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 07:36:52 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 -n 12 -name n1fv_12 -include n1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 48 FP additions, 20 FP multiplications, cannam@95: * (or, 30 additions, 2 multiplications, 18 fused multiply/add), cannam@95: * 49 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "n1f.h" cannam@95: cannam@95: static void n1fv_12(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(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT i; cannam@95: const R *xi; cannam@95: R *xo; cannam@95: xi = ri; cannam@95: xo = ro; cannam@95: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(24, is), MAKE_VOLATILE_STRIDE(24, os)) { cannam@95: V T1, T6, Tk, Tn, Tc, Td, Tf, Tr, T4, Ts, T9, Tg, Te, Tl; cannam@95: { cannam@95: V T2, T3, T7, T8; cannam@95: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@95: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@95: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@95: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@95: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@95: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@95: Tk = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tc = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@95: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tr = VSUB(T3, T2); cannam@95: T4 = VADD(T2, T3); cannam@95: Ts = VSUB(T8, T7); cannam@95: T9 = VADD(T7, T8); cannam@95: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: } cannam@95: Te = VSUB(Tc, Td); cannam@95: Tl = VADD(Td, Tc); cannam@95: { cannam@95: V T5, TF, TB, Tt, Ta, TG, Th, To, Tm, TI; cannam@95: T5 = VFNMS(LDK(KP500000000), T4, T1); cannam@95: TF = VADD(T1, T4); cannam@95: TB = VADD(Tr, Ts); cannam@95: Tt = VSUB(Tr, Ts); cannam@95: Ta = VFNMS(LDK(KP500000000), T9, T6); cannam@95: TG = VADD(T6, T9); cannam@95: Th = VSUB(Tf, Tg); cannam@95: To = VADD(Tf, Tg); cannam@95: Tm = VFNMS(LDK(KP500000000), Tl, Tk); cannam@95: TI = VADD(Tk, Tl); cannam@95: { cannam@95: V TH, TL, Tb, Tx, TJ, Tp, Ti, TA; cannam@95: TH = VSUB(TF, TG); cannam@95: TL = VADD(TF, TG); cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tx = VADD(T5, Ta); cannam@95: TJ = VADD(Tn, To); cannam@95: Tp = VFNMS(LDK(KP500000000), To, Tn); cannam@95: Ti = VADD(Te, Th); cannam@95: TA = VSUB(Te, Th); cannam@95: { cannam@95: V Tq, Ty, TK, TM; cannam@95: Tq = VSUB(Tm, Tp); cannam@95: Ty = VADD(Tm, Tp); cannam@95: TK = VSUB(TI, TJ); cannam@95: TM = VADD(TI, TJ); cannam@95: { cannam@95: V TC, TE, Tj, Tv; cannam@95: TC = VMUL(LDK(KP866025403), VSUB(TA, TB)); cannam@95: TE = VMUL(LDK(KP866025403), VADD(TB, TA)); cannam@95: Tj = VFMA(LDK(KP866025403), Ti, Tb); cannam@95: Tv = VFNMS(LDK(KP866025403), Ti, Tb); cannam@95: { cannam@95: V Tz, TD, Tu, Tw; cannam@95: Tz = VSUB(Tx, Ty); cannam@95: TD = VADD(Tx, Ty); cannam@95: Tu = VFNMS(LDK(KP866025403), Tt, Tq); cannam@95: Tw = VFMA(LDK(KP866025403), Tt, Tq); cannam@95: ST(&(xo[0]), VADD(TL, TM), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 6)]), VSUB(TL, TM), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 3)]), VFMAI(TK, TH), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 9)]), VFNMSI(TK, TH), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 4)]), VFMAI(TE, TD), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 8)]), VFNMSI(TE, TD), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 10)]), VFNMSI(TC, Tz), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 2)]), VFMAI(TC, Tz), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 5)]), VFNMSI(Tw, Tv), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 7)]), VFMAI(Tw, Tv), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 11)]), VFMAI(Tu, Tj), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 1)]), VFNMSI(Tu, Tj), ovs, &(xo[WS(os, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 12, XSIMD_STRING("n1fv_12"), {30, 2, 18, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n1fv_12) (planner *p) { cannam@95: X(kdft_register) (p, n1fv_12, &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 -n 12 -name n1fv_12 -include n1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 48 FP additions, 8 FP multiplications, cannam@95: * (or, 44 additions, 4 multiplications, 4 fused multiply/add), cannam@95: * 27 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "n1f.h" cannam@95: cannam@95: static void n1fv_12(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(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: { cannam@95: INT i; cannam@95: const R *xi; cannam@95: R *xo; cannam@95: xi = ri; cannam@95: xo = ro; cannam@95: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(24, is), MAKE_VOLATILE_STRIDE(24, os)) { cannam@95: V T5, Ta, TJ, Ty, Tq, Tp, Tg, Tl, TI, TA, Tz, Tu; cannam@95: { cannam@95: V T1, T6, T4, Tw, T9, Tx; cannam@95: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@95: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@95: { cannam@95: V T2, T3, T7, T8; cannam@95: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@95: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@95: T4 = VADD(T2, T3); cannam@95: Tw = VSUB(T3, T2); cannam@95: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@95: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@95: T9 = VADD(T7, T8); cannam@95: Tx = VSUB(T8, T7); cannam@95: } cannam@95: T5 = VADD(T1, T4); cannam@95: Ta = VADD(T6, T9); cannam@95: TJ = VADD(Tw, Tx); cannam@95: Ty = VMUL(LDK(KP866025403), VSUB(Tw, Tx)); cannam@95: Tq = VFNMS(LDK(KP500000000), T9, T6); cannam@95: Tp = VFNMS(LDK(KP500000000), T4, T1); cannam@95: } cannam@95: { cannam@95: V Tc, Th, Tf, Ts, Tk, Tt; cannam@95: Tc = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@95: Th = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@95: { cannam@95: V Td, Te, Ti, Tj; cannam@95: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@95: Te = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tf = VADD(Td, Te); cannam@95: Ts = VSUB(Te, Td); cannam@95: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tk = VADD(Ti, Tj); cannam@95: Tt = VSUB(Tj, Ti); cannam@95: } cannam@95: Tg = VADD(Tc, Tf); cannam@95: Tl = VADD(Th, Tk); cannam@95: TI = VADD(Ts, Tt); cannam@95: TA = VFNMS(LDK(KP500000000), Tk, Th); cannam@95: Tz = VFNMS(LDK(KP500000000), Tf, Tc); cannam@95: Tu = VMUL(LDK(KP866025403), VSUB(Ts, Tt)); cannam@95: } cannam@95: { cannam@95: V Tb, Tm, Tn, To; cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tm = VBYI(VSUB(Tg, Tl)); cannam@95: ST(&(xo[WS(os, 9)]), VSUB(Tb, Tm), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 3)]), VADD(Tb, Tm), ovs, &(xo[WS(os, 1)])); cannam@95: Tn = VADD(T5, Ta); cannam@95: To = VADD(Tg, Tl); cannam@95: ST(&(xo[WS(os, 6)]), VSUB(Tn, To), ovs, &(xo[0])); cannam@95: ST(&(xo[0]), VADD(Tn, To), ovs, &(xo[0])); cannam@95: } cannam@95: { cannam@95: V Tv, TE, TC, TD, Tr, TB; cannam@95: Tr = VSUB(Tp, Tq); cannam@95: Tv = VSUB(Tr, Tu); cannam@95: TE = VADD(Tr, Tu); cannam@95: TB = VSUB(Tz, TA); cannam@95: TC = VBYI(VADD(Ty, TB)); cannam@95: TD = VBYI(VSUB(Ty, TB)); cannam@95: ST(&(xo[WS(os, 5)]), VSUB(Tv, TC), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 11)]), VSUB(TE, TD), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 7)]), VADD(TC, Tv), ovs, &(xo[WS(os, 1)])); cannam@95: ST(&(xo[WS(os, 1)]), VADD(TD, TE), ovs, &(xo[WS(os, 1)])); cannam@95: } cannam@95: { cannam@95: V TK, TM, TH, TL, TF, TG; cannam@95: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); cannam@95: TM = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); cannam@95: TF = VADD(Tp, Tq); cannam@95: TG = VADD(Tz, TA); cannam@95: TH = VSUB(TF, TG); cannam@95: TL = VADD(TF, TG); cannam@95: ST(&(xo[WS(os, 10)]), VSUB(TH, TK), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 4)]), VADD(TL, TM), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 2)]), VADD(TH, TK), ovs, &(xo[0])); cannam@95: ST(&(xo[WS(os, 8)]), VSUB(TL, TM), ovs, &(xo[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const kdft_desc desc = { 12, XSIMD_STRING("n1fv_12"), {44, 4, 4, 0}, &GENUS, 0, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n1fv_12) (planner *p) { cannam@95: X(kdft_register) (p, n1fv_12, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */