Chris@42: /* Chris@42: * Copyright (c) 2003, 2007-14 Matteo Frigo Chris@42: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology Chris@42: * Chris@42: * This program is free software; you can redistribute it and/or modify Chris@42: * it under the terms of the GNU General Public License as published by Chris@42: * the Free Software Foundation; either version 2 of the License, or Chris@42: * (at your option) any later version. Chris@42: * Chris@42: * This program is distributed in the hope that it will be useful, Chris@42: * but WITHOUT ANY WARRANTY; without even the implied warranty of Chris@42: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the Chris@42: * GNU General Public License for more details. Chris@42: * Chris@42: * You should have received a copy of the GNU General Public License Chris@42: * along with this program; if not, write to the Free Software Chris@42: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Chris@42: * Chris@42: */ Chris@42: Chris@42: /* This file was automatically generated --- DO NOT EDIT */ Chris@42: /* Generated on Sat Jul 30 16:44:20 EDT 2016 */ Chris@42: Chris@42: #include "codelet-dft.h" Chris@42: Chris@42: #ifdef HAVE_FMA Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1bv_12 -include t1b.h -sign 1 */ Chris@42: Chris@42: /* Chris@42: * This function contains 59 FP additions, 42 FP multiplications, Chris@42: * (or, 41 additions, 24 multiplications, 18 fused multiply/add), Chris@42: * 41 stack variables, 2 constants, and 24 memory accesses Chris@42: */ Chris@42: #include "t1b.h" Chris@42: Chris@42: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: R *x; Chris@42: x = ii; Chris@42: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { Chris@42: V TI, Ti, TA, T7, Tm, TE, Tw, Tk, Tf, TB, TU, TM; Chris@42: { Chris@42: V T9, TK, Tj, TL, Te; Chris@42: { Chris@42: V T1, T4, T2, Tp, Tt, Tr; Chris@42: T1 = LD(&(x[0]), ms, &(x[0])); Chris@42: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@42: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@42: Tp = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@42: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@42: Tr = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@42: { Chris@42: V T5, T3, Tq, Tu, Ts, Td, Tb, T8, Tc, Ta; Chris@42: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@42: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@42: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@42: T5 = BYTW(&(W[TWVL * 14]), T4); Chris@42: T3 = BYTW(&(W[TWVL * 6]), T2); Chris@42: Tq = BYTW(&(W[TWVL * 16]), Tp); Chris@42: Tu = BYTW(&(W[TWVL * 8]), Tt); Chris@42: Ts = BYTW(&(W[0]), Tr); Chris@42: T9 = BYTW(&(W[TWVL * 10]), T8); Chris@42: Td = BYTW(&(W[TWVL * 2]), Tc); Chris@42: Tb = BYTW(&(W[TWVL * 18]), Ta); Chris@42: { Chris@42: V Th, T6, Tl, Tv; Chris@42: Th = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@42: TK = VSUB(T3, T5); Chris@42: T6 = VADD(T3, T5); Chris@42: Tl = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@42: Tv = VADD(Ts, Tu); Chris@42: TI = VSUB(Tu, Ts); Chris@42: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@42: TL = VSUB(Tb, Td); Chris@42: Te = VADD(Tb, Td); Chris@42: Ti = BYTW(&(W[TWVL * 4]), Th); Chris@42: TA = VFNMS(LDK(KP500000000), T6, T1); Chris@42: T7 = VADD(T1, T6); Chris@42: Tm = BYTW(&(W[TWVL * 20]), Tl); Chris@42: TE = VFNMS(LDK(KP500000000), Tv, Tq); Chris@42: Tw = VADD(Tq, Tv); Chris@42: } Chris@42: } Chris@42: } Chris@42: Tk = BYTW(&(W[TWVL * 12]), Tj); Chris@42: Tf = VADD(T9, Te); Chris@42: TB = VFNMS(LDK(KP500000000), Te, T9); Chris@42: TU = VSUB(TK, TL); Chris@42: TM = VADD(TK, TL); Chris@42: } Chris@42: { Chris@42: V Tn, TH, TC, TQ, Ty, Tg; Chris@42: Tn = VADD(Tk, Tm); Chris@42: TH = VSUB(Tk, Tm); Chris@42: TC = VADD(TA, TB); Chris@42: TQ = VSUB(TA, TB); Chris@42: Ty = VADD(T7, Tf); Chris@42: Tg = VSUB(T7, Tf); Chris@42: { Chris@42: V To, TD, TJ, TR; Chris@42: To = VADD(Ti, Tn); Chris@42: TD = VFNMS(LDK(KP500000000), Tn, Ti); Chris@42: TJ = VSUB(TH, TI); Chris@42: TR = VADD(TH, TI); Chris@42: { Chris@42: V TP, TN, TW, TS, TO, TG, TX, TV; Chris@42: { Chris@42: V Tz, Tx, TF, TT; Chris@42: Tz = VADD(To, Tw); Chris@42: Tx = VSUB(To, Tw); Chris@42: TF = VADD(TD, TE); Chris@42: TT = VSUB(TD, TE); Chris@42: TP = VMUL(LDK(KP866025403), VADD(TM, TJ)); Chris@42: TN = VMUL(LDK(KP866025403), VSUB(TJ, TM)); Chris@42: TW = VFMA(LDK(KP866025403), TR, TQ); Chris@42: TS = VFNMS(LDK(KP866025403), TR, TQ); Chris@42: ST(&(x[WS(rs, 6)]), VSUB(Ty, Tz), ms, &(x[0])); Chris@42: ST(&(x[0]), VADD(Ty, Tz), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 9)]), VFMAI(Tx, Tg), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 3)]), VFNMSI(Tx, Tg), ms, &(x[WS(rs, 1)])); Chris@42: TO = VADD(TC, TF); Chris@42: TG = VSUB(TC, TF); Chris@42: TX = VFNMS(LDK(KP866025403), TU, TT); Chris@42: TV = VFMA(LDK(KP866025403), TU, TT); Chris@42: } Chris@42: ST(&(x[WS(rs, 8)]), VFNMSI(TP, TO), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 4)]), VFMAI(TP, TO), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 2)]), VFMAI(TN, TG), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 10)]), VFNMSI(TN, TG), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 5)]), VFMAI(TX, TW), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 7)]), VFNMSI(TX, TW), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 11)]), VFNMSI(TV, TS), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 1)]), VFMAI(TV, TS), ms, &(x[WS(rs, 1)])); Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: VTW(0, 1), Chris@42: VTW(0, 2), Chris@42: VTW(0, 3), Chris@42: VTW(0, 4), Chris@42: VTW(0, 5), Chris@42: VTW(0, 6), Chris@42: VTW(0, 7), Chris@42: VTW(0, 8), Chris@42: VTW(0, 9), Chris@42: VTW(0, 10), Chris@42: VTW(0, 11), Chris@42: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {41, 24, 18, 0}, 0, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_t1bv_12) (planner *p) { Chris@42: X(kdft_dit_register) (p, t1bv_12, &desc); Chris@42: } Chris@42: #else /* HAVE_FMA */ Chris@42: Chris@42: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 12 -name t1bv_12 -include t1b.h -sign 1 */ Chris@42: Chris@42: /* Chris@42: * This function contains 59 FP additions, 30 FP multiplications, Chris@42: * (or, 55 additions, 26 multiplications, 4 fused multiply/add), Chris@42: * 28 stack variables, 2 constants, and 24 memory accesses Chris@42: */ Chris@42: #include "t1b.h" Chris@42: Chris@42: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@42: { Chris@42: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@42: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@42: { Chris@42: INT m; Chris@42: R *x; Chris@42: x = ii; Chris@42: for (m = mb, W = W + (mb * ((TWVL / VL) * 22)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 22), MAKE_VOLATILE_STRIDE(12, rs)) { Chris@42: V T1, Tt, T6, T7, TB, Tq, TC, TD, T9, Tu, Te, Tf, Tx, Tl, Ty; Chris@42: V Tz; Chris@42: { Chris@42: V T5, T3, T4, T2; Chris@42: T1 = LD(&(x[0]), ms, &(x[0])); Chris@42: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@42: T5 = BYTW(&(W[TWVL * 14]), T4); Chris@42: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@42: T3 = BYTW(&(W[TWVL * 6]), T2); Chris@42: Tt = VSUB(T3, T5); Chris@42: T6 = VADD(T3, T5); Chris@42: T7 = VFNMS(LDK(KP500000000), T6, T1); Chris@42: } Chris@42: { Chris@42: V Tn, Tp, Tm, TA, To; Chris@42: Tm = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@42: Tn = BYTW(&(W[0]), Tm); Chris@42: TA = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@42: TB = BYTW(&(W[TWVL * 16]), TA); Chris@42: To = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@42: Tp = BYTW(&(W[TWVL * 8]), To); Chris@42: Tq = VSUB(Tn, Tp); Chris@42: TC = VADD(Tn, Tp); Chris@42: TD = VFNMS(LDK(KP500000000), TC, TB); Chris@42: } Chris@42: { Chris@42: V Td, Tb, T8, Tc, Ta; Chris@42: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@42: T9 = BYTW(&(W[TWVL * 10]), T8); Chris@42: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@42: Td = BYTW(&(W[TWVL * 2]), Tc); Chris@42: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@42: Tb = BYTW(&(W[TWVL * 18]), Ta); Chris@42: Tu = VSUB(Tb, Td); Chris@42: Te = VADD(Tb, Td); Chris@42: Tf = VFNMS(LDK(KP500000000), Te, T9); Chris@42: } Chris@42: { Chris@42: V Ti, Tk, Th, Tw, Tj; Chris@42: Th = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@42: Ti = BYTW(&(W[TWVL * 12]), Th); Chris@42: Tw = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@42: Tx = BYTW(&(W[TWVL * 4]), Tw); Chris@42: Tj = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@42: Tk = BYTW(&(W[TWVL * 20]), Tj); Chris@42: Tl = VSUB(Ti, Tk); Chris@42: Ty = VADD(Ti, Tk); Chris@42: Tz = VFNMS(LDK(KP500000000), Ty, Tx); Chris@42: } Chris@42: { Chris@42: V Ts, TG, TF, TH; Chris@42: { Chris@42: V Tg, Tr, Tv, TE; Chris@42: Tg = VSUB(T7, Tf); Chris@42: Tr = VMUL(LDK(KP866025403), VSUB(Tl, Tq)); Chris@42: Ts = VSUB(Tg, Tr); Chris@42: TG = VADD(Tg, Tr); Chris@42: Tv = VMUL(LDK(KP866025403), VSUB(Tt, Tu)); Chris@42: TE = VSUB(Tz, TD); Chris@42: TF = VBYI(VADD(Tv, TE)); Chris@42: TH = VBYI(VSUB(TE, Tv)); Chris@42: } Chris@42: ST(&(x[WS(rs, 11)]), VSUB(Ts, TF), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 5)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 1)]), VADD(Ts, TF), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 7)]), VSUB(TG, TH), ms, &(x[WS(rs, 1)])); Chris@42: } Chris@42: { Chris@42: V TS, TW, TV, TX; Chris@42: { Chris@42: V TQ, TR, TT, TU; Chris@42: TQ = VADD(T1, T6); Chris@42: TR = VADD(T9, Te); Chris@42: TS = VSUB(TQ, TR); Chris@42: TW = VADD(TQ, TR); Chris@42: TT = VADD(Tx, Ty); Chris@42: TU = VADD(TB, TC); Chris@42: TV = VBYI(VSUB(TT, TU)); Chris@42: TX = VADD(TT, TU); Chris@42: } Chris@42: ST(&(x[WS(rs, 3)]), VSUB(TS, TV), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[0]), VADD(TW, TX), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 9)]), VADD(TS, TV), ms, &(x[WS(rs, 1)])); Chris@42: ST(&(x[WS(rs, 6)]), VSUB(TW, TX), ms, &(x[0])); Chris@42: } Chris@42: { Chris@42: V TK, TO, TN, TP; Chris@42: { Chris@42: V TI, TJ, TL, TM; Chris@42: TI = VADD(Tl, Tq); Chris@42: TJ = VADD(Tt, Tu); Chris@42: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); Chris@42: TO = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); Chris@42: TL = VADD(T7, Tf); Chris@42: TM = VADD(Tz, TD); Chris@42: TN = VSUB(TL, TM); Chris@42: TP = VADD(TL, TM); Chris@42: } Chris@42: ST(&(x[WS(rs, 2)]), VADD(TK, TN), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 8)]), VSUB(TP, TO), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 10)]), VSUB(TN, TK), ms, &(x[0])); Chris@42: ST(&(x[WS(rs, 4)]), VADD(TO, TP), ms, &(x[0])); Chris@42: } Chris@42: } Chris@42: } Chris@42: VLEAVE(); Chris@42: } Chris@42: Chris@42: static const tw_instr twinstr[] = { Chris@42: VTW(0, 1), Chris@42: VTW(0, 2), Chris@42: VTW(0, 3), Chris@42: VTW(0, 4), Chris@42: VTW(0, 5), Chris@42: VTW(0, 6), Chris@42: VTW(0, 7), Chris@42: VTW(0, 8), Chris@42: VTW(0, 9), Chris@42: VTW(0, 10), Chris@42: VTW(0, 11), Chris@42: {TW_NEXT, VL, 0} Chris@42: }; Chris@42: Chris@42: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {55, 26, 4, 0}, 0, 0, 0 }; Chris@42: Chris@42: void XSIMD(codelet_t1bv_12) (planner *p) { Chris@42: X(kdft_dit_register) (p, t1bv_12, &desc); Chris@42: } Chris@42: #endif /* HAVE_FMA */