1 /* 2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include <arm_neon.h> 12 #include <assert.h> 13 14 #include "./vp9_rtcd.h" 15 #include "./vpx_config.h" 16 #include "vp9/common/vp9_common.h" 17 #include "vpx_dsp/arm/transpose_neon.h" 18 19 static int16_t cospi_2_64 = 16305; 20 static int16_t cospi_4_64 = 16069; 21 static int16_t cospi_6_64 = 15679; 22 static int16_t cospi_8_64 = 15137; 23 static int16_t cospi_10_64 = 14449; 24 static int16_t cospi_12_64 = 13623; 25 static int16_t cospi_14_64 = 12665; 26 static int16_t cospi_16_64 = 11585; 27 static int16_t cospi_18_64 = 10394; 28 static int16_t cospi_20_64 = 9102; 29 static int16_t cospi_22_64 = 7723; 30 static int16_t cospi_24_64 = 6270; 31 static int16_t cospi_26_64 = 4756; 32 static int16_t cospi_28_64 = 3196; 33 static int16_t cospi_30_64 = 1606; 34 35 static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16, 36 int16x8_t *q10s16, int16x8_t *q11s16, 37 int16x8_t *q12s16, int16x8_t *q13s16, 38 int16x8_t *q14s16, int16x8_t *q15s16) { 39 int16x4_t d0s16, d1s16, d2s16, d3s16; 40 int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; 41 int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; 42 int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; 43 int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; 44 int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32; 45 int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32; 46 47 d0s16 = vdup_n_s16(cospi_28_64); 48 d1s16 = vdup_n_s16(cospi_4_64); 49 d2s16 = vdup_n_s16(cospi_12_64); 50 d3s16 = vdup_n_s16(cospi_20_64); 51 52 d16s16 = vget_low_s16(*q8s16); 53 d17s16 = vget_high_s16(*q8s16); 54 d18s16 = vget_low_s16(*q9s16); 55 d19s16 = vget_high_s16(*q9s16); 56 d20s16 = vget_low_s16(*q10s16); 57 d21s16 = vget_high_s16(*q10s16); 58 d22s16 = vget_low_s16(*q11s16); 59 d23s16 = vget_high_s16(*q11s16); 60 d24s16 = vget_low_s16(*q12s16); 61 d25s16 = vget_high_s16(*q12s16); 62 d26s16 = vget_low_s16(*q13s16); 63 d27s16 = vget_high_s16(*q13s16); 64 d28s16 = vget_low_s16(*q14s16); 65 d29s16 = vget_high_s16(*q14s16); 66 d30s16 = vget_low_s16(*q15s16); 67 d31s16 = vget_high_s16(*q15s16); 68 69 q2s32 = vmull_s16(d18s16, d0s16); 70 q3s32 = vmull_s16(d19s16, d0s16); 71 q5s32 = vmull_s16(d26s16, d2s16); 72 q6s32 = vmull_s16(d27s16, d2s16); 73 74 q2s32 = vmlsl_s16(q2s32, d30s16, d1s16); 75 q3s32 = vmlsl_s16(q3s32, d31s16, d1s16); 76 q5s32 = vmlsl_s16(q5s32, d22s16, d3s16); 77 q6s32 = vmlsl_s16(q6s32, d23s16, d3s16); 78 79 d8s16 = vrshrn_n_s32(q2s32, 14); 80 d9s16 = vrshrn_n_s32(q3s32, 14); 81 d10s16 = vrshrn_n_s32(q5s32, 14); 82 d11s16 = vrshrn_n_s32(q6s32, 14); 83 q4s16 = vcombine_s16(d8s16, d9s16); 84 q5s16 = vcombine_s16(d10s16, d11s16); 85 86 q2s32 = vmull_s16(d18s16, d1s16); 87 q3s32 = vmull_s16(d19s16, d1s16); 88 q9s32 = vmull_s16(d26s16, d3s16); 89 q13s32 = vmull_s16(d27s16, d3s16); 90 91 q2s32 = vmlal_s16(q2s32, d30s16, d0s16); 92 q3s32 = vmlal_s16(q3s32, d31s16, d0s16); 93 q9s32 = vmlal_s16(q9s32, d22s16, d2s16); 94 q13s32 = vmlal_s16(q13s32, d23s16, d2s16); 95 96 d14s16 = vrshrn_n_s32(q2s32, 14); 97 d15s16 = vrshrn_n_s32(q3s32, 14); 98 d12s16 = vrshrn_n_s32(q9s32, 14); 99 d13s16 = vrshrn_n_s32(q13s32, 14); 100 q6s16 = vcombine_s16(d12s16, d13s16); 101 q7s16 = vcombine_s16(d14s16, d15s16); 102 103 d0s16 = vdup_n_s16(cospi_16_64); 104 105 q2s32 = vmull_s16(d16s16, d0s16); 106 q3s32 = vmull_s16(d17s16, d0s16); 107 q13s32 = vmull_s16(d16s16, d0s16); 108 q15s32 = vmull_s16(d17s16, d0s16); 109 110 q2s32 = vmlal_s16(q2s32, d24s16, d0s16); 111 q3s32 = vmlal_s16(q3s32, d25s16, d0s16); 112 q13s32 = vmlsl_s16(q13s32, d24s16, d0s16); 113 q15s32 = vmlsl_s16(q15s32, d25s16, d0s16); 114 115 d0s16 = vdup_n_s16(cospi_24_64); 116 d1s16 = vdup_n_s16(cospi_8_64); 117 118 d18s16 = vrshrn_n_s32(q2s32, 14); 119 d19s16 = vrshrn_n_s32(q3s32, 14); 120 d22s16 = vrshrn_n_s32(q13s32, 14); 121 d23s16 = vrshrn_n_s32(q15s32, 14); 122 *q9s16 = vcombine_s16(d18s16, d19s16); 123 *q11s16 = vcombine_s16(d22s16, d23s16); 124 125 q2s32 = vmull_s16(d20s16, d0s16); 126 q3s32 = vmull_s16(d21s16, d0s16); 127 q8s32 = vmull_s16(d20s16, d1s16); 128 q12s32 = vmull_s16(d21s16, d1s16); 129 130 q2s32 = vmlsl_s16(q2s32, d28s16, d1s16); 131 q3s32 = vmlsl_s16(q3s32, d29s16, d1s16); 132 q8s32 = vmlal_s16(q8s32, d28s16, d0s16); 133 q12s32 = vmlal_s16(q12s32, d29s16, d0s16); 134 135 d26s16 = vrshrn_n_s32(q2s32, 14); 136 d27s16 = vrshrn_n_s32(q3s32, 14); 137 d30s16 = vrshrn_n_s32(q8s32, 14); 138 d31s16 = vrshrn_n_s32(q12s32, 14); 139 *q13s16 = vcombine_s16(d26s16, d27s16); 140 *q15s16 = vcombine_s16(d30s16, d31s16); 141 142 q0s16 = vaddq_s16(*q9s16, *q15s16); 143 q1s16 = vaddq_s16(*q11s16, *q13s16); 144 q2s16 = vsubq_s16(*q11s16, *q13s16); 145 q3s16 = vsubq_s16(*q9s16, *q15s16); 146 147 *q13s16 = vsubq_s16(q4s16, q5s16); 148 q4s16 = vaddq_s16(q4s16, q5s16); 149 *q14s16 = vsubq_s16(q7s16, q6s16); 150 q7s16 = vaddq_s16(q7s16, q6s16); 151 d26s16 = vget_low_s16(*q13s16); 152 d27s16 = vget_high_s16(*q13s16); 153 d28s16 = vget_low_s16(*q14s16); 154 d29s16 = vget_high_s16(*q14s16); 155 156 d16s16 = vdup_n_s16(cospi_16_64); 157 158 q9s32 = vmull_s16(d28s16, d16s16); 159 q10s32 = vmull_s16(d29s16, d16s16); 160 q11s32 = vmull_s16(d28s16, d16s16); 161 q12s32 = vmull_s16(d29s16, d16s16); 162 163 q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); 164 q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); 165 q11s32 = vmlal_s16(q11s32, d26s16, d16s16); 166 q12s32 = vmlal_s16(q12s32, d27s16, d16s16); 167 168 d10s16 = vrshrn_n_s32(q9s32, 14); 169 d11s16 = vrshrn_n_s32(q10s32, 14); 170 d12s16 = vrshrn_n_s32(q11s32, 14); 171 d13s16 = vrshrn_n_s32(q12s32, 14); 172 q5s16 = vcombine_s16(d10s16, d11s16); 173 q6s16 = vcombine_s16(d12s16, d13s16); 174 175 *q8s16 = vaddq_s16(q0s16, q7s16); 176 *q9s16 = vaddq_s16(q1s16, q6s16); 177 *q10s16 = vaddq_s16(q2s16, q5s16); 178 *q11s16 = vaddq_s16(q3s16, q4s16); 179 *q12s16 = vsubq_s16(q3s16, q4s16); 180 *q13s16 = vsubq_s16(q2s16, q5s16); 181 *q14s16 = vsubq_s16(q1s16, q6s16); 182 *q15s16 = vsubq_s16(q0s16, q7s16); 183 } 184 185 static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16, 186 int16x8_t *q10s16, int16x8_t *q11s16, 187 int16x8_t *q12s16, int16x8_t *q13s16, 188 int16x8_t *q14s16, int16x8_t *q15s16) { 189 int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16; 190 int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; 191 int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; 192 int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; 193 int16x8_t q2s16, q4s16, q5s16, q6s16; 194 int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q7s32, q8s32; 195 int32x4_t q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32; 196 197 d16s16 = vget_low_s16(*q8s16); 198 d17s16 = vget_high_s16(*q8s16); 199 d18s16 = vget_low_s16(*q9s16); 200 d19s16 = vget_high_s16(*q9s16); 201 d20s16 = vget_low_s16(*q10s16); 202 d21s16 = vget_high_s16(*q10s16); 203 d22s16 = vget_low_s16(*q11s16); 204 d23s16 = vget_high_s16(*q11s16); 205 d24s16 = vget_low_s16(*q12s16); 206 d25s16 = vget_high_s16(*q12s16); 207 d26s16 = vget_low_s16(*q13s16); 208 d27s16 = vget_high_s16(*q13s16); 209 d28s16 = vget_low_s16(*q14s16); 210 d29s16 = vget_high_s16(*q14s16); 211 d30s16 = vget_low_s16(*q15s16); 212 d31s16 = vget_high_s16(*q15s16); 213 214 d14s16 = vdup_n_s16(cospi_2_64); 215 d15s16 = vdup_n_s16(cospi_30_64); 216 217 q1s32 = vmull_s16(d30s16, d14s16); 218 q2s32 = vmull_s16(d31s16, d14s16); 219 q3s32 = vmull_s16(d30s16, d15s16); 220 q4s32 = vmull_s16(d31s16, d15s16); 221 222 d30s16 = vdup_n_s16(cospi_18_64); 223 d31s16 = vdup_n_s16(cospi_14_64); 224 225 q1s32 = vmlal_s16(q1s32, d16s16, d15s16); 226 q2s32 = vmlal_s16(q2s32, d17s16, d15s16); 227 q3s32 = vmlsl_s16(q3s32, d16s16, d14s16); 228 q4s32 = vmlsl_s16(q4s32, d17s16, d14s16); 229 230 q5s32 = vmull_s16(d22s16, d30s16); 231 q6s32 = vmull_s16(d23s16, d30s16); 232 q7s32 = vmull_s16(d22s16, d31s16); 233 q8s32 = vmull_s16(d23s16, d31s16); 234 235 q5s32 = vmlal_s16(q5s32, d24s16, d31s16); 236 q6s32 = vmlal_s16(q6s32, d25s16, d31s16); 237 q7s32 = vmlsl_s16(q7s32, d24s16, d30s16); 238 q8s32 = vmlsl_s16(q8s32, d25s16, d30s16); 239 240 q11s32 = vaddq_s32(q1s32, q5s32); 241 q12s32 = vaddq_s32(q2s32, q6s32); 242 q1s32 = vsubq_s32(q1s32, q5s32); 243 q2s32 = vsubq_s32(q2s32, q6s32); 244 245 d22s16 = vrshrn_n_s32(q11s32, 14); 246 d23s16 = vrshrn_n_s32(q12s32, 14); 247 *q11s16 = vcombine_s16(d22s16, d23s16); 248 249 q12s32 = vaddq_s32(q3s32, q7s32); 250 q15s32 = vaddq_s32(q4s32, q8s32); 251 q3s32 = vsubq_s32(q3s32, q7s32); 252 q4s32 = vsubq_s32(q4s32, q8s32); 253 254 d2s16 = vrshrn_n_s32(q1s32, 14); 255 d3s16 = vrshrn_n_s32(q2s32, 14); 256 d24s16 = vrshrn_n_s32(q12s32, 14); 257 d25s16 = vrshrn_n_s32(q15s32, 14); 258 d6s16 = vrshrn_n_s32(q3s32, 14); 259 d7s16 = vrshrn_n_s32(q4s32, 14); 260 *q12s16 = vcombine_s16(d24s16, d25s16); 261 262 d0s16 = vdup_n_s16(cospi_10_64); 263 d1s16 = vdup_n_s16(cospi_22_64); 264 q4s32 = vmull_s16(d26s16, d0s16); 265 q5s32 = vmull_s16(d27s16, d0s16); 266 q2s32 = vmull_s16(d26s16, d1s16); 267 q6s32 = vmull_s16(d27s16, d1s16); 268 269 d30s16 = vdup_n_s16(cospi_26_64); 270 d31s16 = vdup_n_s16(cospi_6_64); 271 272 q4s32 = vmlal_s16(q4s32, d20s16, d1s16); 273 q5s32 = vmlal_s16(q5s32, d21s16, d1s16); 274 q2s32 = vmlsl_s16(q2s32, d20s16, d0s16); 275 q6s32 = vmlsl_s16(q6s32, d21s16, d0s16); 276 277 q0s32 = vmull_s16(d18s16, d30s16); 278 q13s32 = vmull_s16(d19s16, d30s16); 279 280 q0s32 = vmlal_s16(q0s32, d28s16, d31s16); 281 q13s32 = vmlal_s16(q13s32, d29s16, d31s16); 282 283 q10s32 = vmull_s16(d18s16, d31s16); 284 q9s32 = vmull_s16(d19s16, d31s16); 285 286 q10s32 = vmlsl_s16(q10s32, d28s16, d30s16); 287 q9s32 = vmlsl_s16(q9s32, d29s16, d30s16); 288 289 q14s32 = vaddq_s32(q2s32, q10s32); 290 q15s32 = vaddq_s32(q6s32, q9s32); 291 q2s32 = vsubq_s32(q2s32, q10s32); 292 q6s32 = vsubq_s32(q6s32, q9s32); 293 294 d28s16 = vrshrn_n_s32(q14s32, 14); 295 d29s16 = vrshrn_n_s32(q15s32, 14); 296 d4s16 = vrshrn_n_s32(q2s32, 14); 297 d5s16 = vrshrn_n_s32(q6s32, 14); 298 *q14s16 = vcombine_s16(d28s16, d29s16); 299 300 q9s32 = vaddq_s32(q4s32, q0s32); 301 q10s32 = vaddq_s32(q5s32, q13s32); 302 q4s32 = vsubq_s32(q4s32, q0s32); 303 q5s32 = vsubq_s32(q5s32, q13s32); 304 305 d30s16 = vdup_n_s16(cospi_8_64); 306 d31s16 = vdup_n_s16(cospi_24_64); 307 308 d18s16 = vrshrn_n_s32(q9s32, 14); 309 d19s16 = vrshrn_n_s32(q10s32, 14); 310 d8s16 = vrshrn_n_s32(q4s32, 14); 311 d9s16 = vrshrn_n_s32(q5s32, 14); 312 *q9s16 = vcombine_s16(d18s16, d19s16); 313 314 q5s32 = vmull_s16(d2s16, d30s16); 315 q6s32 = vmull_s16(d3s16, d30s16); 316 q7s32 = vmull_s16(d2s16, d31s16); 317 q0s32 = vmull_s16(d3s16, d31s16); 318 319 q5s32 = vmlal_s16(q5s32, d6s16, d31s16); 320 q6s32 = vmlal_s16(q6s32, d7s16, d31s16); 321 q7s32 = vmlsl_s16(q7s32, d6s16, d30s16); 322 q0s32 = vmlsl_s16(q0s32, d7s16, d30s16); 323 324 q1s32 = vmull_s16(d4s16, d30s16); 325 q3s32 = vmull_s16(d5s16, d30s16); 326 q10s32 = vmull_s16(d4s16, d31s16); 327 q2s32 = vmull_s16(d5s16, d31s16); 328 329 q1s32 = vmlsl_s16(q1s32, d8s16, d31s16); 330 q3s32 = vmlsl_s16(q3s32, d9s16, d31s16); 331 q10s32 = vmlal_s16(q10s32, d8s16, d30s16); 332 q2s32 = vmlal_s16(q2s32, d9s16, d30s16); 333 334 *q8s16 = vaddq_s16(*q11s16, *q9s16); 335 *q11s16 = vsubq_s16(*q11s16, *q9s16); 336 q4s16 = vaddq_s16(*q12s16, *q14s16); 337 *q12s16 = vsubq_s16(*q12s16, *q14s16); 338 339 q14s32 = vaddq_s32(q5s32, q1s32); 340 q15s32 = vaddq_s32(q6s32, q3s32); 341 q5s32 = vsubq_s32(q5s32, q1s32); 342 q6s32 = vsubq_s32(q6s32, q3s32); 343 344 d18s16 = vrshrn_n_s32(q14s32, 14); 345 d19s16 = vrshrn_n_s32(q15s32, 14); 346 d10s16 = vrshrn_n_s32(q5s32, 14); 347 d11s16 = vrshrn_n_s32(q6s32, 14); 348 *q9s16 = vcombine_s16(d18s16, d19s16); 349 350 q1s32 = vaddq_s32(q7s32, q10s32); 351 q3s32 = vaddq_s32(q0s32, q2s32); 352 q7s32 = vsubq_s32(q7s32, q10s32); 353 q0s32 = vsubq_s32(q0s32, q2s32); 354 355 d28s16 = vrshrn_n_s32(q1s32, 14); 356 d29s16 = vrshrn_n_s32(q3s32, 14); 357 d14s16 = vrshrn_n_s32(q7s32, 14); 358 d15s16 = vrshrn_n_s32(q0s32, 14); 359 *q14s16 = vcombine_s16(d28s16, d29s16); 360 361 d30s16 = vdup_n_s16(cospi_16_64); 362 363 d22s16 = vget_low_s16(*q11s16); 364 d23s16 = vget_high_s16(*q11s16); 365 q2s32 = vmull_s16(d22s16, d30s16); 366 q3s32 = vmull_s16(d23s16, d30s16); 367 q13s32 = vmull_s16(d22s16, d30s16); 368 q1s32 = vmull_s16(d23s16, d30s16); 369 370 d24s16 = vget_low_s16(*q12s16); 371 d25s16 = vget_high_s16(*q12s16); 372 q2s32 = vmlal_s16(q2s32, d24s16, d30s16); 373 q3s32 = vmlal_s16(q3s32, d25s16, d30s16); 374 q13s32 = vmlsl_s16(q13s32, d24s16, d30s16); 375 q1s32 = vmlsl_s16(q1s32, d25s16, d30s16); 376 377 d4s16 = vrshrn_n_s32(q2s32, 14); 378 d5s16 = vrshrn_n_s32(q3s32, 14); 379 d24s16 = vrshrn_n_s32(q13s32, 14); 380 d25s16 = vrshrn_n_s32(q1s32, 14); 381 q2s16 = vcombine_s16(d4s16, d5s16); 382 *q12s16 = vcombine_s16(d24s16, d25s16); 383 384 q13s32 = vmull_s16(d10s16, d30s16); 385 q1s32 = vmull_s16(d11s16, d30s16); 386 q11s32 = vmull_s16(d10s16, d30s16); 387 q0s32 = vmull_s16(d11s16, d30s16); 388 389 q13s32 = vmlal_s16(q13s32, d14s16, d30s16); 390 q1s32 = vmlal_s16(q1s32, d15s16, d30s16); 391 q11s32 = vmlsl_s16(q11s32, d14s16, d30s16); 392 q0s32 = vmlsl_s16(q0s32, d15s16, d30s16); 393 394 d20s16 = vrshrn_n_s32(q13s32, 14); 395 d21s16 = vrshrn_n_s32(q1s32, 14); 396 d12s16 = vrshrn_n_s32(q11s32, 14); 397 d13s16 = vrshrn_n_s32(q0s32, 14); 398 *q10s16 = vcombine_s16(d20s16, d21s16); 399 q6s16 = vcombine_s16(d12s16, d13s16); 400 401 q5s16 = vdupq_n_s16(0); 402 403 *q9s16 = vsubq_s16(q5s16, *q9s16); 404 *q11s16 = vsubq_s16(q5s16, q2s16); 405 *q13s16 = vsubq_s16(q5s16, q6s16); 406 *q15s16 = vsubq_s16(q5s16, q4s16); 407 } 408 409 void vp9_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest, int stride, 410 int tx_type) { 411 int i; 412 uint8_t *d1, *d2; 413 uint8x8_t d0u8, d1u8, d2u8, d3u8; 414 uint64x1_t d0u64, d1u64, d2u64, d3u64; 415 int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; 416 uint16x8_t q8u16, q9u16, q10u16, q11u16; 417 418 q8s16 = vld1q_s16(input); 419 q9s16 = vld1q_s16(input + 8); 420 q10s16 = vld1q_s16(input + 8 * 2); 421 q11s16 = vld1q_s16(input + 8 * 3); 422 q12s16 = vld1q_s16(input + 8 * 4); 423 q13s16 = vld1q_s16(input + 8 * 5); 424 q14s16 = vld1q_s16(input + 8 * 6); 425 q15s16 = vld1q_s16(input + 8 * 7); 426 427 transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 428 &q15s16); 429 430 switch (tx_type) { 431 case 0: // idct_idct is not supported. Fall back to C 432 vp9_iht8x8_64_add_c(input, dest, stride, tx_type); 433 return; 434 case 1: // iadst_idct 435 // generate IDCT constants 436 // GENERATE_IDCT_CONSTANTS 437 438 // first transform rows 439 IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 440 &q15s16); 441 442 // transpose the matrix 443 transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, 444 &q14s16, &q15s16); 445 446 // generate IADST constants 447 // GENERATE_IADST_CONSTANTS 448 449 // then transform columns 450 IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 451 &q15s16); 452 break; 453 case 2: // idct_iadst 454 // generate IADST constants 455 // GENERATE_IADST_CONSTANTS 456 457 // first transform rows 458 IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 459 &q15s16); 460 461 // transpose the matrix 462 transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, 463 &q14s16, &q15s16); 464 465 // generate IDCT constants 466 // GENERATE_IDCT_CONSTANTS 467 468 // then transform columns 469 IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 470 &q15s16); 471 break; 472 case 3: // iadst_iadst 473 // generate IADST constants 474 // GENERATE_IADST_CONSTANTS 475 476 // first transform rows 477 IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 478 &q15s16); 479 480 // transpose the matrix 481 transpose_s16_8x8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, 482 &q14s16, &q15s16); 483 484 // then transform columns 485 IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16, 486 &q15s16); 487 break; 488 default: // iadst_idct 489 assert(0); 490 break; 491 } 492 493 q8s16 = vrshrq_n_s16(q8s16, 5); 494 q9s16 = vrshrq_n_s16(q9s16, 5); 495 q10s16 = vrshrq_n_s16(q10s16, 5); 496 q11s16 = vrshrq_n_s16(q11s16, 5); 497 q12s16 = vrshrq_n_s16(q12s16, 5); 498 q13s16 = vrshrq_n_s16(q13s16, 5); 499 q14s16 = vrshrq_n_s16(q14s16, 5); 500 q15s16 = vrshrq_n_s16(q15s16, 5); 501 502 for (d1 = d2 = dest, i = 0; i < 2; i++) { 503 if (i != 0) { 504 q8s16 = q12s16; 505 q9s16 = q13s16; 506 q10s16 = q14s16; 507 q11s16 = q15s16; 508 } 509 510 d0u64 = vld1_u64((uint64_t *)d1); 511 d1 += stride; 512 d1u64 = vld1_u64((uint64_t *)d1); 513 d1 += stride; 514 d2u64 = vld1_u64((uint64_t *)d1); 515 d1 += stride; 516 d3u64 = vld1_u64((uint64_t *)d1); 517 d1 += stride; 518 519 q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u64(d0u64)); 520 q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u64(d1u64)); 521 q10u16 = 522 vaddw_u8(vreinterpretq_u16_s16(q10s16), vreinterpret_u8_u64(d2u64)); 523 q11u16 = 524 vaddw_u8(vreinterpretq_u16_s16(q11s16), vreinterpret_u8_u64(d3u64)); 525 526 d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); 527 d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); 528 d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); 529 d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); 530 531 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); 532 d2 += stride; 533 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); 534 d2 += stride; 535 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); 536 d2 += stride; 537 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); 538 d2 += stride; 539 } 540 } 541
