Chris@82: /* Chris@82: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@82: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@82: * Chris@82: * This program is free software; you can redistribute it and/or modify Chris@82: * it under the terms of the GNU General Public License as published by Chris@82: * the Free Software Foundation; either version 2 of the License, or Chris@82: * (at your option) any later version. Chris@82: * Chris@82: * This program is distributed in the hope that it will be useful, Chris@82: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@82: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@82: * GNU General Public License for more details. Chris@82: * Chris@82: * You should have received a copy of the GNU General Public License Chris@82: * along with this program; if not, write to the Free Software Chris@82: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@82: * Chris@82: */ Chris@82: Chris@82: /* This file was automatically generated --- DO NOT EDIT */ Chris@82: /* Generated on Thu May 24 08:05:07 EDT 2018 */ Chris@82: Chris@82: #include "dft/codelet-dft.h" Chris@82: Chris@82: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) Chris@82: Chris@82: /* 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 */ Chris@82: Chris@82: /* Chris@82: * This function contains 48 FP additions, 20 FP multiplications, Chris@82: * (or, 30 additions, 2 multiplications, 18 fused multiply/add), Chris@82: * 33 stack variables, 2 constants, and 30 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n2f.h" Chris@82: Chris@82: 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@82: { Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: 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@82: V T5, Ta, TG, TF, TB, Tt, Ti, Tm, TJ, TI, TA, Tp; Chris@82: { Chris@82: V T1, T6, T4, Tr, T9, Ts; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: { Chris@82: V T2, T3, T7, T8; Chris@82: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@82: T4 = VADD(T2, T3); Chris@82: Tr = VSUB(T3, T2); Chris@82: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@82: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: T9 = VADD(T7, T8); Chris@82: Ts = VSUB(T8, T7); Chris@82: } Chris@82: T5 = VFNMS(LDK(KP500000000), T4, T1); Chris@82: Ta = VFNMS(LDK(KP500000000), T9, T6); Chris@82: TG = VADD(T6, T9); Chris@82: TF = VADD(T1, T4); Chris@82: TB = VADD(Tr, Ts); Chris@82: Tt = VSUB(Tr, Ts); Chris@82: } Chris@82: { Chris@82: V Tk, Tn, Te, Tl, Th, To; Chris@82: Tk = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tn = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@82: { Chris@82: V Tc, Td, Tf, Tg; Chris@82: Tc = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@82: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: Te = VSUB(Tc, Td); Chris@82: Tl = VADD(Td, Tc); Chris@82: Tf = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tg = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: Th = VSUB(Tf, Tg); Chris@82: To = VADD(Tf, Tg); Chris@82: } Chris@82: Ti = VADD(Te, Th); Chris@82: Tm = VFNMS(LDK(KP500000000), Tl, Tk); Chris@82: TJ = VADD(Tn, To); Chris@82: TI = VADD(Tk, Tl); Chris@82: TA = VSUB(Te, Th); Chris@82: Tp = VFNMS(LDK(KP500000000), To, Tn); Chris@82: } Chris@82: { Chris@82: V TN, TO, TP, TQ, TT, TU; Chris@82: { Chris@82: V TH, TK, TL, TM; Chris@82: TH = VSUB(TF, TG); Chris@82: TK = VSUB(TI, TJ); Chris@82: TN = VFNMSI(TK, TH); Chris@82: STM2(&(xo[18]), TN, ovs, &(xo[2])); Chris@82: TO = VFMAI(TK, TH); Chris@82: STM2(&(xo[6]), TO, ovs, &(xo[2])); Chris@82: TL = VADD(TF, TG); Chris@82: TM = VADD(TI, TJ); Chris@82: TP = VSUB(TL, TM); Chris@82: STM2(&(xo[12]), TP, ovs, &(xo[0])); Chris@82: TQ = VADD(TL, TM); Chris@82: STM2(&(xo[0]), TQ, ovs, &(xo[0])); Chris@82: } Chris@82: { Chris@82: V Tj, Tv, Tu, Tw, Tb, Tq, TR, TS; Chris@82: Tb = VSUB(T5, Ta); Chris@82: Tj = VFMA(LDK(KP866025403), Ti, Tb); Chris@82: Tv = VFNMS(LDK(KP866025403), Ti, Tb); Chris@82: Tq = VSUB(Tm, Tp); Chris@82: Tu = VFNMS(LDK(KP866025403), Tt, Tq); Chris@82: Tw = VFMA(LDK(KP866025403), Tt, Tq); Chris@82: TR = VFNMSI(Tu, Tj); Chris@82: STM2(&(xo[2]), TR, ovs, &(xo[2])); Chris@82: STN2(&(xo[0]), TQ, TR, ovs); Chris@82: TS = VFMAI(Tw, Tv); Chris@82: STM2(&(xo[14]), TS, ovs, &(xo[2])); Chris@82: STN2(&(xo[12]), TP, TS, ovs); Chris@82: TT = VFMAI(Tu, Tj); Chris@82: STM2(&(xo[22]), TT, ovs, &(xo[2])); Chris@82: TU = VFNMSI(Tw, Tv); Chris@82: STM2(&(xo[10]), TU, ovs, &(xo[2])); Chris@82: } Chris@82: { Chris@82: V TC, TE, Tz, TD, Tx, Ty; Chris@82: TC = VMUL(LDK(KP866025403), VSUB(TA, TB)); Chris@82: TE = VMUL(LDK(KP866025403), VADD(TB, TA)); Chris@82: Tx = VADD(T5, Ta); Chris@82: Ty = VADD(Tm, Tp); Chris@82: Tz = VSUB(Tx, Ty); Chris@82: TD = VADD(Tx, Ty); Chris@82: { Chris@82: V TV, TW, TX, TY; Chris@82: TV = VFMAI(TC, Tz); Chris@82: STM2(&(xo[4]), TV, ovs, &(xo[0])); Chris@82: STN2(&(xo[4]), TV, TO, ovs); Chris@82: TW = VFNMSI(TE, TD); Chris@82: STM2(&(xo[16]), TW, ovs, &(xo[0])); Chris@82: STN2(&(xo[16]), TW, TN, ovs); Chris@82: TX = VFNMSI(TC, Tz); Chris@82: STM2(&(xo[20]), TX, ovs, &(xo[0])); Chris@82: STN2(&(xo[20]), TX, TT, ovs); Chris@82: TY = VFMAI(TE, TD); Chris@82: STM2(&(xo[8]), TY, ovs, &(xo[0])); Chris@82: STN2(&(xo[8]), TY, TU, ovs); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {30, 2, 18, 0}, &GENUS, 0, 2, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n2fv_12) (planner *p) { Chris@82: X(kdft_register) (p, n2fv_12, &desc); Chris@82: } Chris@82: Chris@82: #else Chris@82: Chris@82: /* 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 */ Chris@82: Chris@82: /* Chris@82: * This function contains 48 FP additions, 8 FP multiplications, Chris@82: * (or, 44 additions, 4 multiplications, 4 fused multiply/add), Chris@82: * 33 stack variables, 2 constants, and 30 memory accesses Chris@82: */ Chris@82: #include "dft/simd/n2f.h" Chris@82: Chris@82: 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@82: { Chris@82: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@82: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@82: { Chris@82: INT i; Chris@82: const R *xi; Chris@82: R *xo; Chris@82: xi = ri; Chris@82: xo = ro; Chris@82: 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@82: V T5, Ta, TJ, Ty, Tq, Tp, Tg, Tl, TI, TA, Tz, Tu; Chris@82: { Chris@82: V T1, T6, T4, Tw, T9, Tx; Chris@82: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@82: T6 = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@82: { Chris@82: V T2, T3, T7, T8; Chris@82: T2 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@82: T3 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@82: T4 = VADD(T2, T3); Chris@82: Tw = VSUB(T3, T2); Chris@82: T7 = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@82: T8 = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@82: T9 = VADD(T7, T8); Chris@82: Tx = VSUB(T8, T7); Chris@82: } Chris@82: T5 = VADD(T1, T4); Chris@82: Ta = VADD(T6, T9); Chris@82: TJ = VADD(Tw, Tx); Chris@82: Ty = VMUL(LDK(KP866025403), VSUB(Tw, Tx)); Chris@82: Tq = VFNMS(LDK(KP500000000), T9, T6); Chris@82: Tp = VFNMS(LDK(KP500000000), T4, T1); Chris@82: } Chris@82: { Chris@82: V Tc, Th, Tf, Ts, Tk, Tt; Chris@82: Tc = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@82: Th = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@82: { Chris@82: V Td, Te, Ti, Tj; Chris@82: Td = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@82: Te = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tf = VADD(Td, Te); Chris@82: Ts = VSUB(Te, Td); Chris@82: Ti = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tj = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@82: Tk = VADD(Ti, Tj); Chris@82: Tt = VSUB(Tj, Ti); Chris@82: } Chris@82: Tg = VADD(Tc, Tf); Chris@82: Tl = VADD(Th, Tk); Chris@82: TI = VADD(Ts, Tt); Chris@82: TA = VFNMS(LDK(KP500000000), Tk, Th); Chris@82: Tz = VFNMS(LDK(KP500000000), Tf, Tc); Chris@82: Tu = VMUL(LDK(KP866025403), VSUB(Ts, Tt)); Chris@82: } Chris@82: { Chris@82: V TN, TO, TP, TQ, TR, TS; Chris@82: { Chris@82: V Tb, Tm, Tn, To; Chris@82: Tb = VSUB(T5, Ta); Chris@82: Tm = VBYI(VSUB(Tg, Tl)); Chris@82: TN = VSUB(Tb, Tm); Chris@82: STM2(&(xo[18]), TN, ovs, &(xo[2])); Chris@82: TO = VADD(Tb, Tm); Chris@82: STM2(&(xo[6]), TO, ovs, &(xo[2])); Chris@82: Tn = VADD(T5, Ta); Chris@82: To = VADD(Tg, Tl); Chris@82: TP = VSUB(Tn, To); Chris@82: STM2(&(xo[12]), TP, ovs, &(xo[0])); Chris@82: TQ = VADD(Tn, To); Chris@82: STM2(&(xo[0]), TQ, ovs, &(xo[0])); Chris@82: } Chris@82: { Chris@82: V Tv, TE, TC, TD, Tr, TB, TT, TU; Chris@82: Tr = VSUB(Tp, Tq); Chris@82: Tv = VSUB(Tr, Tu); Chris@82: TE = VADD(Tr, Tu); Chris@82: TB = VSUB(Tz, TA); Chris@82: TC = VBYI(VADD(Ty, TB)); Chris@82: TD = VBYI(VSUB(Ty, TB)); Chris@82: TR = VSUB(Tv, TC); Chris@82: STM2(&(xo[10]), TR, ovs, &(xo[2])); Chris@82: TS = VSUB(TE, TD); Chris@82: STM2(&(xo[22]), TS, ovs, &(xo[2])); Chris@82: TT = VADD(TC, Tv); Chris@82: STM2(&(xo[14]), TT, ovs, &(xo[2])); Chris@82: STN2(&(xo[12]), TP, TT, ovs); Chris@82: TU = VADD(TD, TE); Chris@82: STM2(&(xo[2]), TU, ovs, &(xo[2])); Chris@82: STN2(&(xo[0]), TQ, TU, ovs); Chris@82: } Chris@82: { Chris@82: V TK, TM, TH, TL, TF, TG; Chris@82: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); Chris@82: TM = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); Chris@82: TF = VADD(Tp, Tq); Chris@82: TG = VADD(Tz, TA); Chris@82: TH = VSUB(TF, TG); Chris@82: TL = VADD(TF, TG); Chris@82: { Chris@82: V TV, TW, TX, TY; Chris@82: TV = VSUB(TH, TK); Chris@82: STM2(&(xo[20]), TV, ovs, &(xo[0])); Chris@82: STN2(&(xo[20]), TV, TS, ovs); Chris@82: TW = VADD(TL, TM); Chris@82: STM2(&(xo[8]), TW, ovs, &(xo[0])); Chris@82: STN2(&(xo[8]), TW, TR, ovs); Chris@82: TX = VADD(TH, TK); Chris@82: STM2(&(xo[4]), TX, ovs, &(xo[0])); Chris@82: STN2(&(xo[4]), TX, TO, ovs); Chris@82: TY = VSUB(TL, TM); Chris@82: STM2(&(xo[16]), TY, ovs, &(xo[0])); Chris@82: STN2(&(xo[16]), TY, TN, ovs); Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: } Chris@82: VLEAVE(); Chris@82: } Chris@82: Chris@82: static const kdft_desc desc = { 12, XSIMD_STRING("n2fv_12"), {44, 4, 4, 0}, &GENUS, 0, 2, 0, 0 }; Chris@82: Chris@82: void XSIMD(codelet_n2fv_12) (planner *p) { Chris@82: X(kdft_register) (p, n2fv_12, &desc); Chris@82: } Chris@82: Chris@82: #endif