KP707106781), TK, TH); cannam@127: TO = VFNMS(LDK(KP414213562), TN, TM); cannam@127: TX = VFMA(LDK(KP414213562), TM, TN); cannam@127: TC = VADD(Ta, Td); cannam@127: Te = VSUB(Ta, Td); cannam@127: TP = VSUB(Tf, Tg); cannam@127: Th = VADD(Tf, Tg); cannam@127: TQ = VSUB(Tj, Ti); cannam@127: Tk = VADD(Ti, Tj); cannam@127: TW = VFMA(LDK(KP707106781), TV, TU); cannam@127: T16 = VFNMS(LDK(KP707106781), TV, TU); cannam@127: } cannam@127: { cannam@127: V TY, TR, Tl, TD; cannam@127: TY = VFMA(LDK(KP414213562), TP, TQ); cannam@127: TR = VFNMS(LDK(KP414213562), TQ, TP); cannam@127: Tl = VSUB(Th, Tk); cannam@127: TD = VADD(Th, Tk); cannam@127: { cannam@127: V TS, T17, TZ, T14; cannam@127: TS = VADD(TO, TR); cannam@127: T17 = VSUB(TO, TR); cannam@127: TZ = VSUB(TX, TY); cannam@127: T14 = VADD(TX, TY); cannam@127: { cannam@127: V TE, TG, Tm, Tv; cannam@127: TE = VSUB(TC, TD); cannam@127: TG = VADD(TC, TD); cannam@127: Tm = VADD(Te, Tl); cannam@127: Tv = VSUB(Te, Tl); cannam@127: { cannam@127: V T18, T1a, TT, T11; cannam@127: T18 = VFMA(LDK(KP923879532), T17, T16); cannam@127: T1a = VFNMS(LDK(KP923879532), T17, T16); cannam@127: TT = VFNMS(LDK(KP923879532), TS, TL); cannam@127: T11 = VFMA(LDK(KP923879532), TS, TL); cannam@127: { cannam@127: V T15, T19, T10, T12; cannam@127: T15 = VFNMS(LDK(KP923879532), T14, T13); cannam@127: T19 = VFMA(LDK(KP923879532), T14, T13); cannam@127: T10 = VFNMS(LDK(KP923879532), TZ, TW); cannam@127: T12 = VFMA(LDK(KP923879532), TZ, TW); cannam@127: ST(&(xo[0]), VADD(TF, TG), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 8)]), VSUB(TF, TG), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 4)]), VFMAI(TE, TB), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 12)]), VFNMSI(TE, TB), ovs, &(xo[0])); cannam@127: { cannam@127: V Tw, Ty, Tn, Tx; cannam@127: Tw = VFNMS(LDK(KP707106781), Tv, Tu); cannam@127: Ty = VFMA(LDK(KP707106781), Tv, Tu); cannam@127: Tn = VFNMS(LDK(KP707106781), Tm, T7); cannam@127: Tx = VFMA(LDK(KP707106781), Tm, T7); cannam@127: ST(&(xo[WS(os, 3)]), VFNMSI(T1a, T19), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 13)]), VFMAI(T1a, T19), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 11)]), VFNMSI(T18, T15), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 5)]), VFMAI(T18, T15), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 15)]), VFNMSI(T12, T11), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 1)]), VFMAI(T12, T11), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 9)]), VFMAI(T10, TT), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 7)]), VFNMSI(T10, TT), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 2)]), VFMAI(Ty, Tx), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 14)]), VFNMSI(Ty, Tx), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 10)]), VFMAI(Tw, Tn), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 6)]), VFNMSI(Tw, Tn), ovs, &(xo[0])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 16, XSIMD_STRING("n1bv_16"), {38, 0, 34, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n1bv_16) (planner *p) { cannam@127: X(kdft_register) (p, n1bv_16, &desc); cannam@127: } cannam@127: cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 16 -name n1bv_16 -include n1b.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 72 FP additions, 12 FP multiplications, cannam@127: * (or, 68 additions, 8 multiplications, 4 fused multiply/add), cannam@127: * 30 stack variables, 3 constants, and 32 memory accesses cannam@127: */ cannam@127: #include "n1b.h" cannam@127: cannam@127: static void n1bv_16(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DVK(KP382683432, +0.382683432365089771728459984030398866761344562); cannam@127: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT i; cannam@127: const R *xi; cannam@127: R *xo; cannam@127: xi = ii; cannam@127: xo = io; cannam@127: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { cannam@127: V Tp, T13, Tu, TY, Tm, T14, Tv, TU, T7, T16, Tx, TN, Te, T17, Ty; cannam@127: V TQ; cannam@127: { cannam@127: V Tn, To, TX, Ts, Tt, TW; cannam@127: Tn = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@127: To = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); cannam@127: TX = VADD(Tn, To); cannam@127: Ts = LD(&(xi[0]), ivs, &(xi[0])); cannam@127: Tt = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@127: TW = VADD(Ts, Tt); cannam@127: Tp = VSUB(Tn, To); cannam@127: T13 = VADD(TW, TX); cannam@127: Tu = VSUB(Ts, Tt); cannam@127: TY = VSUB(TW, TX); cannam@127: } cannam@127: { cannam@127: V Ti, TS, Tl, TT; cannam@127: { cannam@127: V Tg, Th, Tj, Tk; cannam@127: Tg = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@127: Th = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@127: Ti = VSUB(Tg, Th); cannam@127: TS = VADD(Tg, Th); cannam@127: Tj = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); cannam@127: Tk = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@127: Tl = VSUB(Tj, Tk); cannam@127: TT = VADD(Tj, Tk); cannam@127: } cannam@127: Tm = VMUL(LDK(KP707106781), VSUB(Ti, Tl)); cannam@127: T14 = VADD(TS, TT); cannam@127: Tv = VMUL(LDK(KP707106781), VADD(Ti, Tl)); cannam@127: TU = VSUB(TS, TT); cannam@127: } cannam@127: { cannam@127: V T3, TL, T6, TM; cannam@127: { cannam@127: V T1, T2, T4, T5; cannam@127: T1 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@127: T2 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@127: T3 = VSUB(T1, T2); cannam@127: TL = VADD(T1, T2); cannam@127: T4 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@127: T5 = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); cannam@127: T6 = VSUB(T4, T5); cannam@127: TM = VADD(T4, T5); cannam@127: } cannam@127: T7 = VFNMS(LDK(KP382683432), T6, VMUL(LDK(KP923879532), T3)); cannam@127: T16 = VADD(TL, TM); cannam@127: Tx = VFMA(LDK(KP382683432), T3, VMUL(LDK(KP923879532), T6)); cannam@127: TN = VSUB(TL, TM); cannam@127: } cannam@127: { cannam@127: V Ta, TO, Td, TP; cannam@127: { cannam@127: V T8, T9, Tb, Tc; cannam@127: T8 = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); cannam@127: T9 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@127: Ta = VSUB(T8, T9); cannam@127: TO = VADD(T8, T9); cannam@127: Tb = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tc = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@127: Td = VSUB(Tb, Tc); cannam@127: TP = VADD(Tb, Tc); cannam@127: } cannam@127: Te = VFMA(LDK(KP923879532), Ta, VMUL(LDK(KP382683432), Td)); cannam@127: T17 = VADD(TO, TP); cannam@127: Ty = VFNMS(LDK(KP382683432), Ta, VMUL(LDK(KP923879532), Td)); cannam@127: TQ = VSUB(TO, TP); cannam@127: } cannam@127: { cannam@127: V T15, T18, T19, T1a; cannam@127: T15 = VSUB(T13, T14); cannam@127: T18 = VBYI(VSUB(T16, T17)); cannam@127: ST(&(xo[WS(os, 12)]), VSUB(T15, T18), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 4)]), VADD(T15, T18), ovs, &(xo[0])); cannam@127: T19 = VADD(T13, T14); cannam@127: T1a = VADD(T16, T17); cannam@127: ST(&(xo[WS(os, 8)]), VSUB(T19, T1a), ovs, &(xo[0])); cannam@127: ST(&(xo[0]), VADD(T19, T1a), ovs, &(xo[0])); cannam@127: } cannam@127: { cannam@127: V TV, T11, T10, T12, TR, TZ; cannam@127: TR = VMUL(LDK(KP707106781), VSUB(TN, TQ)); cannam@127: TV = VBYI(VSUB(TR, TU)); cannam@127: T11 = VBYI(VADD(TU, TR)); cannam@127: TZ = VMUL(LDK(KP707106781), VADD(TN, TQ)); cannam@127: T10 = VSUB(TY, TZ); cannam@127: T12 = VADD(TY, TZ); cannam@127: ST(&(xo[WS(os, 6)]), VADD(TV, T10), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 14)]), VSUB(T12, T11), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 10)]), VSUB(T10, TV), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 2)]), VADD(T11, T12), ovs, &(xo[0])); cannam@127: } cannam@127: { cannam@127: V Tr, TB, TA, TC; cannam@127: { cannam@127: V Tf, Tq, Tw, Tz; cannam@127: Tf = VSUB(T7, Te); cannam@127: Tq = VSUB(Tm, Tp); cannam@127: Tr = VBYI(VSUB(Tf, Tq)); cannam@127: TB = VBYI(VADD(Tq, Tf)); cannam@127: Tw = VSUB(Tu, Tv); cannam@127: Tz = VSUB(Tx, Ty); cannam@127: TA = VSUB(Tw, Tz); cannam@127: TC = VADD(Tw, Tz); cannam@127: } cannam@127: ST(&(xo[WS(os, 5)]), VADD(Tr, TA), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 13)]), VSUB(TC, TB), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 11)]), VSUB(TA, Tr), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 3)]), VADD(TB, TC), ovs, &(xo[WS(os, 1)])); cannam@127: } cannam@127: { cannam@127: V TF, TJ, TI, TK; cannam@127: { cannam@127: V TD, TE, TG, TH; cannam@127: TD = VADD(Tu, Tv); cannam@127: TE = VADD(T7, Te); cannam@127: TF = VADD(TD, TE); cannam@127: TJ = VSUB(TD, TE); cannam@127: TG = VADD(Tp, Tm); cannam@127: TH = VADD(Tx, Ty); cannam@127: TI = VBYI(VADD(TG, TH)); cannam@127: TK = VBYI(VSUB(TH, TG)); cannam@127: } cannam@127: ST(&(xo[WS(os, 15)]), VSUB(TF, TI), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 7)]), VADD(TJ, TK), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 1)]), VADD(TF, TI), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 9)]), VSUB(TJ, TK), ovs, &(xo[WS(os, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 16, XSIMD_STRING("n1bv_16"), {68, 8, 4, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n1bv_16) (planner *p) { cannam@127: X(kdft_register) (p, n1bv_16, &desc); cannam@127: } cannam@127: cannam@127: #endif /* HAVE_FMA */