Chris@10: /* Chris@10: * Copyright (c) 2003, 2007-11 Matteo Frigo Chris@10: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology Chris@10: * Chris@10: * This program is free software; you can redistribute it and/or modify Chris@10: * it under the terms of the GNU General Public License as published by Chris@10: * the Free Software Foundation; either version 2 of the License, or Chris@10: * (at your option) any later version. Chris@10: * Chris@10: * This program is distributed in the hope that it will be useful, Chris@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@10: * GNU General Public License for more details. Chris@10: * Chris@10: * You should have received a copy of the GNU General Public License Chris@10: * along with this program; if not, write to the Free Software Chris@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@10: * Chris@10: */ Chris@10: Chris@10: /* This file was automatically generated --- DO NOT EDIT */ Chris@10: /* Generated on Sun Nov 25 07:39:26 EST 2012 */ Chris@10: Chris@10: #include "codelet-dft.h" Chris@10: Chris@10: #ifdef HAVE_FMA Chris@10: Chris@10: /* 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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 24 FP additions, 16 FP multiplications, Chris@10: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), Chris@10: * 37 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "ts.h" Chris@10: Chris@10: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: V T2, T6, T3, T5, T1, Tx, T8, Tc, Tf, Ta, T4, Th, Tj, Tl; Chris@10: T2 = LDW(&(W[0])); Chris@10: T6 = LDW(&(W[TWVL * 3])); Chris@10: T3 = LDW(&(W[TWVL * 2])); Chris@10: T5 = LDW(&(W[TWVL * 1])); Chris@10: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@10: Tx = LD(&(ii[0]), ms, &(ii[0])); Chris@10: T8 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@10: Tc = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@10: Tf = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Chris@10: Ta = VMUL(T2, T6); Chris@10: T4 = VMUL(T2, T3); Chris@10: Th = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Chris@10: Tj = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Chris@10: Tl = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Chris@10: { Chris@10: V Tg, Tb, T7, Tp, Tk, Tr, Ti; Chris@10: Tg = VMUL(T2, Tf); Chris@10: Tb = VFNMS(T5, T3, Ta); Chris@10: T7 = VFMA(T5, T6, T4); Chris@10: Tp = VMUL(T2, Th); Chris@10: Tk = VMUL(T3, Tj); Chris@10: Tr = VMUL(T3, Tl); Chris@10: Ti = VFMA(T5, Th, Tg); Chris@10: { Chris@10: V Tv, T9, Tq, Tm, Ts, Tw, Td; Chris@10: Tv = VMUL(T7, Tc); Chris@10: T9 = VMUL(T7, T8); Chris@10: Tq = VFNMS(T5, Tf, Tp); Chris@10: Tm = VFMA(T6, Tl, Tk); Chris@10: Ts = VFNMS(T6, Tj, Tr); Chris@10: Tw = VFNMS(Tb, T8, Tv); Chris@10: Td = VFMA(Tb, Tc, T9); Chris@10: { Chris@10: V Tn, TA, Tu, Tt; Chris@10: Tn = VADD(Ti, Tm); Chris@10: TA = VSUB(Ti, Tm); Chris@10: Tu = VADD(Tq, Ts); Chris@10: Tt = VSUB(Tq, Ts); Chris@10: { Chris@10: V Ty, Tz, Te, To; Chris@10: Ty = VADD(Tw, Tx); Chris@10: Tz = VSUB(Tx, Tw); Chris@10: Te = VADD(T1, Td); Chris@10: To = VSUB(T1, Td); Chris@10: ST(&(ii[WS(rs, 3)]), VADD(TA, Tz), ms, &(ii[WS(rs, 1)])); Chris@10: ST(&(ii[WS(rs, 1)]), VSUB(Tz, TA), ms, &(ii[WS(rs, 1)])); Chris@10: ST(&(ii[WS(rs, 2)]), VSUB(Ty, Tu), ms, &(ii[0])); Chris@10: ST(&(ii[0]), VADD(Tu, Ty), ms, &(ii[0])); Chris@10: ST(&(ri[WS(rs, 1)]), VADD(To, Tt), ms, &(ri[WS(rs, 1)])); Chris@10: ST(&(ri[WS(rs, 3)]), VSUB(To, Tt), ms, &(ri[WS(rs, 1)])); Chris@10: ST(&(ri[0]), VADD(Te, Tn), ms, &(ri[0])); Chris@10: ST(&(ri[WS(rs, 2)]), VSUB(Te, Tn), ms, &(ri[0])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 3), Chris@10: {TW_NEXT, (2 * VL), 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t2sv_4) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2sv_4, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* 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 */ Chris@10: Chris@10: /* Chris@10: * This function contains 24 FP additions, 16 FP multiplications, Chris@10: * (or, 16 additions, 8 multiplications, 8 fused multiply/add), Chris@10: * 21 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "ts.h" Chris@10: Chris@10: static void t2sv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: { Chris@10: INT m; Chris@10: 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)) { Chris@10: V T2, T4, T3, T5, T6, T8; Chris@10: T2 = LDW(&(W[0])); Chris@10: T4 = LDW(&(W[TWVL * 1])); Chris@10: T3 = LDW(&(W[TWVL * 2])); Chris@10: T5 = LDW(&(W[TWVL * 3])); Chris@10: T6 = VFMA(T2, T3, VMUL(T4, T5)); Chris@10: T8 = VFNMS(T4, T3, VMUL(T2, T5)); Chris@10: { Chris@10: V T1, Tp, Ta, To, Te, Tk, Th, Tl, T7, T9; Chris@10: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@10: Tp = LD(&(ii[0]), ms, &(ii[0])); Chris@10: T7 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@10: T9 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@10: Ta = VFMA(T6, T7, VMUL(T8, T9)); Chris@10: To = VFNMS(T8, T7, VMUL(T6, T9)); Chris@10: { Chris@10: V Tc, Td, Tf, Tg; Chris@10: Tc = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Chris@10: Td = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Chris@10: Te = VFMA(T2, Tc, VMUL(T4, Td)); Chris@10: Tk = VFNMS(T4, Tc, VMUL(T2, Td)); Chris@10: Tf = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Chris@10: Tg = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Chris@10: Th = VFMA(T3, Tf, VMUL(T5, Tg)); Chris@10: Tl = VFNMS(T5, Tf, VMUL(T3, Tg)); Chris@10: } Chris@10: { Chris@10: V Tb, Ti, Tn, Tq; Chris@10: Tb = VADD(T1, Ta); Chris@10: Ti = VADD(Te, Th); Chris@10: ST(&(ri[WS(rs, 2)]), VSUB(Tb, Ti), ms, &(ri[0])); Chris@10: ST(&(ri[0]), VADD(Tb, Ti), ms, &(ri[0])); Chris@10: Tn = VADD(Tk, Tl); Chris@10: Tq = VADD(To, Tp); Chris@10: ST(&(ii[0]), VADD(Tn, Tq), ms, &(ii[0])); Chris@10: ST(&(ii[WS(rs, 2)]), VSUB(Tq, Tn), ms, &(ii[0])); Chris@10: } Chris@10: { Chris@10: V Tj, Tm, Tr, Ts; Chris@10: Tj = VSUB(T1, Ta); Chris@10: Tm = VSUB(Tk, Tl); Chris@10: ST(&(ri[WS(rs, 3)]), VSUB(Tj, Tm), ms, &(ri[WS(rs, 1)])); Chris@10: ST(&(ri[WS(rs, 1)]), VADD(Tj, Tm), ms, &(ri[WS(rs, 1)])); Chris@10: Tr = VSUB(Tp, To); Chris@10: Ts = VSUB(Te, Th); Chris@10: ST(&(ii[WS(rs, 1)]), VSUB(Tr, Ts), ms, &(ii[WS(rs, 1)])); Chris@10: ST(&(ii[WS(rs, 3)]), VADD(Ts, Tr), ms, &(ii[WS(rs, 1)])); Chris@10: } Chris@10: } Chris@10: } Chris@10: } Chris@10: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 3), Chris@10: {TW_NEXT, (2 * VL), 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 4, XSIMD_STRING("t2sv_4"), twinstr, &GENUS, {16, 8, 8, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t2sv_4) (planner *p) { Chris@10: X(kdft_dit_register) (p, t2sv_4, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */