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 07:37:22 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 n2fv_12 -with-ostride 2 -include n2f.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 "n2f.h" cannam@95: cannam@95: static void n2fv_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: { cannam@95: V TN, TO, TP, TQ; cannam@95: TN = VADD(TL, TM); cannam@95: STM2(&(xo[0]), TN, ovs, &(xo[0])); cannam@95: TO = VSUB(TL, TM); cannam@95: STM2(&(xo[12]), TO, ovs, &(xo[0])); cannam@95: TP = VFMAI(TK, TH); cannam@95: STM2(&(xo[6]), TP, ovs, &(xo[2])); cannam@95: TQ = VFNMSI(TK, TH); cannam@95: STM2(&(xo[18]), TQ, ovs, &(xo[2])); cannam@95: { cannam@95: V TR, TS, TT, TU; cannam@95: TR = VFMAI(TE, TD); cannam@95: STM2(&(xo[8]), TR, ovs, &(xo[0])); cannam@95: TS = VFNMSI(TE, TD); cannam@95: STM2(&(xo[16]), TS, ovs, &(xo[0])); cannam@95: STN2(&(xo[16]), TS, TQ, ovs); cannam@95: TT = VFNMSI(TC, Tz); cannam@95: STM2(&(xo[20]), TT, ovs, &(xo[0])); cannam@95: TU = VFMAI(TC, Tz); cannam@95: STM2(&(xo[4]), TU, ovs, &(xo[0])); cannam@95: STN2(&(xo[4]), TU, TP, ovs); cannam@95: { cannam@95: V TV, TW, TX, TY; cannam@95: TV = VFNMSI(Tw, Tv); cannam@95: STM2(&(xo[10]), TV, ovs, &(xo[2])); cannam@95: STN2(&(xo[8]), TR, TV, ovs); cannam@95: TW = VFMAI(Tw, Tv); cannam@95: STM2(&(xo[14]), TW, ovs, &(xo[2])); cannam@95: STN2(&(xo[12]), TO, TW, ovs); cannam@95: TX = VFMAI(Tu, Tj); cannam@95: STM2(&(xo[22]), TX, ovs, &(xo[2])); cannam@95: STN2(&(xo[20]), TT, TX, ovs); cannam@95: TY = VFNMSI(Tu, Tj); cannam@95: STM2(&(xo[2]), TY, ovs, &(xo[2])); cannam@95: STN2(&(xo[0]), TN, TY, ovs); cannam@95: } 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("n2fv_12"), {30, 2, 18, 0}, &GENUS, 0, 2, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n2fv_12) (planner *p) { cannam@95: X(kdft_register) (p, n2fv_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 n2fv_12 -with-ostride 2 -include n2f.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 "n2f.h" cannam@95: cannam@95: static void n2fv_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 TN, TO, TP, TQ, TR, TS; cannam@95: { cannam@95: V Tb, Tm, Tn, To; cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tm = VBYI(VSUB(Tg, Tl)); cannam@95: TN = VSUB(Tb, Tm); cannam@95: STM2(&(xo[18]), TN, ovs, &(xo[2])); cannam@95: TO = VADD(Tb, Tm); cannam@95: STM2(&(xo[6]), TO, ovs, &(xo[2])); cannam@95: Tn = VADD(T5, Ta); cannam@95: To = VADD(Tg, Tl); cannam@95: TP = VSUB(Tn, To); cannam@95: STM2(&(xo[12]), TP, ovs, &(xo[0])); cannam@95: TQ = VADD(Tn, To); cannam@95: STM2(&(xo[0]), TQ, ovs, &(xo[0])); cannam@95: } cannam@95: { cannam@95: V Tv, TE, TC, TD, Tr, TB, TT, TU; 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: TR = VSUB(Tv, TC); cannam@95: STM2(&(xo[10]), TR, ovs, &(xo[2])); cannam@95: TS = VSUB(TE, TD); cannam@95: STM2(&(xo[22]), TS, ovs, &(xo[2])); cannam@95: TT = VADD(TC, Tv); cannam@95: STM2(&(xo[14]), TT, ovs, &(xo[2])); cannam@95: STN2(&(xo[12]), TP, TT, ovs); cannam@95: TU = VADD(TD, TE); cannam@95: STM2(&(xo[2]), TU, ovs, &(xo[2])); cannam@95: STN2(&(xo[0]), TQ, TU, ovs); 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: { cannam@95: V TV, TW, TX, TY; cannam@95: TV = VSUB(TH, TK); cannam@95: STM2(&(xo[20]), TV, ovs, &(xo[0])); cannam@95: STN2(&(xo[20]), TV, TS, ovs); cannam@95: TW = VADD(TL, TM); cannam@95: STM2(&(xo[8]), TW, ovs, &(xo[0])); cannam@95: STN2(&(xo[8]), TW, TR, ovs); cannam@95: TX = VADD(TH, TK); cannam@95: STM2(&(xo[4]), TX, ovs, &(xo[0])); cannam@95: STN2(&(xo[4]), TX, TO, ovs); cannam@95: TY = VSUB(TL, TM); cannam@95: STM2(&(xo[16]), TY, ovs, &(xo[0])); cannam@95: STN2(&(xo[16]), TY, TN, 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("n2fv_12"), {44, 4, 4, 0}, &GENUS, 0, 2, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_n2fv_12) (planner *p) { cannam@95: X(kdft_register) (p, n2fv_12, &desc); cannam@95: } cannam@95: cannam@95: #endif /* HAVE_FMA */