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 16:39:05 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 10 -name n1bv_10 -include n1b.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 42 FP additions, 22 FP multiplications, cannam@127: * (or, 24 additions, 4 multiplications, 18 fused multiply/add), cannam@127: * 43 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "n1b.h" cannam@127: cannam@127: static void n1bv_10(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(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); 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(20, is), MAKE_VOLATILE_STRIDE(20, os)) { cannam@127: V Tb, Tr, T3, Ts, T6, Tw, Tg, Tt, T9, Tc, T1, T2; cannam@127: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@127: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@127: { cannam@127: V T4, T5, Te, Tf, T7, T8; cannam@127: T4 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@127: T5 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@127: Te = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@127: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@127: T7 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@127: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tb = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@127: Tr = VADD(T1, T2); cannam@127: T3 = VSUB(T1, T2); cannam@127: Ts = VADD(T4, T5); cannam@127: T6 = VSUB(T4, T5); cannam@127: Tw = VADD(Te, Tf); cannam@127: Tg = VSUB(Te, Tf); cannam@127: Tt = VADD(T7, T8); cannam@127: T9 = VSUB(T7, T8); cannam@127: Tc = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@127: } cannam@127: { cannam@127: V TD, Tu, Tm, Ta, Td, Tv; cannam@127: TD = VSUB(Ts, Tt); cannam@127: Tu = VADD(Ts, Tt); cannam@127: Tm = VSUB(T6, T9); cannam@127: Ta = VADD(T6, T9); cannam@127: Td = VSUB(Tb, Tc); cannam@127: Tv = VADD(Tb, Tc); cannam@127: { cannam@127: V TC, Tx, Tn, Th; cannam@127: TC = VSUB(Tv, Tw); cannam@127: Tx = VADD(Tv, Tw); cannam@127: Tn = VSUB(Td, Tg); cannam@127: Th = VADD(Td, Tg); cannam@127: { cannam@127: V Ty, TA, TE, TG, Ti, Tk, To, Tq, Tz, Tj; cannam@127: Ty = VADD(Tu, Tx); cannam@127: TA = VSUB(Tu, Tx); cannam@127: TE = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TD, TC)); cannam@127: TG = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TC, TD)); cannam@127: Ti = VADD(Ta, Th); cannam@127: Tk = VSUB(Ta, Th); cannam@127: To = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tn, Tm)); cannam@127: Tq = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tm, Tn)); cannam@127: Tz = VFNMS(LDK(KP250000000), Ty, Tr); cannam@127: ST(&(xo[0]), VADD(Tr, Ty), ovs, &(xo[0])); cannam@127: Tj = VFNMS(LDK(KP250000000), Ti, T3); cannam@127: ST(&(xo[WS(os, 5)]), VADD(T3, Ti), ovs, &(xo[WS(os, 1)])); cannam@127: { cannam@127: V TB, TF, Tl, Tp; cannam@127: TB = VFNMS(LDK(KP559016994), TA, Tz); cannam@127: TF = VFMA(LDK(KP559016994), TA, Tz); cannam@127: Tl = VFMA(LDK(KP559016994), Tk, Tj); cannam@127: Tp = VFNMS(LDK(KP559016994), Tk, Tj); cannam@127: ST(&(xo[WS(os, 4)]), VFNMSI(TG, TF), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 6)]), VFMAI(TG, TF), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 8)]), VFMAI(TE, TB), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 2)]), VFNMSI(TE, TB), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 3)]), VFMAI(Tq, Tp), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 7)]), VFNMSI(Tq, Tp), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 9)]), VFNMSI(To, Tl), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 1)]), VFMAI(To, Tl), ovs, &(xo[WS(os, 1)])); 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 = { 10, XSIMD_STRING("n1bv_10"), {24, 4, 18, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n1bv_10) (planner *p) { cannam@127: X(kdft_register) (p, n1bv_10, &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 10 -name n1bv_10 -include n1b.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 42 FP additions, 12 FP multiplications, cannam@127: * (or, 36 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 33 stack variables, 4 constants, and 20 memory accesses cannam@127: */ cannam@127: #include "n1b.h" cannam@127: cannam@127: static void n1bv_10(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(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); 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(20, is), MAKE_VOLATILE_STRIDE(20, os)) { cannam@127: V Tl, Ty, T7, Te, Tw, Tt, Tz, TA, TB, Tg, Th, Tm, Tj, Tk; cannam@127: Tj = LD(&(xi[0]), ivs, &(xi[0])); cannam@127: Tk = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@127: Tl = VSUB(Tj, Tk); cannam@127: Ty = VADD(Tj, Tk); cannam@127: { cannam@127: V T3, Tr, Td, Tv, T6, Ts, Ta, Tu; cannam@127: { cannam@127: V T1, T2, Tb, Tc; cannam@127: T1 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@127: T2 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@127: T3 = VSUB(T1, T2); cannam@127: Tr = VADD(T1, T2); cannam@127: Tb = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@127: Tc = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@127: Td = VSUB(Tb, Tc); cannam@127: Tv = VADD(Tb, Tc); cannam@127: } cannam@127: { cannam@127: V T4, T5, T8, T9; cannam@127: T4 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@127: T5 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@127: T6 = VSUB(T4, T5); cannam@127: Ts = VADD(T4, T5); cannam@127: T8 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@127: T9 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@127: Ta = VSUB(T8, T9); cannam@127: Tu = VADD(T8, T9); cannam@127: } cannam@127: T7 = VSUB(T3, T6); cannam@127: Te = VSUB(Ta, Td); cannam@127: Tw = VSUB(Tu, Tv); cannam@127: Tt = VSUB(Tr, Ts); cannam@127: Tz = VADD(Tr, Ts); cannam@127: TA = VADD(Tu, Tv); cannam@127: TB = VADD(Tz, TA); cannam@127: Tg = VADD(T3, T6); cannam@127: Th = VADD(Ta, Td); cannam@127: Tm = VADD(Tg, Th); cannam@127: } cannam@127: ST(&(xo[WS(os, 5)]), VADD(Tl, Tm), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[0]), VADD(Ty, TB), ovs, &(xo[0])); cannam@127: { cannam@127: V Tf, Tq, To, Tp, Ti, Tn; cannam@127: Tf = VBYI(VFMA(LDK(KP951056516), T7, VMUL(LDK(KP587785252), Te))); cannam@127: Tq = VBYI(VFNMS(LDK(KP951056516), Te, VMUL(LDK(KP587785252), T7))); cannam@127: Ti = VMUL(LDK(KP559016994), VSUB(Tg, Th)); cannam@127: Tn = VFNMS(LDK(KP250000000), Tm, Tl); cannam@127: To = VADD(Ti, Tn); cannam@127: Tp = VSUB(Tn, Ti); cannam@127: ST(&(xo[WS(os, 1)]), VADD(Tf, To), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 7)]), VADD(Tq, Tp), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 9)]), VSUB(To, Tf), ovs, &(xo[WS(os, 1)])); cannam@127: ST(&(xo[WS(os, 3)]), VSUB(Tp, Tq), ovs, &(xo[WS(os, 1)])); cannam@127: } cannam@127: { cannam@127: V Tx, TG, TE, TF, TC, TD; cannam@127: Tx = VBYI(VFNMS(LDK(KP951056516), Tw, VMUL(LDK(KP587785252), Tt))); cannam@127: TG = VBYI(VFMA(LDK(KP951056516), Tt, VMUL(LDK(KP587785252), Tw))); cannam@127: TC = VFNMS(LDK(KP250000000), TB, Ty); cannam@127: TD = VMUL(LDK(KP559016994), VSUB(Tz, TA)); cannam@127: TE = VSUB(TC, TD); cannam@127: TF = VADD(TD, TC); cannam@127: ST(&(xo[WS(os, 2)]), VADD(Tx, TE), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 6)]), VADD(TG, TF), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 8)]), VSUB(TE, Tx), ovs, &(xo[0])); cannam@127: ST(&(xo[WS(os, 4)]), VSUB(TF, TG), ovs, &(xo[0])); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 10, XSIMD_STRING("n1bv_10"), {36, 6, 6, 0}, &GENUS, 0, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n1bv_10) (planner *p) { cannam@127: X(kdft_register) (p, n1bv_10, &desc); cannam@127: } cannam@127: cannam@127: #endif /* HAVE_FMA */