cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:39:04 EST 2012 */ cannam@95: cannam@95: #include "codelet-dft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 10 -name t1bv_10 -include t1b.h -sign 1 */ cannam@95: cannam@95: /* cannam@95: * This function contains 51 FP additions, 40 FP multiplications, cannam@95: * (or, 33 additions, 22 multiplications, 18 fused multiply/add), cannam@95: * 43 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "t1b.h" cannam@95: cannam@95: static void t1bv_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ii; cannam@95: 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)) { cannam@95: V Td, TA, T4, Ta, Tk, TE, Tp, TF, TB, T9, T1, T2, Tb; cannam@95: T1 = LD(&(x[0]), ms, &(x[0])); cannam@95: T2 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V Tg, Tn, Ti, Tl; cannam@95: Tg = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Tn = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Ti = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: Tl = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: { cannam@95: V T6, T8, T5, Tc; cannam@95: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: { cannam@95: V T3, Th, To, Tj, Tm, T7; cannam@95: T7 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: T3 = BYTW(&(W[TWVL * 8]), T2); cannam@95: Th = BYTW(&(W[TWVL * 6]), Tg); cannam@95: To = BYTW(&(W[0]), Tn); cannam@95: Tj = BYTW(&(W[TWVL * 16]), Ti); cannam@95: Tm = BYTW(&(W[TWVL * 10]), Tl); cannam@95: T6 = BYTW(&(W[TWVL * 2]), T5); cannam@95: Td = BYTW(&(W[TWVL * 4]), Tc); cannam@95: T8 = BYTW(&(W[TWVL * 12]), T7); cannam@95: TA = VADD(T1, T3); cannam@95: T4 = VSUB(T1, T3); cannam@95: Ta = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: Tk = VSUB(Th, Tj); cannam@95: TE = VADD(Th, Tj); cannam@95: Tp = VSUB(Tm, To); cannam@95: TF = VADD(Tm, To); cannam@95: } cannam@95: TB = VADD(T6, T8); cannam@95: T9 = VSUB(T6, T8); cannam@95: } cannam@95: } cannam@95: Tb = BYTW(&(W[TWVL * 14]), Ta); cannam@95: { cannam@95: V TL, TG, Tw, Tq, TC, Te; cannam@95: TL = VSUB(TE, TF); cannam@95: TG = VADD(TE, TF); cannam@95: Tw = VSUB(Tk, Tp); cannam@95: Tq = VADD(Tk, Tp); cannam@95: TC = VADD(Tb, Td); cannam@95: Te = VSUB(Tb, Td); cannam@95: { cannam@95: V TM, TD, Tv, Tf; cannam@95: TM = VSUB(TB, TC); cannam@95: TD = VADD(TB, TC); cannam@95: Tv = VSUB(T9, Te); cannam@95: Tf = VADD(T9, Te); cannam@95: { cannam@95: V TP, TN, TH, TJ, Tz, Tx, Tr, Tt, TI, Ts; cannam@95: TP = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TL, TM)); cannam@95: TN = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TM, TL)); cannam@95: TH = VADD(TD, TG); cannam@95: TJ = VSUB(TD, TG); cannam@95: Tz = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tv, Tw)); cannam@95: Tx = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tw, Tv)); cannam@95: Tr = VADD(Tf, Tq); cannam@95: Tt = VSUB(Tf, Tq); cannam@95: ST(&(x[0]), VADD(TA, TH), ms, &(x[0])); cannam@95: TI = VFNMS(LDK(KP250000000), TH, TA); cannam@95: ST(&(x[WS(rs, 5)]), VADD(T4, Tr), ms, &(x[WS(rs, 1)])); cannam@95: Ts = VFNMS(LDK(KP250000000), Tr, T4); cannam@95: { cannam@95: V TK, TO, Tu, Ty; cannam@95: TK = VFNMS(LDK(KP559016994), TJ, TI); cannam@95: TO = VFMA(LDK(KP559016994), TJ, TI); cannam@95: Tu = VFMA(LDK(KP559016994), Tt, Ts); cannam@95: Ty = VFNMS(LDK(KP559016994), Tt, Ts); cannam@95: ST(&(x[WS(rs, 8)]), VFMAI(TN, TK), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 2)]), VFNMSI(TN, TK), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 6)]), VFMAI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 4)]), VFNMSI(TP, TO), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 9)]), VFNMSI(Tx, Tu), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 1)]), VFMAI(Tx, Tu), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VFNMSI(Tz, Ty), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 3)]), VFMAI(Tz, Ty), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: VTW(0, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 10, XSIMD_STRING("t1bv_10"), twinstr, &GENUS, {33, 22, 18, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1bv_10) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1bv_10, &desc); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 10 -name t1bv_10 -include t1b.h -sign 1 */ cannam@95: cannam@95: /* cannam@95: * This function contains 51 FP additions, 30 FP multiplications, cannam@95: * (or, 45 additions, 24 multiplications, 6 fused multiply/add), cannam@95: * 32 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "t1b.h" cannam@95: cannam@95: static void t1bv_10(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: { cannam@95: INT m; cannam@95: R *x; cannam@95: x = ii; cannam@95: 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)) { cannam@95: V Tu, TH, Tg, Tl, Tp, TD, TE, TJ, T5, Ta, To, TA, TB, TI, Tr; cannam@95: V Tt, Ts; cannam@95: Tr = LD(&(x[0]), ms, &(x[0])); cannam@95: Ts = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@95: Tt = BYTW(&(W[TWVL * 8]), Ts); cannam@95: Tu = VSUB(Tr, Tt); cannam@95: TH = VADD(Tr, Tt); cannam@95: { cannam@95: V Td, Tk, Tf, Ti; cannam@95: { cannam@95: V Tc, Tj, Te, Th; cannam@95: Tc = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@95: Td = BYTW(&(W[TWVL * 6]), Tc); cannam@95: Tj = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@95: Tk = BYTW(&(W[0]), Tj); cannam@95: Te = LD(&(x[WS(rs, 9)]), ms, &(x[WS(rs, 1)])); cannam@95: Tf = BYTW(&(W[TWVL * 16]), Te); cannam@95: Th = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@95: Ti = BYTW(&(W[TWVL * 10]), Th); cannam@95: } cannam@95: Tg = VSUB(Td, Tf); cannam@95: Tl = VSUB(Ti, Tk); cannam@95: Tp = VADD(Tg, Tl); cannam@95: TD = VADD(Td, Tf); cannam@95: TE = VADD(Ti, Tk); cannam@95: TJ = VADD(TD, TE); cannam@95: } cannam@95: { cannam@95: V T2, T9, T4, T7; cannam@95: { cannam@95: V T1, T8, T3, T6; cannam@95: T1 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@95: T2 = BYTW(&(W[TWVL * 2]), T1); cannam@95: T8 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@95: T9 = BYTW(&(W[TWVL * 4]), T8); cannam@95: T3 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@95: T4 = BYTW(&(W[TWVL * 12]), T3); cannam@95: T6 = LD(&(x[WS(rs, 8)]), ms, &(x[0])); cannam@95: T7 = BYTW(&(W[TWVL * 14]), T6); cannam@95: } cannam@95: T5 = VSUB(T2, T4); cannam@95: Ta = VSUB(T7, T9); cannam@95: To = VADD(T5, Ta); cannam@95: TA = VADD(T2, T4); cannam@95: TB = VADD(T7, T9); cannam@95: TI = VADD(TA, TB); cannam@95: } cannam@95: { cannam@95: V Tq, Tv, Tw, Tn, Tz, Tb, Tm, Ty, Tx; cannam@95: Tq = VMUL(LDK(KP559016994), VSUB(To, Tp)); cannam@95: Tv = VADD(To, Tp); cannam@95: Tw = VFNMS(LDK(KP250000000), Tv, Tu); cannam@95: Tb = VSUB(T5, Ta); cannam@95: Tm = VSUB(Tg, Tl); cannam@95: Tn = VBYI(VFMA(LDK(KP951056516), Tb, VMUL(LDK(KP587785252), Tm))); cannam@95: Tz = VBYI(VFNMS(LDK(KP951056516), Tm, VMUL(LDK(KP587785252), Tb))); cannam@95: ST(&(x[WS(rs, 5)]), VADD(Tu, Tv), ms, &(x[WS(rs, 1)])); cannam@95: Ty = VSUB(Tw, Tq); cannam@95: ST(&(x[WS(rs, 3)]), VSUB(Ty, Tz), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 7)]), VADD(Tz, Ty), ms, &(x[WS(rs, 1)])); cannam@95: Tx = VADD(Tq, Tw); cannam@95: ST(&(x[WS(rs, 1)]), VADD(Tn, Tx), ms, &(x[WS(rs, 1)])); cannam@95: ST(&(x[WS(rs, 9)]), VSUB(Tx, Tn), ms, &(x[WS(rs, 1)])); cannam@95: } cannam@95: { cannam@95: V TM, TK, TL, TG, TP, TC, TF, TO, TN; cannam@95: TM = VMUL(LDK(KP559016994), VSUB(TI, TJ)); cannam@95: TK = VADD(TI, TJ); cannam@95: TL = VFNMS(LDK(KP250000000), TK, TH); cannam@95: TC = VSUB(TA, TB); cannam@95: TF = VSUB(TD, TE); cannam@95: TG = VBYI(VFNMS(LDK(KP951056516), TF, VMUL(LDK(KP587785252), TC))); cannam@95: TP = VBYI(VFMA(LDK(KP951056516), TC, VMUL(LDK(KP587785252), TF))); cannam@95: ST(&(x[0]), VADD(TH, TK), ms, &(x[0])); cannam@95: TO = VADD(TM, TL); cannam@95: ST(&(x[WS(rs, 4)]), VSUB(TO, TP), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 6)]), VADD(TP, TO), ms, &(x[0])); cannam@95: TN = VSUB(TL, TM); cannam@95: ST(&(x[WS(rs, 2)]), VADD(TG, TN), ms, &(x[0])); cannam@95: ST(&(x[WS(rs, 8)]), VSUB(TN, TG), ms, &(x[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(0, 1), cannam@95: VTW(0, 2), cannam@95: VTW(0, 3), cannam@95: VTW(0, 4), cannam@95: VTW(0, 5), cannam@95: VTW(0, 6), cannam@95: VTW(0, 7), cannam@95: VTW(0, 8), cannam@95: VTW(0, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const ct_desc desc = { 10, XSIMD_STRING("t1bv_10"), twinstr, &GENUS, {45, 24, 6, 0}, 0, 0, 0 }; cannam@95: cannam@95: void XSIMD(codelet_t1bv_10) (planner *p) { cannam@95: X(kdft_dit_register) (p, t1bv_10, &desc); cannam@95: } cannam@95: #endif /* HAVE_FMA */