cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:41:20 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_notw.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name n2sv_8 -with-ostride 1 -include n2s.h -store-multiple 4 */ cannam@127: cannam@127: /* cannam@127: * This function contains 52 FP additions, 8 FP multiplications, cannam@127: * (or, 44 additions, 0 multiplications, 8 fused multiply/add), cannam@127: * 58 stack variables, 1 constants, and 36 memory accesses cannam@127: */ cannam@127: #include "n2s.h" cannam@127: cannam@127: static void n2sv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT i; cannam@127: for (i = v; i > 0; i = i - (2 * VL), ri = ri + ((2 * VL) * ivs), ii = ii + ((2 * VL) * ivs), ro = ro + ((2 * VL) * ovs), io = io + ((2 * VL) * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { cannam@127: V TF, TJ, TD, TR, TS, TT, TU, TV, TW, TE, TX, TY, TK, TI, TZ; cannam@127: V T10, T11, T12; cannam@127: { cannam@127: V Tb, Tn, T3, TC, Ti, TB, T6, To, Tl, Tc, Tw, Tx, T8, T9, Tr; cannam@127: V Ts; cannam@127: { cannam@127: V T1, T2, Tg, Th, T4, T5, Tj, Tk; cannam@127: T1 = LD(&(ri[0]), ivs, &(ri[0])); cannam@127: T2 = LD(&(ri[WS(is, 4)]), ivs, &(ri[0])); cannam@127: Tg = LD(&(ii[0]), ivs, &(ii[0])); cannam@127: Th = LD(&(ii[WS(is, 4)]), ivs, &(ii[0])); cannam@127: T4 = LD(&(ri[WS(is, 2)]), ivs, &(ri[0])); cannam@127: T5 = LD(&(ri[WS(is, 6)]), ivs, &(ri[0])); cannam@127: Tj = LD(&(ii[WS(is, 2)]), ivs, &(ii[0])); cannam@127: Tk = LD(&(ii[WS(is, 6)]), ivs, &(ii[0])); cannam@127: Tb = LD(&(ri[WS(is, 7)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tn = VSUB(T1, T2); cannam@127: T3 = VADD(T1, T2); cannam@127: TC = VSUB(Tg, Th); cannam@127: Ti = VADD(Tg, Th); cannam@127: TB = VSUB(T4, T5); cannam@127: T6 = VADD(T4, T5); cannam@127: To = VSUB(Tj, Tk); cannam@127: Tl = VADD(Tj, Tk); cannam@127: Tc = LD(&(ri[WS(is, 3)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tw = LD(&(ii[WS(is, 7)]), ivs, &(ii[WS(is, 1)])); cannam@127: Tx = LD(&(ii[WS(is, 3)]), ivs, &(ii[WS(is, 1)])); cannam@127: T8 = LD(&(ri[WS(is, 1)]), ivs, &(ri[WS(is, 1)])); cannam@127: T9 = LD(&(ri[WS(is, 5)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tr = LD(&(ii[WS(is, 1)]), ivs, &(ii[WS(is, 1)])); cannam@127: Ts = LD(&(ii[WS(is, 5)]), ivs, &(ii[WS(is, 1)])); cannam@127: } cannam@127: { cannam@127: V TL, T7, TP, Tm, Tz, TH, Te, Tf, TO, TQ, TG, Tu, Tp, TA; cannam@127: { cannam@127: V Td, Tv, TN, Ty, Ta, Tq, TM, Tt; cannam@127: TL = VSUB(T3, T6); cannam@127: T7 = VADD(T3, T6); cannam@127: Td = VADD(Tb, Tc); cannam@127: Tv = VSUB(Tb, Tc); cannam@127: TN = VADD(Tw, Tx); cannam@127: Ty = VSUB(Tw, Tx); cannam@127: Ta = VADD(T8, T9); cannam@127: Tq = VSUB(T8, T9); cannam@127: TM = VADD(Tr, Ts); cannam@127: Tt = VSUB(Tr, Ts); cannam@127: TP = VADD(Ti, Tl); cannam@127: Tm = VSUB(Ti, Tl); cannam@127: Tz = VSUB(Tv, Ty); cannam@127: TH = VADD(Tv, Ty); cannam@127: Te = VADD(Ta, Td); cannam@127: Tf = VSUB(Td, Ta); cannam@127: TO = VSUB(TM, TN); cannam@127: TQ = VADD(TM, TN); cannam@127: TG = VSUB(Tt, Tq); cannam@127: Tu = VADD(Tq, Tt); cannam@127: } cannam@127: TF = VSUB(Tn, To); cannam@127: Tp = VADD(Tn, To); cannam@127: TJ = VSUB(TC, TB); cannam@127: TD = VADD(TB, TC); cannam@127: TR = VSUB(Tm, Tf); cannam@127: STM4(&(io[6]), TR, ovs, &(io[0])); cannam@127: TS = VADD(Tf, Tm); cannam@127: STM4(&(io[2]), TS, ovs, &(io[0])); cannam@127: TT = VADD(T7, Te); cannam@127: STM4(&(ro[0]), TT, ovs, &(ro[0])); cannam@127: TU = VSUB(T7, Te); cannam@127: STM4(&(ro[4]), TU, ovs, &(ro[0])); cannam@127: TV = VADD(TP, TQ); cannam@127: STM4(&(io[0]), TV, ovs, &(io[0])); cannam@127: TW = VSUB(TP, TQ); cannam@127: STM4(&(io[4]), TW, ovs, &(io[0])); cannam@127: TE = VSUB(Tz, Tu); cannam@127: TA = VADD(Tu, Tz); cannam@127: TX = VADD(TL, TO); cannam@127: STM4(&(ro[2]), TX, ovs, &(ro[0])); cannam@127: TY = VSUB(TL, TO); cannam@127: STM4(&(ro[6]), TY, ovs, &(ro[0])); cannam@127: TK = VADD(TG, TH); cannam@127: TI = VSUB(TG, TH); cannam@127: TZ = VFMA(LDK(KP707106781), TA, Tp); cannam@127: STM4(&(ro[1]), TZ, ovs, &(ro[1])); cannam@127: T10 = VFNMS(LDK(KP707106781), TA, Tp); cannam@127: STM4(&(ro[5]), T10, ovs, &(ro[1])); cannam@127: } cannam@127: } cannam@127: T11 = VFMA(LDK(KP707106781), TK, TJ); cannam@127: STM4(&(io[1]), T11, ovs, &(io[1])); cannam@127: T12 = VFNMS(LDK(KP707106781), TK, TJ); cannam@127: STM4(&(io[5]), T12, ovs, &(io[1])); cannam@127: { cannam@127: V T13, T14, T15, T16; cannam@127: T13 = VFMA(LDK(KP707106781), TE, TD); cannam@127: STM4(&(io[3]), T13, ovs, &(io[1])); cannam@127: STN4(&(io[0]), TV, T11, TS, T13, ovs); cannam@127: T14 = VFNMS(LDK(KP707106781), TE, TD); cannam@127: STM4(&(io[7]), T14, ovs, &(io[1])); cannam@127: STN4(&(io[4]), TW, T12, TR, T14, ovs); cannam@127: T15 = VFMA(LDK(KP707106781), TI, TF); cannam@127: STM4(&(ro[3]), T15, ovs, &(ro[1])); cannam@127: STN4(&(ro[0]), TT, TZ, TX, T15, ovs); cannam@127: T16 = VFNMS(LDK(KP707106781), TI, TF); cannam@127: STM4(&(ro[7]), T16, ovs, &(ro[1])); cannam@127: STN4(&(ro[4]), TU, T10, TY, T16, ovs); cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 8, XSIMD_STRING("n2sv_8"), {44, 0, 8, 0}, &GENUS, 0, 1, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n2sv_8) (planner *p) { cannam@127: X(kdft_register) (p, n2sv_8, &desc); cannam@127: } cannam@127: cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_notw.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name n2sv_8 -with-ostride 1 -include n2s.h -store-multiple 4 */ cannam@127: cannam@127: /* cannam@127: * This function contains 52 FP additions, 4 FP multiplications, cannam@127: * (or, 52 additions, 4 multiplications, 0 fused multiply/add), cannam@127: * 34 stack variables, 1 constants, and 36 memory accesses cannam@127: */ cannam@127: #include "n2s.h" cannam@127: cannam@127: static void n2sv_8(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) cannam@127: { cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT i; cannam@127: for (i = v; i > 0; i = i - (2 * VL), ri = ri + ((2 * VL) * ivs), ii = ii + ((2 * VL) * ivs), ro = ro + ((2 * VL) * ovs), io = io + ((2 * VL) * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { cannam@127: V T3, Tn, Ti, TC, T6, TB, Tl, To, Td, TN, Tz, TH, Ta, TM, Tu; cannam@127: V TG; cannam@127: { cannam@127: V T1, T2, Tj, Tk; cannam@127: T1 = LD(&(ri[0]), ivs, &(ri[0])); cannam@127: T2 = LD(&(ri[WS(is, 4)]), ivs, &(ri[0])); cannam@127: T3 = VADD(T1, T2); cannam@127: Tn = VSUB(T1, T2); cannam@127: { cannam@127: V Tg, Th, T4, T5; cannam@127: Tg = LD(&(ii[0]), ivs, &(ii[0])); cannam@127: Th = LD(&(ii[WS(is, 4)]), ivs, &(ii[0])); cannam@127: Ti = VADD(Tg, Th); cannam@127: TC = VSUB(Tg, Th); cannam@127: T4 = LD(&(ri[WS(is, 2)]), ivs, &(ri[0])); cannam@127: T5 = LD(&(ri[WS(is, 6)]), ivs, &(ri[0])); cannam@127: T6 = VADD(T4, T5); cannam@127: TB = VSUB(T4, T5); cannam@127: } cannam@127: Tj = LD(&(ii[WS(is, 2)]), ivs, &(ii[0])); cannam@127: Tk = LD(&(ii[WS(is, 6)]), ivs, &(ii[0])); cannam@127: Tl = VADD(Tj, Tk); cannam@127: To = VSUB(Tj, Tk); cannam@127: { cannam@127: V Tb, Tc, Tv, Tw, Tx, Ty; cannam@127: Tb = LD(&(ri[WS(is, 7)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tc = LD(&(ri[WS(is, 3)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tv = VSUB(Tb, Tc); cannam@127: Tw = LD(&(ii[WS(is, 7)]), ivs, &(ii[WS(is, 1)])); cannam@127: Tx = LD(&(ii[WS(is, 3)]), ivs, &(ii[WS(is, 1)])); cannam@127: Ty = VSUB(Tw, Tx); cannam@127: Td = VADD(Tb, Tc); cannam@127: TN = VADD(Tw, Tx); cannam@127: Tz = VSUB(Tv, Ty); cannam@127: TH = VADD(Tv, Ty); cannam@127: } cannam@127: { cannam@127: V T8, T9, Tq, Tr, Ts, Tt; cannam@127: T8 = LD(&(ri[WS(is, 1)]), ivs, &(ri[WS(is, 1)])); cannam@127: T9 = LD(&(ri[WS(is, 5)]), ivs, &(ri[WS(is, 1)])); cannam@127: Tq = VSUB(T8, T9); cannam@127: Tr = LD(&(ii[WS(is, 1)]), ivs, &(ii[WS(is, 1)])); cannam@127: Ts = LD(&(ii[WS(is, 5)]), ivs, &(ii[WS(is, 1)])); cannam@127: Tt = VSUB(Tr, Ts); cannam@127: Ta = VADD(T8, T9); cannam@127: TM = VADD(Tr, Ts); cannam@127: Tu = VADD(Tq, Tt); cannam@127: TG = VSUB(Tt, Tq); cannam@127: } cannam@127: } cannam@127: { cannam@127: V TR, TS, TT, TU, TV, TW, TX, TY; cannam@127: { cannam@127: V T7, Te, TP, TQ; cannam@127: T7 = VADD(T3, T6); cannam@127: Te = VADD(Ta, Td); cannam@127: TR = VSUB(T7, Te); cannam@127: STM4(&(ro[4]), TR, ovs, &(ro[0])); cannam@127: TS = VADD(T7, Te); cannam@127: STM4(&(ro[0]), TS, ovs, &(ro[0])); cannam@127: TP = VADD(Ti, Tl); cannam@127: TQ = VADD(TM, TN); cannam@127: TT = VSUB(TP, TQ); cannam@127: STM4(&(io[4]), TT, ovs, &(io[0])); cannam@127: TU = VADD(TP, TQ); cannam@127: STM4(&(io[0]), TU, ovs, &(io[0])); cannam@127: } cannam@127: { cannam@127: V Tf, Tm, TL, TO; cannam@127: Tf = VSUB(Td, Ta); cannam@127: Tm = VSUB(Ti, Tl); cannam@127: TV = VADD(Tf, Tm); cannam@127: STM4(&(io[2]), TV, ovs, &(io[0])); cannam@127: TW = VSUB(Tm, Tf); cannam@127: STM4(&(io[6]), TW, ovs, &(io[0])); cannam@127: TL = VSUB(T3, T6); cannam@127: TO = VSUB(TM, TN); cannam@127: TX = VSUB(TL, TO); cannam@127: STM4(&(ro[6]), TX, ovs, &(ro[0])); cannam@127: TY = VADD(TL, TO); cannam@127: STM4(&(ro[2]), TY, ovs, &(ro[0])); cannam@127: } cannam@127: { cannam@127: V TZ, T10, T11, T12; cannam@127: { cannam@127: V Tp, TA, TJ, TK; cannam@127: Tp = VADD(Tn, To); cannam@127: TA = VMUL(LDK(KP707106781), VADD(Tu, Tz)); cannam@127: TZ = VSUB(Tp, TA); cannam@127: STM4(&(ro[5]), TZ, ovs, &(ro[1])); cannam@127: T10 = VADD(Tp, TA); cannam@127: STM4(&(ro[1]), T10, ovs, &(ro[1])); cannam@127: TJ = VSUB(TC, TB); cannam@127: TK = VMUL(LDK(KP707106781), VADD(TG, TH)); cannam@127: T11 = VSUB(TJ, TK); cannam@127: STM4(&(io[5]), T11, ovs, &(io[1])); cannam@127: T12 = VADD(TJ, TK); cannam@127: STM4(&(io[1]), T12, ovs, &(io[1])); cannam@127: } cannam@127: { cannam@127: V TD, TE, T13, T14; cannam@127: TD = VADD(TB, TC); cannam@127: TE = VMUL(LDK(KP707106781), VSUB(Tz, Tu)); cannam@127: T13 = VSUB(TD, TE); cannam@127: STM4(&(io[7]), T13, ovs, &(io[1])); cannam@127: STN4(&(io[4]), TT, T11, TW, T13, ovs); cannam@127: T14 = VADD(TD, TE); cannam@127: STM4(&(io[3]), T14, ovs, &(io[1])); cannam@127: STN4(&(io[0]), TU, T12, TV, T14, ovs); cannam@127: } cannam@127: { cannam@127: V TF, TI, T15, T16; cannam@127: TF = VSUB(Tn, To); cannam@127: TI = VMUL(LDK(KP707106781), VSUB(TG, TH)); cannam@127: T15 = VSUB(TF, TI); cannam@127: STM4(&(ro[7]), T15, ovs, &(ro[1])); cannam@127: STN4(&(ro[4]), TR, TZ, TX, T15, ovs); cannam@127: T16 = VADD(TF, TI); cannam@127: STM4(&(ro[3]), T16, ovs, &(ro[1])); cannam@127: STN4(&(ro[0]), TS, T10, TY, T16, ovs); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const kdft_desc desc = { 8, XSIMD_STRING("n2sv_8"), {52, 4, 0, 0}, &GENUS, 0, 1, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_n2sv_8) (planner *p) { cannam@127: X(kdft_register) (p, n2sv_8, &desc); cannam@127: } cannam@127: cannam@127: #endif /* HAVE_FMA */