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 07:39:04 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 t1bv_12 -include t1b.h -sign 1 */ 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 "t1b.h" cannam@95: cannam@95: static void t1bv_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 = ii; 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 TI, Ti, TA, T7, Tm, TE, Tw, Tk, Tf, TB, TU, TM; cannam@95: { cannam@95: V T9, TK, Tj, TL, Te; cannam@95: { cannam@95: V T1, T4, T2, Tp, Tt, Tr; 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: Tp = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tr = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V T5, T3, Tq, Tu, Ts, 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 = BYTW(&(W[TWVL * 14]), T4); cannam@95: T3 = BYTW(&(W[TWVL * 6]), T2); cannam@95: Tq = BYTW(&(W[TWVL * 16]), Tp); cannam@95: Tu = BYTW(&(W[TWVL * 8]), Tt); cannam@95: Ts = BYTW(&(W[0]), Tr); cannam@95: T9 = BYTW(&(W[TWVL * 10]), T8); cannam@95: Td = BYTW(&(W[TWVL * 2]), Tc); cannam@95: Tb = BYTW(&(W[TWVL * 18]), Ta); cannam@95: { cannam@95: V Th, T6, Tl, Tv; cannam@95: Th = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: TK = VSUB(T3, T5); cannam@95: T6 = VADD(T3, T5); cannam@95: Tl = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@95: Tv = VADD(Ts, Tu); cannam@95: TI = VSUB(Tu, Ts); cannam@95: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: TL = VSUB(Tb, Td); cannam@95: Te = VADD(Tb, Td); cannam@95: Ti = BYTW(&(W[TWVL * 4]), Th); cannam@95: TA = VFNMS(LDK(KP500000000), T6, T1); cannam@95: T7 = VADD(T1, T6); cannam@95: Tm = BYTW(&(W[TWVL * 20]), Tl); cannam@95: TE = VFNMS(LDK(KP500000000), Tv, Tq); cannam@95: Tw = VADD(Tq, Tv); cannam@95: } cannam@95: } cannam@95: } cannam@95: Tk = BYTW(&(W[TWVL * 12]), Tj); cannam@95: Tf = VADD(T9, Te); cannam@95: TB = VFNMS(LDK(KP500000000), Te, T9); cannam@95: TU = VSUB(TK, TL); cannam@95: TM = VADD(TK, TL); cannam@95: } cannam@95: { cannam@95: V Tn, TH, TC, TQ, Ty, Tg; cannam@95: Tn = VADD(Tk, Tm); cannam@95: TH = VSUB(Tk, Tm); cannam@95: TC = VADD(TA, TB); cannam@95: TQ = VSUB(TA, TB); cannam@95: Ty = VADD(T7, Tf); cannam@95: Tg = VSUB(T7, Tf); cannam@95: { cannam@95: V To, TD, TJ, TR; cannam@95: To = VADD(Ti, Tn); cannam@95: TD = VFNMS(LDK(KP500000000), Tn, Ti); cannam@95: TJ = VSUB(TH, TI); cannam@95: TR = VADD(TH, TI); cannam@95: { cannam@95: V TP, TN, TW, TS, TO, TG, TX, TV; cannam@95: { cannam@95: V Tz, Tx, TF, TT; cannam@95: Tz = VADD(To, Tw); cannam@95: Tx = VSUB(To, Tw); cannam@95: TF = VADD(TD, TE); cannam@95: TT = VSUB(TD, TE); cannam@95: TP = VMUL(LDK(KP866025403), VADD(TM, TJ)); cannam@95: TN = VMUL(LDK(KP866025403), VSUB(TJ, TM)); cannam@95: TW = VFMA(LDK(KP866025403), TR, TQ); cannam@95: TS = VFNMS(LDK(KP866025403), TR, TQ); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(Ty, Tz), ms, &(x[0])); cannam@95: ST(&(x[0]), VADD(Ty, Tz), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 9)]), VFMAI(Tx, Tg), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 3)]), VFNMSI(Tx, Tg), ms, &(x[WS(rs, 1)])); cannam@95: TO = VADD(TC, TF); cannam@95: TG = VSUB(TC, TF); cannam@95: TX = VFNMS(LDK(KP866025403), TU, TT); cannam@95: TV = VFMA(LDK(KP866025403), TU, TT); cannam@95: } cannam@95: ST(&(x[WS(rs, 8)]), VFNMSI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 4)]), VFMAI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 2)]), VFMAI(TN, TG), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 10)]), VFNMSI(TN, TG), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 5)]), VFMAI(TX, TW), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VFNMSI(TX, TW), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 11)]), VFNMSI(TV, TS), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VFMAI(TV, 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("t1bv_12"), twinstr, &GENUS, {41, 24, 18, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1bv_12) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1bv_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 t1bv_12 -include t1b.h -sign 1 */ 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 "t1b.h" cannam@95: cannam@95: static void t1bv_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 = ii; 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, Tt, T6, T7, TB, Tq, TC, TD, T9, Tu, Te, Tf, Tx, Tl, Ty; cannam@95: V Tz; 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 = BYTW(&(W[TWVL * 14]), T4); cannam@95: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: T3 = BYTW(&(W[TWVL * 6]), T2); cannam@95: Tt = VSUB(T3, T5); cannam@95: T6 = VADD(T3, T5); cannam@95: T7 = VFNMS(LDK(KP500000000), T6, T1); cannam@95: } cannam@95: { cannam@95: V Tn, Tp, Tm, TA, To; cannam@95: Tm = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Tn = BYTW(&(W[0]), Tm); cannam@95: TA = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: TB = BYTW(&(W[TWVL * 16]), TA); cannam@95: To = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tp = BYTW(&(W[TWVL * 8]), To); cannam@95: Tq = VSUB(Tn, Tp); cannam@95: TC = VADD(Tn, Tp); cannam@95: TD = VFNMS(LDK(KP500000000), TC, TB); 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 = BYTW(&(W[TWVL * 10]), T8); cannam@95: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Td = BYTW(&(W[TWVL * 2]), Tc); cannam@95: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); cannam@95: Tb = BYTW(&(W[TWVL * 18]), Ta); cannam@95: Tu = VSUB(Tb, Td); cannam@95: Te = VADD(Tb, Td); cannam@95: Tf = VFNMS(LDK(KP500000000), Te, T9); cannam@95: } cannam@95: { cannam@95: V Ti, Tk, Th, Tw, Tj; cannam@95: Th = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: Ti = BYTW(&(W[TWVL * 12]), Th); cannam@95: Tw = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: Tx = BYTW(&(W[TWVL * 4]), Tw); cannam@95: Tj = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); cannam@95: Tk = BYTW(&(W[TWVL * 20]), Tj); cannam@95: Tl = VSUB(Ti, Tk); cannam@95: Ty = VADD(Ti, Tk); cannam@95: Tz = VFNMS(LDK(KP500000000), Ty, Tx); cannam@95: } cannam@95: { cannam@95: V Ts, TG, TF, TH; cannam@95: { cannam@95: V Tg, Tr, Tv, TE; cannam@95: Tg = VSUB(T7, Tf); cannam@95: Tr = VMUL(LDK(KP866025403), VSUB(Tl, Tq)); cannam@95: Ts = VSUB(Tg, Tr); cannam@95: TG = VADD(Tg, Tr); cannam@95: Tv = VMUL(LDK(KP866025403), VSUB(Tt, Tu)); cannam@95: TE = VSUB(Tz, TD); cannam@95: TF = VBYI(VADD(Tv, TE)); cannam@95: TH = VBYI(VSUB(TE, Tv)); cannam@95: } cannam@95: ST(&(x[WS(rs, 11)]), VSUB(Ts, TF), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 5)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VADD(Ts, TF), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VSUB(TG, TH), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V TS, TW, TV, TX; cannam@95: { cannam@95: V TQ, TR, TT, TU; cannam@95: TQ = VADD(T1, T6); cannam@95: TR = VADD(T9, Te); cannam@95: TS = VSUB(TQ, TR); cannam@95: TW = VADD(TQ, TR); cannam@95: TT = VADD(Tx, Ty); cannam@95: TU = VADD(TB, TC); cannam@95: TV = VBYI(VSUB(TT, TU)); cannam@95: TX = VADD(TT, TU); cannam@95: } cannam@95: ST(&(x[WS(rs, 3)]), VSUB(TS, TV), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[0]), VADD(TW, TX), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 9)]), VADD(TS, TV), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 6)]), VSUB(TW, TX), ms, &(x[0])); cannam@95: } cannam@95: { cannam@95: V TK, TO, TN, TP; cannam@95: { cannam@95: V TI, TJ, TL, TM; cannam@95: TI = VADD(Tl, Tq); cannam@95: TJ = VADD(Tt, Tu); cannam@95: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); cannam@95: TO = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); cannam@95: TL = VADD(T7, Tf); cannam@95: TM = VADD(Tz, TD); cannam@95: TN = VSUB(TL, TM); cannam@95: TP = VADD(TL, TM); cannam@95: } cannam@95: ST(&(x[WS(rs, 2)]), VADD(TK, TN), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 8)]), VSUB(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 10)]), VSUB(TN, TK), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 4)]), VADD(TO, TP), ms, &(x[0])); 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("t1bv_12"), twinstr, &GENUS, {55, 26, 4, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1bv_12) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1bv_12, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */