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 07:39:26 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.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 4 -name t2sv_4 -include ts.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 37 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "ts.h" cannam@95: cannam@95: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 4); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 4), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@95: V T2, T6, T3, T5, T1, Tx, T8, Tc, Tf, Ta, T4, Th, Tj, Tl; cannam@95: T2 = LDW(&(W[0])); cannam@95: T6 = LDW(&(W[TWVL * 3])); cannam@95: T3 = LDW(&(W[TWVL * 2])); cannam@95: T5 = LDW(&(W[TWVL * 1])); cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: Tx = LD(&(ii[0]), ms, &(ii[0])); cannam@95: T8 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: Tc = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: Tf = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: Ta = VMUL(T2, T6); cannam@95: T4 = VMUL(T2, T3); cannam@95: Th = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: Tj = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: Tl = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: { cannam@95: V Tg, Tb, T7, Tp, Tk, Tr, Ti; cannam@95: Tg = VMUL(T2, Tf); cannam@95: Tb = VFNMS(T5, T3, Ta); cannam@95: T7 = VFMA(T5, T6, T4); cannam@95: Tp = VMUL(T2, Th); cannam@95: Tk = VMUL(T3, Tj); cannam@95: Tr = VMUL(T3, Tl); cannam@95: Ti = VFMA(T5, Th, Tg); cannam@95: { cannam@95: V Tv, T9, Tq, Tm, Ts, Tw, Td; cannam@95: Tv = VMUL(T7, Tc); cannam@95: T9 = VMUL(T7, T8); cannam@95: Tq = VFNMS(T5, Tf, Tp); cannam@95: Tm = VFMA(T6, Tl, Tk); cannam@95: Ts = VFNMS(T6, Tj, Tr); cannam@95: Tw = VFNMS(Tb, T8, Tv); cannam@95: Td = VFMA(Tb, Tc, T9); cannam@95: { cannam@95: V Tn, TA, Tu, Tt; cannam@95: Tn = VADD(Ti, Tm); cannam@95: TA = VSUB(Ti, Tm); cannam@95: Tu = VADD(Tq, Ts); cannam@95: Tt = VSUB(Tq, Ts); cannam@95: { cannam@95: V Ty, Tz, Te, To; cannam@95: Ty = VADD(Tw, Tx); cannam@95: Tz = VSUB(Tx, Tw); cannam@95: Te = VADD(T1, Td); cannam@95: To = VSUB(T1, Td); cannam@95: ST(&(ii[WS(rs, 3)]), VADD(TA, Tz), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 1)]), VSUB(Tz, TA), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 2)]), VSUB(Ty, Tu), ms, &(ii[0])); cannam@95: ST(&(ii[0]), VADD(Tu, Ty), ms, &(ii[0])); cannam@95: ST(&(ri[WS(rs, 1)]), VADD(To, Tt), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 3)]), VSUB(To, Tt), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[0]), VADD(Te, Tn), ms, &(ri[0])); cannam@95: ST(&(ri[WS(rs, 2)]), VSUB(Te, Tn), ms, &(ri[0])); cannam@95: } 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, 3), cannam@95: {TW_NEXT, (2 * VL), 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t2sv_4) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2sv_4, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -n 4 -name t2sv_4 -include ts.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 24 FP additions, 16 FP multiplications, cannam@95: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@95: * 21 stack variables, 0 constants, and 16 memory accesses cannam@95: */ cannam@95: #include "ts.h" cannam@95: cannam@95: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + (mb * 4); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 4), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@95: V T2, T4, T3, T5, T6, T8; cannam@95: T2 = LDW(&(W[0])); cannam@95: T4 = LDW(&(W[TWVL * 1])); cannam@95: T3 = LDW(&(W[TWVL * 2])); cannam@95: T5 = LDW(&(W[TWVL * 3])); cannam@95: T6 = VFMA(T2, T3, VMUL(T4, T5)); cannam@95: T8 = VFNMS(T4, T3, VMUL(T2, T5)); cannam@95: { cannam@95: V T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; cannam@95: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@95: Tp = LD(&(ii[0]), ms, &(ii[0])); cannam@95: T7 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@95: T9 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@95: Ta = VFMA(T6, T7, VMUL(T8, T9)); cannam@95: To = VFNMS(T8, T7, VMUL(T6, T9)); cannam@95: { cannam@95: V Tc, Td, Tf, Tg; cannam@95: Tc = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@95: Td = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@95: Te = VFMA(T2, Tc, VMUL(T4, Td)); cannam@95: Tk = VFNMS(T4, Tc, VMUL(T2, Td)); cannam@95: Tf = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@95: Tg = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@95: Th = VFMA(T3, Tf, VMUL(T5, Tg)); cannam@95: Tl = VFNMS(T5, Tf, VMUL(T3, Tg)); cannam@95: } cannam@95: { cannam@95: V Tb, Ti, Tn, Tq; cannam@95: Tb = VADD(T1, Ta); cannam@95: Ti = VADD(Te, Th); cannam@95: ST(&(ri[WS(rs, 2)]), VSUB(Tb, Ti), ms, &(ri[0])); cannam@95: ST(&(ri[0]), VADD(Tb, Ti), ms, &(ri[0])); cannam@95: Tn = VADD(Tk, Tl); cannam@95: Tq = VADD(To, Tp); cannam@95: ST(&(ii[0]), VADD(Tn, Tq), ms, &(ii[0])); cannam@95: ST(&(ii[WS(rs, 2)]), VSUB(Tq, Tn), ms, &(ii[0])); cannam@95: } cannam@95: { cannam@95: V Tj, Tm, Tr, Ts; cannam@95: Tj = VSUB(T1, Ta); cannam@95: Tm = VSUB(Tk, Tl); cannam@95: ST(&(ri[WS(rs, 3)]), VSUB(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@95: ST(&(ri[WS(rs, 1)]), VADD(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@95: Tr = VSUB(Tp, To); cannam@95: Ts = VSUB(Te, Th); cannam@95: ST(&(ii[WS(rs, 1)]), VSUB(Tr, Ts), ms, &(ii[WS(rs, 1)])); cannam@95: ST(&(ii[WS(rs, 3)]), VADD(Ts, Tr), ms, &(ii[WS(rs, 1)])); 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, 3), cannam@95: {TW_NEXT, (2 * VL), 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t2sv_4) (planner *p) { cannam@95: X(kdft_dit_register) (p, t2sv_4, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */