cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:06:09 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle.native -fma -simd -compact -variables 4 -pipeline-latency 8 -n 4 -name t1sv_4 -include dft/simd/ts.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 22 FP additions, 12 FP multiplications, cannam@167: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@167: * 15 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "dft/simd/ts.h" cannam@167: cannam@167: static void t1sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: 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@167: V T1, Tv, T7, Tu, Te, To, Tk, Tq; cannam@167: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@167: Tv = LD(&(ii[0]), ms, &(ii[0])); cannam@167: { cannam@167: V T3, T6, T4, Tt, T2, T5; cannam@167: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@167: T6 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@167: T2 = LDW(&(W[TWVL * 2])); cannam@167: T4 = VMUL(T2, T3); cannam@167: Tt = VMUL(T2, T6); cannam@167: T5 = LDW(&(W[TWVL * 3])); cannam@167: T7 = VFMA(T5, T6, T4); cannam@167: Tu = VFNMS(T5, T3, Tt); cannam@167: } cannam@167: { cannam@167: V Ta, Td, Tb, Tn, T9, Tc; cannam@167: Ta = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@167: Td = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@167: T9 = LDW(&(W[0])); cannam@167: Tb = VMUL(T9, Ta); cannam@167: Tn = VMUL(T9, Td); cannam@167: Tc = LDW(&(W[TWVL * 1])); cannam@167: Te = VFMA(Tc, Td, Tb); cannam@167: To = VFNMS(Tc, Ta, Tn); cannam@167: } cannam@167: { cannam@167: V Tg, Tj, Th, Tp, Tf, Ti; cannam@167: Tg = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@167: Tj = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@167: Tf = LDW(&(W[TWVL * 4])); cannam@167: Th = VMUL(Tf, Tg); cannam@167: Tp = VMUL(Tf, Tj); cannam@167: Ti = LDW(&(W[TWVL * 5])); cannam@167: Tk = VFMA(Ti, Tj, Th); cannam@167: Tq = VFNMS(Ti, Tg, Tp); cannam@167: } cannam@167: { cannam@167: V T8, Tl, Ts, Tw; cannam@167: T8 = VADD(T1, T7); cannam@167: Tl = VADD(Te, Tk); cannam@167: ST(&(ri[WS(rs, 2)]), VSUB(T8, Tl), ms, &(ri[0])); cannam@167: ST(&(ri[0]), VADD(T8, Tl), ms, &(ri[0])); cannam@167: Ts = VADD(To, Tq); cannam@167: Tw = VADD(Tu, Tv); cannam@167: ST(&(ii[0]), VADD(Ts, Tw), ms, &(ii[0])); cannam@167: ST(&(ii[WS(rs, 2)]), VSUB(Tw, Ts), ms, &(ii[0])); cannam@167: } cannam@167: { cannam@167: V Tm, Tr, Tx, Ty; cannam@167: Tm = VSUB(T1, T7); cannam@167: Tr = VSUB(To, Tq); cannam@167: ST(&(ri[WS(rs, 3)]), VSUB(Tm, Tr), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 1)]), VADD(Tm, Tr), ms, &(ri[WS(rs, 1)])); cannam@167: Tx = VSUB(Tv, Tu); cannam@167: Ty = VSUB(Te, Tk); cannam@167: ST(&(ii[WS(rs, 1)]), VSUB(Tx, Ty), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 3)]), VADD(Ty, Tx), ms, &(ii[WS(rs, 1)])); cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 2), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, (2 * VL), 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t1sv_4) (planner *p) { cannam@167: X(kdft_dit_register) (p, t1sv_4, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle.native -simd -compact -variables 4 -pipeline-latency 8 -n 4 -name t1sv_4 -include dft/simd/ts.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 22 FP additions, 12 FP multiplications, cannam@167: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), cannam@167: * 13 stack variables, 0 constants, and 16 memory accesses cannam@167: */ cannam@167: #include "dft/simd/ts.h" cannam@167: cannam@167: static void t1sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: 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@167: V T1, Tp, T6, To, Tc, Tk, Th, Tl; cannam@167: T1 = LD(&(ri[0]), ms, &(ri[0])); cannam@167: Tp = LD(&(ii[0]), ms, &(ii[0])); cannam@167: { cannam@167: V T3, T5, T2, T4; cannam@167: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); cannam@167: T5 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); cannam@167: T2 = LDW(&(W[TWVL * 2])); cannam@167: T4 = LDW(&(W[TWVL * 3])); cannam@167: T6 = VFMA(T2, T3, VMUL(T4, T5)); cannam@167: To = VFNMS(T4, T3, VMUL(T2, T5)); cannam@167: } cannam@167: { cannam@167: V T9, Tb, T8, Ta; cannam@167: T9 = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); cannam@167: Tb = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); cannam@167: T8 = LDW(&(W[0])); cannam@167: Ta = LDW(&(W[TWVL * 1])); cannam@167: Tc = VFMA(T8, T9, VMUL(Ta, Tb)); cannam@167: Tk = VFNMS(Ta, T9, VMUL(T8, Tb)); cannam@167: } cannam@167: { cannam@167: V Te, Tg, Td, Tf; cannam@167: Te = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); cannam@167: Tg = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); cannam@167: Td = LDW(&(W[TWVL * 4])); cannam@167: Tf = LDW(&(W[TWVL * 5])); cannam@167: Th = VFMA(Td, Te, VMUL(Tf, Tg)); cannam@167: Tl = VFNMS(Tf, Te, VMUL(Td, Tg)); cannam@167: } cannam@167: { cannam@167: V T7, Ti, Tn, Tq; cannam@167: T7 = VADD(T1, T6); cannam@167: Ti = VADD(Tc, Th); cannam@167: ST(&(ri[WS(rs, 2)]), VSUB(T7, Ti), ms, &(ri[0])); cannam@167: ST(&(ri[0]), VADD(T7, Ti), ms, &(ri[0])); cannam@167: Tn = VADD(Tk, Tl); cannam@167: Tq = VADD(To, Tp); cannam@167: ST(&(ii[0]), VADD(Tn, Tq), ms, &(ii[0])); cannam@167: ST(&(ii[WS(rs, 2)]), VSUB(Tq, Tn), ms, &(ii[0])); cannam@167: } cannam@167: { cannam@167: V Tj, Tm, Tr, Ts; cannam@167: Tj = VSUB(T1, T6); cannam@167: Tm = VSUB(Tk, Tl); cannam@167: ST(&(ri[WS(rs, 3)]), VSUB(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@167: ST(&(ri[WS(rs, 1)]), VADD(Tj, Tm), ms, &(ri[WS(rs, 1)])); cannam@167: Tr = VSUB(Tp, To); cannam@167: Ts = VSUB(Tc, Th); cannam@167: ST(&(ii[WS(rs, 1)]), VSUB(Tr, Ts), ms, &(ii[WS(rs, 1)])); cannam@167: ST(&(ii[WS(rs, 3)]), VADD(Ts, Tr), ms, &(ii[WS(rs, 1)])); cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 2), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, (2 * VL), 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t1sv_4) (planner *p) { cannam@167: X(kdft_dit_register) (p, t1sv_4, &desc); cannam@167: } cannam@167: #endif