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 07:37:30 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 12 -name n2bv_12 -with-ostride 2 -include n2b.h -store-multiple 2 */ 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: * 61 stack variables, 2 constants, and 30 memory accesses cannam@95: */ cannam@95: #include "n2b.h" cannam@95: cannam@95: static void n2bv_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 = 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(24, is), MAKE_VOLATILE_STRIDE(24, os)) { cannam@95: V T1, T6, Tc, Th, Td, Te, Ti, Tz, T4, TA, T9, Tj, Tf, Tw; 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: 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: 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: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tz = VSUB(T2, T3); cannam@95: T4 = VADD(T2, T3); cannam@95: TA = VSUB(T7, T8); cannam@95: T9 = VADD(T7, T8); cannam@95: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: } cannam@95: Tf = VADD(Td, Te); cannam@95: Tw = VSUB(Td, Te); cannam@95: { cannam@95: V T5, Tp, TJ, TB, Ta, Tq, Tk, Tx, Tg, Ts; cannam@95: T5 = VADD(T1, T4); cannam@95: Tp = VFNMS(LDK(KP500000000), T4, T1); cannam@95: TJ = VSUB(Tz, TA); cannam@95: TB = VADD(Tz, TA); cannam@95: Ta = VADD(T6, T9); cannam@95: Tq = VFNMS(LDK(KP500000000), T9, T6); cannam@95: Tk = VADD(Ti, Tj); cannam@95: Tx = VSUB(Tj, Ti); cannam@95: Tg = VADD(Tc, Tf); cannam@95: Ts = VFNMS(LDK(KP500000000), Tf, Tc); cannam@95: { cannam@95: V Tr, TF, Tb, Tn, TG, Ty, Tl, Tt; cannam@95: Tr = VADD(Tp, Tq); cannam@95: TF = VSUB(Tp, Tq); cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tn = VADD(T5, Ta); cannam@95: TG = VADD(Tw, Tx); cannam@95: Ty = VSUB(Tw, Tx); cannam@95: Tl = VADD(Th, Tk); cannam@95: Tt = VFNMS(LDK(KP500000000), Tk, Th); cannam@95: { cannam@95: V TC, TE, TH, TL, Tu, TI, Tm, To; cannam@95: TC = VMUL(LDK(KP866025403), VSUB(Ty, TB)); cannam@95: TE = VMUL(LDK(KP866025403), VADD(TB, Ty)); cannam@95: TH = VFNMS(LDK(KP866025403), TG, TF); cannam@95: TL = VFMA(LDK(KP866025403), TG, TF); cannam@95: Tu = VADD(Ts, Tt); cannam@95: TI = VSUB(Ts, Tt); cannam@95: Tm = VSUB(Tg, Tl); cannam@95: To = VADD(Tg, Tl); cannam@95: { cannam@95: V TK, TM, Tv, TD; cannam@95: TK = VFMA(LDK(KP866025403), TJ, TI); cannam@95: TM = VFNMS(LDK(KP866025403), TJ, TI); cannam@95: Tv = VSUB(Tr, Tu); cannam@95: TD = VADD(Tr, Tu); cannam@95: { cannam@95: V TN, TO, TP, TQ; cannam@95: TN = VADD(Tn, To); cannam@95: STM2(&(xo[0]), TN, ovs, &(xo[0])); cannam@95: TO = VSUB(Tn, To); cannam@95: STM2(&(xo[12]), TO, ovs, &(xo[0])); cannam@95: TP = VFMAI(Tm, Tb); cannam@95: STM2(&(xo[18]), TP, ovs, &(xo[2])); cannam@95: TQ = VFNMSI(Tm, Tb); cannam@95: STM2(&(xo[6]), TQ, ovs, &(xo[2])); cannam@95: { cannam@95: V TR, TS, TT, TU; cannam@95: TR = VFMAI(TM, TL); cannam@95: STM2(&(xo[10]), TR, ovs, &(xo[2])); cannam@95: TS = VFNMSI(TM, TL); cannam@95: STM2(&(xo[14]), TS, ovs, &(xo[2])); cannam@95: STN2(&(xo[12]), TO, TS, ovs); cannam@95: TT = VFNMSI(TK, TH); cannam@95: STM2(&(xo[22]), TT, ovs, &(xo[2])); cannam@95: TU = VFMAI(TK, TH); cannam@95: STM2(&(xo[2]), TU, ovs, &(xo[2])); cannam@95: STN2(&(xo[0]), TN, TU, ovs); cannam@95: { cannam@95: V TV, TW, TX, TY; cannam@95: TV = VFNMSI(TE, TD); cannam@95: STM2(&(xo[16]), TV, ovs, &(xo[0])); cannam@95: STN2(&(xo[16]), TV, TP, ovs); cannam@95: TW = VFMAI(TE, TD); cannam@95: STM2(&(xo[8]), TW, ovs, &(xo[0])); cannam@95: STN2(&(xo[8]), TW, TR, ovs); cannam@95: TX = VFMAI(TC, Tv); cannam@95: STM2(&(xo[4]), TX, ovs, &(xo[0])); cannam@95: STN2(&(xo[4]), TX, TQ, ovs); cannam@95: TY = VFNMSI(TC, Tv); cannam@95: STM2(&(xo[20]), TY, ovs, &(xo[0])); cannam@95: STN2(&(xo[20]), TY, TT, ovs); cannam@95: } cannam@95: } 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("n2bv_12"), {30, 2, 18, 0}, &GENUS, 0, 2, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n2bv_12) (planner *p) { cannam@95: X(kdft_register) (p, n2bv_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 -sign 1 -n 12 -name n2bv_12 -with-ostride 