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 16:44:07 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 5 -name t1buv_5 -include t1bu.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 20 FP additions, 19 FP multiplications, cannam@127: * (or, 11 additions, 10 multiplications, 9 fused multiply/add), cannam@127: * 26 stack variables, 4 constants, and 10 memory accesses cannam@127: */ cannam@127: #include "t1bu.h" cannam@127: cannam@127: static void t1buv_5(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(5, rs)) { cannam@127: V T1, T2, T9, T4, T7; cannam@127: T1 = LD(&(x[0]), ms, &(x[0])); cannam@127: T2 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T9 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T4 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: T7 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: { cannam@127: V T3, Ta, T5, T8; cannam@127: T3 = BYTW(&(W[0]), T2); cannam@127: Ta = BYTW(&(W[TWVL * 4]), T9); cannam@127: T5 = BYTW(&(W[TWVL * 6]), T4); cannam@127: T8 = BYTW(&(W[TWVL * 2]), T7); cannam@127: { cannam@127: V T6, Tg, Tb, Th; cannam@127: T6 = VADD(T3, T5); cannam@127: Tg = VSUB(T3, T5); cannam@127: Tb = VADD(T8, Ta); cannam@127: Th = VSUB(T8, Ta); cannam@127: { cannam@127: V Te, Tc, Tk, Ti, Td, Tj, Tf; cannam@127: Te = VSUB(T6, Tb); cannam@127: Tc = VADD(T6, Tb); cannam@127: Tk = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), Tg, Th)); cannam@127: Ti = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Th, Tg)); cannam@127: Td = VFNMS(LDK(KP250000000), Tc, T1); cannam@127: ST(&(x[0]), VADD(T1, Tc), ms, &(x[0])); cannam@127: Tj = VFNMS(LDK(KP559016994), Te, Td); cannam@127: Tf = VFMA(LDK(KP559016994), Te, Td); cannam@127: ST(&(x[WS(rs, 2)]), VFNMSI(Tk, Tj), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 3)]), VFMAI(Tk, Tj), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 4)]), VFNMSI(Ti, Tf), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 1)]), VFMAI(Ti, Tf), ms, &(x[WS(rs, 1)])); 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: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, XSIMD_STRING("t1buv_5"), twinstr, &GENUS, {11, 10, 9, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1buv_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1buv_5, &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 5 -name t1buv_5 -include t1bu.h -sign 1 */ cannam@127: cannam@127: /* cannam@127: * This function contains 20 FP additions, 14 FP multiplications, cannam@127: * (or, 17 additions, 11 multiplications, 3 fused multiply/add), cannam@127: * 20 stack variables, 4 constants, and 10 memory accesses cannam@127: */ cannam@127: #include "t1bu.h" cannam@127: cannam@127: static void t1buv_5(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@127: { cannam@127: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@127: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@127: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@127: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@127: { cannam@127: INT m; cannam@127: R *x; cannam@127: x = ii; cannam@127: for (m = mb, W = W + (mb * ((TWVL / VL) * 8)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 8), MAKE_VOLATILE_STRIDE(5, rs)) { cannam@127: V Tf, T5, Ta, Tc, Td, Tg; cannam@127: Tf = LD(&(x[0]), ms, &(x[0])); cannam@127: { cannam@127: V T2, T9, T4, T7; cannam@127: { cannam@127: V T1, T8, T3, T6; cannam@127: T1 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@127: T2 = BYTW(&(W[0]), T1); cannam@127: T8 = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@127: T9 = BYTW(&(W[TWVL * 4]), T8); cannam@127: T3 = LD(&(x[WS(rs, 4)]), ms, &(x[0])); cannam@127: T4 = BYTW(&(W[TWVL * 6]), T3); cannam@127: T6 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@127: T7 = BYTW(&(W[TWVL * 2]), T6); cannam@127: } cannam@127: T5 = VSUB(T2, T4); cannam@127: Ta = VSUB(T7, T9); cannam@127: Tc = VADD(T2, T4); cannam@127: Td = VADD(T7, T9); cannam@127: Tg = VADD(Tc, Td); cannam@127: } cannam@127: ST(&(x[0]), VADD(Tf, Tg), ms, &(x[0])); cannam@127: { cannam@127: V Tb, Tj, Ti, Tk, Te, Th; cannam@127: Tb = VBYI(VFMA(LDK(KP951056516), T5, VMUL(LDK(KP587785252), Ta))); cannam@127: Tj = VBYI(VFNMS(LDK(KP951056516), Ta, VMUL(LDK(KP587785252), T5))); cannam@127: Te = VMUL(LDK(KP559016994), VSUB(Tc, Td)); cannam@127: Th = VFNMS(LDK(KP250000000), Tg, Tf); cannam@127: Ti = VADD(Te, Th); cannam@127: Tk = VSUB(Th, Te); cannam@127: ST(&(x[WS(rs, 1)]), VADD(Tb, Ti), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 3)]), VSUB(Tk, Tj), ms, &(x[WS(rs, 1)])); cannam@127: ST(&(x[WS(rs, 4)]), VSUB(Ti, Tb), ms, &(x[0])); cannam@127: ST(&(x[WS(rs, 2)]), VADD(Tj, Tk), ms, &(x[0])); 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: {TW_NEXT, VL, 0} cannam@127: }; cannam@127: cannam@127: static const ct_desc desc = { 5, XSIMD_STRING("t1buv_5"), twinstr, &GENUS, {17, 11, 3, 0}, 0, 0, 0 }; cannam@127: cannam@127: void XSIMD(codelet_t1buv_5) (planner *p) { cannam@127: X(kdft_dit_register) (p, t1buv_5, &desc); cannam@127: } cannam@127: #endif /* HAVE_FMA */