cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:36:59 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* 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 */ cannam@95: cannam@95: /* cannam@95: * This function contains 26 FP additions, 10 FP multiplications, cannam@95: * (or, 16 additions, 0 multiplications, 10 fused multiply/add), cannam@95: * 30 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "n1b.h" cannam@95: cannam@95: static void n1bv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT i; cannam@95: const R *xi; cannam@95: R *xo; cannam@95: xi = ii; cannam@95: xo = io; cannam@95: 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)) { cannam@95: V T1, T2, Tc, Td, T4, T5, T7, T8; cannam@95: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@95: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@95: Tc = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@95: Td = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@95: T4 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: T5 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: T7 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@95: T8 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@95: { cannam@95: V T3, Tj, Te, Tk, T6, Tm, T9, Tn, Tp, Tl; cannam@95: T3 = VSUB(T1, T2); cannam@95: Tj = VADD(T1, T2); cannam@95: Te = VSUB(Tc, Td); cannam@95: Tk = VADD(Tc, Td); cannam@95: T6 = VSUB(T4, T5); cannam@95: Tm = VADD(T4, T5); cannam@95: T9 = VSUB(T7, T8); cannam@95: Tn = VADD(T7, T8); cannam@95: Tp = VADD(Tj, Tk); cannam@95: Tl = VSUB(Tj, Tk); cannam@95: { cannam@95: V Tq, To, Ta, Tf; cannam@95: Tq = VADD(Tm, Tn); cannam@95: To = VSUB(Tm, Tn); cannam@95: Ta = VADD(T6, T9); cannam@95: Tf = VSUB(T6, T9); cannam@95: { cannam@95: V Tg, Ti, Tb, Th; cannam@95: ST(&(xo[WS(os, 2)]), VFMAI(To, Tl), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 6)]), VFNMSI(To, Tl), ovs, &(xo[0])); cannam@95: ST(&(xo[0]), VADD(Tp, Tq), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 4)]), VSUB(Tp, Tq), ovs, &(xo[0])); cannam@95: Tg = VFNMS(LDK(KP707106781), Tf, Te); cannam@95: Ti = VFMA(LDK(KP707106781), Tf, Te); cannam@95: Tb = VFNMS(LDK(KP707106781), Ta, T3); cannam@95: Th = VFMA(LDK(KP707106781), Ta, T3); cannam@95: ST(&(xo[WS(os, 7)]), VFNMSI(Ti, Th), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 1)]), VFMAI(Ti, Th), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 5)]), VFMAI(Tg, Tb), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 3)]), VFNMSI(Tg, Tb), ovs, &(xo[WS(os, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 8, XSIMD_STRING("n1bv_8"), {16, 0, 10, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n1bv_8) (planner *p) { cannam@95: X(kdft_register) (p, n1bv_8, &desc); cannam@95: } cannam@95: cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -sign 1 -n 8 -name n1bv_8 -include n1b.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 26 FP additions, 2 FP multiplications, cannam@95: * (or, 26 additions, 2 multiplications, 0 fused multiply/add), cannam@95: * 22 stack variables, 1 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "n1b.h" cannam@95: cannam@95: static void n1bv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@95: { cannam@95: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@95: { cannam@95: INT i; cannam@95: const R *xi; cannam@95: R *xo; cannam@95: xi = ii; cannam@95: xo = io; cannam@95: 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)) { cannam@95: V Ta, Tk, Te, Tj, T7, Tn, Tf, Tm; cannam@95: { cannam@95: V T8, T9, Tc, Td; cannam@95: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@95: T9 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@95: Ta = VSUB(T8, T9); cannam@95: Tk = VADD(T8, T9); cannam@95: Tc = LD(&(xi[0]), ivs, &(xi[0])); cannam@95: Td = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@95: Te = VSUB(Tc, Td); cannam@95: Tj = VADD(Tc, Td); cannam@95: { cannam@95: V T1, T2, T3, T4, T5, T6; cannam@95: T1 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: T2 = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: T3 = VSUB(T1, T2); cannam@95: T4 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@95: T5 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@95: T6 = VSUB(T4, T5); cannam@95: T7 = VMUL(LDK(KP707106781), VSUB(T3, T6)); cannam@95: Tn = VADD(T4, T5); cannam@95: Tf = VMUL(LDK(KP707106781), VADD(T3, T6)); cannam@95: Tm = VADD(T1, T2); cannam@95: } cannam@95: } cannam@95: { cannam@95: V Tb, Tg, Tp, Tq; cannam@95: Tb = VBYI(VSUB(T7, Ta)); cannam@95: Tg = VSUB(Te, Tf); cannam@95: ST(&(xo[WS(os, 3)]), VADD(Tb, Tg), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 5)]), VSUB(Tg, Tb), ovs, &(xo[WS(os, 1)])); cannam@95: Tp = VADD(Tj, Tk); cannam@95: Tq = VADD(Tm, Tn); cannam@95: ST(&(xo[WS(os, 4)]), VSUB(Tp, Tq), ovs, &(xo[0])); cannam@95: ST(&(xo[0]), VADD(Tp, Tq), ovs, &(xo[0])); cannam@95: } cannam@95: { cannam@95: V Th, Ti, Tl, To; cannam@95: Th = VBYI(VADD(Ta, T7)); cannam@95: Ti = VADD(Te, Tf); cannam@95: ST(&(xo[WS(os, 1)]), VADD(Th, Ti), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 7)]), VSUB(Ti, Th), ovs, &(xo[WS(os, 1)])); cannam@95: Tl = VSUB(Tj, Tk); cannam@95: To = VBYI(VSUB(Tm, Tn)); cannam@95: ST(&(xo[WS(os, 6)]), VSUB(Tl, To), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 2)]), VADD(Tl, To), ovs, &(xo[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 8, XSIMD_STRING("n1bv_8"), {26, 2, 0, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n1bv_8) (planner *p) { cannam@95: X(kdft_register) (p, n1bv_8, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */