| OLD | NEW |
|---|
| (Empty) | |
| 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 |
| 13 #include "./vpx_config.h" |
| 14 |
| 15 static int16_t cospi_4_64 = 16069; |
| 16 static int16_t cospi_8_64 = 15137; |
| 17 static int16_t cospi_12_64 = 13623; |
| 18 static int16_t cospi_16_64 = 11585; |
| 19 static int16_t cospi_20_64 = 9102; |
| 20 static int16_t cospi_24_64 = 6270; |
| 21 static int16_t cospi_28_64 = 3196; |
| 22 |
| 23 static INLINE void TRANSPOSE8X8( |
| 24 int16x8_t *q8s16, |
| 25 int16x8_t *q9s16, |
| 26 int16x8_t *q10s16, |
| 27 int16x8_t *q11s16, |
| 28 int16x8_t *q12s16, |
| 29 int16x8_t *q13s16, |
| 30 int16x8_t *q14s16, |
| 31 int16x8_t *q15s16) { |
| 32 int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; |
| 33 int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; |
| 34 int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32; |
| 35 int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16; |
| 36 |
| 37 d16s16 = vget_low_s16(*q8s16); |
| 38 d17s16 = vget_high_s16(*q8s16); |
| 39 d18s16 = vget_low_s16(*q9s16); |
| 40 d19s16 = vget_high_s16(*q9s16); |
| 41 d20s16 = vget_low_s16(*q10s16); |
| 42 d21s16 = vget_high_s16(*q10s16); |
| 43 d22s16 = vget_low_s16(*q11s16); |
| 44 d23s16 = vget_high_s16(*q11s16); |
| 45 d24s16 = vget_low_s16(*q12s16); |
| 46 d25s16 = vget_high_s16(*q12s16); |
| 47 d26s16 = vget_low_s16(*q13s16); |
| 48 d27s16 = vget_high_s16(*q13s16); |
| 49 d28s16 = vget_low_s16(*q14s16); |
| 50 d29s16 = vget_high_s16(*q14s16); |
| 51 d30s16 = vget_low_s16(*q15s16); |
| 52 d31s16 = vget_high_s16(*q15s16); |
| 53 |
| 54 *q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24 |
| 55 *q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26 |
| 56 *q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28 |
| 57 *q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30 |
| 58 *q12s16 = vcombine_s16(d17s16, d25s16); |
| 59 *q13s16 = vcombine_s16(d19s16, d27s16); |
| 60 *q14s16 = vcombine_s16(d21s16, d29s16); |
| 61 *q15s16 = vcombine_s16(d23s16, d31s16); |
| 62 |
| 63 q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q8s16), |
| 64 vreinterpretq_s32_s16(*q10s16)); |
| 65 q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q9s16), |
| 66 vreinterpretq_s32_s16(*q11s16)); |
| 67 q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q12s16), |
| 68 vreinterpretq_s32_s16(*q14s16)); |
| 69 q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q13s16), |
| 70 vreinterpretq_s32_s16(*q15s16)); |
| 71 |
| 72 q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8 |
| 73 vreinterpretq_s16_s32(q1x2s32.val[0])); // q9 |
| 74 q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10 |
| 75 vreinterpretq_s16_s32(q1x2s32.val[1])); // q11 |
| 76 q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12 |
| 77 vreinterpretq_s16_s32(q3x2s32.val[0])); // q13 |
| 78 q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14 |
| 79 vreinterpretq_s16_s32(q3x2s32.val[1])); // q15 |
| 80 |
| 81 *q8s16 = q0x2s16.val[0]; |
| 82 *q9s16 = q0x2s16.val[1]; |
| 83 *q10s16 = q1x2s16.val[0]; |
| 84 *q11s16 = q1x2s16.val[1]; |
| 85 *q12s16 = q2x2s16.val[0]; |
| 86 *q13s16 = q2x2s16.val[1]; |
| 87 *q14s16 = q3x2s16.val[0]; |
| 88 *q15s16 = q3x2s16.val[1]; |
| 89 return; |
| 90 } |
| 91 |
| 92 static INLINE void IDCT8x8_1D( |
| 93 int16x8_t *q8s16, |
| 94 int16x8_t *q9s16, |
| 95 int16x8_t *q10s16, |
| 96 int16x8_t *q11s16, |
| 97 int16x8_t *q12s16, |
| 98 int16x8_t *q13s16, |
| 99 int16x8_t *q14s16, |
| 100 int16x8_t *q15s16) { |
| 101 int16x4_t d0s16, d1s16, d2s16, d3s16; |
| 102 int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16; |
| 103 int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16; |
| 104 int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16; |
| 105 int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; |
| 106 int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32; |
| 107 int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32; |
| 108 |
| 109 d0s16 = vdup_n_s16(cospi_28_64); |
| 110 d1s16 = vdup_n_s16(cospi_4_64); |
| 111 d2s16 = vdup_n_s16(cospi_12_64); |
| 112 d3s16 = vdup_n_s16(cospi_20_64); |
| 113 |
| 114 d16s16 = vget_low_s16(*q8s16); |
| 115 d17s16 = vget_high_s16(*q8s16); |
| 116 d18s16 = vget_low_s16(*q9s16); |
| 117 d19s16 = vget_high_s16(*q9s16); |
| 118 d20s16 = vget_low_s16(*q10s16); |
| 119 d21s16 = vget_high_s16(*q10s16); |
| 120 d22s16 = vget_low_s16(*q11s16); |
| 121 d23s16 = vget_high_s16(*q11s16); |
| 122 d24s16 = vget_low_s16(*q12s16); |
| 123 d25s16 = vget_high_s16(*q12s16); |
| 124 d26s16 = vget_low_s16(*q13s16); |
| 125 d27s16 = vget_high_s16(*q13s16); |
| 126 d28s16 = vget_low_s16(*q14s16); |
| 127 d29s16 = vget_high_s16(*q14s16); |
| 128 d30s16 = vget_low_s16(*q15s16); |
| 129 d31s16 = vget_high_s16(*q15s16); |
| 130 |
| 131 q2s32 = vmull_s16(d18s16, d0s16); |
| 132 q3s32 = vmull_s16(d19s16, d0s16); |
| 133 q5s32 = vmull_s16(d26s16, d2s16); |
| 134 q6s32 = vmull_s16(d27s16, d2s16); |
| 135 |
| 136 q2s32 = vmlsl_s16(q2s32, d30s16, d1s16); |
| 137 q3s32 = vmlsl_s16(q3s32, d31s16, d1s16); |
| 138 q5s32 = vmlsl_s16(q5s32, d22s16, d3s16); |
| 139 q6s32 = vmlsl_s16(q6s32, d23s16, d3s16); |
| 140 |
| 141 d8s16 = vqrshrn_n_s32(q2s32, 14); |
| 142 d9s16 = vqrshrn_n_s32(q3s32, 14); |
| 143 d10s16 = vqrshrn_n_s32(q5s32, 14); |
| 144 d11s16 = vqrshrn_n_s32(q6s32, 14); |
| 145 q4s16 = vcombine_s16(d8s16, d9s16); |
| 146 q5s16 = vcombine_s16(d10s16, d11s16); |
| 147 |
| 148 q2s32 = vmull_s16(d18s16, d1s16); |
| 149 q3s32 = vmull_s16(d19s16, d1s16); |
| 150 q9s32 = vmull_s16(d26s16, d3s16); |
| 151 q13s32 = vmull_s16(d27s16, d3s16); |
| 152 |
| 153 q2s32 = vmlal_s16(q2s32, d30s16, d0s16); |
| 154 q3s32 = vmlal_s16(q3s32, d31s16, d0s16); |
| 155 q9s32 = vmlal_s16(q9s32, d22s16, d2s16); |
| 156 q13s32 = vmlal_s16(q13s32, d23s16, d2s16); |
| 157 |
| 158 d14s16 = vqrshrn_n_s32(q2s32, 14); |
| 159 d15s16 = vqrshrn_n_s32(q3s32, 14); |
| 160 d12s16 = vqrshrn_n_s32(q9s32, 14); |
| 161 d13s16 = vqrshrn_n_s32(q13s32, 14); |
| 162 q6s16 = vcombine_s16(d12s16, d13s16); |
| 163 q7s16 = vcombine_s16(d14s16, d15s16); |
| 164 |
| 165 d0s16 = vdup_n_s16(cospi_16_64); |
| 166 |
| 167 q2s32 = vmull_s16(d16s16, d0s16); |
| 168 q3s32 = vmull_s16(d17s16, d0s16); |
| 169 q13s32 = vmull_s16(d16s16, d0s16); |
| 170 q15s32 = vmull_s16(d17s16, d0s16); |
| 171 |
| 172 q2s32 = vmlal_s16(q2s32, d24s16, d0s16); |
| 173 q3s32 = vmlal_s16(q3s32, d25s16, d0s16); |
| 174 q13s32 = vmlsl_s16(q13s32, d24s16, d0s16); |
| 175 q15s32 = vmlsl_s16(q15s32, d25s16, d0s16); |
| 176 |
| 177 d0s16 = vdup_n_s16(cospi_24_64); |
| 178 d1s16 = vdup_n_s16(cospi_8_64); |
| 179 |
| 180 d18s16 = vqrshrn_n_s32(q2s32, 14); |
| 181 d19s16 = vqrshrn_n_s32(q3s32, 14); |
| 182 d22s16 = vqrshrn_n_s32(q13s32, 14); |
| 183 d23s16 = vqrshrn_n_s32(q15s32, 14); |
| 184 *q9s16 = vcombine_s16(d18s16, d19s16); |
| 185 *q11s16 = vcombine_s16(d22s16, d23s16); |
| 186 |
| 187 q2s32 = vmull_s16(d20s16, d0s16); |
| 188 q3s32 = vmull_s16(d21s16, d0s16); |
| 189 q8s32 = vmull_s16(d20s16, d1s16); |
| 190 q12s32 = vmull_s16(d21s16, d1s16); |
| 191 |
| 192 q2s32 = vmlsl_s16(q2s32, d28s16, d1s16); |
| 193 q3s32 = vmlsl_s16(q3s32, d29s16, d1s16); |
| 194 q8s32 = vmlal_s16(q8s32, d28s16, d0s16); |
| 195 q12s32 = vmlal_s16(q12s32, d29s16, d0s16); |
| 196 |
| 197 d26s16 = vqrshrn_n_s32(q2s32, 14); |
| 198 d27s16 = vqrshrn_n_s32(q3s32, 14); |
| 199 d30s16 = vqrshrn_n_s32(q8s32, 14); |
| 200 d31s16 = vqrshrn_n_s32(q12s32, 14); |
| 201 *q13s16 = vcombine_s16(d26s16, d27s16); |
| 202 *q15s16 = vcombine_s16(d30s16, d31s16); |
| 203 |
| 204 q0s16 = vaddq_s16(*q9s16, *q15s16); |
| 205 q1s16 = vaddq_s16(*q11s16, *q13s16); |
| 206 q2s16 = vsubq_s16(*q11s16, *q13s16); |
| 207 q3s16 = vsubq_s16(*q9s16, *q15s16); |
| 208 |
| 209 *q13s16 = vsubq_s16(q4s16, q5s16); |
| 210 q4s16 = vaddq_s16(q4s16, q5s16); |
| 211 *q14s16 = vsubq_s16(q7s16, q6s16); |
| 212 q7s16 = vaddq_s16(q7s16, q6s16); |
| 213 d26s16 = vget_low_s16(*q13s16); |
| 214 d27s16 = vget_high_s16(*q13s16); |
| 215 d28s16 = vget_low_s16(*q14s16); |
| 216 d29s16 = vget_high_s16(*q14s16); |
| 217 |
| 218 d16s16 = vdup_n_s16(cospi_16_64); |
| 219 |
| 220 q9s32 = vmull_s16(d28s16, d16s16); |
| 221 q10s32 = vmull_s16(d29s16, d16s16); |
| 222 q11s32 = vmull_s16(d28s16, d16s16); |
| 223 q12s32 = vmull_s16(d29s16, d16s16); |
| 224 |
| 225 q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); |
| 226 q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); |
| 227 q11s32 = vmlal_s16(q11s32, d26s16, d16s16); |
| 228 q12s32 = vmlal_s16(q12s32, d27s16, d16s16); |
| 229 |
| 230 d10s16 = vqrshrn_n_s32(q9s32, 14); |
| 231 d11s16 = vqrshrn_n_s32(q10s32, 14); |
| 232 d12s16 = vqrshrn_n_s32(q11s32, 14); |
| 233 d13s16 = vqrshrn_n_s32(q12s32, 14); |
| 234 q5s16 = vcombine_s16(d10s16, d11s16); |
| 235 q6s16 = vcombine_s16(d12s16, d13s16); |
| 236 |
| 237 *q8s16 = vaddq_s16(q0s16, q7s16); |
| 238 *q9s16 = vaddq_s16(q1s16, q6s16); |
| 239 *q10s16 = vaddq_s16(q2s16, q5s16); |
| 240 *q11s16 = vaddq_s16(q3s16, q4s16); |
| 241 *q12s16 = vsubq_s16(q3s16, q4s16); |
| 242 *q13s16 = vsubq_s16(q2s16, q5s16); |
| 243 *q14s16 = vsubq_s16(q1s16, q6s16); |
| 244 *q15s16 = vsubq_s16(q0s16, q7s16); |
| 245 return; |
| 246 } |
| 247 |
| 248 void vp9_idct8x8_64_add_neon( |
| 249 int16_t *input, |
| 250 uint8_t *dest, |
| 251 int dest_stride) { |
| 252 uint8_t *d1, *d2; |
| 253 uint8x8_t d0u8, d1u8, d2u8, d3u8; |
| 254 uint64x1_t d0u64, d1u64, d2u64, d3u64; |
| 255 int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; |
| 256 uint16x8_t q8u16, q9u16, q10u16, q11u16; |
| 257 |
| 258 q8s16 = vld1q_s16(input); |
| 259 q9s16 = vld1q_s16(input + 8); |
| 260 q10s16 = vld1q_s16(input + 16); |
| 261 q11s16 = vld1q_s16(input + 24); |
| 262 q12s16 = vld1q_s16(input + 32); |
| 263 q13s16 = vld1q_s16(input + 40); |
| 264 q14s16 = vld1q_s16(input + 48); |
| 265 q15s16 = vld1q_s16(input + 56); |
| 266 |
| 267 TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, |
| 268 &q12s16, &q13s16, &q14s16, &q15s16); |
| 269 |
| 270 IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, |
| 271 &q12s16, &q13s16, &q14s16, &q15s16); |
| 272 |
| 273 TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, |
| 274 &q12s16, &q13s16, &q14s16, &q15s16); |
| 275 |
| 276 IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, |
| 277 &q12s16, &q13s16, &q14s16, &q15s16); |
| 278 |
| 279 q8s16 = vrshrq_n_s16(q8s16, 5); |
| 280 q9s16 = vrshrq_n_s16(q9s16, 5); |
| 281 q10s16 = vrshrq_n_s16(q10s16, 5); |
| 282 q11s16 = vrshrq_n_s16(q11s16, 5); |
| 283 q12s16 = vrshrq_n_s16(q12s16, 5); |
| 284 q13s16 = vrshrq_n_s16(q13s16, 5); |
| 285 q14s16 = vrshrq_n_s16(q14s16, 5); |
| 286 q15s16 = vrshrq_n_s16(q15s16, 5); |
| 287 |
| 288 d1 = d2 = dest; |
| 289 |
| 290 d0u64 = vld1_u64((uint64_t *)d1); |
| 291 d1 += dest_stride; |
| 292 d1u64 = vld1_u64((uint64_t *)d1); |
| 293 d1 += dest_stride; |
| 294 d2u64 = vld1_u64((uint64_t *)d1); |
| 295 d1 += dest_stride; |
| 296 d3u64 = vld1_u64((uint64_t *)d1); |
| 297 d1 += dest_stride; |
| 298 |
| 299 q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), |
| 300 vreinterpret_u8_u64(d0u64)); |
| 301 q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), |
| 302 vreinterpret_u8_u64(d1u64)); |
| 303 q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), |
| 304 vreinterpret_u8_u64(d2u64)); |
| 305 q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), |
| 306 vreinterpret_u8_u64(d3u64)); |
| 307 |
| 308 d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); |
| 309 d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); |
| 310 d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); |
| 311 d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); |
| 312 |
| 313 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); |
| 314 d2 += dest_stride; |
