OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved. | 2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
| 11 #include <assert.h> |
| 12 |
11 #include "./vpx_dsp_rtcd.h" | 13 #include "./vpx_dsp_rtcd.h" |
12 #include "vpx_dsp/vpx_dsp_common.h" | 14 #include "vpx_dsp/vpx_dsp_common.h" |
13 #include "vpx_ports/mem.h" | 15 #include "vpx_ports/mem.h" |
14 | 16 |
15 void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, | 17 void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, |
16 uint8_t *dst, ptrdiff_t dst_stride, | 18 uint8_t *dst, ptrdiff_t dst_stride, |
17 const int16_t *filter_x, int x_step_q4, | 19 const int16_t *filter_x, int x_step_q4, |
18 const int16_t *filter_y, int y_step_q4, | 20 const int16_t *filter_y, int y_step_q4, |
19 int w, int h) { | 21 int w, int h) { |
20 /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the | 22 /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the |
21 * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4). | 23 * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4). |
22 */ | 24 */ |
23 DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); | 25 DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); |
24 | 26 |
25 // Account for the vertical phase needing 3 lines prior and 4 lines post | 27 // Account for the vertical phase needing 3 lines prior and 4 lines post |
26 int intermediate_height = h + 7; | 28 int intermediate_height = h + 7; |
27 | 29 |
28 if (x_step_q4 != 16 || y_step_q4 != 16) { | 30 assert(y_step_q4 == 16); |
29 vpx_convolve8_c(src, src_stride, | 31 assert(x_step_q4 == 16); |
30 dst, dst_stride, | |
31 filter_x, x_step_q4, | |
32 filter_y, y_step_q4, | |
33 w, h); | |
34 return; | |
35 } | |
36 | 32 |
37 /* Filter starting 3 lines back. The neon implementation will ignore the | 33 /* Filter starting 3 lines back. The neon implementation will ignore the |
38 * given height and filter a multiple of 4 lines. Since this goes in to | 34 * given height and filter a multiple of 4 lines. Since this goes in to |
39 * the temp buffer which has lots of extra room and is subsequently discarded | 35 * the temp buffer which has lots of extra room and is subsequently discarded |
40 * this is safe if somewhat less than ideal. | 36 * this is safe if somewhat less than ideal. |
41 */ | 37 */ |
42 vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, | 38 vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, |
43 temp, 64, | 39 temp, 64, |
44 filter_x, x_step_q4, filter_y, y_step_q4, | 40 filter_x, x_step_q4, filter_y, y_step_q4, |
45 w, intermediate_height); | 41 w, intermediate_height); |
46 | 42 |
47 /* Step into the temp buffer 3 lines to get the actual frame data */ | 43 /* Step into the temp buffer 3 lines to get the actual frame data */ |
48 vpx_convolve8_vert_neon(temp + 64 * 3, 64, | 44 vpx_convolve8_vert_neon(temp + 64 * 3, 64, |
49 dst, dst_stride, | 45 dst, dst_stride, |
50 filter_x, x_step_q4, filter_y, y_step_q4, | 46 filter_x, x_step_q4, filter_y, y_step_q4, |
51 w, h); | 47 w, h); |
52 } | 48 } |
53 | 49 |
54 void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride, | 50 void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride, |
55 uint8_t *dst, ptrdiff_t dst_stride, | 51 uint8_t *dst, ptrdiff_t dst_stride, |
56 const int16_t *filter_x, int x_step_q4, | 52 const int16_t *filter_x, int x_step_q4, |
57 const int16_t *filter_y, int y_step_q4, | 53 const int16_t *filter_y, int y_step_q4, |
58 int w, int h) { | 54 int w, int h) { |
59 DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); | 55 DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]); |
60 int intermediate_height = h + 7; | 56 int intermediate_height = h + 7; |
61 | 57 |
62 if (x_step_q4 != 16 || y_step_q4 != 16) { | 58 assert(y_step_q4 == 16); |
63 vpx_convolve8_avg_c(src, src_stride, | 59 assert(x_step_q4 == 16); |
64 dst, dst_stride, | |
65 filter_x, x_step_q4, | |
66 filter_y, y_step_q4, | |
67 w, h); | |
68 return; | |
69 } | |
70 | 60 |
71 /* This implementation has the same issues as above. In addition, we only want | 61 /* This implementation has the same issues as above. In addition, we only want |
72 * to average the values after both passes. | 62 * to average the values after both passes. |
73 */ | 63 */ |
74 vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, | 64 vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, |
75 temp, 64, | 65 temp, 64, |
76 filter_x, x_step_q4, filter_y, y_step_q4, | 66 filter_x, x_step_q4, filter_y, y_step_q4, |
77 w, intermediate_height); | 67 w, intermediate_height); |
78 vpx_convolve8_avg_vert_neon(temp + 64 * 3, | 68 vpx_convolve8_avg_vert_neon(temp + 64 * 3, |
79 64, dst, dst_stride, | 69 64, dst, dst_stride, |
80 filter_x, x_step_q4, filter_y, y_step_q4, | 70 filter_x, x_step_q4, filter_y, y_step_q4, |
81 w, h); | 71 w, h); |
82 } | 72 } |
OLD | NEW |