| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2012, 2014 Free Software Foundation, Inc. | ||
| 4 | * | ||
| 5 | * This file is part of VOLK | ||
| 6 | * | ||
| 7 | * SPDX-License-Identifier: LGPL-3.0-or-later | ||
| 8 | */ | ||
| 9 | |||
| 10 | /*! | ||
| 11 | * \page volk_16i_branch_4_state_8 | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * <FIXME> | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_16i_branch_4_state_8(short* target, short* src0, char** permuters, short* | ||
| 20 | * cntl2, short* cntl3, short* scalars) \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li src0: <FIXME> | ||
| 24 | * \li permuters: <FIXME> | ||
| 25 | * \li cntl2: <FIXME> | ||
| 26 | * \li cntl3: <FIXME> | ||
| 27 | * \li scalars: <FIXME> | ||
| 28 | * | ||
| 29 | * \b Outputs | ||
| 30 | * \li target: <FIXME> | ||
| 31 | * | ||
| 32 | * \b Example | ||
| 33 | * \code | ||
| 34 | * int N = 10000; | ||
| 35 | * | ||
| 36 | * volk_16i_branch_4_state_8(); | ||
| 37 | * | ||
| 38 | * volk_free(x); | ||
| 39 | * volk_free(t); | ||
| 40 | * \endcode | ||
| 41 | */ | ||
| 42 | |||
| 43 | #ifndef INCLUDED_volk_16i_branch_4_state_8_a_H | ||
| 44 | #define INCLUDED_volk_16i_branch_4_state_8_a_H | ||
| 45 | |||
| 46 | #include <inttypes.h> | ||
| 47 | #include <stdio.h> | ||
| 48 | |||
| 49 | #ifdef LV_HAVE_SSSE3 | ||
| 50 | |||
| 51 | #include <emmintrin.h> | ||
| 52 | #include <tmmintrin.h> | ||
| 53 | #include <xmmintrin.h> | ||
| 54 | |||
| 55 | ✗ | static inline void volk_16i_branch_4_state_8_a_ssse3(short* target, | |
| 56 | short* src0, | ||
| 57 | char** permuters, | ||
| 58 | short* cntl2, | ||
| 59 | short* cntl3, | ||
| 60 | short* scalars) | ||
| 61 | { | ||
| 62 | __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10, xmm11; | ||
| 63 | __m128i *p_target, *p_src0, *p_cntl2, *p_cntl3, *p_scalars; | ||
| 64 | |||
| 65 | ✗ | p_target = (__m128i*)target; | |
| 66 | ✗ | p_src0 = (__m128i*)src0; | |
| 67 | ✗ | p_cntl2 = (__m128i*)cntl2; | |
| 68 | ✗ | p_cntl3 = (__m128i*)cntl3; | |
| 69 | ✗ | p_scalars = (__m128i*)scalars; | |
| 70 | |||
| 71 | ✗ | xmm0 = _mm_load_si128(p_scalars); | |
| 72 | |||
| 73 | ✗ | xmm1 = _mm_shufflelo_epi16(xmm0, 0); | |
| 74 | ✗ | xmm2 = _mm_shufflelo_epi16(xmm0, 0x55); | |
| 75 | ✗ | xmm3 = _mm_shufflelo_epi16(xmm0, 0xaa); | |
| 76 | ✗ | xmm4 = _mm_shufflelo_epi16(xmm0, 0xff); | |
| 77 | |||
| 78 | ✗ | xmm1 = _mm_shuffle_epi32(xmm1, 0x00); | |
| 79 | ✗ | xmm2 = _mm_shuffle_epi32(xmm2, 0x00); | |
| 80 | ✗ | xmm3 = _mm_shuffle_epi32(xmm3, 0x00); | |
| 81 | ✗ | xmm4 = _mm_shuffle_epi32(xmm4, 0x00); | |
| 82 | |||
| 83 | ✗ | xmm0 = _mm_load_si128((__m128i*)permuters[0]); | |
| 84 | ✗ | xmm6 = _mm_load_si128((__m128i*)permuters[1]); | |
| 85 | ✗ | xmm8 = _mm_load_si128((__m128i*)permuters[2]); | |
| 86 | ✗ | xmm10 = _mm_load_si128((__m128i*)permuters[3]); | |
| 87 | |||
| 88 | ✗ | xmm5 = _mm_load_si128(p_src0); | |
| 89 | ✗ | xmm0 = _mm_shuffle_epi8(xmm5, xmm0); | |
| 90 | ✗ | xmm6 = _mm_shuffle_epi8(xmm5, xmm6); | |
| 91 | ✗ | xmm8 = _mm_shuffle_epi8(xmm5, xmm8); | |
| 92 | ✗ | xmm10 = _mm_shuffle_epi8(xmm5, xmm10); | |
| 93 | |||
| 94 | ✗ | xmm5 = _mm_add_epi16(xmm1, xmm2); | |
| 95 | |||
| 96 | ✗ | xmm6 = _mm_add_epi16(xmm2, xmm6); | |
| 97 | ✗ | xmm8 = _mm_add_epi16(xmm1, xmm8); | |
| 98 | |||
| 99 | ✗ | xmm7 = _mm_load_si128(p_cntl2); | |
| 100 | ✗ | xmm9 = _mm_load_si128(p_cntl3); | |
| 101 | |||
| 102 | ✗ | xmm0 = _mm_add_epi16(xmm5, xmm0); | |
| 103 | |||
| 104 | ✗ | xmm7 = _mm_and_si128(xmm7, xmm3); | |
| 105 | ✗ | xmm9 = _mm_and_si128(xmm9, xmm4); | |
| 106 | |||
| 107 | ✗ | xmm5 = _mm_load_si128(&p_cntl2[1]); | |
| 108 | ✗ | xmm11 = _mm_load_si128(&p_cntl3[1]); | |
| 109 | |||
| 110 | ✗ | xmm7 = _mm_add_epi16(xmm7, xmm9); | |
| 111 | |||
| 112 | ✗ | xmm5 = _mm_and_si128(xmm5, xmm3); | |
| 113 | ✗ | xmm11 = _mm_and_si128(xmm11, xmm4); | |
| 114 | |||
| 115 | ✗ | xmm0 = _mm_add_epi16(xmm0, xmm7); | |
| 116 | |||
| 117 | |||
| 118 | ✗ | xmm7 = _mm_load_si128(&p_cntl2[2]); | |
| 119 | ✗ | xmm9 = _mm_load_si128(&p_cntl3[2]); | |
| 120 | |||
| 121 | ✗ | xmm5 = _mm_add_epi16(xmm5, xmm11); | |
| 122 | |||
| 123 | ✗ | xmm7 = _mm_and_si128(xmm7, xmm3); | |
| 124 | ✗ | xmm9 = _mm_and_si128(xmm9, xmm4); | |
| 125 | |||
| 126 | ✗ | xmm6 = _mm_add_epi16(xmm6, xmm5); | |
| 127 | |||
| 128 | |||
| 129 | ✗ | xmm5 = _mm_load_si128(&p_cntl2[3]); | |
| 130 | ✗ | xmm11 = _mm_load_si128(&p_cntl3[3]); | |
| 131 | |||
| 132 | ✗ | xmm7 = _mm_add_epi16(xmm7, xmm9); | |
| 133 | |||
| 134 | ✗ | xmm5 = _mm_and_si128(xmm5, xmm3); | |
| 135 | ✗ | xmm11 = _mm_and_si128(xmm11, xmm4); | |
| 136 | |||
| 137 | ✗ | xmm8 = _mm_add_epi16(xmm8, xmm7); | |
| 138 | |||
| 139 | ✗ | xmm5 = _mm_add_epi16(xmm5, xmm11); | |
| 140 | |||
| 141 | _mm_store_si128(p_target, xmm0); | ||
| 142 | ✗ | _mm_store_si128(&p_target[1], xmm6); | |
| 143 | |||
| 144 | ✗ | xmm10 = _mm_add_epi16(xmm5, xmm10); | |
| 145 | |||
| 146 | ✗ | _mm_store_si128(&p_target[2], xmm8); | |
| 147 | |||
| 148 | ✗ | _mm_store_si128(&p_target[3], xmm10); | |
| 149 | ✗ | } | |
| 150 | |||
| 151 | |||
| 152 | #endif /*LV_HAVE_SSEs*/ | ||
| 153 | |||
| 154 | #ifdef LV_HAVE_GENERIC | ||
| 155 | ✗ | static inline void volk_16i_branch_4_state_8_generic(short* target, | |
| 156 | short* src0, | ||
| 157 | char** permuters, | ||
| 158 | short* cntl2, | ||
| 159 | short* cntl3, | ||
| 160 | short* scalars) | ||
| 161 | { | ||
| 162 | ✗ | int i = 0; | |
| 163 | |||
| 164 | ✗ | int bound = 4; | |
| 165 | |||
| 166 | ✗ | for (; i < bound; ++i) { | |
| 167 | ✗ | target[i * 8] = src0[((char)permuters[i][0]) / 2] + ((i + 1) % 2 * scalars[0]) + | |
| 168 | ✗ | (((i >> 1) ^ 1) * scalars[1]) + (cntl2[i * 8] & scalars[2]) + | |
| 169 | ✗ | (cntl3[i * 8] & scalars[3]); | |
| 170 | ✗ | target[i * 8 + 1] = src0[((char)permuters[i][1 * 2]) / 2] + | |
| 171 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 172 | ✗ | (cntl2[i * 8 + 1] & scalars[2]) + | |
| 173 | ✗ | (cntl3[i * 8 + 1] & scalars[3]); | |
| 174 | ✗ | target[i * 8 + 2] = src0[((char)permuters[i][2 * 2]) / 2] + | |
| 175 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 176 | ✗ | (cntl2[i * 8 + 2] & scalars[2]) + | |
| 177 | ✗ | (cntl3[i * 8 + 2] & scalars[3]); | |
| 178 | ✗ | target[i * 8 + 3] = src0[((char)permuters[i][3 * 2]) / 2] + | |
| 179 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 180 | ✗ | (cntl2[i * 8 + 3] & scalars[2]) + | |
| 181 | ✗ | (cntl3[i * 8 + 3] & scalars[3]); | |
| 182 | ✗ | target[i * 8 + 4] = src0[((char)permuters[i][4 * 2]) / 2] + | |
| 183 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 184 | ✗ | (cntl2[i * 8 + 4] & scalars[2]) + | |
| 185 | ✗ | (cntl3[i * 8 + 4] & scalars[3]); | |
| 186 | ✗ | target[i * 8 + 5] = src0[((char)permuters[i][5 * 2]) / 2] + | |
| 187 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 188 | ✗ | (cntl2[i * 8 + 5] & scalars[2]) + | |
| 189 | ✗ | (cntl3[i * 8 + 5] & scalars[3]); | |
| 190 | ✗ | target[i * 8 + 6] = src0[((char)permuters[i][6 * 2]) / 2] + | |
| 191 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 192 | ✗ | (cntl2[i * 8 + 6] & scalars[2]) + | |
| 193 | ✗ | (cntl3[i * 8 + 6] & scalars[3]); | |
| 194 | ✗ | target[i * 8 + 7] = src0[((char)permuters[i][7 * 2]) / 2] + | |
| 195 | ✗ | ((i + 1) % 2 * scalars[0]) + (((i >> 1) ^ 1) * scalars[1]) + | |
| 196 | ✗ | (cntl2[i * 8 + 7] & scalars[2]) + | |
| 197 | ✗ | (cntl3[i * 8 + 7] & scalars[3]); | |
| 198 | } | ||
| 199 | ✗ | } | |
| 200 | |||
| 201 | #endif /*LV_HAVE_GENERIC*/ | ||
| 202 | |||
| 203 | |||
| 204 | #endif /*INCLUDED_volk_16i_branch_4_state_8_a_H*/ | ||
| 205 |