1 /*
2 * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.com>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21 #include <string.h>
22
23 #include "libavutil/intreadwrite.h"
24
25 #include "libavcodec/avcodec.h"
26
27 #include "libavcodec/hevcdsp.h"
28
29 #include "checkasm.h"
30
31 static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
32 static const uint32_t sao_size[5] = {8, 16, 32, 48, 64};
33
34 #define SIZEOF_PIXEL ((bit_depth + 7) / 8)
35 #define PIXEL_STRIDE (2*MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride
36 #define BUF_SIZE (PIXEL_STRIDE * (64+2) * 2) //+2 for top and bottom row, *2 for high bit depth
37 #define OFFSET_THRESH (1 << (bit_depth - 5))
38 #define OFFSET_LENGTH 5
39
40 #define randomize_buffers(buf0, buf1, size) \
41 do { \
42 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
43 int k; \
44 for (k = 0; k < size; k += 4) { \
45 uint32_t r = rnd() & mask; \
46 AV_WN32A(buf0 + k, r); \
47 AV_WN32A(buf1 + k, r); \
48 } \
49 } while (0)
50
51 #define randomize_buffers2(buf, size) \
52 do { \
53 uint32_t max_offset = OFFSET_THRESH; \
54 int k; \
55 if (bit_depth == 8) { \
56 for (k = 0; k < size; k++) { \
57 uint8_t r = rnd() % max_offset; \
58 buf[k] = r; \
59 } \
60 } else { \
61 for (k = 0; k < size; k++) { \
62 uint16_t r = rnd() % max_offset; \
63 buf[k] = r; \
64 } \
65 } \
66 } while (0)
67
check_sao_band(HEVCDSPContext h,int bit_depth)68 static void check_sao_band(HEVCDSPContext h, int bit_depth)
69 {
70 int i;
71 LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
72 LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
73 LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
74 LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
75 int16_t offset_val[OFFSET_LENGTH];
76 int left_class = rnd()%32;
77
78 for (i = 0; i <= 4; i++) {
79 int block_size = sao_size[i];
80 ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
81 declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
82 int16_t *sao_offset_val, int sao_left_class, int width, int height);
83
84 randomize_buffers(src0, src1, BUF_SIZE);
85 randomize_buffers2(offset_val, OFFSET_LENGTH);
86 memset(dst0, 0, BUF_SIZE);
87 memset(dst1, 0, BUF_SIZE);
88
89 if (check_func(h.sao_band_filter[i], "hevc_sao_band_%dx%d_%d", block_size, block_size, bit_depth)) {
90 call_ref(dst0, src0, stride, stride, offset_val, left_class, block_size, block_size);
91 call_new(dst1, src1, stride, stride, offset_val, left_class, block_size, block_size);
92 if (memcmp(dst0, dst1, BUF_SIZE))
93 fail();
94 bench_new(dst1, src1, stride, stride, offset_val, left_class, block_size, block_size);
95 }
96 }
97 }
98
check_sao_edge(HEVCDSPContext h,int bit_depth)99 static void check_sao_edge(HEVCDSPContext h, int bit_depth)
100 {
101 int i;
102 LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
103 LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
104 LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
105 LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
106 int16_t offset_val[OFFSET_LENGTH];
107 int eo = rnd()%4;
108
109 for (i = 0; i <= 4; i++) {
110 int block_size = sao_size[i];
111 ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
112 int offset = (AV_INPUT_BUFFER_PADDING_SIZE + PIXEL_STRIDE)*SIZEOF_PIXEL;
113 declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst,
114 int16_t *sao_offset_val, int eo, int width, int height);
115
116 randomize_buffers(src0, src1, BUF_SIZE);
117 randomize_buffers2(offset_val, OFFSET_LENGTH);
118 memset(dst0, 0, BUF_SIZE);
119 memset(dst1, 0, BUF_SIZE);
120
121 if (check_func(h.sao_edge_filter[i], "hevc_sao_edge_%dx%d_%d", block_size, block_size, bit_depth)) {
122 call_ref(dst0, src0 + offset, stride, offset_val, eo, block_size, block_size);
123 call_new(dst1, src1 + offset, stride, offset_val, eo, block_size, block_size);
124 if (memcmp(dst0, dst1, BUF_SIZE))
125 fail();
126 bench_new(dst1, src1 + offset, stride, offset_val, eo, block_size, block_size);
127 }
128 }
129 }
130
checkasm_check_hevc_sao(void)131 void checkasm_check_hevc_sao(void)
132 {
133 int bit_depth;
134
135 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
136 HEVCDSPContext h;
137
138 ff_hevc_dsp_init(&h, bit_depth);
139 check_sao_band(h, bit_depth);
140 }
141 report("sao_band");
142
143 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
144 HEVCDSPContext h;
145
146 ff_hevc_dsp_init(&h, bit_depth);
147 check_sao_edge(h, bit_depth);
148 }
149 report("sao_edge");
150 }
151