2 -include n2b.h -store-multiple 2 */ 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: * 33 stack variables, 2 constants, and 30 memory accesses cannam@95: */ cannam@95: #include "n2b.h" cannam@95: cannam@95: static void n2bv_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 = 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(24, is), MAKE_VOLATILE_STRIDE(24, os)) { cannam@95: V T5, Ta, TG, TF, Ty, Tm, Ti, Tp, TJ, TI, Tx, Ts; cannam@95: { cannam@95: V T1, T6, T4, Tk, T9, Tl; 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: Tk = VSUB(T2, T3); 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: Tl = VSUB(T7, T8); cannam@95: } cannam@95: T5 = VFNMS(LDK(KP500000000), T4, T1); cannam@95: Ta = VFNMS(LDK(KP500000000), T9, T6); cannam@95: TG = VADD(T6, T9); cannam@95: TF = VADD(T1, T4); cannam@95: Ty = VADD(Tk, Tl); cannam@95: Tm = VMUL(LDK(KP866025403), VSUB(Tk, Tl)); cannam@95: } cannam@95: { cannam@95: V Tn, Tq, Te, To, Th, Tr; cannam@95: Tn = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tq = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@95: { cannam@95: V Tc, Td, Tf, Tg; cannam@95: Tc = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@95: Td = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@95: Te = VSUB(Tc, Td); cannam@95: To = VADD(Tc, Td); cannam@95: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@95: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@95: Th = VSUB(Tf, Tg); cannam@95: Tr = VADD(Tf, Tg); cannam@95: } cannam@95: Ti = VMUL(LDK(KP866025403), VSUB(Te, Th)); cannam@95: Tp = VFNMS(LDK(KP500000000), To, Tn); cannam@95: TJ = VADD(Tq, Tr); cannam@95: TI = VADD(Tn, To); cannam@95: Tx = VADD(Te, Th); cannam@95: Ts = VFNMS(LDK(KP500000000), Tr, Tq); cannam@95: } cannam@95: { cannam@95: V TN, TO, TP, TQ, TR, TS; cannam@95: { cannam@95: V TH, TK, TL, TM; cannam@95: TH = VSUB(TF, TG); cannam@95: TK = VBYI(VSUB(TI, TJ)); cannam@95: TN = VSUB(TH, TK); cannam@95: STM2(&(xo[6]), TN, ovs, &(xo[2])); cannam@95: TO = VADD(TH, TK); cannam@95: STM2(&(xo[18]), TO, ovs, &(xo[2])); cannam@95: TL = VADD(TF, TG); cannam@95: TM = VADD(TI, TJ); cannam@95: TP = VSUB(TL, TM); cannam@95: STM2(&(xo[12]), TP, ovs, &(xo[0])); cannam@95: TQ = VADD(TL, TM); cannam@95: STM2(&(xo[0]), TQ, ovs, &(xo[0])); cannam@95: } cannam@95: { cannam@95: V Tj, Tv, Tu, Tw, Tb, Tt, TT, TU; cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tj = VSUB(Tb, Ti); cannam@95: Tv = VADD(Tb, Ti); cannam@95: Tt = VSUB(Tp, Ts); cannam@95: Tu = VBYI(VADD(Tm, Tt)); cannam@95: Tw = VBYI(VSUB(Tt, Tm)); cannam@95: TR = VSUB(Tj, Tu); cannam@95: STM2(&(xo[22]), TR, ovs, &(xo[2])); cannam@95: TS = VADD(Tv, Tw); cannam@95: STM2(&(xo[10]), TS, ovs, &(xo[2])); cannam@95: TT = VADD(Tj, Tu); cannam@95: STM2(&(xo[2]), TT, ovs, &(xo[2])); cannam@95: STN2(&(xo[0]), TQ, TT, ovs); cannam@95: TU = VSUB(Tv, Tw); cannam@95: STM2(&(xo[14]), TU, ovs, &(xo[2])); cannam@95: STN2(&(xo[12]), TP, TU, ovs); cannam@95: } cannam@95: { cannam@95: V Tz, TD, TC, TE, TA, TB; cannam@95: Tz = VBYI(VMUL(LDK(KP866025403), VSUB(Tx, Ty))); cannam@95: TD = VBYI(VMUL(LDK(KP866025403), VADD(Ty, Tx))); cannam@95: TA = VADD(T5, Ta); cannam@95: TB = VADD(Tp, Ts); cannam@95: TC = VSUB(TA, TB); cannam@95: TE = VADD(TA, TB); cannam@95: { cannam@95: V TV, TW, TX, TY; cannam@95: TV = VADD(Tz, TC); cannam@95: STM2(&(xo[4]), TV, ovs, &(xo[0])); cannam@95: STN2(&(xo[4]), TV, TN, ovs); cannam@95: TW = VSUB(TE, TD); cannam@95: STM2(&(xo[16]), TW, ovs, &(xo[0])); cannam@95: STN2(&(xo[16]), TW, TO, ovs); cannam@95: TX = VSUB(TC, Tz); cannam@95: STM2(&(xo[20]), TX, ovs, &(xo[0])); cannam@95: STN2(&(xo[20]), TX, TR, ovs); cannam@95: TY = VADD(TD, TE); cannam@95: STM2(&(xo[8]), TY, ovs, &(xo[0])); cannam@95: STN2(&(xo[8]), TY, TS, ovs); 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("n2bv_12"), {44, 4, 4, 0}, &GENUS, 0, 2, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n2bv_12) (planner *p) { cannam@95: X(kdft_register) (p, n2bv_12, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */