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 07:38:03 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_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 10 -name t1fv_10 -include t1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 51 FP additions, 40 FP multiplications, Chris@10: * (or, 33 additions, 22 multiplications, 18 fused multiply/add), Chris@10: * 43 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "t1f.h" Chris@10: Chris@10: static void t1fv_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 18)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: V Td, TA, T4, Ta, Tk, TE, Tp, TF, TB, T9, T1, T2, Tb; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: T2 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V Tg, Tn, Ti, Tl; Chris@10: Tg = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: Tn = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: Ti = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: Tl = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: { Chris@10: V T6, T8, T5, Tc; Chris@10: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V T3, Th, To, Tj, Tm, T7; Chris@10: T7 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: T3 = BYTWJ(&(W[TWVL * 8]), T2); Chris@10: Th = BYTWJ(&(W[TWVL * 6]), Tg); Chris@10: To = BYTWJ(&(W[0]), Tn); Chris@10: Tj = BYTWJ(&(W[TWVL * 16]), Ti); Chris@10: Tm = BYTWJ(&(W[TWVL * 10]), Tl); Chris@10: T6 = BYTWJ(&(W[TWVL * 2]), T5); Chris@10: Td = BYTWJ(&(W[TWVL * 4]), Tc); Chris@10: T8 = BYTWJ(&(W[TWVL * 12]), T7); Chris@10: TA = VADD(T1, T3); Chris@10: T4 = VSUB(T1, T3); Chris@10: Ta = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: Tk = VSUB(Th, Tj); Chris@10: TE = VADD(Th, Tj); Chris@10: Tp = VSUB(Tm, To); Chris@10: TF = VADD(Tm, To); Chris@10: } Chris@10: TB = VADD(T6, T8); Chris@10: T9 = VSUB(T6, T8); Chris@10: } Chris@10: } Chris@10: Tb = BYTWJ(&(W[TWVL * 14]), Ta); Chris@10: { Chris@10: V TL, TG, Tw, Tq, TC, Te; Chris@10: TL = VSUB(TE, TF); Chris@10: TG = VADD(TE, TF); Chris@10: Tw = VSUB(Tk, Tp); Chris@10: Tq = VADD(Tk, Tp); Chris@10: TC = VADD(Tb, Td); Chris@10: Te = VSUB(Tb, Td); Chris@10: { Chris@10: V TM, TD, Tv, Tf; Chris@10: TM = VSUB(TB, TC); Chris@10: TD = VADD(TB, TC); Chris@10: Tv = VSUB(T9, Te); Chris@10: Tf = VADD(T9, Te); Chris@10: { Chris@10: V TP, TN, TH, TJ, Tz, Tx, Tr, Tt, TI, Ts; Chris@10: TP = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TL, TM)); Chris@10: TN = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TM, TL)); Chris@10: TH = VADD(TD, TG); Chris@10: TJ = VSUB(TD, TG); Chris@10: Tz = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tv, Tw)); Chris@10: Tx = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tw, Tv)); Chris@10: Tr = VADD(Tf, Tq); Chris@10: Tt = VSUB(Tf, Tq); Chris@10: ST(&(x[0]), VADD(TA, TH), ms, &(x[0])); Chris@10: TI = VFNMS(LDK(KP250000000), TH, TA); Chris@10: ST(&(x[WS(rs, 5)]), VADD(T4, Tr), ms, &(x[WS(rs, 1)])); Chris@10: Ts = VFNMS(LDK(KP250000000), Tr, T4); Chris@10: { Chris@10: V TK, TO, Tu, Ty; Chris@10: TK = VFNMS(LDK(KP559016994), TJ, TI); Chris@10: TO = VFMA(LDK(KP559016994), TJ, TI); Chris@10: Tu = VFMA(LDK(KP559016994), Tt, Ts); Chris@10: Ty = VFNMS(LDK(KP559016994), Tt, Ts); Chris@10: ST(&(x[WS(rs, 8)]), VFNMSI(TN, TK), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 2)]), VFMAI(TN, TK), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 6)]), VFNMSI(TP, TO), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 4)]), VFMAI(TP, TO), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 9)]), VFMAI(Tx, Tu), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 1)]), VFNMSI(Tx, Tu), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VFMAI(Tz, Ty), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 3)]), VFNMSI(Tz, Ty), ms, &(x[WS(rs, 1)])); 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, 2), Chris@10: VTW(0, 3), Chris@10: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: VTW(0, 9), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 10, XSIMD_STRING("t1fv_10"), twinstr, &GENUS, {33, 22, 18, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1fv_10) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1fv_10, &desc); Chris@10: } Chris@10: #else /* HAVE_FMA */ Chris@10: Chris@10: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 10 -name t1fv_10 -include t1f.h */ Chris@10: Chris@10: /* Chris@10: * This function contains 51 FP additions, 30 FP multiplications, Chris@10: * (or, 45 additions, 24 multiplications, 6 fused multiply/add), Chris@10: * 32 stack variables, 4 constants, and 20 memory accesses Chris@10: */ Chris@10: #include "t1f.h" Chris@10: Chris@10: static void t1fv_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); Chris@10: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); Chris@10: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); Chris@10: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ri; Chris@10: for (m = mb, W = W + (mb * ((TWVL / VL) * 18)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(10, rs)) { Chris@10: V Tr, TH, Tg, Tl, Tm, TA, TB, TJ, T5, Ta, Tb, TD, TE, TI, To; Chris@10: V Tq, Tp; Chris@10: To = LD(&(x[0]), ms, &(x[0])); Chris@10: Tp = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tq = BYTWJ(&(W[TWVL * 8]), Tp); Chris@10: Tr = VSUB(To, Tq); Chris@10: TH = VADD(To, Tq); Chris@10: { Chris@10: V Td, Tk, Tf, Ti; Chris@10: { Chris@10: V Tc, Tj, Te, Th; Chris@10: Tc = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: Td = BYTWJ(&(W[TWVL * 6]), Tc); Chris@10: Tj = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: Tk = BYTWJ(&(W[0]), Tj); Chris@10: Te = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: Tf = BYTWJ(&(W[TWVL * 16]), Te); Chris@10: Th = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: Ti = BYTWJ(&(W[TWVL * 10]), Th); Chris@10: } Chris@10: Tg = VSUB(Td, Tf); Chris@10: Tl = VSUB(Ti, Tk); Chris@10: Tm = VADD(Tg, Tl); Chris@10: TA = VADD(Td, Tf); Chris@10: TB = VADD(Ti, Tk); Chris@10: TJ = VADD(TA, TB); Chris@10: } Chris@10: { Chris@10: V T2, T9, T4, T7; Chris@10: { Chris@10: V T1, T8, T3, T6; Chris@10: T1 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: T2 = BYTWJ(&(W[TWVL * 2]), T1); Chris@10: T8 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: T9 = BYTWJ(&(W[TWVL * 4]), T8); Chris@10: T3 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: T4 = BYTWJ(&(W[TWVL * 12]), T3); Chris@10: T6 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: T7 = BYTWJ(&(W[TWVL * 14]), T6); Chris@10: } Chris@10: T5 = VSUB(T2, T4); Chris@10: Ta = VSUB(T7, T9); Chris@10: Tb = VADD(T5, Ta); Chris@10: TD = VADD(T2, T4); Chris@10: TE = VADD(T7, T9); Chris@10: TI = VADD(TD, TE); Chris@10: } Chris@10: { Chris@10: V Tn, Ts, Tt, Tx, Tz, Tv, Tw, Ty, Tu; Chris@10: Tn = VMUL(LDK(KP559016994), VSUB(Tb, Tm)); Chris@10: Ts = VADD(Tb, Tm); Chris@10: Tt = VFNMS(LDK(KP250000000), Ts, Tr); Chris@10: Tv = VSUB(T5, Ta); Chris@10: Tw = VSUB(Tg, Tl); Chris@10: Tx = VBYI(VFMA(LDK(KP951056516), Tv, VMUL(LDK(KP587785252), Tw))); Chris@10: Tz = VBYI(VFNMS(LDK(KP587785252), Tv, VMUL(LDK(KP951056516), Tw))); Chris@10: ST(&(x[WS(rs, 5)]), VADD(Tr, Ts), ms, &(x[WS(rs, 1)])); Chris@10: Ty = VSUB(Tt, Tn); Chris@10: ST(&(x[WS(rs, 3)]), VSUB(Ty, Tz), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VADD(Tz, Ty), ms, &(x[WS(rs, 1)])); Chris@10: Tu = VADD(Tn, Tt); Chris@10: ST(&(x[WS(rs, 1)]), VSUB(Tu, Tx), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 9)]), VADD(Tx, Tu), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V TM, TK, TL, TG, TO, TC, TF, TP, TN; Chris@10: TM = VMUL(LDK(KP559016994), VSUB(TI, TJ)); Chris@10: TK = VADD(TI, TJ); Chris@10: TL = VFNMS(LDK(KP250000000), TK, TH); Chris@10: TC = VSUB(TA, TB); Chris@10: TF = VSUB(TD, TE); Chris@10: TG = VBYI(VFNMS(LDK(KP587785252), TF, VMUL(LDK(KP951056516), TC))); Chris@10: TO = VBYI(VFMA(LDK(KP951056516), TF, VMUL(LDK(KP587785252), TC))); Chris@10: ST(&(x[0]), VADD(TH, TK), ms, &(x[0])); Chris@10: TP = VADD(TM, TL); Chris@10: ST(&(x[WS(rs, 4)]), VADD(TO, TP), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 6)]), VSUB(TP, TO), ms, &(x[0])); Chris@10: TN = VSUB(TL, TM); Chris@10: ST(&(x[WS(rs, 2)]), VADD(TG, TN), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 8)]), VSUB(TN, TG), ms, &(x[0])); 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: VTW(0, 4), Chris@10: VTW(0, 5), Chris@10: VTW(0, 6), Chris@10: VTW(0, 7), Chris@10: VTW(0, 8), Chris@10: VTW(0, 9), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 10, XSIMD_STRING("t1fv_10"), twinstr, &GENUS, {45, 24, 6, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1fv_10) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1fv_10, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */