cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:05:07 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_notw_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name n2fv_12 -with-ostride 2 -include dft/simd/n2f.h -store-multiple 2 */ cannam@167: cannam@167: /* cannam@167: * This function contains 48 FP additions, 20 FP multiplications, cannam@167: * (or, 30 additions, 2 multiplications, 18 fused multiply/add), cannam@167: * 33 stack variables, 2 constants, and 30 memory accesses cannam@167: */ cannam@167: #include "dft/simd/n2f.h" cannam@167: cannam@167: 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@167: { cannam@167: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: { cannam@167: INT i; cannam@167: const R *xi; cannam@167: R *xo; cannam@167: xi = ri; cannam@167: xo = ro; cannam@167: 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@167: V T5, Ta, TG, TF, TB, Tt, Ti, Tm, TJ, TI, TA, Tp; cannam@167: { cannam@167: V T1, T6, T4, Tr, T9, Ts; cannam@167: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@167: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@167: { cannam@167: V T2, T3, T7, T8; cannam@167: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@167: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@167: T4 = VADD(T2, T3); cannam@167: Tr = VSUB(T3, T2); cannam@167: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@167: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@167: T9 = VADD(T7, T8); cannam@167: Ts = VSUB(T8, T7); cannam@167: } cannam@167: T5 = VFNMS(LDK(KP500000000), T4, T1); cannam@167: Ta = VFNMS(LDK(KP500000000), T9, T6); cannam@167: TG = VADD(T6, T9); cannam@167: TF = VADD(T1, T4); cannam@167: TB = VADD(Tr, Ts); cannam@167: Tt = VSUB(Tr, Ts); cannam@167: } cannam@167: { cannam@167: V Tk, Tn, Te, Tl, Th, To; cannam@167: Tk = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@167: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@167: { cannam@167: V Tc, Td, Tf, Tg; cannam@167: Tc = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@167: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@167: Te = VSUB(Tc, Td); cannam@167: Tl = VADD(Td, Tc); cannam@167: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@167: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@167: Th = VSUB(Tf, Tg); cannam@167: To = VADD(Tf, Tg); cannam@167: } cannam@167: Ti = VADD(Te, Th); cannam@167: Tm = VFNMS(LDK(KP500000000), Tl, Tk); cannam@167: TJ = VADD(Tn, To); cannam@167: TI = VADD(Tk, Tl); cannam@167: TA = VSUB(Te, Th); cannam@167: Tp = VFNMS(LDK(KP500000000), To, Tn); cannam@167: } cannam@167: { cannam@167: V TN, TO, TP, TQ, TT, TU; cannam@167: { cannam@167: V TH, TK, TL, TM; cannam@167: TH = VSUB(TF, TG); cannam@167: TK = VSUB(TI, TJ); cannam@167: TN = VFNMSI(TK, TH); cannam@167: STM2(&(xo[18]), TN, ovs, &(xo[2])); cannam@167: TO = VFMAI(TK, TH); cannam@167: STM2(&(xo[6]), TO, ovs, &(xo[2])); cannam@167: TL = VADD(TF, TG); cannam@167: TM = VADD(TI, TJ); cannam@167: TP = VSUB(TL, TM); cannam@167: STM2(&(xo[12]), TP, ovs, &(xo[0])); cannam@167: TQ = VADD(TL, TM); cannam@167: STM2(&(xo[0]), TQ, ovs, &(xo[0])); cannam@167: } cannam@167: { cannam@167: V Tj, Tv, Tu, Tw, Tb, Tq, TR, TS; cannam@167: Tb = VSUB(T5, Ta); cannam@167: Tj = VFMA(LDK(KP866025403), Ti, Tb); cannam@167: Tv = VFNMS(LDK(KP866025403), Ti, Tb); cannam@167: Tq = VSUB(Tm, Tp); cannam@167: Tu = VFNMS(LDK(KP866025403), Tt, Tq); cannam@167: Tw = VFMA(LDK(KP866025403), Tt, Tq); cannam@167: TR = VFNMSI(Tu, Tj); cannam@167: STM2(&(xo[2]), TR, ovs, &(xo[2])); cannam@167: STN2(&(xo[0]), TQ, TR, ovs); cannam@167: TS = VFMAI(Tw, Tv); cannam@167: STM2(&(xo[14]), TS, ovs, &(xo[2])); cannam@167: STN2(&(xo[12]), TP, TS, ovs); cannam@167: TT = VFMAI(Tu, Tj); cannam@167: STM2(&(xo[22]), TT, ovs, &(xo[2])); cannam@167: TU = VFNMSI(Tw, Tv); cannam@167: STM2(&(xo[10]), TU, ovs, &(xo[2])); cannam@167: } cannam@167: { cannam@167: V TC, TE, Tz, TD, Tx, Ty; cannam@167: TC = VMUL(LDK(KP866025403), VSUB(TA, TB)); cannam@167: TE = VMUL(LDK(KP866025403), VADD(TB, TA)); cannam@167: Tx = VADD(T5, Ta); cannam@167: Ty = VADD(Tm, Tp); cannam@167: Tz = VSUB(Tx, Ty); cannam@167: TD = VADD(Tx, Ty); cannam@167: { cannam@167: V TV, TW, TX, TY; cannam@167: TV = VFMAI(TC, Tz); cannam@167: STM2(&(xo[4]), TV, ovs, &(xo[0])); cannam@167: STN2(&(xo[4]), TV, TO, ovs); cannam@167: TW = VFNMSI(TE, TD); cannam@167: STM2(&(xo[16]), TW, ovs, &(xo[0])); cannam@167: STN2(&(xo[16]), TW, TN, ovs); cannam@167: TX = VFNMSI(TC, Tz); cannam@167: STM2(&(xo[20]), TX, ovs, &(xo[0])); cannam@167: STN2(&(xo[20]), TX, TT, ovs); cannam@167: TY = VFMAI(TE, TD); cannam@167: STM2(&(xo[8]), TY, ovs, &(xo[0])); cannam@167: STN2(&(xo[8]), TY, TU, ovs); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {30, 2, 18, 0}, &GENUS, 0, 2, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_n2fv_12) (planner *p) { cannam@167: X(kdft_register) (p, n2fv_12, &desc); cannam@167: } cannam@167: cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name n2fv_12 -with-ostride 2 -include dft/simd/n2f.h -store-multiple 2 */ cannam@167: cannam@167: /* cannam@167: * This function contains 48 FP additions, 8 FP multiplications, cannam@167: * (or, 44 additions, 4 multiplications, 4 fused multiply/add), cannam@167: * 33 stack variables, 2 constants, and 30 memory accesses cannam@167: */ cannam@167: #include "dft/simd/n2f.h" cannam@167: cannam@167: 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@167: { cannam@167: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@167: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@167: { cannam@167: INT i; cannam@167: const R *xi; cannam@167: R *xo; cannam@167: xi = ri; cannam@167: xo = ro; cannam@167: 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@167: V T5, Ta, TJ, Ty, Tq, Tp, Tg, Tl, TI, TA, Tz, Tu; cannam@167: { cannam@167: V T1, T6, T4, Tw, T9, Tx; cannam@167: T1 = LD(&(xi[0]), ivs, &(xi[0])); cannam@167: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); cannam@167: { cannam@167: V T2, T3, T7, T8; cannam@167: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); cannam@167: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); cannam@167: T4 = VADD(T2, T3); cannam@167: Tw = VSUB(T3, T2); cannam@167: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); cannam@167: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); cannam@167: T9 = VADD(T7, T8); cannam@167: Tx = VSUB(T8, T7); cannam@167: } cannam@167: T5 = VADD(T1, T4); cannam@167: Ta = VADD(T6, T9); cannam@167: TJ = VADD(Tw, Tx); cannam@167: Ty = VMUL(LDK(KP866025403), VSUB(Tw, Tx)); cannam@167: Tq = VFNMS(LDK(KP500000000), T9, T6); cannam@167: Tp = VFNMS(LDK(KP500000000), T4, T1); cannam@167: } cannam@167: { cannam@167: V Tc, Th, Tf, Ts, Tk, Tt; cannam@167: Tc = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); cannam@167: Th = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); cannam@167: { cannam@167: V Td, Te, Ti, Tj; cannam@167: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); cannam@167: Te = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); cannam@167: Tf = VADD(Td, Te); cannam@167: Ts = VSUB(Te, Td); cannam@167: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); cannam@167: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); cannam@167: Tk = VADD(Ti, Tj); cannam@167: Tt = VSUB(Tj, Ti); cannam@167: } cannam@167: Tg = VADD(Tc, Tf); cannam@167: Tl = VADD(Th, Tk); cannam@167: TI = VADD(Ts, Tt); cannam@167: TA = VFNMS(LDK(KP500000000), Tk, Th); cannam@167: Tz = VFNMS(LDK(KP500000000), Tf, Tc); cannam@167: Tu = VMUL(LDK(KP866025403), VSUB(Ts, Tt)); cannam@167: } cannam@167: { cannam@167: V TN, TO, TP, TQ, TR, TS; cannam@167: { cannam@167: V Tb, Tm, Tn, To; cannam@167: Tb = VSUB(T5, Ta); cannam@167: Tm = VBYI(VSUB(Tg, Tl)); cannam@167: TN = VSUB(Tb, Tm); cannam@167: STM2(&(xo[18]), TN, ovs, &(xo[2])); cannam@167: TO = VADD(Tb, Tm); cannam@167: STM2(&(xo[6]), TO, ovs, &(xo[2])); cannam@167: Tn = VADD(T5, Ta); cannam@167: To = VADD(Tg, Tl); cannam@167: TP = VSUB(Tn, To); cannam@167: STM2(&(xo[12]), TP, ovs, &(xo[0])); cannam@167: TQ = VADD(Tn, To); cannam@167: STM2(&(xo[0]), TQ, ovs, &(xo[0])); cannam@167: } cannam@167: { cannam@167: V Tv, TE, TC, TD, Tr, TB, TT, TU; cannam@167: Tr = VSUB(Tp, Tq); cannam@167: Tv = VSUB(Tr, Tu); cannam@167: TE = VADD(Tr, Tu); cannam@167: TB = VSUB(Tz, TA); cannam@167: TC = VBYI(VADD(Ty, TB)); cannam@167: TD = VBYI(VSUB(Ty, TB)); cannam@167: TR = VSUB(Tv, TC); cannam@167: STM2(&(xo[10]), TR, ovs, &(xo[2])); cannam@167: TS = VSUB(TE, TD); cannam@167: STM2(&(xo[22]), TS, ovs, &(xo[2])); cannam@167: TT = VADD(TC, Tv); cannam@167: STM2(&(xo[14]), TT, ovs, &(xo[2])); cannam@167: STN2(&(xo[12]), TP, TT, ovs); cannam@167: TU = VADD(TD, TE); cannam@167: STM2(&(xo[2]), TU, ovs, &(xo[2])); cannam@167: STN2(&(xo[0]), TQ, TU, ovs); cannam@167: } cannam@167: { cannam@167: V TK, TM, TH, TL, TF, TG; cannam@167: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); cannam@167: TM = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); cannam@167: TF = VADD(Tp, Tq); cannam@167: TG = VADD(Tz, TA); cannam@167: TH = VSUB(TF, TG); cannam@167: TL = VADD(TF, TG); cannam@167: { cannam@167: V TV, TW, TX, TY; cannam@167: TV = VSUB(TH, TK); cannam@167: STM2(&(xo[20]), TV, ovs, &(xo[0])); cannam@167: STN2(&(xo[20]), TV, TS, ovs); cannam@167: TW = VADD(TL, TM); cannam@167: STM2(&(xo[8]), TW, ovs, &(xo[0])); cannam@167: STN2(&(xo[8]), TW, TR, ovs); cannam@167: TX = VADD(TH, TK); cannam@167: STM2(&(xo[4]), TX, ovs, &(xo[0])); cannam@167: STN2(&(xo[4]), TX, TO, ovs); cannam@167: TY = VSUB(TL, TM); cannam@167: STM2(&(xo[16]), TY, ovs, &(xo[0])); cannam@167: STN2(&(xo[16]), TY, TN, ovs); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {44, 4, 4, 0}, &GENUS, 0, 2, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_n2fv_12) (planner *p) { cannam@167: X(kdft_register) (p, n2fv_12, &desc); cannam@167: } cannam@167: cannam@167: #endif