| OLD | NEW |
|---|
| (Empty) |
| 1 ; | |
| 2 ; Copyright (c) 2013 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 EXPORT |vp9_iht8x8_64_add_neon| | |
| 12 ARM | |
| 13 REQUIRE8 | |
| 14 PRESERVE8 | |
| 15 | |
| 16 AREA ||.text||, CODE, READONLY, ALIGN=2 | |
| 17 | |
| 18 ; Generate IADST constants in r0 - r12 for the IADST. | |
| 19 MACRO | |
| 20 GENERATE_IADST_CONSTANTS | |
| 21 ; generate cospi_2_64 = 16305 | |
| 22 mov r0, #0x3f00 | |
| 23 add r0, #0xb1 | |
| 24 | |
| 25 ; generate cospi_30_64 = 1606 | |
| 26 mov r1, #0x600 | |
| 27 add r1, #0x46 | |
| 28 | |
| 29 ; generate cospi_10_64 = 14449 | |
| 30 mov r2, #0x3800 | |
| 31 add r2, #0x71 | |
| 32 | |
| 33 ; generate cospi_22_64 = 7723 | |
| 34 mov r3, #0x1e00 | |
| 35 add r3, #0x2b | |
| 36 | |
| 37 ; generate cospi_18_64 = 10394 | |
| 38 mov r4, #0x2800 | |
| 39 add r4, #0x9a | |
| 40 | |
| 41 ; generate cospi_14_64 = 12665 | |
| 42 mov r5, #0x3100 | |
| 43 add r5, #0x79 | |
| 44 | |
| 45 ; generate cospi_26_64 = 4756 | |
| 46 mov r6, #0x1200 | |
| 47 add r6, #0x94 | |
| 48 | |
| 49 ; generate cospi_6_64 = 15679 | |
| 50 mov r7, #0x3d00 | |
| 51 add r7, #0x3f | |
| 52 | |
| 53 ; generate cospi_8_64 = 15137 | |
| 54 mov r8, #0x3b00 | |
| 55 add r8, #0x21 | |
| 56 | |
| 57 ; generate cospi_24_64 = 6270 | |
| 58 mov r9, #0x1800 | |
| 59 add r9, #0x7e | |
| 60 | |
| 61 ; generate 0 | |
| 62 mov r10, #0 | |
| 63 | |
| 64 ; generate cospi_16_64 = 11585 | |
| 65 mov r12, #0x2d00 | |
| 66 add r12, #0x41 | |
| 67 MEND | |
| 68 | |
| 69 ; Generate IDCT constants in r3 - r9 for the IDCT. | |
| 70 MACRO | |
| 71 GENERATE_IDCT_CONSTANTS | |
| 72 ; generate cospi_28_64 = 3196 | |
| 73 mov r3, #0x0c00 | |
| 74 add r3, #0x7c | |
| 75 | |
| 76 ; generate cospi_4_64 = 16069 | |
| 77 mov r4, #0x3e00 | |
| 78 add r4, #0xc5 | |
| 79 | |
| 80 ; generate cospi_12_64 = 13623 | |
| 81 mov r5, #0x3500 | |
| 82 add r5, #0x37 | |
| 83 | |
| 84 ; generate cospi_20_64 = 9102 | |
| 85 mov r6, #0x2300 | |
| 86 add r6, #0x8e | |
| 87 | |
| 88 ; generate cospi_16_64 = 11585 | |
| 89 mov r7, #0x2d00 | |
| 90 add r7, #0x41 | |
| 91 | |
| 92 ; generate cospi_24_64 = 6270 | |
| 93 mov r8, #0x1800 | |
| 94 add r8, #0x7e | |
| 95 | |
| 96 ; generate cospi_8_64 = 15137 | |
| 97 mov r9, #0x3b00 | |
| 98 add r9, #0x21 | |
| 99 MEND | |
| 100 | |
| 101 ; Transpose a 8x8 16bits data matrix. Datas are loaded in q8-q15. | |
| 102 MACRO | |
| 103 TRANSPOSE8X8 | |
| 104 vswp d17, d24 | |
| 105 vswp d23, d30 | |
| 106 vswp d21, d28 | |
| 107 vswp d19, d26 | |
| 108 vtrn.32 q8, q10 | |
| 109 vtrn.32 q9, q11 | |
| 110 vtrn.32 q12, q14 | |
| 111 vtrn.32 q13, q15 | |
| 112 vtrn.16 q8, q9 | |
| 113 vtrn.16 q10, q11 | |
| 114 vtrn.16 q12, q13 | |
| 115 vtrn.16 q14, q15 | |
| 116 MEND | |
| 117 | |
| 118 ; Parallel 1D IDCT on all the columns of a 8x8 16bits data matrix which are | |
| 119 ; loaded in q8-q15. The IDCT constants are loaded in r3 - r9. The output | |
| 120 ; will be stored back into q8-q15 registers. This macro will touch q0-q7 | |
| 121 ; registers and use them as buffer during calculation. | |
| 122 MACRO | |
| 123 IDCT8x8_1D | |
| 124 ; stage 1 | |
| 125 vdup.16 d0, r3 ; duplicate cospi_28_64 | |
| 126 vdup.16 d1, r4 ; duplicate cospi_4_64 | |
| 127 vdup.16 d2, r5 ; duplicate cospi_12_64 | |
| 128 vdup.16 d3, r6 ; duplicate cospi_20_64 | |
| 129 | |
| 130 ; input[1] * cospi_28_64 | |
| 131 vmull.s16 q2, d18, d0 | |
| 132 vmull.s16 q3, d19, d0 | |
| 133 | |
| 134 ; input[5] * cospi_12_64 | |
| 135 vmull.s16 q5, d26, d2 | |
| 136 vmull.s16 q6, d27, d2 | |
| 137 | |
| 138 ; input[1]*cospi_28_64-input[7]*cospi_4_64 | |
| 139 vmlsl.s16 q2, d30, d1 | |
| 140 vmlsl.s16 q3, d31, d1 | |
| 141 | |
| 142 ; input[5] * cospi_12_64 - input[3] * cospi_20_64 | |
| 143 vmlsl.s16 q5, d22, d3 | |
| 144 vmlsl.s16 q6, d23, d3 | |
| 145 | |
| 146 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 147 vqrshrn.s32 d8, q2, #14 ; >> 14 | |
| 148 vqrshrn.s32 d9, q3, #14 ; >> 14 | |
| 149 | |
| 150 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 151 vqrshrn.s32 d10, q5, #14 ; >> 14 | |
| 152 vqrshrn.s32 d11, q6, #14 ; >> 14 | |
| 153 | |
| 154 ; input[1] * cospi_4_64 | |
| 155 vmull.s16 q2, d18, d1 | |
| 156 vmull.s16 q3, d19, d1 | |
| 157 | |
| 158 ; input[5] * cospi_20_64 | |
| 159 vmull.s16 q9, d26, d3 | |
| 160 vmull.s16 q13, d27, d3 | |
| 161 | |
| 162 ; input[1]*cospi_4_64+input[7]*cospi_28_64 | |
| 163 vmlal.s16 q2, d30, d0 | |
| 164 vmlal.s16 q3, d31, d0 | |
| 165 | |
| 166 ; input[5] * cospi_20_64 + input[3] * cospi_12_64 | |
| 167 vmlal.s16 q9, d22, d2 | |
| 168 vmlal.s16 q13, d23, d2 | |
| 169 | |
| 170 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 171 vqrshrn.s32 d14, q2, #14 ; >> 14 | |
| 172 vqrshrn.s32 d15, q3, #14 ; >> 14 | |
| 173 | |
| 174 ; stage 2 & stage 3 - even half | |
| 175 vdup.16 d0, r7 ; duplicate cospi_16_64 | |
| 176 | |
| 177 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 178 vqrshrn.s32 d12, q9, #14 ; >> 14 | |
| 179 vqrshrn.s32 d13, q13, #14 ; >> 14 | |
| 180 | |
| 181 ; input[0] * cospi_16_64 | |
| 182 vmull.s16 q2, d16, d0 | |
| 183 vmull.s16 q3, d17, d0 | |
| 184 | |
| 185 ; input[0] * cospi_16_64 | |
| 186 vmull.s16 q13, d16, d0 | |
| 187 vmull.s16 q15, d17, d0 | |
| 188 | |
| 189 ; (input[0] + input[2]) * cospi_16_64 | |
| 190 vmlal.s16 q2, d24, d0 | |
| 191 vmlal.s16 q3, d25, d0 | |
| 192 | |
| 193 ; (input[0] - input[2]) * cospi_16_64 | |
| 194 vmlsl.s16 q13, d24, d0 | |
| 195 vmlsl.s16 q15, d25, d0 | |
| 196 | |
| 197 vdup.16 d0, r8 ; duplicate cospi_24_64 | |
| 198 vdup.16 d1, r9 ; duplicate cospi_8_64 | |
| 199 | |
| 200 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 201 vqrshrn.s32 d18, q2, #14 ; >> 14 | |
| 202 vqrshrn.s32 d19, q3, #14 ; >> 14 | |
| 203 | |
| 204 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 205 vqrshrn.s32 d22, q13, #14 ; >> 14 | |
| 206 vqrshrn.s32 d23, q15, #14 ; >> 14 | |
| 207 | |
| 208 ; input[1] * cospi_24_64 | |
| 209 vmull.s16 q2, d20, d0 | |
| 210 vmull.s16 q3, d21, d0 | |
| 211 | |
| 212 ; input[1] * cospi_8_64 | |
| 213 vmull.s16 q8, d20, d1 | |
| 214 vmull.s16 q12, d21, d1 | |
| 215 | |
| 216 ; input[1] * cospi_24_64 - input[3] * cospi_8_64 | |
| 217 vmlsl.s16 q2, d28, d1 | |
| 218 vmlsl.s16 q3, d29, d1 | |
| 219 | |
| 220 ; input[1] * cospi_8_64 + input[3] * cospi_24_64 | |
| 221 vmlal.s16 q8, d28, d0 | |
| 222 vmlal.s16 q12, d29, d0 | |
| 223 | |
| 224 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 225 vqrshrn.s32 d26, q2, #14 ; >> 14 | |
| 226 vqrshrn.s32 d27, q3, #14 ; >> 14 | |
| 227 | |
| 228 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 229 vqrshrn.s32 d30, q8, #14 ; >> 14 | |
| 230 vqrshrn.s32 d31, q12, #14 ; >> 14 | |
| 231 | |
| 232 vadd.s16 q0, q9, q15 ; output[0] = step[0] + step[3] | |
| 233 vadd.s16 q1, q11, q13 ; output[1] = step[1] + step[2] | |
| 234 vsub.s16 q2, q11, q13 ; output[2] = step[1] - step[2] | |
| 235 vsub.s16 q3, q9, q15 ; output[3] = step[0] - step[3] | |
| 236 | |
| 237 ; stage 3 -odd half | |
| 238 vdup.16 d16, r7 ; duplicate cospi_16_64 | |
| 239 | |
| 240 ; stage 2 - odd half | |
| 241 vsub.s16 q13, q4, q5 ; step2[5] = step1[4] - step1[5] | |
| 242 vadd.s16 q4, q4, q5 ; step2[4] = step1[4] + step1[5] | |
| 243 vsub.s16 q14, q7, q6 ; step2[6] = -step1[6] + step1[7] | |
| 244 vadd.s16 q7, q7, q6 ; step2[7] = step1[6] + step1[7] | |
| 245 | |
| 246 ; step2[6] * cospi_16_64 | |
| 247 vmull.s16 q9, d28, d16 | |
| 248 vmull.s16 q10, d29, d16 | |
| 249 | |
| 250 ; step2[6] * cospi_16_64 | |
| 251 vmull.s16 q11, d28, d16 | |
| 252 vmull.s16 q12, d29, d16 | |
| 253 | |
| 254 ; (step2[6] - step2[5]) * cospi_16_64 | |
| 255 vmlsl.s16 q9, d26, d16 | |
| 256 vmlsl.s16 q10, d27, d16 | |
| 257 | |
| 258 ; (step2[5] + step2[6]) * cospi_16_64 | |
| 259 vmlal.s16 q11, d26, d16 | |
| 260 vmlal.s16 q12, d27, d16 | |
| 261 | |
| 262 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 263 vqrshrn.s32 d10, q9, #14 ; >> 14 | |
| 264 vqrshrn.s32 d11, q10, #14 ; >> 14 | |
| 265 | |
| 266 ; dct_const_round_shift(input_dc * cospi_16_64) | |
| 267 vqrshrn.s32 d12, q11, #14 ; >> 14 | |
| 268 vqrshrn.s32 d13, q12, #14 ; >> 14 | |
| 269 | |
| 270 ; stage 4 | |
| 271 vadd.s16 q8, q0, q7 ; output[0] = step1[0] + step1[7]; | |
| 272 vadd.s16 q9, q1, q6 ; output[1] = step1[1] + step1[6]; | |
| 273 vadd.s16 q10, q2, q5 ; output[2] = step1[2] + step1[5]; | |
| 274 vadd.s16 q11, q3, q4 ; output[3] = step1[3] + step1[4]; | |
| 275 vsub.s16 q12, q3, q4 ; output[4] = step1[3] - step1[4]; | |
| 276 vsub.s16 q13, q2, q5 ; output[5] = step1[2] - step1[5]; | |
| 277 vsub.s16 q14, q1, q6 ; output[6] = step1[1] - step1[6]; | |
| 278 vsub.s16 q15, q0, q7 ; output[7] = step1[0] - step1[7]; | |
| 279 MEND | |
| 280 | |
| 281 ; Parallel 1D IADST on all the columns of a 8x8 16bits data matrix which | |
| 282 ; loaded in q8-q15. IADST constants are loaded in r0 - r12 registers. The | |
| 283 ; output will be stored back into q8-q15 registers. This macro will touch | |
| 284 ; q0 - q7 registers and use them as buffer during calculation. | |
| 285 MACRO | |
| 286 IADST8X8_1D | |
| 287 vdup.16 d14, r0 ; duplicate cospi_2_64 | |
| 288 vdup.16 d15, r1 ; duplicate cospi_30_64 | |
| 289 | |
| 290 ; cospi_2_64 * x0 | |
| 291 vmull.s16 q1, d30, d14 | |
| 292 vmull.s16 q2, d31, d14 | |
| 293 | |
| 294 ; cospi_30_64 * x0 | |
| 295 vmull.s16 q3, d30, d15 | |
| 296 vmull.s16 q4, d31, d15 | |
| 297 | |
| 298 vdup.16 d30, r4 ; duplicate cospi_18_64 | |
| 299 vdup.16 d31, r5 ; duplicate cospi_14_64 | |
| 300 | |
| 301 ; s0 = cospi_2_64 * x0 + cospi_30_64 * x1; | |
| 302 vmlal.s16 q1, d16, d15 | |
| 303 vmlal.s16 q2, d17, d15 | |
| 304 | |
| 305 ; s1 = cospi_30_64 * x0 - cospi_2_64 * x1 | |
| 306 vmlsl.s16 q3, d16, d14 | |
| 307 vmlsl.s16 q4, d17, d14 | |
| 308 | |
| 309 ; cospi_18_64 * x4 | |
| 310 vmull.s16 q5, d22, d30 | |
| 311 vmull.s16 q6, d23, d30 | |
| 312 | |
| 313 ; cospi_14_64 * x4 | |
| 314 vmull.s16 q7, d22, d31 | |
| 315 vmull.s16 q8, d23, d31 | |
| 316 | |
| 317 ; s4 = cospi_18_64 * x4 + cospi_14_64 * x5; | |
| 318 vmlal.s16 q5, d24, d31 | |
| 319 vmlal.s16 q6, d25, d31 | |
| 320 | |
| 321 ; s5 = cospi_14_64 * x4 - cospi_18_64 * x5 | |
| 322 vmlsl.s16 q7, d24, d30 | |
| 323 vmlsl.s16 q8, d25, d30 | |
| 324 | |
| 325 ; (s0 + s4) | |
| 326 vadd.s32 q11, q1, q5 | |
| 327 vadd.s32 q12, q2, q6 | |
| 328 | |
| 329 vdup.16 d0, r2 ; duplicate cospi_10_64 | |
| 330 vdup.16 d1, r3 ; duplicate cospi_22_64 | |
| 331 | |
| 332 ; (s0 - s4) | |
| 333 vsub.s32 q1, q1, q5 | |
| 334 vsub.s32 q2, q2, q6 | |
| 335 | |
| 336 ; x0 = dct_const_round_shift(s0 + s4); | |
| 337 vqrshrn.s32 d22, q11, #14 ; >> 14 | |
| 338 vqrshrn.s32 d23, q12, #14 ; >> 14 | |
| 339 | |
| 340 ; (s1 + s5) | |
| 341 vadd.s32 q12, q3, q7 | |
| 342 vadd.s32 q15, q4, q8 | |
| 343 | |
| 344 ; (s1 - s5) | |
| 345 vsub.s32 q3, q3, q7 | |
| 346 vsub.s32 q4, q4, q8 | |
| 347 | |
| 348 ; x4 = dct_const_round_shift(s0 - s4); | |
| 349 vqrshrn.s32 d2, q1, #14 ; >> 14 | |
| 350 vqrshrn.s32 d3, q2, #14 ; >> 14 | |
| 351 | |
| 352 ; x1 = dct_const_round_shift(s1 + s5); | |
| 353 vqrshrn.s32 d24, q12, #14 ; >> 14 | |
| 354 vqrshrn.s32 d25, q15, #14 ; >> 14 | |
| 355 | |
| 356 ; x5 = dct_const_round_shift(s1 - s5); | |
| 357 vqrshrn.s32 d6, q3, #14 ; >> 14 | |
| 358 vqrshrn.s32 d7, q4, #14 ; >> 14 | |
| 359 | |
| 360 ; cospi_10_64 * x2 | |
| 361 vmull.s16 q4, d26, d0 | |
| 362 vmull.s16 q5, d27, d0 | |
| 363 | |
| 364 ; cospi_22_64 * x2 | |
| 365 vmull.s16 q2, d26, d1 | |
| 366 vmull.s16 q6, d27, d1 | |
| 367 | |
| 368 vdup.16 d30, r6 ; duplicate cospi_26_64 | |
| 369 vdup.16 d31, r7 ; duplicate cospi_6_64 | |
| 370 | |
| 371 ; s2 = cospi_10_64 * x2 + cospi_22_64 * x3; | |
| 372 vmlal.s16 q4, d20, d1 | |
| 373 vmlal.s16 q5, d21, d1 | |
| 374 | |
| 375 ; s3 = cospi_22_64 * x2 - cospi_10_64 * x3; | |
| 376 vmlsl.s16 q2, d20, d0 | |
| 377 vmlsl.s16 q6, d21, d0 | |
| 378 | |
| 379 ; cospi_26_64 * x6 | |
| 380 vmull.s16 q0, d18, d30 | |
| 381 vmull.s16 q13, d19, d30 | |
| 382 | |
| 383 ; s6 = cospi_26_64 * x6 + cospi_6_64 * x7; | |
| 384 vmlal.s16 q0, d28, d31 | |
| 385 vmlal.s16 q13, d29, d31 | |
| 386 | |
| 387 ; cospi_6_64 * x6 | |
| 388 vmull.s16 q10, d18, d31 | |
| 389 vmull.s16 q9, d19, d31 | |
| 390 | |
| 391 ; s7 = cospi_6_64 * x6 - cospi_26_64 * x7; | |
| 392 vmlsl.s16 q10, d28, d30 | |
| 393 vmlsl.s16 q9, d29, d30 | |
| 394 | |
| 395 ; (s3 + s7) | |
| 396 vadd.s32 q14, q2, q10 | |
| 397 vadd.s32 q15, q6, q9 | |
| 398 | |
| 399 ; (s3 - s7) | |
| 400 vsub.s32 q2, q2, q10 | |
| 401 vsub.s32 q6, q6, q9 | |
| 402 | |
| 403 ; x3 = dct_const_round_shift(s3 + s7); | |
| 404 vqrshrn.s32 d28, q14, #14 ; >> 14 | |
| 405 vqrshrn.s32 d29, q15, #14 ; >> 14 | |
| 406 | |
| 407 ; x7 = dct_const_round_shift(s3 - s7); | |
| 408 vqrshrn.s32 d4, q2, #14 ; >> 14 | |
| 409 vqrshrn.s32 d5, q6, #14 ; >> 14 | |
| 410 | |
| 411 ; (s2 + s6) | |
| 412 vadd.s32 q9, q4, q0 | |
| 413 vadd.s32 q10, q5, q13 | |
| 414 | |
| 415 ; (s2 - s6) | |
| 416 vsub.s32 q4, q4, q0 | |
| 417 vsub.s32 q5, q5, q13 | |
| 418 | |
| 419 vdup.16 d30, r8 ; duplicate cospi_8_64 | |
| 420 vdup.16 d31, r9 ; duplicate cospi_24_64 | |
| 421 | |
| 422 ; x2 = dct_const_round_shift(s2 + s6); | |
| 423 vqrshrn.s32 d18, q9, #14 ; >> 14 | |
| 424 vqrshrn.s32 d19, q10, #14 ; >> 14 | |
| 425 | |
| 426 ; x6 = dct_const_round_shift(s2 - s6); | |
| 427 vqrshrn.s32 d8, q4, #14 ; >> 14 | |
| 428 vqrshrn.s32 d9, q5, #14 ; >> 14 | |
| 429 | |
| 430 ; cospi_8_64 * x4 | |
| 431 vmull.s16 q5, d2, d30 | |
| 432 vmull.s16 q6, d3, d30 | |
| 433 | |
| 434 ; cospi_24_64 * x4 | |
| 435 vmull.s16 q7, d2, d31 | |
| 436 vmull.s16 q0, d3, d31 | |
| 437 | |
| 438 ; s4 = cospi_8_64 * x4 + cospi_24_64 * x5; | |
| 439 vmlal.s16 q5, d6, d31 | |
| 440 vmlal.s16 q6, d7, d31 | |
| 441 | |
| 442 ; s5 = cospi_24_64 * x4 - cospi_8_64 * x5; | |
| 443 vmlsl.s16 q7, d6, d30 | |
| 444 vmlsl.s16 q0, d7, d30 | |
| 445 | |
| 446 ; cospi_8_64 * x7 | |
| 447 vmull.s16 q1, d4, d30 | |
| 448 vmull.s16 q3, d5, d30 | |
| 449 | |
| 450 ; cospi_24_64 * x7 | |
| 451 vmull.s16 q10, d4, d31 | |
| 452 vmull.s16 q2, d5, d31 | |
| 453 | |
| 454 ; s6 = -cospi_24_64 * x6 + cospi_8_64 * x7; | |
| 455 vmlsl.s16 q1, d8, d31 | |
| 456 vmlsl.s16 q3, d9, d31 | |
| 457 | |
| 458 ; s7 = cospi_8_64 * x6 + cospi_24_64 * x7; | |
| 459 vmlal.s16 q10, d8, d30 | |
| 460 vmlal.s16 q2, d9, d30 | |
| 461 | |
| 462 vadd.s16 q8, q11, q9 ; x0 = s0 + s2; | |
| 463 | |
| 464 vsub.s16 q11, q11, q9 ; x2 = s0 - s2; | |
| 465 | |
| 466 vadd.s16 q4, q12, q14 ; x1 = s1 + s3; | |
| 467 | |
| 468 vsub.s16 q12, q12, q14 ; x3 = s1 - s3; | |
| 469 | |
| 470 ; (s4 + s6) | |
| 471 vadd.s32 q14, q5, q1 | |
| 472 vadd.s32 q15, q6, q3 | |
| 473 | |
| 474 ; (s4 - s6) | |
| 475 vsub.s32 q5, q5, q1 | |
| 476 vsub.s32 q6, q6, q3 | |
| 477 | |
| 478 ; x4 = dct_const_round_shift(s4 + s6); | |
| 479 vqrshrn.s32 d18, q14, #14 ; >> 14 | |
| 480 vqrshrn.s32 d19, q15, #14 ; >> 14 | |
| 481 | |
| 482 ; x6 = dct_const_round_shift(s4 - s6); | |
| 483 vqrshrn.s32 d10, q5, #14 ; >> 14 | |
| 484 vqrshrn.s32 d11, q6, #14 ; >> 14 | |
| 485 | |
| 486 ; (s5 + s7) | |
| 487 vadd.s32 q1, q7, q10 | |
| 488 vadd.s32 q3, q0, q2 | |
| 489 | |
| 490 ; (s5 - s7)) | |
| 491 vsub.s32 q7, q7, q10 | |
| 492 vsub.s32 q0, q0, q2 | |
| 493 | |
| 494 ; x5 = dct_const_round_shift(s5 + s7); | |
| 495 vqrshrn.s32 d28, q1, #14 ; >> 14 | |
| 496 vqrshrn.s32 d29, q3, #14 ; >> 14 | |
| 497 | |
| 498 ; x7 = dct_const_round_shift(s5 - s7); | |
| 499 vqrshrn.s32 d14, q7, #14 ; >> 14 | |
| 500 vqrshrn.s32 d15, q0, #14 ; >> 14 | |
| 501 | |
| 502 vdup.16 d30, r12 ; duplicate cospi_16_64 | |
| 503 | |
| 504 ; cospi_16_64 * x2 | |
| 505 vmull.s16 q2, d22, d30 | |
| 506 vmull.s16 q3, d23, d30 | |
| 507 | |
| 508 ; cospi_6_64 * x6 | |
| 509 vmull.s16 q13, d22, d30 | |
| 510 vmull.s16 q1, d23, d30 | |
| 511 | |
| 512 ; cospi_16_64 * x2 + cospi_16_64 * x3; | |
| 513 vmlal.s16 q2, d24, d30 | |
| 514 vmlal.s16 q3, d25, d30 | |
| 515 | |
| 516 ; cospi_16_64 * x2 - cospi_16_64 * x3; | |
