cannam@167: /* cannam@167: * Copyright (c) 2003, 2007-14 Matteo Frigo cannam@167: * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology cannam@167: * cannam@167: * This program is free software; you can redistribute it and/or modify cannam@167: * it under the terms of the GNU General Public License as published by cannam@167: * the Free Software Foundation; either version 2 of the License, or cannam@167: * (at your option) any later version. cannam@167: * cannam@167: * This program is distributed in the hope that it will be useful, cannam@167: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@167: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@167: * GNU General Public License for more details. cannam@167: * cannam@167: * You should have received a copy of the GNU General Public License cannam@167: * along with this program; if not, write to the Free Software cannam@167: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@167: * cannam@167: */ cannam@167: cannam@167: /* This file was automatically generated --- DO NOT EDIT */ cannam@167: /* Generated on Thu May 24 08:05:51 EDT 2018 */ cannam@167: cannam@167: #include "dft/codelet-dft.h" cannam@167: cannam@167: #if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA) cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle_c.native -fma -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 4 -name t3fv_4 -include dft/simd/t3f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 12 FP additions, 10 FP multiplications, cannam@167: * (or, 10 additions, 8 multiplications, 2 fused multiply/add), cannam@167: * 16 stack variables, 0 constants, and 8 memory accesses cannam@167: */ cannam@167: #include "dft/simd/t3f.h" cannam@167: cannam@167: static void t3fv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 4)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 4), MAKE_VOLATILE_STRIDE(4, rs)) { cannam@167: V T2, T3, T4; cannam@167: T2 = LDW(&(W[0])); cannam@167: T3 = LDW(&(W[TWVL * 2])); cannam@167: T4 = VZMULJ(T2, T3); cannam@167: { cannam@167: V T1, Tb, T6, T9, Ta, T5, T8; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: Ta = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: Tb = VZMULJ(T3, Ta); cannam@167: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T6 = VZMULJ(T4, T5); cannam@167: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T9 = VZMULJ(T2, T8); cannam@167: { cannam@167: V T7, Tc, Td, Te; cannam@167: T7 = VSUB(T1, T6); cannam@167: Tc = VSUB(T9, Tb); cannam@167: ST(&(x[WS(rs, 1)]), VFNMSI(Tc, T7), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 3)]), VFMAI(Tc, T7), ms, &(x[WS(rs, 1)])); cannam@167: Td = VADD(T1, T6); cannam@167: Te = VADD(T9, Tb); cannam@167: ST(&(x[WS(rs, 2)]), VSUB(Td, Te), ms, &(x[0])); cannam@167: ST(&(x[0]), VADD(Td, Te), ms, &(x[0])); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("t3fv_4"), twinstr, &GENUS, {10, 8, 2, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t3fv_4) (planner *p) { cannam@167: X(kdft_dit_register) (p, t3fv_4, &desc); cannam@167: } cannam@167: #else cannam@167: cannam@167: /* Generated by: ../../../genfft/gen_twiddle_c.native -simd -compact -variables 4 -pipeline-latency 8 -twiddle-log3 -precompute-twiddles -no-generate-bytw -n 4 -name t3fv_4 -include dft/simd/t3f.h */ cannam@167: cannam@167: /* cannam@167: * This function contains 12 FP additions, 8 FP multiplications, cannam@167: * (or, 12 additions, 8 multiplications, 0 fused multiply/add), cannam@167: * 16 stack variables, 0 constants, and 8 memory accesses cannam@167: */ cannam@167: #include "dft/simd/t3f.h" cannam@167: cannam@167: static void t3fv_4(R *ri, R *ii, const R *W, stride rs, INT mb, INT me, INT ms) cannam@167: { cannam@167: { cannam@167: INT m; cannam@167: R *x; cannam@167: x = ri; cannam@167: for (m = mb, W = W + (mb * ((TWVL / VL) * 4)); m < me; m = m + VL, x = x + (VL * ms), W = W + (TWVL * 4), MAKE_VOLATILE_STRIDE(4, rs)) { cannam@167: V T2, T3, T4; cannam@167: T2 = LDW(&(W[0])); cannam@167: T3 = LDW(&(W[TWVL * 2])); cannam@167: T4 = VZMULJ(T2, T3); cannam@167: { cannam@167: V T1, Tb, T6, T9, Ta, T5, T8; cannam@167: T1 = LD(&(x[0]), ms, &(x[0])); cannam@167: Ta = LD(&(x[WS(rs, 3)]), ms, &(x[WS(rs, 1)])); cannam@167: Tb = VZMULJ(T3, Ta); cannam@167: T5 = LD(&(x[WS(rs, 2)]), ms, &(x[0])); cannam@167: T6 = VZMULJ(T4, T5); cannam@167: T8 = LD(&(x[WS(rs, 1)]), ms, &(x[WS(rs, 1)])); cannam@167: T9 = VZMULJ(T2, T8); cannam@167: { cannam@167: V T7, Tc, Td, Te; cannam@167: T7 = VSUB(T1, T6); cannam@167: Tc = VBYI(VSUB(T9, Tb)); cannam@167: ST(&(x[WS(rs, 1)]), VSUB(T7, Tc), ms, &(x[WS(rs, 1)])); cannam@167: ST(&(x[WS(rs, 3)]), VADD(T7, Tc), ms, &(x[WS(rs, 1)])); cannam@167: Td = VADD(T1, T6); cannam@167: Te = VADD(T9, Tb); cannam@167: ST(&(x[WS(rs, 2)]), VSUB(Td, Te), ms, &(x[0])); cannam@167: ST(&(x[0]), VADD(Td, Te), ms, &(x[0])); cannam@167: } cannam@167: } cannam@167: } cannam@167: } cannam@167: VLEAVE(); cannam@167: } cannam@167: cannam@167: static const tw_instr twinstr[] = { cannam@167: VTW(0, 1), cannam@167: VTW(0, 3), cannam@167: {TW_NEXT, VL, 0} cannam@167: }; cannam@167: cannam@167: static const ct_desc desc = { 4, XSIMD_STRING("t3fv_4"), twinstr, &GENUS, {12, 8, 0, 0}, 0, 0, 0 }; cannam@167: cannam@167: void XSIMD(codelet_t3fv_4) (planner *p) { cannam@167: X(kdft_dit_register) (p, t3fv_4, &desc); cannam@167: } cannam@167: #endif