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 8 -name n1bv_8 -include n1b.h */ 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: * 30 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_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 Tg, Ti, Tb, Th; Chris@10: ST(&(xo[WS(os, 2)]), VFMAI(To, Tl), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VFNMSI(To, Tl), ovs, &(xo[0])); Chris@10: ST(&(xo[0]), VADD(Tp, Tq), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VSUB(Tp, Tq), 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: ST(&(xo[WS(os, 7)]), VFNMSI(Ti, Th), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VFMAI(Ti, Th), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VFMAI(Tg, Tb), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 3)]), VFNMSI(Tg, Tb), ovs, &(xo[WS(os, 1)])); 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("n1bv_8"), {16, 0, 10, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_8) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_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 n1bv_8 -include n1b.h */ 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: * 22 stack variables, 1 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "n1b.h" Chris@10: Chris@10: static void n1bv_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; 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 Tb, Tg, Tp, Tq; Chris@10: Tb = VBYI(VSUB(T7, Ta)); Chris@10: Tg = VSUB(Te, Tf); Chris@10: ST(&(xo[WS(os, 3)]), VADD(Tb, Tg), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VSUB(Tg, Tb), ovs, &(xo[WS(os, 1)])); Chris@10: Tp = VADD(Tj, Tk); Chris@10: Tq = VADD(Tm, Tn); Chris@10: ST(&(xo[WS(os, 4)]), VSUB(Tp, Tq), ovs, &(xo[0])); Chris@10: ST(&(xo[0]), VADD(Tp, Tq), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V Th, Ti, Tl, To; Chris@10: Th = VBYI(VADD(Ta, T7)); Chris@10: Ti = VADD(Te, Tf); Chris@10: ST(&(xo[WS(os, 1)]), VADD(Th, Ti), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VSUB(Ti, Th), ovs, &(xo[WS(os, 1)])); Chris@10: Tl = VSUB(Tj, Tk); Chris@10: To = VBYI(VSUB(Tm, Tn)); Chris@10: ST(&(xo[WS(os, 6)]), VSUB(Tl, To), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 2)]), VADD(Tl, To), ovs, &(xo[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 8, XSIMD_STRING("n1bv_8"), {26, 2, 0, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1bv_8) (planner *p) { Chris@10: X(kdft_register) (p, n1bv_8, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */