cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:38:03 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1fv_12 -include t1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 59 FP additions, 42 FP multiplications, cannam@95: * (or, 41 additions, 24 multiplications, 18 fused multiply/add), cannam@95: * 41 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "t1f.h" cannam@95: cannam@95: static void t1fv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { cannam@95: V Tq, Ti, T7, TQ, Tu, TA, TU, Tk, TR, Tf, TE, TM; cannam@95: { cannam@95: V T9, TC, Tj, TD, Te; cannam@95: { cannam@95: V T1, T4, T2, Tm, Tx, To; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Tm = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Tx = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: To = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V T5, T3, Tn, Ty, Tp, Td, Tb, T8, Tc, Ta; cannam@95: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); cannam@95: T5 = BYTWJ(&(W[TWVL * 14]), T4); cannam@95: T3 = BYTWJ(&(W[TWVL * 6]), T2); cannam@95: Tn = BYTWJ(&(W[0]), Tm); cannam@95: Ty = BYTWJ(&(W[TWVL * 16]), Tx); cannam@95: Tp = BYTWJ(&(W[TWVL * 8]), To); cannam@95: T9 = BYTWJ(&(W[TWVL * 10]), T8); cannam@95: Td = BYTWJ(&(W[TWVL * 2]), Tc); cannam@95: Tb = BYTWJ(&(W[TWVL * 18]), Ta); cannam@95: { cannam@95: V Th, T6, Tt, Tz; cannam@95: Th = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@95: TC = VSUB(T5, T3); cannam@95: T6 = VADD(T3, T5); cannam@95: Tt = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: Tz = VADD(Tn, Tp); cannam@95: Tq = VSUB(Tn, Tp); cannam@95: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: TD = VSUB(Td, Tb); cannam@95: Te = VADD(Tb, Td); cannam@95: Ti = BYTWJ(&(W[TWVL * 20]), Th); cannam@95: T7 = VFNMS(LDK(KP500000000), T6, T1); cannam@95: TQ = VADD(T1, T6); cannam@95: Tu = BYTWJ(&(W[TWVL * 4]), Tt); cannam@95: TA = VFNMS(LDK(KP500000000), Tz, Ty); cannam@95: TU = VADD(Ty, Tz); cannam@95: } cannam@95: } cannam@95: } cannam@95: Tk = BYTWJ(&(W[TWVL * 12]), Tj); cannam@95: TR = VADD(T9, Te); cannam@95: Tf = VFNMS(LDK(KP500000000), Te, T9); cannam@95: TE = VSUB(TC, TD); cannam@95: TM = VADD(TC, TD); cannam@95: } cannam@95: { cannam@95: V Tv, Tl, TI, Tg, TW, TS; cannam@95: Tv = VADD(Tk, Ti); cannam@95: Tl = VSUB(Ti, Tk); cannam@95: TI = VADD(T7, Tf); cannam@95: Tg = VSUB(T7, Tf); cannam@95: TW = VADD(TQ, TR); cannam@95: TS = VSUB(TQ, TR); cannam@95: { cannam@95: V TT, Tw, TL, Tr; cannam@95: TT = VADD(Tu, Tv); cannam@95: Tw = VFNMS(LDK(KP500000000), Tv, Tu); cannam@95: TL = VSUB(Tl, Tq); cannam@95: Tr = VADD(Tl, Tq); cannam@95: { cannam@95: V TP, TN, TG, Ts, TO, TK, TH, TF; cannam@95: { cannam@95: V TX, TV, TJ, TB; cannam@95: TX = VADD(TT, TU); cannam@95: TV = VSUB(TT, TU); cannam@95: TJ = VADD(Tw, TA); cannam@95: TB = VSUB(Tw, TA); cannam@95: TP = VMUL(LDK(KP866025403), VADD(TM, TL)); cannam@95: TN = VMUL(LDK(KP866025403), VSUB(TL, TM)); cannam@95: TG = VFNMS(LDK(KP866025403), Tr, Tg); cannam@95: Ts = VFMA(LDK(KP866025403), Tr, Tg); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(TW, TX), ms, &(x[0])); cannam@95: ST(&(x[0]), VADD(TW, TX), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 3)]), VFMAI(TV, TS), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 9)]), VFNMSI(TV, TS), ms, &(x[WS(rs, 1)])); cannam@95: TO = VADD(TI, TJ); cannam@95: TK = VSUB(TI, TJ); cannam@95: TH = VFMA(LDK(KP866025403), TE, TB); cannam@95: TF = VFNMS(LDK(KP866025403), TE, TB); cannam@95: } cannam@95: ST(&(x[WS(rs, 4)]), VFMAI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 8)]), VFNMSI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 10)]), VFNMSI(TN, TK), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 2)]), VFMAI(TN, TK), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 5)]), VFNMSI(TH, TG), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VFMAI(TH, TG), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 11)]), VFMAI(TF, Ts), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VFNMSI(TF, Ts), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: VTW(0, 9), cannam@95: VTW(0, 10), cannam@95: VTW(0, 11), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 12, XSIMD_STRING("t1fv_12"), twinstr, &GENUS, {41, 24, 18, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fv_12) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fv_12, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1fv_12 -include t1f.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 59 FP additions, 30 FP multiplications, cannam@95: * (or, 55 additions, 26 multiplications, 4 fused multiply/add), cannam@95: * 28 stack variables, 2 constants, and 24 memory accesses cannam@95: */ cannam@95: #include "t1f.h" cannam@95: cannam@95: static void t1fv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); cannam@95: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ri; cannam@95: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { cannam@95: V T1, TH, T6, TA, Tq, TE, Tv, TL, T9, TI, Te, TB, Ti, TD, Tn; cannam@95: V TK; cannam@95: { cannam@95: V T5, T3, T4, T2; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: T5 = BYTWJ(&(W[TWVL * 14]), T4); cannam@95: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: T3 = BYTWJ(&(W[TWVL * 6]), T2); cannam@95: TH = VSUB(T5, T3); cannam@95: T6 = VADD(T3, T5); cannam@95: TA = VFNMS(LDK(KP500000000), T6, T1); cannam@95: } cannam@95: { cannam@95: V Tu, Ts, Tp, Tt, Tr; cannam@95: Tp = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: Tq = BYTWJ(&(W[TWVL * 16]), Tp); cannam@95: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tu = BYTWJ(&(W[TWVL * 8]), Tt); cannam@95: Tr = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Ts = BYTWJ(&(W[0]), Tr); cannam@95: TE = VSUB(Tu, Ts); cannam@95: Tv = VADD(Ts, Tu); cannam@95: TL = VFNMS(LDK(KP500000000), Tv, Tq); cannam@95: } cannam@95: { cannam@95: V Td, Tb, T8, Tc, Ta; cannam@95: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: T9 = BYTWJ(&(W[TWVL * 10]), T8); cannam@95: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Td = BYTWJ(&(W[TWVL * 2]), Tc); cannam@95: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); cannam@95: Tb = BYTWJ(&(W[TWVL * 18]), Ta); cannam@95: TI = VSUB(Td, Tb); cannam@95: Te = VADD(Tb, Td); cannam@95: TB = VFNMS(LDK(KP500000000), Te, T9); cannam@95: } cannam@95: { cannam@95: V Tm, Tk, Th, Tl, Tj; cannam@95: Th = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: Ti = BYTWJ(&(W[TWVL * 4]), Th); cannam@95: Tl = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@95: Tm = BYTWJ(&(W[TWVL * 20]), Tl); cannam@95: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Tk = BYTWJ(&(W[TWVL * 12]), Tj); cannam@95: TD = VSUB(Tm, Tk); cannam@95: Tn = VADD(Tk, Tm); cannam@95: TK = VFNMS(LDK(KP500000000), Tn, Ti); cannam@95: } cannam@95: { cannam@95: V Tg, Ty, Tx, Tz; cannam@95: { cannam@95: V T7, Tf, To, Tw; cannam@95: T7 = VADD(T1, T6); cannam@95: Tf = VADD(T9, Te); cannam@95: Tg = VSUB(T7, Tf); cannam@95: Ty = VADD(T7, Tf); cannam@95: To = VADD(Ti, Tn); cannam@95: Tw = VADD(Tq, Tv); cannam@95: Tx = VBYI(VSUB(To, Tw)); cannam@95: Tz = VADD(To, Tw); cannam@95: } cannam@95: ST(&(x[WS(rs, 9)]), VSUB(Tg, Tx), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[0]), VADD(Ty, Tz), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 3)]), VADD(Tg, Tx), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(Ty, Tz), ms, &(x[0])); cannam@95: } cannam@95: { cannam@95: V TS, TW, TV, TX; cannam@95: { cannam@95: V TQ, TR, TT, TU; cannam@95: TQ = VADD(TA, TB); cannam@95: TR = VADD(TK, TL); cannam@95: TS = VSUB(TQ, TR); cannam@95: TW = VADD(TQ, TR); cannam@95: TT = VADD(TD, TE); cannam@95: TU = VADD(TH, TI); cannam@95: TV = VBYI(VMUL(LDK(KP866025403), VSUB(TT, TU))); cannam@95: TX = VBYI(VMUL(LDK(KP866025403), VADD(TU, TT))); cannam@95: } cannam@95: ST(&(x[WS(rs, 10)]), VSUB(TS, TV), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 4)]), VADD(TW, TX), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 2)]), VADD(TS, TV), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 8)]), VSUB(TW, TX), ms, &(x[0])); cannam@95: } cannam@95: { cannam@95: V TG, TP, TN, TO; cannam@95: { cannam@95: V TC, TF, TJ, TM; cannam@95: TC = VSUB(TA, TB); cannam@95: TF = VMUL(LDK(KP866025403), VSUB(TD, TE)); cannam@95: TG = VSUB(TC, TF); cannam@95: TP = VADD(TC, TF); cannam@95: TJ = VMUL(LDK(KP866025403), VSUB(TH, TI)); cannam@95: TM = VSUB(TK, TL); cannam@95: TN = VBYI(VADD(TJ, TM)); cannam@95: TO = VBYI(VSUB(TJ, TM)); cannam@95: } cannam@95: ST(&(x[WS(rs, 5)]), VSUB(TG, TN), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 11)]), VSUB(TP, TO), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VADD(TN, TG), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VADD(TO, TP), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: VTW(0, 9), cannam@95: VTW(0, 10), cannam@95: VTW(0, 11), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 12, XSIMD_STRING("t1fv_12"), twinstr, &GENUS, {55, 26, 4, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1fv_12) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1fv_12, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */