cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:45:18 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* 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@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 37 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "ts.h" cannam@127: cannam@127: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: V T2, T6, T3, T5, T1, Tx, T8, Tc, Tf, Ta, T4, Th, Tj, Tl; cannam@127: T2 = LDW(&(W[0])); cannam@127: T6 = LDW(&(W[TWVL * 3])); cannam@127: T3 = LDW(&(W[TWVL * 2])); cannam@127: T5 = LDW(&(W[TWVL * 1])); cannam@127: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@127: Tx = LD(&(ii[0]), ms, &(ii[0])); cannam@127: T8 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@127: Tc = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@127: Tf = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@127: Ta = VMUL(T2, T6); cannam@127: T4 = VMUL(T2, T3); cannam@127: Th = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@127: Tj = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@127: Tl = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@127: { cannam@127: V Tg, Tb, T7, Tp, Tk, Tr, Ti; cannam@127: Tg = VMUL(T2, Tf); cannam@127: Tb = VFNMS(T5, T3, Ta); cannam@127: T7 = VFMA(T5, T6, T4); cannam@127: Tp = VMUL(T2, Th); cannam@127: Tk = VMUL(T3, Tj); cannam@127: Tr = VMUL(T3, Tl); cannam@127: Ti = VFMA(T5, Th, Tg); cannam@127: { cannam@127: V Tv, T9, Tq, Tm, Ts, Tw, Td; cannam@127: Tv = VMUL(T7, Tc); cannam@127: T9 = VMUL(T7, T8); cannam@127: Tq = VFNMS(T5, Tf, Tp); cannam@127: Tm = VFMA(T6, Tl, Tk); cannam@127: Ts = VFNMS(T6, Tj, Tr); cannam@127: Tw = VFNMS(Tb, T8, Tv); cannam@127: Td = VFMA(Tb, Tc, T9); cannam@127: { cannam@127: V Tn, TA, Tu, Tt; cannam@127: Tn = VADD(Ti, Tm); cannam@127: TA = VSUB(Ti, Tm); cannam@127: Tu = VADD(Tq, Ts); cannam@127: Tt = VSUB(Tq, Ts); cannam@127: { cannam@127: V Ty, Tz, Te, To; cannam@127: Ty = VADD(Tw, Tx); cannam@127: Tz = VSUB(Tx, Tw); cannam@127: Te = VADD(T1, Td); cannam@127: To = VSUB(T1, Td); cannam@127: ST(&(ii[WS(rs, 3)]), VADD(TA, Tz), ms, &(ii[WS(rs, 1)])); cannam@127: ST(&(ii[WS(rs, 1)]), VSUB(Tz, TA), ms, &(ii[WS(rs, 1)])); cannam@127: ST(&(ii[WS(rs, 2)]), VSUB(Ty, Tu), ms, &(ii[0])); cannam@127: ST(&(ii[0]), VADD(Tu, Ty), ms, &(ii[0])); cannam@127: ST(&(ri[WS(rs, 1)]), VADD(To, Tt), ms, &(ri[WS(rs, 1)])); cannam@127: ST(&(ri[WS(rs, 3)]), VSUB(To, Tt), ms, &(ri[WS(rs, 1)])); cannam@127: ST(&(ri[0]), VADD(Te, Tn), ms, &(ri[0])); cannam@127: ST(&(ri[WS(rs, 2)]), VSUB(Te, Tn), ms, &(ri[0])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 3), cannam@127: {TW_NEXT, (2 * VL), 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t2sv_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2sv_4, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* 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@127: cannam@127: /* cannam@127: * This function contains 24 FP additions, 16 FP multiplications, cannam@127: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), cannam@127: * 21 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "ts.h" cannam@127: cannam@127: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: { cannam@127: INT m; cannam@127: 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@127: V T2, T4, T3, T5, T6, T8; cannam@127: T2 = LDW(&(W[0])); cannam@127: T4 = LDW(&(W[TWVL * 1])); cannam@127: T3 = LDW(&(W[TWVL * 2])); cannam@127: T5 = LDW(&(W[TWVL * 3])); cannam@127: T6 = VFMA(T2, T3, VMUL(T4, T5)); cannam@127: T8 = VFNMS(T4, T3, VMUL(T2, T5)); cannam@127: { cannam@127: V T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; cannam@127: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@127: Tp = LD(&(ii[0]), ms, &(ii[0])); cannam@127: T7 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@127: T9 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@127: Ta = VFMA(T6, T7, VMUL(T8, T9)); cannam@127: To = VFNMS(T8, T7, VMUL(T6, T9)); cannam@127: { cannam@127: V Tc, Td, Tf, Tg; cannam@127: Tc = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@127: Td = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@127: Te = VFMA(T2, Tc, VMUL(T4, Td)); cannam@127: Tk = VFNMS(T4, Tc, VMUL(T2, Td)); cannam@127: Tf = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@127: Tg = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@127: Th = VFMA(T3, Tf, VMUL(T5, Tg)); cannam@127: Tl = VFNMS(T5, Tf, VMUL(T3, Tg)); cannam@127: } cannam@127: { cannam@127: V Tb, Ti, Tn, Tq; cannam@127: Tb = VADD(T1, Ta); cannam@127: Ti = VADD(Te, Th); cannam@127: ST(&(ri[WS(rs, 2)]), VSUB(Tb, Ti), ms, &(ri[0])); cannam@127: ST(&(ri[0]), VADD(Tb, Ti), ms, &(ri[0])); cannam@127: Tn = VADD(Tk, Tl); cannam@127: Tq = VADD(To, Tp); cannam@127: ST(&(ii[0]), VADD(Tn, Tq), ms, &(ii[0])); cannam@127: ST(&(ii[WS(rs, 2)]), VSUB(Tq, Tn), ms, &(ii[0])); cannam@127: } cannam@127: { cannam@127: V Tj, Tm, Tr, Ts; cannam@127: Tj = VSUB(T1, Ta); cannam@127: Tm = VSUB(Tk, Tl); cannam@127: ST(&(ri[WS(rs, 3)]), VSUB(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@127: ST(&(ri[WS(rs, 1)]), VADD(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@127: Tr = VSUB(Tp, To); cannam@127: Ts = VSUB(Te, Th); cannam@127: ST(&(ii[WS(rs, 1)]), VSUB(Tr, Ts), ms, &(ii[WS(rs, 1)])); cannam@127: ST(&(ii[WS(rs, 3)]), VADD(Ts, Tr), ms, &(ii[WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 3), cannam@127: {TW_NEXT, (2 * VL), 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t2sv_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t2sv_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */