Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:40:09 EDT 2016 */ Chris@42: Chris@42: #include "codelet-dft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* 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 */ Chris@42: Chris@42: /* Chris@42: * This function contains 48 FP additions, 20 FP multiplications, Chris@42: * (or, 30 additions, 2 multiplications, 18 fused multiply/add), Chris@42: * 61 stack variables, 2 constants, and 30 memory accesses Chris@42: */ Chris@42: #include "n2f.h" Chris@42: Chris@42: static void n2fv_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@42: { Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT i; Chris@42: const R *xi; Chris@42: R *xo; Chris@42: xi = ri; Chris@42: xo = ro; Chris@42: 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)) { Chris@42: V T1, T6, Tk, Tn, Tc, Td, Tf, Tr, T4, Ts, T9, Tg, Te, Tl; Chris@42: { Chris@42: V T2, T3, T7, T8; Chris@42: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@42: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@42: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@42: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@42: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@42: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@42: Tk = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tc = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@42: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tr = VSUB(T3, T2); Chris@42: T4 = VADD(T2, T3); Chris@42: Ts = VSUB(T8, T7); Chris@42: T9 = VADD(T7, T8); Chris@42: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@42: } Chris@42: Te = VSUB(Tc, Td); Chris@42: Tl = VADD(Td, Tc); Chris@42: { Chris@42: V T5, TF, TB, Tt, Ta, TG, Th, To, Tm, TI; Chris@42: T5 = VFNMS(LDK(KP500000000), T4, T1); Chris@42: TF = VADD(T1, T4); Chris@42: TB = VADD(Tr, Ts); Chris@42: Tt = VSUB(Tr, Ts); Chris@42: Ta = VFNMS(LDK(KP500000000), T9, T6); Chris@42: TG = VADD(T6, T9); Chris@42: Th = VSUB(Tf, Tg); Chris@42: To = VADD(Tf, Tg); Chris@42: Tm = VFNMS(LDK(KP500000000), Tl, Tk); Chris@42: TI = VADD(Tk, Tl); Chris@42: { Chris@42: V TH, TL, Tb, Tx, TJ, Tp, Ti, TA; Chris@42: TH = VSUB(TF, TG); Chris@42: TL = VADD(TF, TG); Chris@42: Tb = VSUB(T5, Ta); Chris@42: Tx = VADD(T5, Ta); Chris@42: TJ = VADD(Tn, To); Chris@42: Tp = VFNMS(LDK(KP500000000), To, Tn); Chris@42: Ti = VADD(Te, Th); Chris@42: TA = VSUB(Te, Th); Chris@42: { Chris@42: V Tq, Ty, TK, TM; Chris@42: Tq = VSUB(Tm, Tp); Chris@42: Ty = VADD(Tm, Tp); Chris@42: TK = VSUB(TI, TJ); Chris@42: TM = VADD(TI, TJ); Chris@42: { Chris@42: V TC, TE, Tj, Tv; Chris@42: TC = VMUL(LDK(KP866025403), VSUB(TA, TB)); Chris@42: TE = VMUL(LDK(KP866025403), VADD(TB, TA)); Chris@42: Tj = VFMA(LDK(KP866025403), Ti, Tb); Chris@42: Tv = VFNMS(LDK(KP866025403), Ti, Tb); Chris@42: { Chris@42: V Tz, TD, Tu, Tw; Chris@42: Tz = VSUB(Tx, Ty); Chris@42: TD = VADD(Tx, Ty); Chris@42: Tu = VFNMS(LDK(KP866025403), Tt, Tq); Chris@42: Tw = VFMA(LDK(KP866025403), Tt, Tq); Chris@42: { Chris@42: V TN, TO, TP, TQ; Chris@42: TN = VADD(TL, TM); Chris@42: STM2(&(xo[0]), TN, ovs, &(xo[0])); Chris@42: TO = VSUB(TL, TM); Chris@42: STM2(&(xo[12]), TO, ovs, &(xo[0])); Chris@42: TP = VFMAI(TK, TH); Chris@42: STM2(&(xo[6]), TP, ovs, &(xo[2])); Chris@42: TQ = VFNMSI(TK, TH); Chris@42: STM2(&(xo[18]), TQ, ovs, &(xo[2])); Chris@42: { Chris@42: V TR, TS, TT, TU; Chris@42: TR = VFMAI(TE, TD); Chris@42: STM2(&(xo[8]), TR, ovs, &(xo[0])); Chris@42: TS = VFNMSI(TE, TD); Chris@42: STM2(&(xo[16]), TS, ovs, &(xo[0])); Chris@42: STN2(&(xo[16]), TS, TQ, ovs); Chris@42: TT = VFNMSI(TC, Tz); Chris@42: STM2(&(xo[20]), TT, ovs, &(xo[0])); Chris@42: TU = VFMAI(TC, Tz); Chris@42: STM2(&(xo[4]), TU, ovs, &(xo[0])); Chris@42: STN2(&(xo[4]), TU, TP, ovs); Chris@42: { Chris@42: V TV, TW, TX, TY; Chris@42: TV = VFNMSI(Tw, Tv); Chris@42: STM2(&(xo[10]), TV, ovs, &(xo[2])); Chris@42: STN2(&(xo[8]), TR, TV, ovs); Chris@42: TW = VFMAI(Tw, Tv); Chris@42: STM2(&(xo[14]), TW, ovs, &(xo[2])); Chris@42: STN2(&(xo[12]), TO, TW, ovs); Chris@42: TX = VFMAI(Tu, Tj); Chris@42: STM2(&(xo[22]), TX, ovs, &(xo[2])); Chris@42: STN2(&(xo[20]), TT, TX, ovs); Chris@42: TY = VFNMSI(Tu, Tj); Chris@42: STM2(&(xo[2]), TY, ovs, &(xo[2])); Chris@42: STN2(&(xo[0]), TN, TY, ovs); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {30, 2, 18, 0}, &GENUS, 0, 2, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_n2fv_12) (planner *p) { Chris@42: X(kdft_register) (p, n2fv_12, &desc); Chris@42: } Chris@42: Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* 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 */ Chris@42: Chris@42: /* Chris@42: * This function contains 48 FP additions, 8 FP multiplications, Chris@42: * (or, 44 additions, 4 multiplications, 4 fused multiply/add), Chris@42: * 33 stack variables, 2 constants, and 30 memory accesses Chris@42: */ Chris@42: #include "n2f.h" Chris@42: Chris@42: static void n2fv_12(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@42: { Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: { Chris@42: INT i; Chris@42: const R *xi; Chris@42: R *xo; Chris@42: xi = ri; Chris@42: xo = ro; Chris@42: 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)) { Chris@42: V T5, Ta, TJ, Ty, Tq, Tp, Tg, Tl, TI, TA, Tz, Tu; Chris@42: { Chris@42: V T1, T6, T4, Tw, T9, Tx; Chris@42: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@42: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@42: { Chris@42: V T2, T3, T7, T8; Chris@42: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@42: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@42: T4 = VADD(T2, T3); Chris@42: Tw = VSUB(T3, T2); Chris@42: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@42: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@42: T9 = VADD(T7, T8); Chris@42: Tx = VSUB(T8, T7); Chris@42: } Chris@42: T5 = VADD(T1, T4); Chris@42: Ta = VADD(T6, T9); Chris@42: TJ = VADD(Tw, Tx); Chris@42: Ty = VMUL(LDK(KP866025403), VSUB(Tw, Tx)); Chris@42: Tq = VFNMS(LDK(KP500000000), T9, T6); Chris@42: Tp = VFNMS(LDK(KP500000000), T4, T1); Chris@42: } Chris@42: { Chris@42: V Tc, Th, Tf, Ts, Tk, Tt; Chris@42: Tc = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@42: Th = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@42: { Chris@42: V Td, Te, Ti, Tj; Chris@42: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@42: Te = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tf = VADD(Td, Te); Chris@42: Ts = VSUB(Te, Td); Chris@42: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@42: Tk = VADD(Ti, Tj); Chris@42: Tt = VSUB(Tj, Ti); Chris@42: } Chris@42: Tg = VADD(Tc, Tf); Chris@42: Tl = VADD(Th, Tk); Chris@42: TI = VADD(Ts, Tt); Chris@42: TA = VFNMS(LDK(KP500000000), Tk, Th); Chris@42: Tz = VFNMS(LDK(KP500000000), Tf, Tc); Chris@42: Tu = VMUL(LDK(KP866025403), VSUB(Ts, Tt)); Chris@42: } Chris@42: { Chris@42: V TN, TO, TP, TQ, TR, TS; Chris@42: { Chris@42: V Tb, Tm, Tn, To; Chris@42: Tb = VSUB(T5, Ta); Chris@42: Tm = VBYI(VSUB(Tg, Tl)); Chris@42: TN = VSUB(Tb, Tm); Chris@42: STM2(&(xo[18]), TN, ovs, &(xo[2])); Chris@42: TO = VADD(Tb, Tm); Chris@42: STM2(&(xo[6]), TO, ovs, &(xo[2])); Chris@42: Tn = VADD(T5, Ta); Chris@42: To = VADD(Tg, Tl); Chris@42: TP = VSUB(Tn, To); Chris@42: STM2(&(xo[12]), TP, ovs, &(xo[0])); Chris@42: TQ = VADD(Tn, To); Chris@42: STM2(&(xo[0]), TQ, ovs, &(xo[0])); Chris@42: } Chris@42: { Chris@42: V Tv, TE, TC, TD, Tr, TB, TT, TU; Chris@42: Tr = VSUB(Tp, Tq); Chris@42: Tv = VSUB(Tr, Tu); Chris@42: TE = VADD(Tr, Tu); Chris@42: TB = VSUB(Tz, TA); Chris@42: TC = VBYI(VADD(Ty, TB)); Chris@42: TD = VBYI(VSUB(Ty, TB)); Chris@42: TR = VSUB(Tv, TC); Chris@42: STM2(&(xo[10]), TR, ovs, &(xo[2])); Chris@42: TS = VSUB(TE, TD); Chris@42: STM2(&(xo[22]), TS, ovs, &(xo[2])); Chris@42: TT = VADD(TC, Tv); Chris@42: STM2(&(xo[14]), TT, ovs, &(xo[2])); Chris@42: STN2(&(xo[12]), TP, TT, ovs); Chris@42: TU = VADD(TD, TE); Chris@42: STM2(&(xo[2]), TU, ovs, &(xo[2])); Chris@42: STN2(&(xo[0]), TQ, TU, ovs); Chris@42: } Chris@42: { Chris@42: V TK, TM, TH, TL, TF, TG; Chris@42: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); Chris@42: TM = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); Chris@42: TF = VADD(Tp, Tq); Chris@42: TG = VADD(Tz, TA); Chris@42: TH = VSUB(TF, TG); Chris@42: TL = VADD(TF, TG); Chris@42: { Chris@42: V TV, TW, TX, TY; Chris@42: TV = VSUB(TH, TK); Chris@42: STM2(&(xo[20]), TV, ovs, &(xo[0])); Chris@42: STN2(&(xo[20]), TV, TS, ovs); Chris@42: TW = VADD(TL, TM); Chris@42: STM2(&(xo[8]), TW, ovs, &(xo[0])); Chris@42: STN2(&(xo[8]), TW, TR, ovs); Chris@42: TX = VADD(TH, TK); Chris@42: STM2(&(xo[4]), TX, ovs, &(xo[0])); Chris@42: STN2(&(xo[4]), TX, TO, ovs); Chris@42: TY = VSUB(TL, TM); Chris@42: STM2(&(xo[16]), TY, ovs, &(xo[0])); Chris@42: STN2(&(xo[16]), TY, TN, ovs); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {44, 4, 4, 0}, &GENUS, 0, 2, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_n2fv_12) (planner *p) { Chris@42: X(kdft_register) (p, n2fv_12, &desc); Chris@42: } Chris@42: Chris@42: #endif /* HAVE_FMA */