Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:40:34 EDT 2016 */ Chris@42: Chris@42: #include "codelet-dft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 8 -name n2bv_8 -with-ostride 2 -include n2b.h -store-multiple 2 */ Chris@42: Chris@42: /* Chris@42: * This function contains 26 FP additions, 10 FP multiplications, Chris@42: * (or, 16 additions, 0 multiplications, 10 fused multiply/add), Chris@42: * 38 stack variables, 1 constants, and 20 memory accesses Chris@42: */ Chris@42: #include "n2b.h" Chris@42: Chris@42: static void n2bv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@42: { Chris@42: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@42: { Chris@42: INT i; Chris@42: const R *xi; Chris@42: R *xo; Chris@42: xi = ii; Chris@42: xo = io; Chris@42: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@42: V T1, T2, Tc, Td, T4, T5, T7, T8; Chris@42: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@42: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@42: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@42: Td = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@42: T4 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@42: T5 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@42: T7 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@42: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@42: { Chris@42: V T3, Tj, Te, Tk, T6, Tm, T9, Tn, Tp, Tl; Chris@42: T3 = VSUB(T1, T2); Chris@42: Tj = VADD(T1, T2); Chris@42: Te = VSUB(Tc, Td); Chris@42: Tk = VADD(Tc, Td); Chris@42: T6 = VSUB(T4, T5); Chris@42: Tm = VADD(T4, T5); Chris@42: T9 = VSUB(T7, T8); Chris@42: Tn = VADD(T7, T8); Chris@42: Tp = VADD(Tj, Tk); Chris@42: Tl = VSUB(Tj, Tk); Chris@42: { Chris@42: V Tq, To, Ta, Tf; Chris@42: Tq = VADD(Tm, Tn); Chris@42: To = VSUB(Tm, Tn); Chris@42: Ta = VADD(T6, T9); Chris@42: Tf = VSUB(T6, T9); Chris@42: { Chris@42: V Tr, Ts, Tt, Tu, Tg, Ti, Tb, Th; Chris@42: Tr = VFMAI(To, Tl); Chris@42: STM2(&(xo[4]), Tr, ovs, &(xo[0])); Chris@42: Ts = VFNMSI(To, Tl); Chris@42: STM2(&(xo[12]), Ts, ovs, &(xo[0])); Chris@42: Tt = VADD(Tp, Tq); Chris@42: STM2(&(xo[0]), Tt, ovs, &(xo[0])); Chris@42: Tu = VSUB(Tp, Tq); Chris@42: STM2(&(xo[8]), Tu, ovs, &(xo[0])); Chris@42: Tg = VFNMS(LDK(KP707106781), Tf, Te); Chris@42: Ti = VFMA(LDK(KP707106781), Tf, Te); Chris@42: Tb = VFNMS(LDK(KP707106781), Ta, T3); Chris@42: Th = VFMA(LDK(KP707106781), Ta, T3); Chris@42: { Chris@42: V Tv, Tw, Tx, Ty; Chris@42: Tv = VFNMSI(Ti, Th); Chris@42: STM2(&(xo[14]), Tv, ovs, &(xo[2])); Chris@42: STN2(&(xo[12]), Ts, Tv, ovs); Chris@42: Tw = VFMAI(Ti, Th); Chris@42: STM2(&(xo[2]), Tw, ovs, &(xo[2])); Chris@42: STN2(&(xo[0]), Tt, Tw, ovs); Chris@42: Tx = VFMAI(Tg, Tb); Chris@42: STM2(&(xo[10]), Tx, ovs, &(xo[2])); Chris@42: STN2(&(xo[8]), Tu, Tx, ovs); Chris@42: Ty = VFNMSI(Tg, Tb); Chris@42: STM2(&(xo[6]), Ty, ovs, &(xo[2])); Chris@42: STN2(&(xo[4]), Tr, Ty, ovs); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const kdft_desc desc = { 8, XSIMD_STRING("n2bv_8"), {16, 0, 10, 0}, &GENUS, 0, 2, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_n2bv_8) (planner *p) { Chris@42: X(kdft_register) (p, n2bv_8, &desc); Chris@42: } Chris@42: Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 8 -name n2bv_8 -with-ostride 2 -include n2b.h -store-multiple 2 */ Chris@42: Chris@42: /* Chris@42: * This function contains 26 FP additions, 2 FP multiplications, Chris@42: * (or, 26 additions, 2 multiplications, 0 fused multiply/add), Chris@42: * 24 stack variables, 1 constants, and 20 memory accesses Chris@42: */ Chris@42: #include "n2b.h" Chris@42: Chris@42: static void n2bv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@42: { Chris@42: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@42: { Chris@42: INT i; Chris@42: const R *xi; Chris@42: R *xo; Chris@42: xi = ii; Chris@42: xo = io; Chris@42: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@42: V Ta, Tk, Te, Tj, T7, Tn, Tf, Tm, Tr, Tu; Chris@42: { Chris@42: V T8, T9, Tc, Td; Chris@42: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@42: T9 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@42: Ta = VSUB(T8, T9); Chris@42: Tk = VADD(T8, T9); Chris@42: Tc = LD(&(xi[0]), ivs, &(xi[0])); Chris@42: Td = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@42: Te = VSUB(Tc, Td); Chris@42: Tj = VADD(Tc, Td); Chris@42: { Chris@42: V T1, T2, T3, T4, T5, T6; Chris@42: T1 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@42: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@42: T3 = VSUB(T1, T2); Chris@42: T4 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@42: T5 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@42: T6 = VSUB(T4, T5); Chris@42: T7 = VMUL(LDK(KP707106781), VSUB(T3, T6)); Chris@42: Tn = VADD(T4, T5); Chris@42: Tf = VMUL(LDK(KP707106781), VADD(T3, T6)); Chris@42: Tm = VADD(T1, T2); Chris@42: } Chris@42: } Chris@42: { Chris@42: V Ts, Tb, Tg, Tp, Tq, Tt; Chris@42: Tb = VBYI(VSUB(T7, Ta)); Chris@42: Tg = VSUB(Te, Tf); Chris@42: Tr = VADD(Tb, Tg); Chris@42: STM2(&(xo[6]), Tr, ovs, &(xo[2])); Chris@42: Ts = VSUB(Tg, Tb); Chris@42: STM2(&(xo[10]), Ts, ovs, &(xo[2])); Chris@42: Tp = VADD(Tj, Tk); Chris@42: Tq = VADD(Tm, Tn); Chris@42: Tt = VSUB(Tp, Tq); Chris@42: STM2(&(xo[8]), Tt, ovs, &(xo[0])); Chris@42: STN2(&(xo[8]), Tt, Ts, ovs); Chris@42: Tu = VADD(Tp, Tq); Chris@42: STM2(&(xo[0]), Tu, ovs, &(xo[0])); Chris@42: } Chris@42: { Chris@42: V Tw, Th, Ti, Tv; Chris@42: Th = VBYI(VADD(Ta, T7)); Chris@42: Ti = VADD(Te, Tf); Chris@42: Tv = VADD(Th, Ti); Chris@42: STM2(&(xo[2]), Tv, ovs, &(xo[2])); Chris@42: STN2(&(xo[0]), Tu, Tv, ovs); Chris@42: Tw = VSUB(Ti, Th); Chris@42: STM2(&(xo[14]), Tw, ovs, &(xo[2])); Chris@42: { Chris@42: V Tl, To, Tx, Ty; Chris@42: Tl = VSUB(Tj, Tk); Chris@42: To = VBYI(VSUB(Tm, Tn)); Chris@42: Tx = VSUB(Tl, To); Chris@42: STM2(&(xo[12]), Tx, ovs, &(xo[0])); Chris@42: STN2(&(xo[12]), Tx, Tw, ovs); Chris@42: Ty = VADD(Tl, To); Chris@42: STM2(&(xo[4]), Ty, ovs, &(xo[0])); Chris@42: STN2(&(xo[4]), Ty, Tr, ovs); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const kdft_desc desc = { 8, XSIMD_STRING("n2bv_8"), {26, 2, 0, 0}, &GENUS, 0, 2, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_n2bv_8) (planner *p) { Chris@42: X(kdft_register) (p, n2bv_8, &desc); Chris@42: } Chris@42: Chris@42: #endif /* HAVE_FMA */