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 07:39:04 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 12 -name t1bv_12 -include t1b.h -sign 1 */ Chris@10: Chris@10: /* Chris@10: * This function contains 59 FP additions, 42 FP multiplications, Chris@10: * (or, 41 additions, 24 multiplications, 18 fused multiply/add), Chris@10: * 41 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "t1b.h" Chris@10: Chris@10: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ii; Chris@10: 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@10: V TI, Ti, TA, T7, Tm, TE, Tw, Tk, Tf, TB, TU, TM; Chris@10: { Chris@10: V T9, TK, Tj, TL, Te; Chris@10: { Chris@10: V T1, T4, T2, Tp, Tt, Tr; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: Tp = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: Tt = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tr = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: { Chris@10: V T5, T3, Tq, Tu, Ts, Td, Tb, T8, Tc, Ta; Chris@10: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@10: T5 = BYTW(&(W[TWVL * 14]), T4); Chris@10: T3 = BYTW(&(W[TWVL * 6]), T2); Chris@10: Tq = BYTW(&(W[TWVL * 16]), Tp); Chris@10: Tu = BYTW(&(W[TWVL * 8]), Tt); Chris@10: Ts = BYTW(&(W[0]), Tr); Chris@10: T9 = BYTW(&(W[TWVL * 10]), T8); Chris@10: Td = BYTW(&(W[TWVL * 2]), Tc); Chris@10: Tb = BYTW(&(W[TWVL * 18]), Ta); Chris@10: { Chris@10: V Th, T6, Tl, Tv; Chris@10: Th = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: TK = VSUB(T3, T5); Chris@10: T6 = VADD(T3, T5); Chris@10: Tl = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@10: Tv = VADD(Ts, Tu); Chris@10: TI = VSUB(Tu, Ts); Chris@10: Tj = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: TL = VSUB(Tb, Td); Chris@10: Te = VADD(Tb, Td); Chris@10: Ti = BYTW(&(W[TWVL * 4]), Th); Chris@10: TA = VFNMS(LDK(KP500000000), T6, T1); Chris@10: T7 = VADD(T1, T6); Chris@10: Tm = BYTW(&(W[TWVL * 20]), Tl); Chris@10: TE = VFNMS(LDK(KP500000000), Tv, Tq); Chris@10: Tw = VADD(Tq, Tv); Chris@10: } Chris@10: } Chris@10: } Chris@10: Tk = BYTW(&(W[TWVL * 12]), Tj); Chris@10: Tf = VADD(T9, Te); Chris@10: TB = VFNMS(LDK(KP500000000), Te, T9); Chris@10: TU = VSUB(TK, TL); Chris@10: TM = VADD(TK, TL); Chris@10: } Chris@10: { Chris@10: V Tn, TH, TC, TQ, Ty, Tg; Chris@10: Tn = VADD(Tk, Tm); Chris@10: TH = VSUB(Tk, Tm); Chris@10: TC = VADD(TA, TB); Chris@10: TQ = VSUB(TA, TB); Chris@10: Ty = VADD(T7, Tf); Chris@10: Tg = VSUB(T7, Tf); Chris@10: { Chris@10: V To, TD, TJ, TR; Chris@10: To = VADD(Ti, Tn); Chris@10: TD = VFNMS(LDK(KP500000000), Tn, Ti); Chris@10: TJ = VSUB(TH, TI); Chris@10: TR = VADD(TH, TI); Chris@10: { Chris@10: V TP, TN, TW, TS, TO, TG, TX, TV; Chris@10: { Chris@10: V Tz, Tx, TF, TT; Chris@10: Tz = VADD(To, Tw); Chris@10: Tx = VSUB(To, Tw); Chris@10: TF = VADD(TD, TE); Chris@10: TT = VSUB(TD, TE); Chris@10: TP = VMUL(LDK(KP866025403), VADD(TM, TJ)); Chris@10: TN = VMUL(LDK(KP866025403), VSUB(TJ, TM)); Chris@10: TW = VFMA(LDK(KP866025403), TR, TQ); Chris@10: TS = VFNMS(LDK(KP866025403), TR, TQ); Chris@10: ST(&(x[WS(rs, 6)]), VSUB(Ty, Tz), ms, &(x[0])); Chris@10: ST(&(x[0]), VADD(Ty, Tz), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 9)]), VFMAI(Tx, Tg), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 3)]), VFNMSI(Tx, Tg), ms, &(x[WS(rs, 1)])); Chris@10: TO = VADD(TC, TF); Chris@10: TG = VSUB(TC, TF); Chris@10: TX = VFNMS(LDK(KP866025403), TU, TT); Chris@10: TV = VFMA(LDK(KP866025403), TU, TT); Chris@10: } Chris@10: ST(&(x[WS(rs, 8)]), 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, 2)]), VFMAI(TN, TG), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 10)]), VFNMSI(TN, TG), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 5)]), VFMAI(TX, TW), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VFNMSI(TX, TW), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 11)]), VFNMSI(TV, TS), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 1)]), VFMAI(TV, TS), ms, &(x[WS(rs, 1)])); 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: VTW(0, 10), Chris@10: VTW(0, 11), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {41, 24, 18, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1bv_12) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1bv_12, &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 12 -name t1bv_12 -include t1b.h -sign 1 */ Chris@10: Chris@10: /* Chris@10: * This function contains 59 FP additions, 30 FP multiplications, Chris@10: * (or, 55 additions, 26 multiplications, 4 fused multiply/add), Chris@10: * 28 stack variables, 2 constants, and 24 memory accesses Chris@10: */ Chris@10: #include "t1b.h" Chris@10: Chris@10: static void t1bv_12(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) Chris@10: { Chris@10: DVK(KP866025403, +0.866025403784438646763723170752936183471402627); Chris@10: DVK(KP500000000, +0.500000000000000000000000000000000000000000000); Chris@10: { Chris@10: INT m; Chris@10: R *x; Chris@10: x = ii; Chris@10: 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@10: V T1, Tt, T6, T7, TB, Tq, TC, TD, T9, Tu, Te, Tf, Tx, Tl, Ty; Chris@10: V Tz; Chris@10: { Chris@10: V T5, T3, T4, T2; Chris@10: T1 = LD(&(x[0]), ms, &(x[0])); Chris@10: T4 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); Chris@10: T5 = BYTW(&(W[TWVL * 14]), T4); Chris@10: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); Chris@10: T3 = BYTW(&(W[TWVL * 6]), T2); Chris@10: Tt = VSUB(T3, T5); Chris@10: T6 = VADD(T3, T5); Chris@10: T7 = VFNMS(LDK(KP500000000), T6, T1); Chris@10: } Chris@10: { Chris@10: V Tn, Tp, Tm, TA, To; Chris@10: Tm = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); Chris@10: Tn = BYTW(&(W[0]), Tm); Chris@10: TA = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); Chris@10: TB = BYTW(&(W[TWVL * 16]), TA); Chris@10: To = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); Chris@10: Tp = BYTW(&(W[TWVL * 8]), To); Chris@10: Tq = VSUB(Tn, Tp); Chris@10: TC = VADD(Tn, Tp); Chris@10: TD = VFNMS(LDK(KP500000000), TC, TB); Chris@10: } Chris@10: { Chris@10: V Td, Tb, T8, Tc, Ta; Chris@10: T8 = LD(&(x[WS(rs, 6)]), ms, &(x[0])); Chris@10: T9 = BYTW(&(W[TWVL * 10]), T8); Chris@10: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); Chris@10: Td = BYTW(&(W[TWVL * 2]), Tc); Chris@10: Ta = LD(&(x[WS(rs, 10)]), ms, &(x[0])); Chris@10: Tb = BYTW(&(W[TWVL * 18]), Ta); Chris@10: Tu = VSUB(Tb, Td); Chris@10: Te = VADD(Tb, Td); Chris@10: Tf = VFNMS(LDK(KP500000000), Te, T9); Chris@10: } Chris@10: { Chris@10: V Ti, Tk, Th, Tw, Tj; Chris@10: Th = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); Chris@10: Ti = BYTW(&(W[TWVL * 12]), Th); Chris@10: Tw = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); Chris@10: Tx = BYTW(&(W[TWVL * 4]), Tw); Chris@10: Tj = LD(&(x[WS(rs, 11)]), ms, &(x[WS(rs, 1)])); Chris@10: Tk = BYTW(&(W[TWVL * 20]), Tj); Chris@10: Tl = VSUB(Ti, Tk); Chris@10: Ty = VADD(Ti, Tk); Chris@10: Tz = VFNMS(LDK(KP500000000), Ty, Tx); Chris@10: } Chris@10: { Chris@10: V Ts, TG, TF, TH; Chris@10: { Chris@10: V Tg, Tr, Tv, TE; Chris@10: Tg = VSUB(T7, Tf); Chris@10: Tr = VMUL(LDK(KP866025403), VSUB(Tl, Tq)); Chris@10: Ts = VSUB(Tg, Tr); Chris@10: TG = VADD(Tg, Tr); Chris@10: Tv = VMUL(LDK(KP866025403), VSUB(Tt, Tu)); Chris@10: TE = VSUB(Tz, TD); Chris@10: TF = VBYI(VADD(Tv, TE)); Chris@10: TH = VBYI(VSUB(TE, Tv)); Chris@10: } Chris@10: ST(&(x[WS(rs, 11)]), VSUB(Ts, TF), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 5)]), VADD(TG, TH), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 1)]), VADD(Ts, TF), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 7)]), VSUB(TG, TH), ms, &(x[WS(rs, 1)])); Chris@10: } Chris@10: { Chris@10: V TS, TW, TV, TX; Chris@10: { Chris@10: V TQ, TR, TT, TU; Chris@10: TQ = VADD(T1, T6); Chris@10: TR = VADD(T9, Te); Chris@10: TS = VSUB(TQ, TR); Chris@10: TW = VADD(TQ, TR); Chris@10: TT = VADD(Tx, Ty); Chris@10: TU = VADD(TB, TC); Chris@10: TV = VBYI(VSUB(TT, TU)); Chris@10: TX = VADD(TT, TU); Chris@10: } Chris@10: ST(&(x[WS(rs, 3)]), VSUB(TS, TV), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[0]), VADD(TW, TX), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 9)]), VADD(TS, TV), ms, &(x[WS(rs, 1)])); Chris@10: ST(&(x[WS(rs, 6)]), VSUB(TW, TX), ms, &(x[0])); Chris@10: } Chris@10: { Chris@10: V TK, TO, TN, TP; Chris@10: { Chris@10: V TI, TJ, TL, TM; Chris@10: TI = VADD(Tl, Tq); Chris@10: TJ = VADD(Tt, Tu); Chris@10: TK = VBYI(VMUL(LDK(KP866025403), VSUB(TI, TJ))); Chris@10: TO = VBYI(VMUL(LDK(KP866025403), VADD(TJ, TI))); Chris@10: TL = VADD(T7, Tf); Chris@10: TM = VADD(Tz, TD); Chris@10: TN = VSUB(TL, TM); Chris@10: TP = VADD(TL, TM); Chris@10: } Chris@10: ST(&(x[WS(rs, 2)]), VADD(TK, TN), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 8)]), VSUB(TP, TO), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 10)]), VSUB(TN, TK), ms, &(x[0])); Chris@10: ST(&(x[WS(rs, 4)]), VADD(TO, TP), 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: VTW(0, 10), Chris@10: VTW(0, 11), Chris@10: {TW_NEXT, VL, 0} Chris@10: }; Chris@10: Chris@10: static const ct_desc desc = { 12, XSIMD_STRING("t1bv_12"), twinstr, &GENUS, {55, 26, 4, 0}, 0, 0, 0 }; Chris@10: Chris@10: void XSIMD(codelet_t1bv_12) (planner *p) { Chris@10: X(kdft_dit_register) (p, t1bv_12, &desc); Chris@10: } Chris@10: #endif /* HAVE_FMA */