| 315 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); |
| 316 d2 += dest_stride; |
| 317 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); |
| 318 d2 += dest_stride; |
| 319 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); |
| 320 d2 += dest_stride; |
| 321 |
| 322 q8s16 = q12s16; |
| 323 q9s16 = q13s16; |
| 324 q10s16 = q14s16; |
| 325 q11s16 = q15s16; |
| 326 |
| 327 d0u64 = vld1_u64((uint64_t *)d1); |
| 328 d1 += dest_stride; |
| 329 d1u64 = vld1_u64((uint64_t *)d1); |
| 330 d1 += dest_stride; |
| 331 d2u64 = vld1_u64((uint64_t *)d1); |
| 332 d1 += dest_stride; |
| 333 d3u64 = vld1_u64((uint64_t *)d1); |
| 334 d1 += dest_stride; |
| 335 |
| 336 q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), |
| 337 vreinterpret_u8_u64(d0u64)); |
| 338 q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), |
| 339 vreinterpret_u8_u64(d1u64)); |
| 340 q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), |
| 341 vreinterpret_u8_u64(d2u64)); |
| 342 q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), |
| 343 vreinterpret_u8_u64(d3u64)); |
| 344 |
| 345 d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); |
| 346 d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); |
| 347 d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); |
| 348 d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); |
| 349 |
| 350 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); |
| 351 d2 += dest_stride; |
| 352 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); |
| 353 d2 += dest_stride; |
| 354 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); |
| 355 d2 += dest_stride; |
| 356 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); |
| 357 d2 += dest_stride; |
| 358 return; |
| 359 } |
| 360 |
| 361 void vp9_idct8x8_12_add_neon( |
| 362 int16_t *input, |
| 363 uint8_t *dest, |
| 364 int dest_stride) { |
| 365 uint8_t *d1, *d2; |
| 366 uint8x8_t d0u8, d1u8, d2u8, d3u8; |
| 367 int16x4_t d10s16, d11s16, d12s16, d13s16, d16s16; |
| 368 int16x4_t d26s16, d27s16, d28s16, d29s16; |
| 369 uint64x1_t d0u64, d1u64, d2u64, d3u64; |
| 370 int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16; |
| 371 int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16; |
| 372 uint16x8_t q8u16, q9u16, q10u16, q11u16; |
| 373 int32x4_t q9s32, q10s32, q11s32, q12s32; |
| 374 |
| 375 q8s16 = vld1q_s16(input); |
| 376 q9s16 = vld1q_s16(input + 8); |
| 377 q10s16 = vld1q_s16(input + 16); |
| 378 q11s16 = vld1q_s16(input + 24); |
| 379 q12s16 = vld1q_s16(input + 32); |
| 380 q13s16 = vld1q_s16(input + 40); |
| 381 q14s16 = vld1q_s16(input + 48); |
| 382 q15s16 = vld1q_s16(input + 56); |
| 383 |
| 384 TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, |
| 385 &q12s16, &q13s16, &q14s16, &q15s16); |
| 386 |
| 387 // First transform rows |
| 388 // stage 1 |
| 389 q0s16 = vdupq_n_s16(cospi_28_64 * 2); |
| 390 q1s16 = vdupq_n_s16(cospi_4_64 * 2); |
| 391 |
| 392 q4s16 = vqrdmulhq_s16(q9s16, q0s16); |
