vp9_reconintra_neon.c 19.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12
/*
 *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <arm_neon.h>

13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
#include "./vp9_rtcd.h"
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"

//------------------------------------------------------------------------------
// DC 8x8

// 'do_above' and 'do_left' facilitate branch removal when inlined.
static INLINE void dc_8x8(uint8_t *dst, ptrdiff_t stride,
                          const uint8_t *above, const uint8_t *left,
                          int do_above, int do_left) {
  uint16x8_t sum_top;
  uint16x8_t sum_left;
  uint8x8_t dc0;

  if (do_above) {
    const uint8x8_t A = vld1_u8(above);  // top row
    const uint16x4_t p0 = vpaddl_u8(A);  // cascading summation of the top
    const uint16x4_t p1 = vpadd_u16(p0, p0);
    const uint16x4_t p2 = vpadd_u16(p1, p1);
    sum_top = vcombine_u16(p2, p2);
  }

  if (do_left) {
    const uint8x8_t L = vld1_u8(left);  // left border
    const uint16x4_t p0 = vpaddl_u8(L);  // cascading summation of the left
    const uint16x4_t p1 = vpadd_u16(p0, p0);
    const uint16x4_t p2 = vpadd_u16(p1, p1);
    sum_left = vcombine_u16(p2, p2);
  }

  if (do_above && do_left) {
    const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
    dc0 = vrshrn_n_u16(sum, 4);
  } else if (do_above) {
    dc0 = vrshrn_n_u16(sum_top, 3);
  } else if (do_left) {
    dc0 = vrshrn_n_u16(sum_left, 3);
  } else {
    dc0 = vdup_n_u8(0x80);
  }

  {
    const uint8x8_t dc = vdup_lane_u8(dc0, 0);
    int i;
    for (i = 0; i < 8; ++i) {
      vst1_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc));
    }
  }
}

void vp9_dc_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
                               const uint8_t *above, const uint8_t *left) {
  dc_8x8(dst, stride, above, left, 1, 1);
}

void vp9_dc_left_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
                                    const uint8_t *above, const uint8_t *left) {
  (void)above;
  dc_8x8(dst, stride, NULL, left, 0, 1);
}

void vp9_dc_top_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
                                   const uint8_t *above, const uint8_t *left) {
  (void)left;
  dc_8x8(dst, stride, above, NULL, 1, 0);
}

void vp9_dc_128_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
                                   const uint8_t *above, const uint8_t *left) {
  (void)above;
  (void)left;
  dc_8x8(dst, stride, NULL, NULL, 0, 0);
}

88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
//------------------------------------------------------------------------------
// DC 16x16

// 'do_above' and 'do_left' facilitate branch removal when inlined.
static INLINE void dc_16x16(uint8_t *dst, ptrdiff_t stride,
                            const uint8_t *above, const uint8_t *left,
                            int do_above, int do_left) {
  uint16x8_t sum_top;
  uint16x8_t sum_left;
  uint8x8_t dc0;

  if (do_above) {
    const uint8x16_t A = vld1q_u8(above);  // top row
    const uint16x8_t p0 = vpaddlq_u8(A);  // cascading summation of the top
    const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
    const uint16x4_t p2 = vpadd_u16(p1, p1);
    const uint16x4_t p3 = vpadd_u16(p2, p2);
    sum_top = vcombine_u16(p3, p3);
  }

  if (do_left) {
    const uint8x16_t L = vld1q_u8(left);  // left row
    const uint16x8_t p0 = vpaddlq_u8(L);  // cascading summation of the left
    const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
    const uint16x4_t p2 = vpadd_u16(p1, p1);
    const uint16x4_t p3 = vpadd_u16(p2, p2);
    sum_left = vcombine_u16(p3, p3);
  }

  if (do_above && do_left) {
    const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
    dc0 = vrshrn_n_u16(sum, 5);
  } else if (do_above) {
    dc0 = vrshrn_n_u16(sum_top, 4);
  } else if (do_left) {
    dc0 = vrshrn_n_u16(sum_left, 4);
  } else {
    dc0 = vdup_n_u8(0x80);
  }

  {
    const uint8x16_t dc = vdupq_lane_u8(dc0, 0);
    int i;
    for (i = 0; i < 16; ++i) {
      vst1q_u8(dst + i * stride, dc);
    }
  }
}

void vp9_dc_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
                                 const uint8_t *above, const uint8_t *left) {
  dc_16x16(dst, stride, above, left, 1, 1);
}

void vp9_dc_left_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
                                      const uint8_t *above,
                                      const uint8_t *left) {
  (void)above;
  dc_16x16(dst, stride, NULL, left, 0, 1);
}

void vp9_dc_top_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
                                     const uint8_t *above,
                                     const uint8_t *left) {
  (void)left;
  dc_16x16(dst, stride, above, NULL, 1, 0);
}

void vp9_dc_128_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
                                     const uint8_t *above,
                                     const uint8_t *left) {
  (void)above;
  (void)left;
  dc_16x16(dst, stride, NULL, NULL, 0, 0);
}

164 165
#if !HAVE_NEON_ASM

166 167 168 169 170
void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
                              const uint8_t *above, const uint8_t *left) {
  int i;
  uint32x2_t d0u32 = vdup_n_u32(0);
  (void)left;
171

172 173 174
  d0u32 = vld1_lane_u32((const uint32_t *)above, d0u32, 0);
  for (i = 0; i < 4; i++, dst += y_stride)
    vst1_lane_u32((uint32_t *)dst, d0u32, 0);
175 176
}

177 178 179 180 181
void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
                              const uint8_t *above, const uint8_t *left) {
  int i;
  uint8x8_t d0u8 = vdup_n_u8(0);
  (void)left;
182

183 184 185
  d0u8 = vld1_u8(above);
  for (i = 0; i < 8; i++, dst += y_stride)
    vst1_u8(dst, d0u8);
186 187
}

188 189 190 191 192
void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
                                const uint8_t *above, const uint8_t *left) {
  int i;
  uint8x16_t q0u8 = vdupq_n_u8(0);
  (void)left;
193

194 195 196
  q0u8 = vld1q_u8(above);
  for (i = 0; i < 16; i++, dst += y_stride)
    vst1q_u8(dst, q0u8);
197 198
}

199 200 201 202 203 204
void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
                                const uint8_t *above, const uint8_t *left) {
  int i;
  uint8x16_t q0u8 = vdupq_n_u8(0);
  uint8x16_t q1u8 = vdupq_n_u8(0);
  (void)left;
205

206 207 208 209 210 211
  q0u8 = vld1q_u8(above);
  q1u8 = vld1q_u8(above + 16);
  for (i = 0; i < 32; i++, dst += y_stride) {
    vst1q_u8(dst, q0u8);
    vst1q_u8(dst + 16, q1u8);
  }
212 213
}

214 215 216 217 218
void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
                              const uint8_t *above, const uint8_t *left) {
  uint8x8_t d0u8 = vdup_n_u8(0);
  uint32x2_t d1u32 = vdup_n_u32(0);
  (void)above;
219

220
  d1u32 = vld1_lane_u32((const uint32_t *)left, d1u32, 0);
221

222 223 224 225 226 227 228 229 230 231 232
  d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0);
  vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1);
  vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2);
  vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3);
  vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
233 234
}

235 236 237 238 239
void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
                              const uint8_t *above, const uint8_t *left) {
  uint8x8_t d0u8 = vdup_n_u8(0);
  uint64x1_t d1u64 = vdup_n_u64(0);
  (void)above;
240

241
  d1u64 = vld1_u64((const uint64_t *)left);
242

243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6);
  vst1_u8(dst, d0u8);
  dst += y_stride;
  d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7);
  vst1_u8(dst, d0u8);
}

void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
                                const uint8_t *above, const uint8_t *left) {
  int j;
  uint8x8_t d2u8 = vdup_n_u8(0);
  uint8x16_t q0u8 = vdupq_n_u8(0);
  uint8x16_t q1u8 = vdupq_n_u8(0);
  (void)above;

  q1u8 = vld1q_u8(left);
  d2u8 = vget_low_u8(q1u8);
  for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {
    q0u8 = vdupq_lane_u8(d2u8, 0);
    vst1q_u8(dst, q0u8);
281
    dst += y_stride;
282 283
    q0u8 = vdupq_lane_u8(d2u8, 1);
    vst1q_u8(dst, q0u8);
284
    dst += y_stride;
285 286
    q0u8 = vdupq_lane_u8(d2u8, 2);
    vst1q_u8(dst, q0u8);
287
    dst += y_stride;
288 289
    q0u8 = vdupq_lane_u8(d2u8, 3);
    vst1q_u8(dst, q0u8);
290
    dst += y_stride;
291 292
    q0u8 = vdupq_lane_u8(d2u8, 4);
    vst1q_u8(dst, q0u8);
293
    dst += y_stride;
294 295
    q0u8 = vdupq_lane_u8(d2u8, 5);
    vst1q_u8(dst, q0u8);
296
    dst += y_stride;
297 298
    q0u8 = vdupq_lane_u8(d2u8, 6);
    vst1q_u8(dst, q0u8);
299
    dst += y_stride;
300 301 302 303
    q0u8 = vdupq_lane_u8(d2u8, 7);
    vst1q_u8(dst, q0u8);
    dst += y_stride;
  }
304 305
}

306 307 308 309 310 311 312
void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
                                const uint8_t *above, const uint8_t *left) {
  int j, k;
  uint8x8_t d2u8 = vdup_n_u8(0);
  uint8x16_t q0u8 = vdupq_n_u8(0);
  uint8x16_t q1u8 = vdupq_n_u8(0);
  (void)above;
313

314
  for (k = 0; k < 2; k++, left += 16) {
315 316 317
    q1u8 = vld1q_u8(left);
    d2u8 = vget_low_u8(q1u8);
    for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
      q0u8 = vdupq_lane_u8(d2u8, 0);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 1);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 2);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 3);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 4);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 5);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 6);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
      q0u8 = vdupq_lane_u8(d2u8, 7);
      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q0u8);
      dst += y_stride;
350
    }
351
  }
352 353
}

354 355 356 357 358 359 360
void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
                               const uint8_t *above, const uint8_t *left) {
  int i;
  uint16x8_t q1u16, q3u16;
  int16x8_t q1s16;
  uint8x8_t d0u8 = vdup_n_u8(0);
  uint32x2_t d2u32 = vdup_n_u32(0);
361

362 363 364 365 366 367 368 369 370 371
  d0u8 = vdup_n_u8(above[-1]);
  d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0);
  q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8);
  for (i = 0; i < 4; i++, dst += y_stride) {
    q1u16 = vdupq_n_u16((uint16_t)left[i]);
    q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16),
                      vreinterpretq_s16_u16(q3u16));
    d0u8 = vqmovun_s16(q1s16);
    vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
  }
372 373
}

374 375 376 377 378 379 380
void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
                               const uint8_t *above, const uint8_t *left) {
  int j;
  uint16x8_t q0u16, q3u16, q10u16;
  int16x8_t q0s16;
  uint16x4_t d20u16;
  uint8x8_t d0u8, d2u8, d30u8;
381

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
  d0u8 = vdup_n_u8(above[-1]);
  d30u8 = vld1_u8(left);
  d2u8 = vld1_u8(above);
  q10u16 = vmovl_u8(d30u8);
  q3u16 = vsubl_u8(d2u8, d0u8);
  d20u16 = vget_low_u16(q10u16);
  for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {
    q0u16 = vdupq_lane_u16(d20u16, 0);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += y_stride;
    q0u16 = vdupq_lane_u16(d20u16, 1);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += y_stride;
    q0u16 = vdupq_lane_u16(d20u16, 2);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += y_stride;
    q0u16 = vdupq_lane_u16(d20u16, 3);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += y_stride;
  }
414 415
}

416 417 418 419 420 421 422 423
void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
                                 const uint8_t *above, const uint8_t *left) {
  int j, k;
  uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16;
  uint8x16_t q0u8, q1u8;
  int16x8_t q0s16, q1s16, q8s16, q11s16;
  uint16x4_t d20u16;
  uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8;
424

425 426 427 428 429 430 431
  q0u8 = vdupq_n_u8(above[-1]);
  q1u8 = vld1q_u8(above);
  q2u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
  q3u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
  for (k = 0; k < 2; k++, left += 8) {
    d18u8 = vld1_u8(left);
    q10u16 = vmovl_u8(d18u8);
432 433
    d20u16 = vget_low_u16(q10u16);
    for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {
434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453
      q0u16 = vdupq_lane_u16(d20u16, 0);
      q8u16 = vdupq_lane_u16(d20u16, 1);
      q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                        vreinterpretq_s16_u16(q2u16));
      q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                        vreinterpretq_s16_u16(q3u16));
      q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
                         vreinterpretq_s16_u16(q2u16));
      q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
                        vreinterpretq_s16_u16(q3u16));
      d2u8 = vqmovun_s16(q1s16);
      d3u8 = vqmovun_s16(q0s16);
      d22u8 = vqmovun_s16(q11s16);
      d23u8 = vqmovun_s16(q8s16);
      vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
      vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
      dst += y_stride;
      vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
      vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
      dst += y_stride;
454

455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
      q0u16 = vdupq_lane_u16(d20u16, 2);
      q8u16 = vdupq_lane_u16(d20u16, 3);
      q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                        vreinterpretq_s16_u16(q2u16));
      q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                        vreinterpretq_s16_u16(q3u16));
      q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
                         vreinterpretq_s16_u16(q2u16));
      q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
                        vreinterpretq_s16_u16(q3u16));
      d2u8 = vqmovun_s16(q1s16);
      d3u8 = vqmovun_s16(q0s16);
      d22u8 = vqmovun_s16(q11s16);
      d23u8 = vqmovun_s16(q8s16);
      vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
      vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
      dst += y_stride;
      vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
      vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
      dst += y_stride;
475
    }
476
  }
477 478
}

479 480 481 482 483 484 485 486
void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
                                 const uint8_t *above, const uint8_t *left) {
  int j, k;
  uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16;
  uint8x16_t q0u8, q1u8, q2u8;
  int16x8_t q12s16, q13s16, q14s16, q15s16;
  uint16x4_t d6u16;
  uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8;
487

488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
  q0u8 = vdupq_n_u8(above[-1]);
  q1u8 = vld1q_u8(above);
  q2u8 = vld1q_u8(above + 16);
  q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
  q9u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
  q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q0u8));
  q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q0u8));
  for (k = 0; k < 4; k++, left += 8) {
    d26u8 = vld1_u8(left);
    q3u16 = vmovl_u8(d26u8);
    d6u16 = vget_low_u16(q3u16);
    for (j = 0; j < 2; j++, d6u16 = vget_high_u16(q3u16)) {
      q0u16 = vdupq_lane_u16(d6u16, 0);
      q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q8u16));
      q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q9u16));
      q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q10u16));
      q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q11u16));
      d0u8 = vqmovun_s16(q12s16);
      d1u8 = vqmovun_s16(q13s16);
      d2u8 = vqmovun_s16(q14s16);
      d3u8 = vqmovun_s16(q15s16);
      q0u8 = vcombine_u8(d0u8, d1u8);
      q1u8 = vcombine_u8(d2u8, d3u8);
      vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
      vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
      dst += y_stride;
518

519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536
      q0u16 = vdupq_lane_u16(d6u16, 1);
      q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q8u16));
      q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q9u16));
      q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q10u16));
      q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q11u16));
      d0u8 = vqmovun_s16(q12s16);
      d1u8 = vqmovun_s16(q13s16);
      d2u8 = vqmovun_s16(q14s16);
      d3u8 = vqmovun_s16(q15s16);
      q0u8 = vcombine_u8(d0u8, d1u8);
      q1u8 = vcombine_u8(d2u8, d3u8);
      vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
      vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
      dst += y_stride;
537

538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555
      q0u16 = vdupq_lane_u16(d6u16, 2);
      q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q8u16));
      q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q9u16));
      q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q10u16));
      q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q11u16));
      d0u8 = vqmovun_s16(q12s16);
      d1u8 = vqmovun_s16(q13s16);
      d2u8 = vqmovun_s16(q14s16);
      d3u8 = vqmovun_s16(q15s16);
      q0u8 = vcombine_u8(d0u8, d1u8);
      q1u8 = vcombine_u8(d2u8, d3u8);
      vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
      vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
      dst += y_stride;
556

557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574
      q0u16 = vdupq_lane_u16(d6u16, 3);
      q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q8u16));
      q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q9u16));
      q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q10u16));
      q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
                         vreinterpretq_s16_u16(q11u16));
      d0u8 = vqmovun_s16(q12s16);
      d1u8 = vqmovun_s16(q13s16);
      d2u8 = vqmovun_s16(q14s16);
      d3u8 = vqmovun_s16(q15s16);
      q0u8 = vcombine_u8(d0u8, d1u8);
      q1u8 = vcombine_u8(d2u8, d3u8);
      vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
      vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
      dst += y_stride;
575
    }
576
  }
577
}
578
#endif  // !HAVE_NEON_ASM