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 07:37:47 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.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 4 -name n2sv_4 -with-ostride 1 -include n2s.h -store-multiple 4 */ Chris@10: Chris@10: /* Chris@10: * This function contains 16 FP additions, 0 FP multiplications, Chris@10: * (or, 16 additions, 0 multiplications, 0 fused multiply/add), Chris@10: * 25 stack variables, 0 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "n2s.h" Chris@10: Chris@10: static void n2sv_4(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - (2 * VL), ri = ri + ((2 * VL) * ivs), ii = ii + ((2 * VL) * ivs), ro = ro + ((2 * VL) * ovs), io = io + ((2 * VL) * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@10: V T1, T2, T7, T8, T4, T5, Tc, Td; Chris@10: T1 = LD(&(ri[0]), ivs, &(ri[0])); Chris@10: T2 = LD(&(ri[WS(is, 2)]), ivs, &(ri[0])); Chris@10: T7 = LD(&(ii[0]), ivs, &(ii[0])); Chris@10: T8 = LD(&(ii[WS(is, 2)]), ivs, &(ii[0])); Chris@10: T4 = LD(&(ri[WS(is, 1)]), ivs, &(ri[WS(is, 1)])); Chris@10: T5 = LD(&(ri[WS(is, 3)]), ivs, &(ri[WS(is, 1)])); Chris@10: Tc = LD(&(ii[WS(is, 1)]), ivs, &(ii[WS(is, 1)])); Chris@10: Td = LD(&(ii[WS(is, 3)]), ivs, &(ii[WS(is, 1)])); Chris@10: { Chris@10: V T3, Tb, T9, Tf, T6, Ta, Te, Tg; Chris@10: T3 = VADD(T1, T2); Chris@10: Tb = VSUB(T1, T2); Chris@10: T9 = VSUB(T7, T8); Chris@10: Tf = VADD(T7, T8); Chris@10: T6 = VADD(T4, T5); Chris@10: Ta = VSUB(T4, T5); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tg = VADD(Tc, Td); Chris@10: { Chris@10: V Th, Ti, Tj, Tk; Chris@10: Th = VADD(Ta, T9); Chris@10: STM4(&(io[3]), Th, ovs, &(io[1])); Chris@10: Ti = VSUB(T9, Ta); Chris@10: STM4(&(io[1]), Ti, ovs, &(io[1])); Chris@10: Tj = VADD(T3, T6); Chris@10: STM4(&(ro[0]), Tj, ovs, &(ro[0])); Chris@10: Tk = VSUB(T3, T6); Chris@10: STM4(&(ro[2]), Tk, ovs, &(ro[0])); Chris@10: { Chris@10: V Tl, Tm, Tn, To; Chris@10: Tl = VADD(Tf, Tg); Chris@10: STM4(&(io[0]), Tl, ovs, &(io[0])); Chris@10: Tm = VSUB(Tf, Tg); Chris@10: STM4(&(io[2]), Tm, ovs, &(io[0])); Chris@10: STN4(&(io[0]), Tl, Ti, Tm, Th, ovs); Chris@10: Tn = VSUB(Tb, Te); Chris@10: STM4(&(ro[3]), Tn, ovs, &(ro[1])); Chris@10: To = VADD(Tb, Te); Chris@10: STM4(&(ro[1]), To, ovs, &(ro[1])); Chris@10: STN4(&(ro[0]), Tj, To, Tk, Tn, ovs); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 4, XSIMD_STRING("n2sv_4"), {16, 0, 0, 0}, &GENUS, 0, 1, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2sv_4) (planner *p) { Chris@10: X(kdft_register) (p, n2sv_4, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw.native -simd -compact -variables 4 -pipeline-latency 8 -n 4 -name n2sv_4 -with-ostride 1 -include n2s.h -store-multiple 4 */ Chris@10: Chris@10: /* Chris@10: * This function contains 16 FP additions, 0 FP multiplications, Chris@10: * (or, 16 additions, 0 multiplications, 0 fused multiply/add), Chris@10: * 17 stack variables, 0 constants, and 18 memory accesses Chris@10: */ Chris@10: #include "n2s.h" Chris@10: Chris@10: static void n2sv_4(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: { Chris@10: INT i; Chris@10: for (i = v; i > 0; i = i - (2 * VL), ri = ri + ((2 * VL) * ivs), ii = ii + ((2 * VL) * ivs), ro = ro + ((2 * VL) * ovs), io = io + ((2 * VL) * ovs), MAKE_VOLATILE_STRIDE(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@10: V T3, Tb, T9, Tf, T6, Ta, Te, Tg; Chris@10: { Chris@10: V T1, T2, T7, T8; Chris@10: T1 = LD(&(ri[0]), ivs, &(ri[0])); Chris@10: T2 = LD(&(ri[WS(is, 2)]), ivs, &(ri[0])); Chris@10: T3 = VADD(T1, T2); Chris@10: Tb = VSUB(T1, T2); Chris@10: T7 = LD(&(ii[0]), ivs, &(ii[0])); Chris@10: T8 = LD(&(ii[WS(is, 2)]), ivs, &(ii[0])); Chris@10: T9 = VSUB(T7, T8); Chris@10: Tf = VADD(T7, T8); Chris@10: } Chris@10: { Chris@10: V T4, T5, Tc, Td; Chris@10: T4 = LD(&(ri[WS(is, 1)]), ivs, &(ri[WS(is, 1)])); Chris@10: T5 = LD(&(ri[WS(is, 3)]), ivs, &(ri[WS(is, 1)])); Chris@10: T6 = VADD(T4, T5); Chris@10: Ta = VSUB(T4, T5); Chris@10: Tc = LD(&(ii[WS(is, 1)]), ivs, &(ii[WS(is, 1)])); Chris@10: Td = LD(&(ii[WS(is, 3)]), ivs, &(ii[WS(is, 1)])); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tg = VADD(Tc, Td); Chris@10: } Chris@10: { Chris@10: V Th, Ti, Tj, Tk; Chris@10: Th = VSUB(T3, T6); Chris@10: STM4(&(ro[2]), Th, ovs, &(ro[0])); Chris@10: Ti = VSUB(Tf, Tg); Chris@10: STM4(&(io[2]), Ti, ovs, &(io[0])); Chris@10: Tj = VADD(T3, T6); Chris@10: STM4(&(ro[0]), Tj, ovs, &(ro[0])); Chris@10: Tk = VADD(Tf, Tg); Chris@10: STM4(&(io[0]), Tk, ovs, &(io[0])); Chris@10: { Chris@10: V Tl, Tm, Tn, To; Chris@10: Tl = VSUB(T9, Ta); Chris@10: STM4(&(io[1]), Tl, ovs, &(io[1])); Chris@10: Tm = VADD(Tb, Te); Chris@10: STM4(&(ro[1]), Tm, ovs, &(ro[1])); Chris@10: Tn = VADD(Ta, T9); Chris@10: STM4(&(io[3]), Tn, ovs, &(io[1])); Chris@10: STN4(&(io[0]), Tk, Tl, Ti, Tn, ovs); Chris@10: To = VSUB(Tb, Te); Chris@10: STM4(&(ro[3]), To, ovs, &(ro[1])); Chris@10: STN4(&(ro[0]), Tj, Tm, Th, To, ovs); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 4, XSIMD_STRING("n2sv_4"), {16, 0, 0, 0}, &GENUS, 0, 1, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2sv_4) (planner *p) { Chris@10: X(kdft_register) (p, n2sv_4, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */