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 16:45:04 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 -n 4 -name t1sv_4 -include ts.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 35 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "ts.h" cannam@127: cannam@127: static void t1sv_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 * 6); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 6), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: V T1, Tv, T3, T6, T5, Ta, Td, Tc, Tg, Tj, Tt, T4, Tf, Ti, Tn; cannam@127: V Tb, T2, T9; cannam@127: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@127: Tv = LD(&(ii[0]), ms, &(ii[0])); cannam@127: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@127: T6 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@127: T2 = LDW(&(W[TWVL * 2])); cannam@127: T5 = LDW(&(W[TWVL * 3])); cannam@127: Ta = 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: T9 = LDW(&(W[0])); cannam@127: Tc = LDW(&(W[TWVL * 1])); cannam@127: Tg = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@127: Tj = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@127: Tt = VMUL(T2, T6); cannam@127: T4 = VMUL(T2, T3); cannam@127: Tf = LDW(&(W[TWVL * 4])); cannam@127: Ti = LDW(&(W[TWVL * 5])); cannam@127: Tn = VMUL(T9, Td); cannam@127: Tb = VMUL(T9, Ta); cannam@127: { cannam@127: V Tu, T7, Tp, Th, To, Te; cannam@127: Tu = VFNMS(T5, T3, Tt); cannam@127: T7 = VFMA(T5, T6, T4); cannam@127: Tp = VMUL(Tf, Tj); cannam@127: Th = VMUL(Tf, Tg); cannam@127: To = VFNMS(Tc, Ta, Tn); cannam@127: Te = VFMA(Tc, Td, Tb); cannam@127: { cannam@127: V Tw, Tx, T8, Tm, Tq, Tk; cannam@127: Tw = VADD(Tu, Tv); cannam@127: Tx = VSUB(Tv, Tu); cannam@127: T8 = VADD(T1, T7); cannam@127: Tm = VSUB(T1, T7); cannam@127: Tq = VFNMS(Ti, Tg, Tp); cannam@127: Tk = VFMA(Ti, Tj, Th); cannam@127: { cannam@127: V Ts, Tr, Tl, Ty; cannam@127: Ts = VADD(To, Tq); cannam@127: Tr = VSUB(To, Tq); cannam@127: Tl = VADD(Te, Tk); cannam@127: Ty = VSUB(Te, Tk); cannam@127: ST(&(ri[WS(rs, 1)]), VADD(Tm, Tr), ms, &(ri[WS(rs, 1)])); cannam@127: ST(&(ri[WS(rs, 3)]), VSUB(Tm, Tr), ms, &(ri[WS(rs, 1)])); cannam@127: ST(&(ii[WS(rs, 2)]), VSUB(Tw, Ts), ms, &(ii[0])); cannam@127: ST(&(ii[0]), VADD(Ts, Tw), ms, &(ii[0])); cannam@127: ST(&(ii[WS(rs, 3)]), VADD(Ty, Tx), ms, &(ii[WS(rs, 1)])); cannam@127: ST(&(ii[WS(rs, 1)]), VSUB(Tx, Ty), ms, &(ii[WS(rs, 1)])); cannam@127: ST(&(ri[0]), VADD(T8, Tl), ms, &(ri[0])); cannam@127: ST(&(ri[WS(rs, 2)]), VSUB(T8, Tl), ms, &(ri[0])); 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, 2), 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("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1sv_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1sv_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 -n 4 -name t1sv_4 -include ts.h */ cannam@127: cannam@127: /* cannam@127: * This function contains 22 FP additions, 12 FP multiplications, cannam@127: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@127: * 13 stack variables, 0 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "ts.h" cannam@127: cannam@127: static void t1sv_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 * 6); m < me; m = m + (2 * VL), ri = ri + ((2 * VL) * ms), ii = ii + ((2 * VL) * ms), W = W + ((2 * VL) * 6), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: V T1, Tp, T6, To, Tc, Tk, Th, Tl; cannam@127: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@127: Tp = LD(&(ii[0]), ms, &(ii[0])); cannam@127: { cannam@127: V T3, T5, T2, T4; cannam@127: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@127: T5 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@127: T2 = LDW(&(W[TWVL * 2])); cannam@127: T4 = LDW(&(W[TWVL * 3])); cannam@127: T6 = VFMA(T2, T3, VMUL(T4, T5)); cannam@127: To = VFNMS(T4, T3, VMUL(T2, T5)); cannam@127: } cannam@127: { cannam@127: V T9, Tb, T8, Ta; cannam@127: T9 = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@127: Tb = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@127: T8 = LDW(&(W[0])); cannam@127: Ta = LDW(&(W[TWVL * 1])); cannam@127: Tc = VFMA(T8, T9, VMUL(Ta, Tb)); cannam@127: Tk = VFNMS(Ta, T9, VMUL(T8, Tb)); cannam@127: } cannam@127: { cannam@127: V Te, Tg, Td, Tf; cannam@127: Te = 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: Td = LDW(&(W[TWVL * 4])); cannam@127: Tf = LDW(&(W[TWVL * 5])); cannam@127: Th = VFMA(Td, Te, VMUL(Tf, Tg)); cannam@127: Tl = VFNMS(Tf, Te, VMUL(Td, Tg)); cannam@127: } cannam@127: { cannam@127: V T7, Ti, Tn, Tq; cannam@127: T7 = VADD(T1, T6); cannam@127: Ti = VADD(Tc, Th); cannam@127: ST(&(ri[WS(rs, 2)]), VSUB(T7, Ti), ms, &(ri[0])); cannam@127: ST(&(ri[0]), VADD(T7, 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, T6); 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(Tc, 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: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), 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("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1sv_4) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1sv_4, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */