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 07:37:22 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 -n 8 -name n2fv_8 -with-ostride 2 -include n2f.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 "n2f.h" Chris@10: Chris@10: static void n2fv_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 = ri; Chris@10: xo = ro; 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 = VSUB(Tj, Tk); Chris@10: Tl = VADD(Tj, Tk); Chris@10: { Chris@10: V Tq, To, Ta, Tf; Chris@10: Tq = VSUB(Tn, Tm); Chris@10: To = VADD(Tm, Tn); Chris@10: Ta = VADD(T6, T9); Chris@10: Tf = VSUB(T9, T6); Chris@10: { Chris@10: V Tr, Ts, Tt, Tu, Tg, Ti, Tb, Th; Chris@10: Tr = VADD(Tl, To); Chris@10: STM2(&(xo[0]), Tr, ovs, &(xo[0])); Chris@10: Ts = VSUB(Tl, To); Chris@10: STM2(&(xo[8]), Ts, ovs, &(xo[0])); Chris@10: Tt = VFMAI(Tq, Tp); Chris@10: STM2(&(xo[4]), Tt, ovs, &(xo[0])); Chris@10: Tu = VFNMSI(Tq, Tp); Chris@10: STM2(&(xo[12]), Tu, ovs, &(xo[0])); Chris@10: Tg = VFNMS(LDK(KP707106781), Tf, Te); Chris@10: Ti = VFMA(LDK(KP707106781), Tf, Te); Chris@10: Tb = VFMA(LDK(KP707106781), Ta, T3); Chris@10: Th = VFNMS(LDK(KP707106781), Ta, T3); Chris@10: { Chris@10: V Tv, Tw, Tx, Ty; Chris@10: Tv = VFMAI(Ti, Th); Chris@10: STM2(&(xo[6]), Tv, ovs, &(xo[2])); Chris@10: STN2(&(xo[4]), Tt, Tv, ovs); Chris@10: Tw = VFNMSI(Ti, Th); Chris@10: STM2(&(xo[10]), Tw, ovs, &(xo[2])); Chris@10: STN2(&(xo[8]), Ts, Tw, ovs); Chris@10: Tx = VFMAI(Tg, Tb); Chris@10: STM2(&(xo[14]), Tx, ovs, &(xo[2])); Chris@10: STN2(&(xo[12]), Tu, Tx, ovs); Chris@10: Ty = VFNMSI(Tg, Tb); Chris@10: STM2(&(xo[2]), Ty, ovs, &(xo[2])); Chris@10: STN2(&(xo[0]), 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("n2fv_8"), {16, 0, 10, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2fv_8) (planner *p) { Chris@10: X(kdft_register) (p, n2fv_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 -n 8 -name n2fv_8 -with-ostride 2 -include n2f.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 "n2f.h" Chris@10: Chris@10: static void n2fv_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 = ri; Chris@10: xo = ro; 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 T3, Tj, Tf, Tk, Ta, Tn, Tc, Tm, Ts, Tu; Chris@10: { Chris@10: V T1, T2, Td, Te; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T3 = VSUB(T1, T2); Chris@10: Tj = VADD(T1, T2); Chris@10: Td = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Te = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Tf = VSUB(Td, Te); Chris@10: Tk = VADD(Td, Te); Chris@10: { Chris@10: V T4, T5, T6, T7, T8, T9; 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: T6 = VSUB(T4, T5); 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: T9 = VSUB(T7, T8); Chris@10: Ta = VMUL(LDK(KP707106781), VADD(T6, T9)); Chris@10: Tn = VADD(T7, T8); Chris@10: Tc = VMUL(LDK(KP707106781), VSUB(T9, T6)); Chris@10: Tm = VADD(T4, T5); Chris@10: } Chris@10: } Chris@10: { Chris@10: V Tr, Tb, Tg, Tp, Tq, Tt; Chris@10: Tb = VADD(T3, Ta); Chris@10: Tg = VBYI(VSUB(Tc, Tf)); Chris@10: Tr = VSUB(Tb, Tg); Chris@10: STM2(&(xo[14]), Tr, ovs, &(xo[2])); Chris@10: Ts = VADD(Tb, Tg); Chris@10: STM2(&(xo[2]), Ts, ovs, &(xo[2])); Chris@10: Tp = VSUB(Tj, Tk); Chris@10: Tq = VBYI(VSUB(Tn, Tm)); Chris@10: Tt = VSUB(Tp, Tq); Chris@10: STM2(&(xo[12]), Tt, ovs, &(xo[0])); Chris@10: STN2(&(xo[12]), Tt, Tr, ovs); Chris@10: Tu = VADD(Tp, Tq); Chris@10: STM2(&(xo[4]), Tu, ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V Tv, Th, Ti, Tw; Chris@10: Th = VSUB(T3, Ta); Chris@10: Ti = VBYI(VADD(Tf, Tc)); Chris@10: Tv = VSUB(Th, Ti); Chris@10: STM2(&(xo[10]), Tv, ovs, &(xo[2])); Chris@10: Tw = VADD(Th, Ti); Chris@10: STM2(&(xo[6]), Tw, ovs, &(xo[2])); Chris@10: STN2(&(xo[4]), Tu, Tw, ovs); Chris@10: { Chris@10: V Tl, To, Tx, Ty; Chris@10: Tl = VADD(Tj, Tk); Chris@10: To = VADD(Tm, Tn); Chris@10: Tx = VSUB(Tl, To); Chris@10: STM2(&(xo[8]), Tx, ovs, &(xo[0])); Chris@10: STN2(&(xo[8]), Tx, Tv, ovs); Chris@10: Ty = VADD(Tl, To); Chris@10: STM2(&(xo[0]), Ty, ovs, &(xo[0])); Chris@10: STN2(&(xo[0]), Ty, Ts, 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("n2fv_8"), {26, 2, 0, 0}, &GENUS, 0, 2, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n2fv_8) (planner *p) { Chris@10: X(kdft_register) (p, n2fv_8, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */