Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:36:53 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 16 -name n1fv_16 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 72 FP additions, 34 FP multiplications, Chris@10: * (or, 38 additions, 0 multiplications, 34 fused multiply/add), Chris@10: * 54 stack variables, 3 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_16(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DVK(KP414213562, +0.414213562373095048801688724209698078569671875); Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ri; Chris@10: xo = ro; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { Chris@10: V T7, Tu, TF, TB, T13, TL, TO, TX, TC, Te, TP, Th, TQ, Tk, TW; Chris@10: V T16; Chris@10: { Chris@10: V TH, TU, Tz, Tf, TK, TV, TA, TM, Ta, TN, Td, Tg, Ti, Tj; Chris@10: { Chris@10: V T1, T2, T4, T5, To, Tp, Tr, Ts; Chris@10: T1 = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: T2 = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: T4 = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: T5 = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: To = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: Tp = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Tr = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Ts = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: { Chris@10: V T8, TJ, Tq, TI, Tt, T9, Tb, Tc, T3, T6; Chris@10: T8 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: TH = VSUB(T1, T2); Chris@10: T3 = VADD(T1, T2); Chris@10: TU = VSUB(T4, T5); Chris@10: T6 = VADD(T4, T5); Chris@10: TJ = VSUB(To, Tp); Chris@10: Tq = VADD(To, Tp); Chris@10: TI = VSUB(Tr, Ts); Chris@10: Tt = VADD(Tr, Ts); Chris@10: T9 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tb = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tc = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: T7 = VSUB(T3, T6); Chris@10: Tz = VADD(T3, T6); Chris@10: Tf = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); Chris@10: TK = VADD(TI, TJ); Chris@10: TV = VSUB(TJ, TI); Chris@10: TA = VADD(Tt, Tq); Chris@10: Tu = VSUB(Tq, Tt); Chris@10: TM = VSUB(T8, T9); Chris@10: Ta = VADD(T8, T9); Chris@10: TN = VSUB(Tb, Tc); Chris@10: Td = VADD(Tb, Tc); Chris@10: Tg = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ti = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tj = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: } Chris@10: } Chris@10: TF = VSUB(Tz, TA); Chris@10: TB = VADD(Tz, TA); Chris@10: T13 = VFNMS(LDK(KP707106781), TK, TH); Chris@10: TL = VFMA(LDK(KP707106781), TK, TH); Chris@10: TO = VFNMS(LDK(KP414213562), TN, TM); Chris@10: TX = VFMA(LDK(KP414213562), TM, TN); Chris@10: TC = VADD(Ta, Td); Chris@10: Te = VSUB(Ta, Td); Chris@10: TP = VSUB(Tf, Tg); Chris@10: Th = VADD(Tf, Tg); Chris@10: TQ = VSUB(Tj, Ti); Chris@10: Tk = VADD(Ti, Tj); Chris@10: TW = VFNMS(LDK(KP707106781), TV, TU); Chris@10: T16 = VFMA(LDK(KP707106781), TV, TU); Chris@10: } Chris@10: { Chris@10: V TY, TR, Tl, TD; Chris@10: TY = VFMA(LDK(KP414213562), TP, TQ); Chris@10: TR = VFNMS(LDK(KP414213562), TQ, TP); Chris@10: Tl = VSUB(Th, Tk); Chris@10: TD = VADD(Th, Tk); Chris@10: { Chris@10: V TS, T17, TZ, T14; Chris@10: TS = VADD(TO, TR); Chris@10: T17 = VSUB(TR, TO); Chris@10: TZ = VSUB(TX, TY); Chris@10: T14 = VADD(TX, TY); Chris@10: { Chris@10: V TE, TG, Tm, Tv; Chris@10: TE = VADD(TC, TD); Chris@10: TG = VSUB(TD, TC); Chris@10: Tm = VADD(Te, Tl); Chris@10: Tv = VSUB(Tl, Te); Chris@10: { Chris@10: V T18, T1a, TT, T11; Chris@10: T18 = VFNMS(LDK(KP923879532), T17, T16); Chris@10: T1a = VFMA(LDK(KP923879532), T17, T16); Chris@10: TT = VFNMS(LDK(KP923879532), TS, TL); Chris@10: T11 = VFMA(LDK(KP923879532), TS, TL); Chris@10: { Chris@10: V T15, T19, T10, T12; Chris@10: T15 = VFNMS(LDK(KP923879532), T14, T13); Chris@10: T19 = VFMA(LDK(KP923879532), T14, T13); Chris@10: T10 = VFNMS(LDK(KP923879532), TZ, TW); Chris@10: T12 = VFMA(LDK(KP923879532), TZ, TW); Chris@10: ST(&(xo[WS(os, 4)]), VFMAI(TG, TF), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 12)]), VFNMSI(TG, TF), ovs, &(xo[0])); Chris@10: ST(&(xo[0]), VADD(TB, TE), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(TB, TE), ovs, &(xo[0])); Chris@10: { Chris@10: V Tw, Ty, Tn, Tx; Chris@10: Tw = VFNMS(LDK(KP707106781), Tv, Tu); Chris@10: Ty = VFMA(LDK(KP707106781), Tv, Tu); Chris@10: Tn = VFNMS(LDK(KP707106781), Tm, T7); Chris@10: Tx = VFMA(LDK(KP707106781), Tm, T7); Chris@10: ST(&(xo[WS(os, 3)]), VFMAI(T1a, T19), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 13)]), VFNMSI(T1a, T19), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 11)]), VFMAI(T18, T15), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VFNMSI(T18, T15), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VFNMSI(T12, T11), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 15)]), VFMAI(T12, T11), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 7)]), VFMAI(T10, TT), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VFNMSI(T10, TT), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 14)]), VFNMSI(Ty, Tx), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 2)]), VFMAI(Ty, Tx), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 10)]), VFMAI(Tw, Tn), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VFNMSI(Tw, Tn), ovs, &(xo[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 16, XSIMD_STRING("n1fv_16"), {38, 0, 34, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_16) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_16, &desc); Chris@10: } Chris@10: Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_notw_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 16 -name n1fv_16 -include n1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 72 FP additions, 12 FP multiplications, Chris@10: * (or, 68 additions, 8 multiplications, 4 fused multiply/add), Chris@10: * 30 stack variables, 3 constants, and 32 memory accesses Chris@10: */ Chris@10: #include "n1f.h" Chris@10: Chris@10: static void n1fv_16(const R *ri, const R *ii, R *ro, R *io, stride is, stride os, INT v, INT ivs, INT ovs) Chris@10: { Chris@10: DVK(KP923879532, +0.923879532511286756128183189396788286822416626); Chris@10: DVK(KP382683432, +0.382683432365089771728459984030398866761344562); Chris@10: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); Chris@10: { Chris@10: INT i; Chris@10: const R *xi; Chris@10: R *xo; Chris@10: xi = ri; Chris@10: xo = ro; Chris@10: for (i = v; i > 0; i = i - VL, xi = xi + (VL * ivs), xo = xo + (VL * ovs), MAKE_VOLATILE_STRIDE(32, is), MAKE_VOLATILE_STRIDE(32, os)) { Chris@10: V Tp, T13, Tu, TN, Tm, T14, Tv, TY, T7, T17, Ty, TT, Te, T16, Tx; Chris@10: V TQ; Chris@10: { Chris@10: V Tn, To, TM, Ts, Tt, TL; Chris@10: Tn = LD(&(xi[WS(is, 4)]), ivs, &(xi[0])); Chris@10: To = LD(&(xi[WS(is, 12)]), ivs, &(xi[0])); Chris@10: TM = VADD(Tn, To); Chris@10: Ts = LD(&(xi[0]), ivs, &(xi[0])); Chris@10: Tt = LD(&(xi[WS(is, 8)]), ivs, &(xi[0])); Chris@10: TL = VADD(Ts, Tt); Chris@10: Tp = VSUB(Tn, To); Chris@10: T13 = VADD(TL, TM); Chris@10: Tu = VSUB(Ts, Tt); Chris@10: TN = VSUB(TL, TM); Chris@10: } Chris@10: { Chris@10: V Ti, TW, Tl, TX; Chris@10: { Chris@10: V Tg, Th, Tj, Tk; Chris@10: Tg = LD(&(xi[WS(is, 14)]), ivs, &(xi[0])); Chris@10: Th = LD(&(xi[WS(is, 6)]), ivs, &(xi[0])); Chris@10: Ti = VSUB(Tg, Th); Chris@10: TW = VADD(Tg, Th); Chris@10: Tj = LD(&(xi[WS(is, 2)]), ivs, &(xi[0])); Chris@10: Tk = LD(&(xi[WS(is, 10)]), ivs, &(xi[0])); Chris@10: Tl = VSUB(Tj, Tk); Chris@10: TX = VADD(Tj, Tk); Chris@10: } Chris@10: Tm = VMUL(LDK(KP707106781), VSUB(Ti, Tl)); Chris@10: T14 = VADD(TX, TW); Chris@10: Tv = VMUL(LDK(KP707106781), VADD(Tl, Ti)); Chris@10: TY = VSUB(TW, TX); Chris@10: } Chris@10: { Chris@10: V T3, TR, T6, TS; Chris@10: { Chris@10: V T1, T2, T4, T5; Chris@10: T1 = LD(&(xi[WS(is, 15)]), ivs, &(xi[WS(is, 1)])); Chris@10: T2 = LD(&(xi[WS(is, 7)]), ivs, &(xi[WS(is, 1)])); Chris@10: T3 = VSUB(T1, T2); Chris@10: TR = VADD(T1, T2); Chris@10: T4 = LD(&(xi[WS(is, 3)]), ivs, &(xi[WS(is, 1)])); Chris@10: T5 = LD(&(xi[WS(is, 11)]), ivs, &(xi[WS(is, 1)])); Chris@10: T6 = VSUB(T4, T5); Chris@10: TS = VADD(T4, T5); Chris@10: } Chris@10: T7 = VFNMS(LDK(KP923879532), T6, VMUL(LDK(KP382683432), T3)); Chris@10: T17 = VADD(TR, TS); Chris@10: Ty = VFMA(LDK(KP923879532), T3, VMUL(LDK(KP382683432), T6)); Chris@10: TT = VSUB(TR, TS); Chris@10: } Chris@10: { Chris@10: V Ta, TO, Td, TP; Chris@10: { Chris@10: V T8, T9, Tb, Tc; Chris@10: T8 = LD(&(xi[WS(is, 1)]), ivs, &(xi[WS(is, 1)])); Chris@10: T9 = LD(&(xi[WS(is, 9)]), ivs, &(xi[WS(is, 1)])); Chris@10: Ta = VSUB(T8, T9); Chris@10: TO = VADD(T8, T9); Chris@10: Tb = LD(&(xi[WS(is, 5)]), ivs, &(xi[WS(is, 1)])); Chris@10: Tc = LD(&(xi[WS(is, 13)]), ivs, &(xi[WS(is, 1)])); Chris@10: Td = VSUB(Tb, Tc); Chris@10: TP = VADD(Tb, Tc); Chris@10: } Chris@10: Te = VFMA(LDK(KP382683432), Ta, VMUL(LDK(KP923879532), Td)); Chris@10: T16 = VADD(TO, TP); Chris@10: Tx = VFNMS(LDK(KP382683432), Td, VMUL(LDK(KP923879532), Ta)); Chris@10: TQ = VSUB(TO, TP); Chris@10: } Chris@10: { Chris@10: V T15, T18, T19, T1a; Chris@10: T15 = VADD(T13, T14); Chris@10: T18 = VADD(T16, T17); Chris@10: ST(&(xo[WS(os, 8)]), VSUB(T15, T18), ovs, &(xo[0])); Chris@10: ST(&(xo[0]), VADD(T15, T18), ovs, &(xo[0])); Chris@10: T19 = VSUB(T13, T14); Chris@10: T1a = VBYI(VSUB(T17, T16)); Chris@10: ST(&(xo[WS(os, 12)]), VSUB(T19, T1a), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 4)]), VADD(T19, T1a), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V TV, T11, T10, T12, TU, TZ; Chris@10: TU = VMUL(LDK(KP707106781), VADD(TQ, TT)); Chris@10: TV = VADD(TN, TU); Chris@10: T11 = VSUB(TN, TU); Chris@10: TZ = VMUL(LDK(KP707106781), VSUB(TT, TQ)); Chris@10: T10 = VBYI(VADD(TY, TZ)); Chris@10: T12 = VBYI(VSUB(TZ, TY)); Chris@10: ST(&(xo[WS(os, 14)]), VSUB(TV, T10), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 6)]), VADD(T11, T12), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 2)]), VADD(TV, T10), ovs, &(xo[0])); Chris@10: ST(&(xo[WS(os, 10)]), VSUB(T11, T12), ovs, &(xo[0])); Chris@10: } Chris@10: { Chris@10: V Tr, TB, TA, TC; Chris@10: { Chris@10: V Tf, Tq, Tw, Tz; Chris@10: Tf = VSUB(T7, Te); Chris@10: Tq = VSUB(Tm, Tp); Chris@10: Tr = VBYI(VSUB(Tf, Tq)); Chris@10: TB = VBYI(VADD(Tq, Tf)); Chris@10: Tw = VADD(Tu, Tv); Chris@10: Tz = VADD(Tx, Ty); Chris@10: TA = VSUB(Tw, Tz); Chris@10: TC = VADD(Tw, Tz); Chris@10: } Chris@10: ST(&(xo[WS(os, 7)]), VADD(Tr, TA), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 15)]), VSUB(TC, TB), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 9)]), VSUB(TA, Tr), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 1)]), VADD(TB, TC), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: { Chris@10: V TF, TJ, TI, TK; Chris@10: { Chris@10: V TD, TE, TG, TH; Chris@10: TD = VSUB(Tu, Tv); Chris@10: TE = VADD(Te, T7); Chris@10: TF = VADD(TD, TE); Chris@10: TJ = VSUB(TD, TE); Chris@10: TG = VADD(Tp, Tm); Chris@10: TH = VSUB(Ty, Tx); Chris@10: TI = VBYI(VADD(TG, TH)); Chris@10: TK = VBYI(VSUB(TH, TG)); Chris@10: } Chris@10: ST(&(xo[WS(os, 13)]), VSUB(TF, TI), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 5)]), VADD(TJ, TK), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 3)]), VADD(TF, TI), ovs, &(xo[WS(os, 1)])); Chris@10: ST(&(xo[WS(os, 11)]), VSUB(TJ, TK), ovs, &(xo[WS(os, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const kdft_desc desc = { 16, XSIMD_STRING("n1fv_16"), {68, 8, 4, 0}, &GENUS, 0, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_n1fv_16) (planner *p) { Chris@10: X(kdft_register) (p, n1fv_16, &desc); Chris@10: } Chris@10: Chris@10: #endif /* HAVE_FMA */