| 517 vmlsl.s16 q13, d24, d30 | |
| 518 vmlsl.s16 q1, d25, d30 | |
| 519 | |
| 520 ; x2 = dct_const_round_shift(s2); | |
| 521 vqrshrn.s32 d4, q2, #14 ; >> 14 | |
| 522 vqrshrn.s32 d5, q3, #14 ; >> 14 | |
| 523 | |
| 524 ;x3 = dct_const_round_shift(s3); | |
| 525 vqrshrn.s32 d24, q13, #14 ; >> 14 | |
| 526 vqrshrn.s32 d25, q1, #14 ; >> 14 | |
| 527 | |
| 528 ; cospi_16_64 * x6 | |
| 529 vmull.s16 q13, d10, d30 | |
| 530 vmull.s16 q1, d11, d30 | |
| 531 | |
| 532 ; cospi_6_64 * x6 | |
| 533 vmull.s16 q11, d10, d30 | |
| 534 vmull.s16 q0, d11, d30 | |
| 535 | |
| 536 ; cospi_16_64 * x6 + cospi_16_64 * x7; | |
| 537 vmlal.s16 q13, d14, d30 | |
| 538 vmlal.s16 q1, d15, d30 | |
| 539 | |
| 540 ; cospi_16_64 * x6 - cospi_16_64 * x7; | |
| 541 vmlsl.s16 q11, d14, d30 | |
| 542 vmlsl.s16 q0, d15, d30 | |
| 543 | |
| 544 ; x6 = dct_const_round_shift(s6); | |
| 545 vqrshrn.s32 d20, q13, #14 ; >> 14 | |
| 546 vqrshrn.s32 d21, q1, #14 ; >> 14 | |
| 547 | |
| 548 ;x7 = dct_const_round_shift(s7); | |
| 549 vqrshrn.s32 d12, q11, #14 ; >> 14 | |
| 550 vqrshrn.s32 d13, q0, #14 ; >> 14 | |
| 551 | |
| 552 vdup.16 q5, r10 ; duplicate 0 | |
| 553 | |
| 554 vsub.s16 q9, q5, q9 ; output[1] = -x4; | |
| 555 vsub.s16 q11, q5, q2 ; output[3] = -x2; | |
| 556 vsub.s16 q13, q5, q6 ; output[5] = -x7; | |
| 557 vsub.s16 q15, q5, q4 ; output[7] = -x1; | |
| 558 MEND | |
| 559 | |
| 560 | |
| 561 AREA Block, CODE, READONLY ; name this block of code | |
| 562 ;void vp9_iht8x8_64_add_neon(int16_t *input, uint8_t *dest, | |
| 563 ; int dest_stride, int tx_type) | |
| 564 ; | |
| 565 ; r0 int16_t input | |
| 566 ; r1 uint8_t *dest | |
| 567 ; r2 int dest_stride | |
| 568 ; r3 int tx_type) | |
| 569 ; This function will only handle tx_type of 1,2,3. | |
| 570 |vp9_iht8x8_64_add_neon| PROC | |
| 571 | |
| 572 ; load the inputs into d16-d19 | |
| 573 vld1.s16 {q8,q9}, [r0]! | |
| 574 vld1.s16 {q10,q11}, [r0]! | |
| 575 vld1.s16 {q12,q13}, [r0]! | |
| 576 vld1.s16 {q14,q15}, [r0]! | |
| 577 | |
| 578 push {r0-r10} | |
| 579 vpush {d8-d15} | |
| 580 | |
| 581 ; transpose the input data | |
| 582 TRANSPOSE8X8 | |
| 583 | |
| 584 ; decide the type of transform | |
| 585 cmp r3, #2 | |
| 586 beq idct_iadst | |
| 587 cmp r3, #3 | |
| 588 beq iadst_iadst | |
| 589 | |
| 590 iadst_idct | |
| 591 ; generate IDCT constants | |
| 592 GENERATE_IDCT_CONSTANTS | |
| 593 | |
| 594 ; first transform rows | |
| 595 IDCT8x8_1D | |
| 596 | |
| 597 ; transpose the matrix | |
| 598 TRANSPOSE8X8 | |
| 599 | |
| 600 ; generate IADST constants | |
| 601 GENERATE_IADST_CONSTANTS | |
| 602 | |
| 603 ; then transform columns | |
| 604 IADST8X8_1D | |
| 605 | |
| 606 b end_vp9_iht8x8_64_add_neon | |
| 607 | |
| 608 idct_iadst | |
| 609 ; generate IADST constants | |
| 610 GENERATE_IADST_CONSTANTS | |
| 611 | |
| 612 ; first transform rows | |
| 613 IADST8X8_1D | |
| 614 | |
| 615 ; transpose the matrix | |
| 616 TRANSPOSE8X8 | |
| 617 | |
| 618 ; generate IDCT constants | |
| 619 GENERATE_IDCT_CONSTANTS | |
| 620 | |
| 621 ; then transform columns | |
| 622 IDCT8x8_1D | |
| 623 | |
| 624 b end_vp9_iht8x8_64_add_neon | |
| 625 | |
| 626 iadst_iadst | |
| 627 ; generate IADST constants | |
| 628 GENERATE_IADST_CONSTANTS | |
| 629 | |
| 630 ; first transform rows | |
| 631 IADST8X8_1D | |
| 632 | |
| 633 ; transpose the matrix | |
| 634 TRANSPOSE8X8 | |
| 635 | |
| 636 ; then transform columns | |
| 637 IADST8X8_1D | |
| 638 | |
| 639 end_vp9_iht8x8_64_add_neon | |
| 640 vpop {d8-d15} | |
| 641 pop {r0-r10} | |
| 642 | |
| 643 ; ROUND_POWER_OF_TWO(temp_out[j], 5) | |
| 644 vrshr.s16 q8, q8, #5 | |
| 645 vrshr.s16 q9, q9, #5 | |
| 646 vrshr.s16 q10, q10, #5 | |
| 647 vrshr.s16 q11, q11, #5 | |
| 648 vrshr.s16 q12, q12, #5 | |
| 649 vrshr.s16 q13, q13, #5 | |
| 650 vrshr.s16 q14, q14, #5 | |
| 651 vrshr.s16 q15, q15, #5 | |
| 652 | |
| 653 ; save dest pointer | |
| 654 mov r0, r1 | |
| 655 | |
| 656 ; load destination data | |
| 657 vld1.64 {d0}, [r1], r2 | |
| 658 vld1.64 {d1}, [r1], r2 | |
| 659 vld1.64 {d2}, [r1], r2 | |
| 660 vld1.64 {d3}, [r1], r2 | |
| 661 vld1.64 {d4}, [r1], r2 | |
| 662 vld1.64 {d5}, [r1], r2 | |
| 663 vld1.64 {d6}, [r1], r2 | |
| 664 vld1.64 {d7}, [r1] | |
| 665 | |
| 666 ; ROUND_POWER_OF_TWO(temp_out[j], 5) + dest[j * dest_stride + i] | |
| 667 vaddw.u8 q8, q8, d0 | |
| 668 vaddw.u8 q9, q9, d1 | |
| 669 vaddw.u8 q10, q10, d2 | |
| 670 vaddw.u8 q11, q11, d3 | |
| 671 vaddw.u8 q12, q12, d4 | |
| 672 vaddw.u8 q13, q13, d5 | |
| 673 vaddw.u8 q14, q14, d6 | |
| 674 vaddw.u8 q15, q15, d7 | |
| 675 | |
| 676 ; clip_pixel | |
| 677 vqmovun.s16 d0, q8 | |
| 678 vqmovun.s16 d1, q9 | |
| 679 vqmovun.s16 d2, q10 | |
| 680 vqmovun.s16 d3, q11 | |
| 681 vqmovun.s16 d4, q12 | |
| 682 vqmovun.s16 d5, q13 | |
| 683 vqmovun.s16 d6, q14 | |
| 684 vqmovun.s16 d7, q15 | |
| 685 | |
| 686 ; store the data | |
| 687 vst1.64 {d0}, [r0], r2 | |
| 688 vst1.64 {d1}, [r0], r2 | |
| 689 vst1.64 {d2}, [r0], r2 | |
| 690 vst1.64 {d3}, [r0], r2 | |
| 691 vst1.64 {d4}, [r0], r2 | |
| 692 vst1.64 {d5}, [r0], r2 | |
| 693 vst1.64 {d6}, [r0], r2 | |
| 694 vst1.64 {d7}, [r0], r2 | |
| 695 bx lr | |
| 696 ENDP ; |vp9_iht8x8_64_add_neon| | |
| 697 | |
| 698 END | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 812033011:
libvpx: Pull from upstream (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/