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 07:37:29 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 8 -name n2bv_8 -with-ostride 2 -include n2b.h -store-multiple 2 */ Chris@10: Chris@10: /* Chris@10: * This function contains 26 FP additions, 10 FP multiplications, Chris@10: * (or, 16 additions, 0 multiplications, 10 fused multiply/add), Chris@10: * 38 stack variables, 1 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "n2b.h" Chris@10: Chris@10: 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@10: { Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); 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(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@10: V T1, T2, Tc, Td, T4, T5, T7, T8; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Td = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: T4 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T5 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: { Chris@10: V T3, Tj, Te, Tk, T6, Tm, T9, Tn, Tp, Tl; Chris@10: T3 = VSUB(T1, T2); Chris@10: Tj = VADD(T1, T2); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tk = VADD(Tc, Td); Chris@10: T6 = VSUB(T4, T5); Chris@10: Tm = VADD(T4, T5); Chris@10: T9 = VSUB(T7, T8); Chris@10: Tn = VADD(T7, T8); Chris@10: Tp = VADD(Tj, Tk); Chris@10: Tl = VSUB(Tj, Tk); Chris@10: { Chris@10: V Tq, To, Ta, Tf; Chris@10: Tq = VADD(Tm, Tn); Chris@10: To = VSUB(Tm, Tn); Chris@10: Ta = VADD(T6, T9); Chris@10: Tf = VSUB(T6, T9); Chris@10: { Chris@10: V Tr, Ts, Tt, Tu, Tg, Ti, Tb, Th; Chris@10: Tr = VFMAI(To, Tl); Chris@10: STM2(&(xo[4]), Tr, ovs, &(xo[0])); Chris@10: Ts = VFNMSI(To, Tl); Chris@10: STM2(&(xo[12]), Ts, ovs, &(xo[0])); Chris@10: Tt = VADD(Tp, Tq); Chris@10: STM2(&(xo[0]), Tt, ovs, &(xo[0])); Chris@10: Tu = VSUB(Tp, Tq); Chris@10: STM2(&(xo[8]), Tu, ovs, &(xo[0])); Chris@10: Tg = VFNMS(LDK(KP707106781), Tf, Te); Chris@10: Ti = VFMA(LDK(KP707106781), Tf, Te); Chris@10: Tb = VFNMS(LDK(KP707106781), Ta, T3); Chris@10: Th = VFMA(LDK(KP707106781), Ta, T3); Chris@10: { Chris@10: V Tv, Tw, Tx, Ty; Chris@10: Tv = VFNMSI(Ti, Th); Chris@10: STM2(&(xo[14]), Tv, ovs, &(xo[2])); Chris@10: STN2(&(xo[12]), Ts, Tv, ovs); Chris@10: Tw = VFMAI(Ti, Th); Chris@10: STM2(&(xo[2]), Tw, ovs, &(xo[2])); Chris@10: STN2(&(xo[0]), Tt, Tw, ovs); Chris@10: Tx = VFMAI(Tg, Tb); Chris@10: STM2(&(xo[10]), Tx, ovs, &(xo[2])); Chris@10: STN2(&(xo[8]), Tu, Tx, ovs); Chris@10: Ty = VFNMSI(Tg, Tb); Chris@10: STM2(&(xo[6]), Ty, ovs, &(xo[2])); Chris@10: STN2(&(xo[4]), Tr, Ty, ovs); 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 = { 8, XSIMD_STRING("n2bv_8"), {16, 0, 10, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2bv_8) (planner *p) { Chris@10: X(kdft_register) (p, n2bv_8, &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 8 -name n2bv_8 -with-ostride 2 -include n2b.h -store-multiple 2 */ Chris@10: Chris@10: /* Chris@10: * This function contains 26 FP additions, 2 FP multiplications, Chris@10: * (or, 26 additions, 2 multiplications, 0 fused multiply/add), Chris@10: * 24 stack variables, 1 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "n2b.h" Chris@10: Chris@10: 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@10: { Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); 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(16, is), MAKE_VOLATILE_STRIDE(16, os)) { Chris@10: V Ta, Tk, Te, Tj, T7, Tn, Tf, Tm, Tr, Tu; Chris@10: { Chris@10: V T8, T9, Tc, Td; Chris@10: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: T9 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Ta = VSUB(T8, T9); Chris@10: Tk = VADD(T8, T9); Chris@10: Tc = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: Td = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: Te = VSUB(Tc, Td); Chris@10: Tj = VADD(Tc, Td); Chris@10: { Chris@10: V T1, T2, T3, T4, T5, T6; Chris@10: T1 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = VSUB(T1, T2); Chris@10: T4 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T5 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T6 = VSUB(T4, T5); Chris@10: T7 = VMUL(LDK(KP707106781), VSUB(T3, T6)); Chris@10: Tn = VADD(T4, T5); Chris@10: Tf = VMUL(LDK(KP707106781), VADD(T3, T6)); Chris@10: Tm = VADD(T1, T2); Chris@10: } Chris@10: } Chris@10: { Chris@10: V Ts, Tb, Tg, Tp, Tq, Tt; Chris@10: Tb = VBYI(VSUB(T7, Ta)); Chris@10: Tg = VSUB(Te, Tf); Chris@10: Tr = VADD(Tb, Tg); Chris@10: STM2(&(xo[6]), Tr, ovs, &(xo[2])); Chris@10: Ts = VSUB(Tg, Tb); Chris@10: STM2(&(xo[10]), Ts, ovs, &(xo[2])); Chris@10: Tp = VADD(Tj, Tk); Chris@10: Tq = VADD(Tm, Tn); Chris@10: Tt = VSUB(Tp, Tq); Chris@10: STM2(&(xo[8]), Tt, ovs, &(xo[0])); Chris@10: STN2(&(xo[8]), Tt, Ts, ovs); Chris@10: Tu = VADD(Tp, Tq); Chris@10: STM2(&(xo[0]), Tu, ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V Tw, Th, Ti, Tv; Chris@10: Th = VBYI(VADD(Ta, T7)); Chris@10: Ti = VADD(Te, Tf); Chris@10: Tv = VADD(Th, Ti); Chris@10: STM2(&(xo[2]), Tv, ovs, &(xo[2])); Chris@10: STN2(&(xo[0]), Tu, Tv, ovs); Chris@10: Tw = VSUB(Ti, Th); Chris@10: STM2(&(xo[14]), Tw, ovs, &(xo[2])); Chris@10: { Chris@10: V Tl, To, Tx, Ty; Chris@10: Tl = VSUB(Tj, Tk); Chris@10: To = VBYI(VSUB(Tm, Tn)); Chris@10: Tx = VSUB(Tl, To); Chris@10: STM2(&(xo[12]), Tx, ovs, &(xo[0])); Chris@10: STN2(&(xo[12]), Tx, Tw, ovs); Chris@10: Ty = VADD(Tl, To); Chris@10: STM2(&(xo[4]), Ty, ovs, &(xo[0])); Chris@10: STN2(&(xo[4]), Ty, Tr, ovs); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 8, XSIMD_STRING("n2bv_8"), {26, 2, 0, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2bv_8) (planner *p) { Chris@10: X(kdft_register) (p, n2bv_8, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */