1 /*
2 * Copyright (c) 2017 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 "./vpx_dsp_rtcd.h"
15 #include "vpx_dsp/arm/mem_neon.h"
16
vpx_comp_avg_pred_neon(uint8_t * comp,const uint8_t * pred,int width,int height,const uint8_t * ref,int ref_stride)17 void vpx_comp_avg_pred_neon(uint8_t *comp, const uint8_t *pred, int width,
18 int height, const uint8_t *ref, int ref_stride) {
19 if (width > 8) {
20 int x, y = height;
21 do {
22 for (x = 0; x < width; x += 16) {
23 const uint8x16_t p = vld1q_u8(pred + x);
24 const uint8x16_t r = vld1q_u8(ref + x);
25 const uint8x16_t avg = vrhaddq_u8(p, r);
26 vst1q_u8(comp + x, avg);
27 }
28 comp += width;
29 pred += width;
30 ref += ref_stride;
31 } while (--y);
32 } else if (width == 8) {
33 int i = width * height;
34 do {
35 const uint8x16_t p = vld1q_u8(pred);
36 uint8x16_t r;
37 const uint8x8_t r_0 = vld1_u8(ref);
38 const uint8x8_t r_1 = vld1_u8(ref + ref_stride);
39 r = vcombine_u8(r_0, r_1);
40 ref += 2 * ref_stride;
41 r = vrhaddq_u8(r, p);
42 vst1q_u8(comp, r);
43
44 pred += 16;
45 comp += 16;
46 i -= 16;
47 } while (i);
48 } else {
49 int i = width * height;
50 assert(width == 4);
51 do {
52 const uint8x16_t p = vld1q_u8(pred);
53 uint8x16_t r;
54
55 r = load_unaligned_u8q(ref, ref_stride);
56 ref += 4 * ref_stride;
57 r = vrhaddq_u8(r, p);
58 vst1q_u8(comp, r);
59
60 pred += 16;
61 comp += 16;
62 i -= 16;
63 } while (i);
64 }
65 }
66