cfl_simd.h (14837B)
1 /* 2 * Copyright (c) 2017, Alliance for Open Media. All rights reserved. 3 * 4 * This source code is subject to the terms of the BSD 2 Clause License and 5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 * was not distributed with this source code in the LICENSE file, you can 7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 * Media Patent License 1.0 was not distributed with this source code in the 9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 */ 11 12 #ifndef AOM_AV1_COMMON_X86_CFL_SIMD_H_ 13 #define AOM_AV1_COMMON_X86_CFL_SIMD_H_ 14 15 #include "av1/common/blockd.h" 16 17 // SSSE3 version is optimal for with == 4, we reuse them in AVX2 18 void cfl_subsample_lbd_420_4x4_ssse3(const uint8_t *cfl_type, int input_stride, 19 uint16_t *output_q3); 20 void cfl_subsample_lbd_420_4x8_ssse3(const uint8_t *cfl_type, int input_stride, 21 uint16_t *output_q3); 22 void cfl_subsample_lbd_420_4x16_ssse3(const uint8_t *cfl_type, int input_stride, 23 uint16_t *output_q3); 24 25 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 26 void cfl_subsample_lbd_420_8x4_ssse3(const uint8_t *cfl_type, int input_stride, 27 uint16_t *output_q3); 28 void cfl_subsample_lbd_420_8x8_ssse3(const uint8_t *cfl_type, int input_stride, 29 uint16_t *output_q3); 30 void cfl_subsample_lbd_420_8x16_ssse3(const uint8_t *cfl_type, int input_stride, 31 uint16_t *output_q3); 32 void cfl_subsample_lbd_420_8x32_ssse3(const uint8_t *cfl_type, int input_stride, 33 uint16_t *output_q3); 34 35 // SSSE3 version is optimal for with == 16, we reuse it in AVX2 36 void cfl_subsample_lbd_420_16x4_ssse3(const uint8_t *cfl_type, int input_stride, 37 uint16_t *output_q3); 38 void cfl_subsample_lbd_420_16x8_ssse3(const uint8_t *cfl_type, int input_stride, 39 uint16_t *output_q3); 40 void cfl_subsample_lbd_420_16x16_ssse3(const uint8_t *cfl_type, 41 int input_stride, uint16_t *output_q3); 42 void cfl_subsample_lbd_420_16x32_ssse3(const uint8_t *cfl_type, 43 int input_stride, uint16_t *output_q3); 44 45 // SSSE3 version is optimal for with == 4, we reuse them in AVX2 46 void cfl_subsample_lbd_422_4x4_ssse3(const uint8_t *cfl_type, int input_stride, 47 uint16_t *output_q3); 48 void cfl_subsample_lbd_422_4x8_ssse3(const uint8_t *cfl_type, int input_stride, 49 uint16_t *output_q3); 50 void cfl_subsample_lbd_422_4x16_ssse3(const uint8_t *cfl_type, int input_stride, 51 uint16_t *output_q3); 52 53 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 54 void cfl_subsample_lbd_422_8x4_ssse3(const uint8_t *cfl_type, int input_stride, 55 uint16_t *output_q3); 56 void cfl_subsample_lbd_422_8x8_ssse3(const uint8_t *cfl_type, int input_stride, 57 uint16_t *output_q3); 58 void cfl_subsample_lbd_422_8x16_ssse3(const uint8_t *cfl_type, int input_stride, 59 uint16_t *output_q3); 60 void cfl_subsample_lbd_422_8x32_ssse3(const uint8_t *cfl_type, int input_stride, 61 uint16_t *output_q3); 62 63 // SSSE3 version is optimal for with == 16, we reuse it in AVX2 64 void cfl_subsample_lbd_422_16x4_ssse3(const uint8_t *cfl_type, int input_stride, 65 uint16_t *output_q3); 66 void cfl_subsample_lbd_422_16x8_ssse3(const uint8_t *cfl_type, int input_stride, 67 uint16_t *output_q3); 68 void cfl_subsample_lbd_422_16x16_ssse3(const uint8_t *cfl_type, 69 int input_stride, uint16_t *output_q3); 70 void cfl_subsample_lbd_422_16x32_ssse3(const uint8_t *cfl_type, 71 int input_stride, uint16_t *output_q3); 72 73 // SSSE3 version is optimal for with == 4, we reuse them in AVX2 74 void cfl_subsample_lbd_444_4x4_ssse3(const uint8_t *cfl_type, int input_stride, 75 uint16_t *output_q3); 76 void cfl_subsample_lbd_444_4x8_ssse3(const uint8_t *cfl_type, int input_stride, 77 uint16_t *output_q3); 78 void cfl_subsample_lbd_444_4x16_ssse3(const uint8_t *cfl_type, int input_stride, 79 uint16_t *output_q3); 80 81 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 82 void cfl_subsample_lbd_444_8x4_ssse3(const uint8_t *cfl_type, int input_stride, 83 uint16_t *output_q3); 84 void cfl_subsample_lbd_444_8x8_ssse3(const uint8_t *cfl_type, int input_stride, 85 uint16_t *output_q3); 86 void cfl_subsample_lbd_444_8x16_ssse3(const uint8_t *cfl_type, int input_stride, 87 uint16_t *output_q3); 88 void cfl_subsample_lbd_444_8x32_ssse3(const uint8_t *cfl_type, int input_stride, 89 uint16_t *output_q3); 90 91 // SSSE3 version is optimal for with == 16, we reuse it in AVX2 92 void cfl_subsample_lbd_444_16x4_ssse3(const uint8_t *cfl_type, int input_stride, 93 uint16_t *output_q3); 94 void cfl_subsample_lbd_444_16x8_ssse3(const uint8_t *cfl_type, int input_stride, 95 uint16_t *output_q3); 96 void cfl_subsample_lbd_444_16x16_ssse3(const uint8_t *cfl_type, 97 int input_stride, uint16_t *output_q3); 98 void cfl_subsample_lbd_444_16x32_ssse3(const uint8_t *cfl_type, 99 int input_stride, uint16_t *output_q3); 100 101 #if CONFIG_AV1_HIGHBITDEPTH 102 void cfl_subsample_hbd_420_4x4_ssse3(const uint16_t *cfl_type, int input_stride, 103 uint16_t *output_q3); 104 void cfl_subsample_hbd_420_4x8_ssse3(const uint16_t *cfl_type, int input_stride, 105 uint16_t *output_q3); 106 void cfl_subsample_hbd_420_4x16_ssse3(const uint16_t *cfl_type, 107 int input_stride, uint16_t *output_q3); 108 109 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 110 void cfl_subsample_hbd_420_8x4_ssse3(const uint16_t *cfl_type, int input_stride, 111 uint16_t *output_q3); 112 void cfl_subsample_hbd_420_8x8_ssse3(const uint16_t *cfl_type, int input_stride, 113 uint16_t *output_q3); 114 void cfl_subsample_hbd_420_8x16_ssse3(const uint16_t *cfl_type, 115 int input_stride, uint16_t *output_q3); 116 void cfl_subsample_hbd_420_8x32_ssse3(const uint16_t *cfl_type, 117 int input_stride, uint16_t *output_q3); 118 119 // SSSE3 version is faster for with == 16, we reuse it in AVX2 120 void cfl_subsample_hbd_420_16x4_ssse3(const uint16_t *cfl_type, 121 int input_stride, uint16_t *output_q3); 122 void cfl_subsample_hbd_420_16x8_ssse3(const uint16_t *cfl_type, 123 int input_stride, uint16_t *output_q3); 124 void cfl_subsample_hbd_420_16x16_ssse3(const uint16_t *cfl_type, 125 int input_stride, uint16_t *output_q3); 126 void cfl_subsample_hbd_420_16x32_ssse3(const uint16_t *cfl_type, 127 int input_stride, uint16_t *output_q3); 128 129 void cfl_subsample_hbd_422_4x4_ssse3(const uint16_t *cfl_type, int input_stride, 130 uint16_t *output_q3); 131 void cfl_subsample_hbd_422_4x8_ssse3(const uint16_t *cfl_type, int input_stride, 132 uint16_t *output_q3); 133 void cfl_subsample_hbd_422_4x16_ssse3(const uint16_t *cfl_type, 134 int input_stride, uint16_t *output_q3); 135 136 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 137 void cfl_subsample_hbd_422_8x4_ssse3(const uint16_t *cfl_type, int input_stride, 138 uint16_t *output_q3); 139 void cfl_subsample_hbd_422_8x8_ssse3(const uint16_t *cfl_type, int input_stride, 140 uint16_t *output_q3); 141 void cfl_subsample_hbd_422_8x16_ssse3(const uint16_t *cfl_type, 142 int input_stride, uint16_t *output_q3); 143 void cfl_subsample_hbd_422_8x32_ssse3(const uint16_t *cfl_type, 144 int input_stride, uint16_t *output_q3); 145 146 // SSSE3 version is faster for with == 16, we reuse it in AVX2 147 void cfl_subsample_hbd_422_16x4_ssse3(const uint16_t *cfl_type, 148 int input_stride, uint16_t *output_q3); 149 void cfl_subsample_hbd_422_16x8_ssse3(const uint16_t *cfl_type, 150 int input_stride, uint16_t *output_q3); 151 void cfl_subsample_hbd_422_16x16_ssse3(const uint16_t *cfl_type, 152 int input_stride, uint16_t *output_q3); 153 void cfl_subsample_hbd_422_16x32_ssse3(const uint16_t *cfl_type, 154 int input_stride, uint16_t *output_q3); 155 156 void cfl_subsample_hbd_444_4x4_ssse3(const uint16_t *cfl_type, int input_stride, 157 uint16_t *output_q3); 158 void cfl_subsample_hbd_444_4x8_ssse3(const uint16_t *cfl_type, int input_stride, 159 uint16_t *output_q3); 160 void cfl_subsample_hbd_444_4x16_ssse3(const uint16_t *cfl_type, 161 int input_stride, uint16_t *output_q3); 162 163 // SSSE3 version is optimal for with == 8, we reuse it in AVX2 164 void cfl_subsample_hbd_444_8x4_ssse3(const uint16_t *cfl_type, int input_stride, 165 uint16_t *output_q3); 166 void cfl_subsample_hbd_444_8x8_ssse3(const uint16_t *cfl_type, int input_stride, 167 uint16_t *output_q3); 168 void cfl_subsample_hbd_444_8x16_ssse3(const uint16_t *cfl_type, 169 int input_stride, uint16_t *output_q3); 170 void cfl_subsample_hbd_444_8x32_ssse3(const uint16_t *cfl_type, 171 int input_stride, uint16_t *output_q3); 172 173 // SSSE3 version is faster for with == 16, we reuse it in AVX2 174 void cfl_subsample_hbd_444_16x4_ssse3(const uint16_t *cfl_type, 175 int input_stride, uint16_t *output_q3); 176 void cfl_subsample_hbd_444_16x8_ssse3(const uint16_t *cfl_type, 177 int input_stride, uint16_t *output_q3); 178 void cfl_subsample_hbd_444_16x16_ssse3(const uint16_t *cfl_type, 179 int input_stride, uint16_t *output_q3); 180 void cfl_subsample_hbd_444_16x32_ssse3(const uint16_t *cfl_type, 181 int input_stride, uint16_t *output_q3); 182 #endif // CONFIG_AV1_HIGHBITDEPTH 183 184 // SSE2 version is optimal for with == 4, we reuse them in AVX2 185 void cfl_subtract_average_4x4_sse2(const uint16_t *src, int16_t *dst); 186 void cfl_subtract_average_4x8_sse2(const uint16_t *src, int16_t *dst); 187 void cfl_subtract_average_4x16_sse2(const uint16_t *src, int16_t *dst); 188 189 // SSE2 version is optimal for with == 8, we reuse them in AVX2 190 void cfl_subtract_average_8x4_sse2(const uint16_t *src, int16_t *dst); 191 void cfl_subtract_average_8x8_sse2(const uint16_t *src, int16_t *dst); 192 void cfl_subtract_average_8x16_sse2(const uint16_t *src, int16_t *dst); 193 void cfl_subtract_average_8x32_sse2(const uint16_t *src, int16_t *dst); 194 195 void cfl_predict_lbd_4x4_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 196 int dst_stride, int alpha_q3); 197 void cfl_predict_lbd_4x8_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 198 int dst_stride, int alpha_q3); 199 void cfl_predict_lbd_4x16_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 200 int dst_stride, int alpha_q3); 201 202 void cfl_predict_lbd_8x4_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 203 int dst_stride, int alpha_q3); 204 void cfl_predict_lbd_8x8_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 205 int dst_stride, int alpha_q3); 206 void cfl_predict_lbd_8x16_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 207 int dst_stride, int alpha_q3); 208 void cfl_predict_lbd_8x32_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 209 int dst_stride, int alpha_q3); 210 211 void cfl_predict_lbd_16x4_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 212 int dst_stride, int alpha_q3); 213 void cfl_predict_lbd_16x8_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 214 int dst_stride, int alpha_q3); 215 void cfl_predict_lbd_16x16_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 216 int dst_stride, int alpha_q3); 217 void cfl_predict_lbd_16x32_ssse3(const int16_t *pred_buf_q3, uint8_t *dst, 218 int dst_stride, int alpha_q3); 219 220 #if CONFIG_AV1_HIGHBITDEPTH 221 void cfl_predict_hbd_4x4_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 222 int dst_stride, int alpha_q3, int bd); 223 void cfl_predict_hbd_4x8_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 224 int dst_stride, int alpha_q3, int bd); 225 void cfl_predict_hbd_4x16_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 226 int dst_stride, int alpha_q3, int bd); 227 228 void cfl_predict_hbd_8x4_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 229 int dst_stride, int alpha_q3, int bd); 230 void cfl_predict_hbd_8x8_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 231 int dst_stride, int alpha_q3, int bd); 232 void cfl_predict_hbd_8x16_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 233 int dst_stride, int alpha_q3, int bd); 234 void cfl_predict_hbd_8x32_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 235 int dst_stride, int alpha_q3, int bd); 236 237 void cfl_predict_hbd_16x4_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 238 int dst_stride, int alpha_q3, int bd); 239 void cfl_predict_hbd_16x8_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 240 int dst_stride, int alpha_q3, int bd); 241 void cfl_predict_hbd_16x16_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 242 int dst_stride, int alpha_q3, int bd); 243 void cfl_predict_hbd_16x32_ssse3(const int16_t *pred_buf_q3, uint16_t *dst, 244 int dst_stride, int alpha_q3, int bd); 245 #endif // CONFIG_AV1_HIGHBITDEPTH 246 #endif // AOM_AV1_COMMON_X86_CFL_SIMD_H_