| 393 |
| 394 q0s16 = vdupq_n_s16(-cospi_20_64 * 2); |
| 395 |
| 396 q7s16 = vqrdmulhq_s16(q9s16, q1s16); |
| 397 |
| 398 q1s16 = vdupq_n_s16(cospi_12_64 * 2); |
| 399 |
| 400 q5s16 = vqrdmulhq_s16(q11s16, q0s16); |
| 401 |
| 402 q0s16 = vdupq_n_s16(cospi_16_64 * 2); |
| 403 |
| 404 q6s16 = vqrdmulhq_s16(q11s16, q1s16); |
| 405 |
| 406 // stage 2 & stage 3 - even half |
| 407 q1s16 = vdupq_n_s16(cospi_24_64 * 2); |
| 408 |
| 409 q9s16 = vqrdmulhq_s16(q8s16, q0s16); |
| 410 |
| 411 q0s16 = vdupq_n_s16(cospi_8_64 * 2); |
| 412 |
| 413 q13s16 = vqrdmulhq_s16(q10s16, q1s16); |
| 414 |
| 415 q15s16 = vqrdmulhq_s16(q10s16, q0s16); |
| 416 |
| 417 // stage 3 -odd half |
| 418 q0s16 = vaddq_s16(q9s16, q15s16); |
| 419 q1s16 = vaddq_s16(q9s16, q13s16); |
| 420 q2s16 = vsubq_s16(q9s16, q13s16); |
| 421 q3s16 = vsubq_s16(q9s16, q15s16); |
| 422 |
| 423 // stage 2 - odd half |
| 424 q13s16 = vsubq_s16(q4s16, q5s16); |
| 425 q4s16 = vaddq_s16(q4s16, q5s16); |
| 426 q14s16 = vsubq_s16(q7s16, q6s16); |
| 427 q7s16 = vaddq_s16(q7s16, q6s16); |
| 428 d26s16 = vget_low_s16(q13s16); |
| 429 d27s16 = vget_high_s16(q13s16); |
| 430 d28s16 = vget_low_s16(q14s16); |
| 431 d29s16 = vget_high_s16(q14s16); |
| 432 |
| 433 d16s16 = vdup_n_s16(cospi_16_64); |
| 434 q9s32 = vmull_s16(d28s16, d16s16); |
| 435 q10s32 = vmull_s16(d29s16, d16s16); |
| 436 q11s32 = vmull_s16(d28s16, d16s16); |
| 437 q12s32 = vmull_s16(d29s16, d16s16); |
| 438 |
| 439 q9s32 = vmlsl_s16(q9s32, d26s16, d16s16); |
| 440 q10s32 = vmlsl_s16(q10s32, d27s16, d16s16); |
| 441 q11s32 = vmlal_s16(q11s32, d26s16, d16s16); |
| 442 q12s32 = vmlal_s16(q12s32, d27s16, d16s16); |
| 443 |
| 444 d10s16 = vqrshrn_n_s32(q9s32, 14); |
| 445 d11s16 = vqrshrn_n_s32(q10s32, 14); |
| 446 d12s16 = vqrshrn_n_s32(q11s32, 14); |
| 447 d13s16 = vqrshrn_n_s32(q12s32, 14); |
| 448 q5s16 = vcombine_s16(d10s16, d11s16); |
| 449 q6s16 = vcombine_s16(d12s16, d13s16); |
| 450 |
| 451 // stage 4 |
| 452 q8s16 = vaddq_s16(q0s16, q7s16); |
| 453 q9s16 = vaddq_s16(q1s16, q6s16); |
| 454 q10s16 = vaddq_s16(q2s16, q5s16); |
| 455 q11s16 = vaddq_s16(q3s16, q4s16); |
| 456 q12s16 = vsubq_s16(q3s16, q4s16); |
| 457 q13s16 = vsubq_s16(q2s16, q5s16); |
| 458 q14s16 = vsubq_s16(q1s16, q6s16); |
| 459 q15s16 = vsubq_s16(q0s16, q7s16); |
| 460 |
| 461 TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, |
| 462 &q12s16, &q13s16, &q14s16, &q15s16); |
| 463 |
| 464 IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, |
| 465 &q12s16, &q13s16, &q14s16, &q15s16); |
| 466 |
| 467 q8s16 = vrshrq_n_s16(q8s16, 5); |
| 468 q9s16 = vrshrq_n_s16(q9s16, 5); |
| 469 q10s16 = vrshrq_n_s16(q10s16, 5); |
| 470 q11s16 = vrshrq_n_s16(q11s16, 5); |
| 471 q12s16 = vrshrq_n_s16(q12s16, 5); |
| 472 q13s16 = vrshrq_n_s16(q13s16, 5); |
| 473 q14s16 = vrshrq_n_s16(q14s16, 5); |
| 474 q15s16 = vrshrq_n_s16(q15s16, 5); |
| 475 |
| 476 d1 = d2 = dest; |
| 477 |
| 478 d0u64 = vld1_u64((uint64_t *)d1); |
| 479 d1 += dest_stride; |
| 480 d1u64 = vld1_u64((uint64_t *)d1); |
| 481 d1 += dest_stride; |
| 482 d2u64 = vld1_u64((uint64_t *)d1); |
| 483 d1 += dest_stride; |
| 484 d3u64 = vld1_u64((uint64_t *)d1); |
| 485 d1 += dest_stride; |
| 486 |
| 487 q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), |
| 488 vreinterpret_u8_u64(d0u64)); |
| 489 q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), |
| 490 vreinterpret_u8_u64(d1u64)); |
| 491 q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), |
| 492 vreinterpret_u8_u64(d2u64)); |
| 493 q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), |
| 494 vreinterpret_u8_u64(d3u64)); |
| 495 |
| 496 d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); |
| 497 d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); |
| 498 d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); |
| 499 d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); |
| 500 |
| 501 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); |
| 502 d2 += dest_stride; |
| 503 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); |
| 504 d2 += dest_stride; |
| 505 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); |
| 506 d2 += dest_stride; |
| 507 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); |
| 508 d2 += dest_stride; |
| 509 |
| 510 q8s16 = q12s16; |
| 511 q9s16 = q13s16; |
| 512 q10s16 = q14s16; |
| 513 q11s16 = q15s16; |
| 514 |
| 515 d0u64 = vld1_u64((uint64_t *)d1); |
| 516 d1 += dest_stride; |
| 517 d1u64 = vld1_u64((uint64_t *)d1); |
| 518 d1 += dest_stride; |
| 519 d2u64 = vld1_u64((uint64_t *)d1); |
| 520 d1 += dest_stride; |
| 521 d3u64 = vld1_u64((uint64_t *)d1); |
| 522 d1 += dest_stride; |
| 523 |
| 524 q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), |
| 525 vreinterpret_u8_u64(d0u64)); |
| 526 q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), |
| 527 vreinterpret_u8_u64(d1u64)); |
| 528 q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16), |
| 529 vreinterpret_u8_u64(d2u64)); |
| 530 q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16), |
| 531 vreinterpret_u8_u64(d3u64)); |
| 532 |
| 533 d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16)); |
| 534 d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16)); |
| 535 d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16)); |
| 536 d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16)); |
| 537 |
| 538 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8)); |
| 539 d2 += dest_stride; |
| 540 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8)); |
| 541 d2 += dest_stride; |
| 542 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8)); |
| 543 d2 += dest_stride; |
| 544 vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8)); |
| 545 d2 += dest_stride; |
| 546 return; |
| 547 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 812033011:
libvpx: Pull from upstream (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/