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 07:39:24 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 -n 4 -name t1sv_4 -include ts.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 12 FP multiplications, Chris@10: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 35 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "ts.h" Chris@10: Chris@10: static void t1sv_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 * 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)) { Chris@10: V T1, Tv, T3, T6, T5, Ta, Td, Tc, Tg, Tj, Tt, T4, Tf, Ti, Tn; Chris@10: V Tb, T2, T9; Chris@10: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@10: Tv = LD(&(ii[0]), ms, &(ii[0])); Chris@10: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@10: T6 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@10: T2 = LDW(&(W[TWVL * 2])); Chris@10: T5 = LDW(&(W[TWVL * 3])); Chris@10: Ta = 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: T9 = LDW(&(W[0])); Chris@10: Tc = LDW(&(W[TWVL * 1])); Chris@10: Tg = LD(&(ri[WS(rs, 3)]), ms, &(ri[WS(rs, 1)])); Chris@10: Tj = LD(&(ii[WS(rs, 3)]), ms, &(ii[WS(rs, 1)])); Chris@10: Tt = VMUL(T2, T6); Chris@10: T4 = VMUL(T2, T3); Chris@10: Tf = LDW(&(W[TWVL * 4])); Chris@10: Ti = LDW(&(W[TWVL * 5])); Chris@10: Tn = VMUL(T9, Td); Chris@10: Tb = VMUL(T9, Ta); Chris@10: { Chris@10: V Tu, T7, Tp, Th, To, Te; Chris@10: Tu = VFNMS(T5, T3, Tt); Chris@10: T7 = VFMA(T5, T6, T4); Chris@10: Tp = VMUL(Tf, Tj); Chris@10: Th = VMUL(Tf, Tg); Chris@10: To = VFNMS(Tc, Ta, Tn); Chris@10: Te = VFMA(Tc, Td, Tb); Chris@10: { Chris@10: V Tw, Tx, T8, Tm, Tq, Tk; Chris@10: Tw = VADD(Tu, Tv); Chris@10: Tx = VSUB(Tv, Tu); Chris@10: T8 = VADD(T1, T7); Chris@10: Tm = VSUB(T1, T7); Chris@10: Tq = VFNMS(Ti, Tg, Tp); Chris@10: Tk = VFMA(Ti, Tj, Th); Chris@10: { Chris@10: V Ts, Tr, Tl, Ty; Chris@10: Ts = VADD(To, Tq); Chris@10: Tr = VSUB(To, Tq); Chris@10: Tl = VADD(Te, Tk); Chris@10: Ty = VSUB(Te, Tk); Chris@10: ST(&(ri[WS(rs, 1)]), VADD(Tm, Tr), ms, &(ri[WS(rs, 1)])); Chris@10: ST(&(ri[WS(rs, 3)]), VSUB(Tm, Tr), ms, &(ri[WS(rs, 1)])); Chris@10: ST(&(ii[WS(rs, 2)]), VSUB(Tw, Ts), ms, &(ii[0])); Chris@10: ST(&(ii[0]), VADD(Ts, Tw), ms, &(ii[0])); Chris@10: ST(&(ii[WS(rs, 3)]), VADD(Ty, Tx), ms, &(ii[WS(rs, 1)])); Chris@10: ST(&(ii[WS(rs, 1)]), VSUB(Tx, Ty), ms, &(ii[WS(rs, 1)])); Chris@10: ST(&(ri[0]), VADD(T8, Tl), ms, &(ri[0])); Chris@10: ST(&(ri[WS(rs, 2)]), VSUB(T8, Tl), ms, &(ri[0])); 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, 2), 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("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1sv_4) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1sv_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 -n 4 -name t1sv_4 -include ts.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 22 FP additions, 12 FP multiplications, Chris@10: * (or, 16 additions, 6 multiplications, 6 fused multiply/add), Chris@10: * 13 stack variables, 0 constants, and 16 memory accesses Chris@10: */ Chris@10: #include "ts.h" Chris@10: Chris@10: static void t1sv_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 * 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)) { Chris@10: V T1, Tp, T6, To, Tc, Tk, Th, Tl; Chris@10: T1 = LD(&(ri[0]), ms, &(ri[0])); Chris@10: Tp = LD(&(ii[0]), ms, &(ii[0])); Chris@10: { Chris@10: V T3, T5, T2, T4; Chris@10: T3 = LD(&(ri[WS(rs, 2)]), ms, &(ri[0])); Chris@10: T5 = LD(&(ii[WS(rs, 2)]), ms, &(ii[0])); Chris@10: T2 = LDW(&(W[TWVL * 2])); Chris@10: T4 = LDW(&(W[TWVL * 3])); Chris@10: T6 = VFMA(T2, T3, VMUL(T4, T5)); Chris@10: To = VFNMS(T4, T3, VMUL(T2, T5)); Chris@10: } Chris@10: { Chris@10: V T9, Tb, T8, Ta; Chris@10: T9 = LD(&(ri[WS(rs, 1)]), ms, &(ri[WS(rs, 1)])); Chris@10: Tb = LD(&(ii[WS(rs, 1)]), ms, &(ii[WS(rs, 1)])); Chris@10: T8 = LDW(&(W[0])); Chris@10: Ta = LDW(&(W[TWVL * 1])); Chris@10: Tc = VFMA(T8, T9, VMUL(Ta, Tb)); Chris@10: Tk = VFNMS(Ta, T9, VMUL(T8, Tb)); Chris@10: } Chris@10: { Chris@10: V Te, Tg, Td, Tf; Chris@10: Te = 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: Td = LDW(&(W[TWVL * 4])); Chris@10: Tf = LDW(&(W[TWVL * 5])); Chris@10: Th = VFMA(Td, Te, VMUL(Tf, Tg)); Chris@10: Tl = VFNMS(Tf, Te, VMUL(Td, Tg)); Chris@10: } Chris@10: { Chris@10: V T7, Ti, Tn, Tq; Chris@10: T7 = VADD(T1, T6); Chris@10: Ti = VADD(Tc, Th); Chris@10: ST(&(ri[WS(rs, 2)]), VSUB(T7, Ti), ms, &(ri[0])); Chris@10: ST(&(ri[0]), VADD(T7, 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, T6); 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(Tc, 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: VLEAVE(); Chris@10: } Chris@10: Chris@10: static const tw_instr twinstr[] = { Chris@10: VTW(0, 1), Chris@10: VTW(0, 2), 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("t1sv_4"), twinstr, &GENUS, {16, 6, 6, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1sv_4) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1sv_4, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */