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 07:36:59 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_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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 42 FP additions, 22 FP multiplications, Chris@10: * (or, 24 additions, 4 multiplications, 18 fused multiply/add), Chris@10: * 43 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_10(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ii; Chris@10: xo = io; Chris@10: 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)) { Chris@10: V Tb, Tr, T3, Ts, T6, Tw, Tg, Tt, T9, Tc, T1, T2; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: { Chris@10: V T4, T5, Te, Tf, T7, T8; Chris@10: T4 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Te = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tb = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: Tr = VADD(T1, T2); Chris@10: T3 = VSUB(T1, T2); Chris@10: Ts = VADD(T4, T5); Chris@10: T6 = VSUB(T4, T5); Chris@10: Tw = VADD(Te, Tf); Chris@10: Tg = VSUB(Te, Tf); Chris@10: Tt = VADD(T7, T8); Chris@10: T9 = VSUB(T7, T8); Chris@10: Tc = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: } Chris@10: { Chris@10: V TD, Tu, Tm, Ta, Td, Tv; Chris@10: TD = VSUB(Ts, Tt); Chris@10: Tu = VADD(Ts, Tt); Chris@10: Tm = VSUB(T6, T9); Chris@10: Ta = VADD(T6, T9); Chris@10: Td = VSUB(Tb, Tc); Chris@10: Tv = VADD(Tb, Tc); Chris@10: { Chris@10: V TC, Tx, Tn, Th; Chris@10: TC = VSUB(Tv, Tw); Chris@10: Tx = VADD(Tv, Tw); Chris@10: Tn = VSUB(Td, Tg); Chris@10: Th = VADD(Td, Tg); Chris@10: { Chris@10: V Ty, TA, TE, TG, Ti, Tk, To, Tq, Tz, Tj; Chris@10: Ty = VADD(Tu, Tx); Chris@10: TA = VSUB(Tu, Tx); Chris@10: TE = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TD, TC)); Chris@10: TG = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TC, TD)); Chris@10: Ti = VADD(Ta, Th); Chris@10: Tk = VSUB(Ta, Th); Chris@10: To = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tn, Tm)); Chris@10: Tq = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tm, Tn)); Chris@10: Tz = VFNMS(LDK(KP250000000), Ty, Tr); Chris@10: ST(&(xo[0]), VADD(Tr, Ty), ovs, &(xo[0])); Chris@10: Tj = VFNMS(LDK(KP250000000), Ti, T3); Chris@10: ST(&(xo[WS(os, 5)]), VADD(T3, Ti), ovs, &(xo[WS(os, 1)])); Chris@10: { Chris@10: V TB, TF, Tl, Tp; Chris@10: TB = VFNMS(LDK(KP559016994), TA, Tz); Chris@10: TF = VFMA(LDK(KP559016994), TA, Tz); Chris@10: Tl = VFMA(LDK(KP559016994), Tk, Tj); Chris@10: Tp = VFNMS(LDK(KP559016994), Tk, Tj); Chris@10: ST(&(xo[WS(os, 4)]), VFNMSI(TG, TF), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VFMAI(TG, TF), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 8)]), VFMAI(TE, TB), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 2)]), VFNMSI(TE, TB), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 3)]), VFMAI(Tq, Tp), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VFNMSI(Tq, Tp), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VFNMSI(To, Tl), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VFMAI(To, Tl), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 10, XSIMD_STRING("n1bv_10"), {24, 4, 18, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_10) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_10, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 10 -name n1bv_10 -include n1b.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 42 FP additions, 12 FP multiplications, Chris@10: * (or, 36 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 33 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_10(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ii; Chris@10: xo = io; Chris@10: 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)) { Chris@10: V Tl, Ty, T7, Te, Tw, Tt, Tz, TA, TB, Tg, Th, Tm, Tj, Tk; Chris@10: Tj = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: Tk = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tl = VSUB(Tj, Tk); Chris@10: Ty = VADD(Tj, Tk); Chris@10: { Chris@10: V T3, Tr, Td, Tv, T6, Ts, Ta, Tu; Chris@10: { Chris@10: V T1, T2, Tb, Tc; Chris@10: T1 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = VSUB(T1, T2); Chris@10: Tr = VADD(T1, T2); Chris@10: Tb = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = VSUB(Tb, Tc); Chris@10: Tv = VADD(Tb, Tc); Chris@10: } Chris@10: { Chris@10: V T4, T5, T8, T9; Chris@10: T4 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T6 = VSUB(T4, T5); Chris@10: Ts = VADD(T4, T5); Chris@10: T8 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T9 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ta = VSUB(T8, T9); Chris@10: Tu = VADD(T8, T9); Chris@10: } Chris@10: T7 = VSUB(T3, T6); Chris@10: Te = VSUB(Ta, Td); Chris@10: Tw = VSUB(Tu, Tv); Chris@10: Tt = VSUB(Tr, Ts); Chris@10: Tz = VADD(Tr, Ts); Chris@10: TA = VADD(Tu, Tv); Chris@10: TB = VADD(Tz, TA); Chris@10: Tg = VADD(T3, T6); Chris@10: Th = VADD(Ta, Td); Chris@10: Tm = VADD(Tg, Th); Chris@10: } Chris@10: ST(&(xo[WS(os, 5)]), VADD(Tl, Tm), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[0]), VADD(Ty, TB), ovs, &(xo[0])); Chris@10: { Chris@10: V Tf, Tq, To, Tp, Ti, Tn; Chris@10: Tf = VBYI(VFMA(LDK(KP951056516), T7, VMUL(LDK(KP587785252), Te))); Chris@10: Tq = VBYI(VFNMS(LDK(KP951056516), Te, VMUL(LDK(KP587785252), T7))); Chris@10: Ti = VMUL(LDK(KP559016994), VSUB(Tg, Th)); Chris@10: Tn = VFNMS(LDK(KP250000000), Tm, Tl); Chris@10: To = VADD(Ti, Tn); Chris@10: Tp = VSUB(Tn, Ti); Chris@10: ST(&(xo[WS(os, 1)]), VADD(Tf, To), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VADD(Tq, Tp), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VSUB(To, Tf), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 3)]), VSUB(Tp, Tq), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: { Chris@10: V Tx, TG, TE, TF, TC, TD; Chris@10: Tx = VBYI(VFNMS(LDK(KP951056516), Tw, VMUL(LDK(KP587785252), Tt))); Chris@10: TG = VBYI(VFMA(LDK(KP951056516), Tt, VMUL(LDK(KP587785252), Tw))); Chris@10: TC = VFNMS(LDK(KP250000000), TB, Ty); Chris@10: TD = VMUL(LDK(KP559016994), VSUB(Tz, TA)); Chris@10: TE = VSUB(TC, TD); Chris@10: TF = VADD(TD, TC); Chris@10: ST(&(xo[WS(os, 2)]), VADD(Tx, TE), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VADD(TG, TF), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(TE, Tx), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VSUB(TF, TG), ovs, &(xo[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 10, XSIMD_STRING("n1bv_10"), {36, 6, 6, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_10) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_10, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */