cannam@127: /* cannam@127: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@127: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@127: * cannam@127: * This program is free software; you can redistribute it and/or modify cannam@127: * it under the terms of the GNU General Public License as published by cannam@127: * the Free Software Foundation; either version 2 of the License, or cannam@127: * (at your option) any later version. cannam@127: * cannam@127: * This program is distributed in the hope that it will be useful, cannam@127: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@127: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@127: * GNU General Public License for more details. cannam@127: * cannam@127: * You should have received a copy of the GNU General Public License cannam@127: * along with this program; if not, write to the Free Software cannam@127: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@127: * cannam@127: */ cannam@127: cannam@127: /* This file was automatically generated --- DO NOT EDIT */ cannam@127: /* Generated on Sat Jul 30 16:44:16 EDT 2016 */ cannam@127: cannam@127: #include "codelet-dft.h" cannam@127: cannam@127: #ifdef HAVE_FMA cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1bv_8 -include t1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 33 FP additions, 24 FP multiplications, cannam@127: * (or, 23 additions, 14 multiplications, 10 fused multiply/add), cannam@127: * 36 stack variables, 1 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t1b.h" cannam@127: cannam@127: static void t1bv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 14)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: V T1, T2, Th, Tj, T5, T7, Ta, Tc; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T2 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: Th = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: Tj = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@127: T5 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T7 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@127: Ta = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@127: Tc = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: { cannam@127: V T3, Ti, Tk, T6, T8, Tb, Td; cannam@127: T3 = BYTW(&(W[TWVL * 6]), T2); cannam@127: Ti = BYTW(&(W[TWVL * 2]), Th); cannam@127: Tk = BYTW(&(W[TWVL * 10]), Tj); cannam@127: T6 = BYTW(&(W[0]), T5); cannam@127: T8 = BYTW(&(W[TWVL * 8]), T7); cannam@127: Tb = BYTW(&(W[TWVL * 12]), Ta); cannam@127: Td = BYTW(&(W[TWVL * 4]), Tc); cannam@127: { cannam@127: V Tq, T4, Tr, Tl, Tt, T9, Tu, Te, Tw, Ts; cannam@127: Tq = VADD(T1, T3); cannam@127: T4 = VSUB(T1, T3); cannam@127: Tr = VADD(Ti, Tk); cannam@127: Tl = VSUB(Ti, Tk); cannam@127: Tt = VADD(T6, T8); cannam@127: T9 = VSUB(T6, T8); cannam@127: Tu = VADD(Tb, Td); cannam@127: Te = VSUB(Tb, Td); cannam@127: Tw = VADD(Tq, Tr); cannam@127: Ts = VSUB(Tq, Tr); cannam@127: { cannam@127: V Tx, Tv, Tm, Tf; cannam@127: Tx = VADD(Tt, Tu); cannam@127: Tv = VSUB(Tt, Tu); cannam@127: Tm = VSUB(T9, Te); cannam@127: Tf = VADD(T9, Te); cannam@127: { cannam@127: V Tp, Tn, To, Tg; cannam@127: ST(&(x[0]), VADD(Tw, Tx), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 4)]), VSUB(Tw, Tx), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 2)]), VFMAI(Tv, Ts), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 6)]), VFNMSI(Tv, Ts), ms, &(x[0])); cannam@127: Tp = VFMA(LDK(KP707106781), Tm, Tl); cannam@127: Tn = VFNMS(LDK(KP707106781), Tm, Tl); cannam@127: To = VFMA(LDK(KP707106781), Tf, T4); cannam@127: Tg = VFNMS(LDK(KP707106781), Tf, T4); cannam@127: ST(&(x[WS(rs, 1)]), VFMAI(Tp, To), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 7)]), VFNMSI(Tp, To), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 5)]), VFMAI(Tn, Tg), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 3)]), VFNMSI(Tn, Tg), ms, &(x[WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: VTW(0, 5), cannam@127: VTW(0, 6), cannam@127: VTW(0, 7), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 8, XSIMD_STRING("t1bv_8"), twinstr, &GENUS, {23, 14, 10, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1bv_8) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1bv_8, &desc); cannam@127: } cannam@127: #else /* HAVE_FMA */ cannam@127: cannam@127: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -n 8 -name t1bv_8 -include t1b.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 33 FP additions, 16 FP multiplications, cannam@127: * (or, 33 additions, 16 multiplications, 0 fused multiply/add), cannam@127: * 24 stack variables, 1 constants, and 16 memory accesses cannam@127: */ cannam@127: #include "t1b.h" cannam@127: cannam@127: static void t1bv_8(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP707106781, +0.707106781186547524400844362104849039284835938); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 14)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 14), MAKE_VOLATILE_STRIDE(8, rs)) { cannam@127: V Tl, Tq, Tg, Tr, T5, Tt, Ta, Tu, Ti, Tk, Tj; cannam@127: Ti = LD(&(x[0]), ms, &(x[0])); cannam@127: Tj = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: Tk = BYTW(&(W[TWVL * 6]), Tj); cannam@127: Tl = VSUB(Ti, Tk); cannam@127: Tq = VADD(Ti, Tk); cannam@127: { cannam@127: V Td, Tf, Tc, Te; cannam@127: Tc = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: Td = BYTW(&(W[TWVL * 2]), Tc); cannam@127: Te = LD(&(x[WS(rs, 6)]), ms, &(x[0])); cannam@127: Tf = BYTW(&(W[TWVL * 10]), Te); cannam@127: Tg = VSUB(Td, Tf); cannam@127: Tr = VADD(Td, Tf); cannam@127: } cannam@127: { cannam@127: V T2, T4, T1, T3; cannam@127: T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T2 = BYTW(&(W[0]), T1); cannam@127: T3 = LD(&(x[WS(rs, 5)]), ms, &(x[WS(rs, 1)])); cannam@127: T4 = BYTW(&(W[TWVL * 8]), T3); cannam@127: T5 = VSUB(T2, T4); cannam@127: Tt = VADD(T2, T4); cannam@127: } cannam@127: { cannam@127: V T7, T9, T6, T8; cannam@127: T6 = LD(&(x[WS(rs, 7)]), ms, &(x[WS(rs, 1)])); cannam@127: T7 = BYTW(&(W[TWVL * 12]), T6); cannam@127: T8 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T9 = BYTW(&(W[TWVL * 4]), T8); cannam@127: Ta = VSUB(T7, T9); cannam@127: Tu = VADD(T7, T9); cannam@127: } cannam@127: { cannam@127: V Ts, Tv, Tw, Tx; cannam@127: Ts = VSUB(Tq, Tr); cannam@127: Tv = VBYI(VSUB(Tt, Tu)); cannam@127: ST(&(x[WS(rs, 6)]), VSUB(Ts, Tv), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 2)]), VADD(Ts, Tv), ms, &(x[0])); cannam@127: Tw = VADD(Tq, Tr); cannam@127: Tx = VADD(Tt, Tu); cannam@127: ST(&(x[WS(rs, 4)]), VSUB(Tw, Tx), ms, &(x[0])); cannam@127: ST(&(x[0]), VADD(Tw, Tx), ms, &(x[0])); cannam@127: { cannam@127: V Th, To, Tn, Tp, Tb, Tm; cannam@127: Tb = VMUL(LDK(KP707106781), VSUB(T5, Ta)); cannam@127: Th = VBYI(VSUB(Tb, Tg)); cannam@127: To = VBYI(VADD(Tg, Tb)); cannam@127: Tm = VMUL(LDK(KP707106781), VADD(T5, Ta)); cannam@127: Tn = VSUB(Tl, Tm); cannam@127: Tp = VADD(Tl, Tm); cannam@127: ST(&(x[WS(rs, 3)]), VADD(Th, Tn), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 7)]), VSUB(Tp, To), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 5)]), VSUB(Tn, Th), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 1)]), VADD(To, Tp), ms, &(x[WS(rs, 1)])); cannam@127: } cannam@127: } cannam@127: } cannam@127: } cannam@127: VLEAVE(); cannam@127: } cannam@127: cannam@127: static const tw_instr twinstr[] = { cannam@127: VTW(0, 1), cannam@127: VTW(0, 2), cannam@127: VTW(0, 3), cannam@127: VTW(0, 4), cannam@127: VTW(0, 5), cannam@127: VTW(0, 6), cannam@127: VTW(0, 7), cannam@127: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 8, XSIMD_STRING("t1bv_8"), twinstr, &GENUS, {33, 16, 0, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1bv_8) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1bv_8, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */