FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/tests/checkasm/vvc_mc.c
Date: 2024-11-20 23:03:26
Exec Total Coverage
Lines: 262 272 96.3%
Functions: 9 9 100.0%
Branches: 168 247 68.0%
Line Branch Exec Source
1 /*
2 * Copyright (c) 2023-2024 Nuo Mi
3 * Copyright (c) 2023-2024 Wu Jianhua
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20 */
21
22 #include <string.h>
23
24 #include "checkasm.h"
25 #include "libavcodec/vvc/ctu.h"
26 #include "libavcodec/vvc/data.h"
27 #include "libavcodec/vvc/dsp.h"
28
29 #include "libavutil/common.h"
30 #include "libavutil/intreadwrite.h"
31 #include "libavutil/mem_internal.h"
32
33 static const uint32_t pixel_mask[] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff, 0x3fff3fff, 0xffffffff };
34 static const int sizes[] = { 2, 4, 8, 16, 32, 64, 128 };
35
36 #define SIZEOF_PIXEL ((bit_depth + 7) / 8)
37 #define PIXEL_STRIDE (MAX_CTU_SIZE * 2)
38 #define EXTRA_BEFORE 3
39 #define EXTRA_AFTER 4
40 #define SRC_EXTRA (EXTRA_BEFORE + EXTRA_AFTER) * 2
41 #define SRC_BUF_SIZE (PIXEL_STRIDE + SRC_EXTRA) * (PIXEL_STRIDE + SRC_EXTRA)
42 #define DST_BUF_SIZE (MAX_CTU_SIZE * MAX_CTU_SIZE * 2)
43 #define SRC_OFFSET ((PIXEL_STRIDE + EXTRA_BEFORE * 2) * EXTRA_BEFORE)
44
45 #define randomize_buffers(buf0, buf1, size, mask) \
46 do { \
47 int k; \
48 for (k = 0; k < size; k += 4 / sizeof(*buf0)) { \
49 uint32_t r = rnd() & mask; \
50 AV_WN32A(buf0 + k, r); \
51 AV_WN32A(buf1 + k, r); \
52 } \
53 } while (0)
54
55 #define randomize_pixels(buf0, buf1, size) \
56 do { \
57 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
58 randomize_buffers(buf0, buf1, size, mask); \
59 } while (0)
60
61 #define randomize_avg_src(buf0, buf1, size) \
62 do { \
63 uint32_t mask = 0x3fff3fff; \
64 randomize_buffers(buf0, buf1, size, mask); \
65 } while (0)
66
67 #define randomize_prof_src(buf0, buf1, size) \
68 do { \
69 const int shift = 14 - bit_depth; \
70 const int mask16 = 0x3fff >> shift << shift; \
71 uint32_t mask = (mask16 << 16) | mask16; \
72 randomize_buffers(buf0, buf1, size, mask); \
73 } while (0)
74
75 13 static void check_put_vvc_luma(void)
76 {
77 13 LOCAL_ALIGNED_32(int16_t, dst0, [DST_BUF_SIZE / 2]);
78 13 LOCAL_ALIGNED_32(int16_t, dst1, [DST_BUF_SIZE / 2]);
79 13 LOCAL_ALIGNED_32(uint8_t, src0, [SRC_BUF_SIZE]);
80 13 LOCAL_ALIGNED_32(uint8_t, src1, [SRC_BUF_SIZE]);
81 VVCDSPContext c;
82
83 13 declare_func(void, int16_t *dst, const uint8_t *src, const ptrdiff_t src_stride,
84 const int height, const int8_t *hf, const int8_t *vf, const int width);
85
86
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
87
2/2
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 39 times.
 710814 randomize_pixels(src0, src1, SRC_BUF_SIZE);
88 39 ff_vvc_dsp_init(&c, bit_depth);
89
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int i = 0; i < 2; i++) {
90
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int j = 0; j < 2; j++) {
91
2/2
✓ Branch 0 taken 936 times.
✓ Branch 1 taken 156 times.
 1092 for (int h = 4; h <= MAX_CTU_SIZE; h *= 2) {
92
2/2
✓ Branch 0 taken 5616 times.
✓ Branch 1 taken 936 times.
 6552 for (int w = 4; w <= MAX_CTU_SIZE; w *= 2) {
93 5616 const int idx = av_log2(w) - 1;
94 5616 const int mx = rnd() % 16;
95 5616 const int my = rnd() % 16;
96 5616 const int8_t *hf = ff_vvc_inter_luma_filters[rnd() % 3][mx];
97 5616 const int8_t *vf = ff_vvc_inter_luma_filters[rnd() % 3][my];
98 const char *type;
99
4/5
✓ Branch 0 taken 1404 times.
✓ Branch 1 taken 1404 times.
✓ Branch 2 taken 1404 times.
✓ Branch 3 taken 1404 times.
✗ Branch 4 not taken.
 5616 switch ((j << 1) | i) {
100 1404 case 0: type = "put_luma_pixels"; break; // 0 0
101 1404 case 1: type = "put_luma_h"; break; // 0 1
102 1404 case 2: type = "put_luma_v"; break; // 1 0
103 1404 case 3: type = "put_luma_hv"; break; // 1 1
104 }
105
2/2
✓ Branch 3 taken 1110 times.
✓ Branch 4 taken 4506 times.
 5616 if (check_func(c.inter.put[LUMA][idx][j][i], "%s_%d_%dx%d", type, bit_depth, w, h)) {
106 1110 memset(dst0, 0, DST_BUF_SIZE);
107 1110 memset(dst1, 0, DST_BUF_SIZE);
108 1110 call_ref(dst0, src0 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
109 1110 call_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
110
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1110 times.
 1110 if (memcmp(dst0, dst1, DST_BUF_SIZE))
111 ✗ fail();
112
2/2
✓ Branch 0 taken 185 times.
✓ Branch 1 taken 925 times.
 1110 if (w == h)
113
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 185 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 185 bench_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
114 }
115 }
116 }
117 }
118 }
119 }
120 13 report("put_luma");
121 13 }
122
123 13 static void check_put_vvc_luma_uni(void)
124 {
125 13 LOCAL_ALIGNED_32(uint8_t, dst0, [DST_BUF_SIZE]);
126 13 LOCAL_ALIGNED_32(uint8_t, dst1, [DST_BUF_SIZE]);
127 13 LOCAL_ALIGNED_32(uint8_t, src0, [SRC_BUF_SIZE]);
128 13 LOCAL_ALIGNED_32(uint8_t, src1, [SRC_BUF_SIZE]);
129
130 VVCDSPContext c;
131 13 declare_func(void, uint8_t *dst, ptrdiff_t dststride,
132 const uint8_t *src, ptrdiff_t srcstride, int height,
133 const int8_t *hf, const int8_t *vf, int width);
134
135
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
136 39 ff_vvc_dsp_init(&c, bit_depth);
137
2/2
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 39 times.
 710814 randomize_pixels(src0, src1, SRC_BUF_SIZE);
138
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int i = 0; i < 2; i++) {
139
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int j = 0; j < 2; j++) {
140
2/2
✓ Branch 0 taken 936 times.
✓ Branch 1 taken 156 times.
 1092 for (int h = 4; h <= MAX_CTU_SIZE; h *= 2) {
141
2/2
✓ Branch 0 taken 5616 times.
✓ Branch 1 taken 936 times.
 6552 for (int w = 4; w <= MAX_CTU_SIZE; w *= 2) {
142 5616 const int idx = av_log2(w) - 1;
143 5616 const int mx = rnd() % VVC_INTER_LUMA_FACTS;
144 5616 const int my = rnd() % VVC_INTER_LUMA_FACTS;
145 5616 const int8_t *hf = ff_vvc_inter_luma_filters[rnd() % VVC_INTER_LUMA_FILTER_TYPES][mx];
146 5616 const int8_t *vf = ff_vvc_inter_luma_filters[rnd() % VVC_INTER_LUMA_FILTER_TYPES][my];
147 const char *type;
148
149
4/5
✓ Branch 0 taken 1404 times.
✓ Branch 1 taken 1404 times.
✓ Branch 2 taken 1404 times.
✓ Branch 3 taken 1404 times.
✗ Branch 4 not taken.
 5616 switch ((j << 1) | i) {
150 1404 case 0: type = "put_uni_pixels"; break; // 0 0
151 1404 case 1: type = "put_uni_h"; break; // 0 1
152 1404 case 2: type = "put_uni_v"; break; // 1 0
153 1404 case 3: type = "put_uni_hv"; break; // 1 1
154 }
155
156
2/2
✓ Branch 3 taken 1110 times.
✓ Branch 4 taken 4506 times.
 5616 if (check_func(c.inter.put_uni[LUMA][idx][j][i], "%s_luma_%d_%dx%d", type, bit_depth, w, h)) {
157 1110 memset(dst0, 0, DST_BUF_SIZE);
158 1110 memset(dst1, 0, DST_BUF_SIZE);
159 1110 call_ref(dst0, PIXEL_STRIDE, src0 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
160 1110 call_new(dst1, PIXEL_STRIDE, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
161
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1110 times.
 1110 if (memcmp(dst0, dst1, DST_BUF_SIZE))
162 ✗ fail();
163
2/2
✓ Branch 0 taken 185 times.
✓ Branch 1 taken 925 times.
 1110 if (w == h)
164
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 185 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 185 bench_new(dst1, PIXEL_STRIDE, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
165 }
166 }
167 }
168 }
169 }
170 }
171 13 report("put_uni_luma");
172 13 }
173
174 13 static void check_put_vvc_chroma(void)
175 {
176 13 LOCAL_ALIGNED_32(int16_t, dst0, [DST_BUF_SIZE / 2]);
177 13 LOCAL_ALIGNED_32(int16_t, dst1, [DST_BUF_SIZE / 2]);
178 13 LOCAL_ALIGNED_32(uint8_t, src0, [SRC_BUF_SIZE]);
179 13 LOCAL_ALIGNED_32(uint8_t, src1, [SRC_BUF_SIZE]);
180 VVCDSPContext c;
181
182 13 declare_func(void, int16_t *dst, const uint8_t *src, const ptrdiff_t src_stride,
183 const int height, const int8_t *hf, const int8_t *vf, const int width);
184
185
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
186
2/2
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 39 times.
 710814 randomize_pixels(src0, src1, SRC_BUF_SIZE);
187 39 ff_vvc_dsp_init(&c, bit_depth);
188
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int i = 0; i < 2; i++) {
189
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int j = 0; j < 2; j++) {
190
2/2
✓ Branch 0 taken 1092 times.
✓ Branch 1 taken 156 times.
 1248 for (int h = 2; h <= MAX_CTU_SIZE; h *= 2) {
191
2/2
✓ Branch 0 taken 7644 times.
✓ Branch 1 taken 1092 times.
 8736 for (int w = 2; w <= MAX_CTU_SIZE; w *= 2) {
192 7644 const int idx = av_log2(w) - 1;
193 7644 const int mx = rnd() % VVC_INTER_CHROMA_FACTS;
194 7644 const int my = rnd() % VVC_INTER_CHROMA_FACTS;
195 7644 const int8_t *hf = ff_vvc_inter_chroma_filters[rnd() % VVC_INTER_CHROMA_FILTER_TYPES][mx];
196 7644 const int8_t *vf = ff_vvc_inter_chroma_filters[rnd() % VVC_INTER_CHROMA_FILTER_TYPES][my];
197 const char *type;
198
4/5
✓ Branch 0 taken 1911 times.
✓ Branch 1 taken 1911 times.
✓ Branch 2 taken 1911 times.
✓ Branch 3 taken 1911 times.
✗ Branch 4 not taken.
 7644 switch ((j << 1) | i) {
199 1911 case 0: type = "put_chroma_pixels"; break; // 0 0
200 1911 case 1: type = "put_chroma_h"; break; // 0 1
201 1911 case 2: type = "put_chroma_v"; break; // 1 0
202 1911 case 3: type = "put_chroma_hv"; break; // 1 1
203 }
204
2/2
✓ Branch 3 taken 1463 times.
✓ Branch 4 taken 6181 times.
 7644 if (check_func(c.inter.put[CHROMA][idx][j][i], "%s_%d_%dx%d", type, bit_depth, w, h)) {
205 1463 memset(dst0, 0, DST_BUF_SIZE);
206 1463 memset(dst1, 0, DST_BUF_SIZE);
207 1463 call_ref(dst0, src0 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
208 1463 call_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
209
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1463 times.
 1463 if (memcmp(dst0, dst1, DST_BUF_SIZE))
210 ✗ fail();
211
2/2
✓ Branch 0 taken 209 times.
✓ Branch 1 taken 1254 times.
 1463 if (w == h)
212
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 209 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 209 bench_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
213 }
214 }
215 }
216 }
217 }
218 }
219 13 report("put_chroma");
220 13 }
221
222 13 static void check_put_vvc_chroma_uni(void)
223 {
224 13 LOCAL_ALIGNED_32(uint8_t, dst0, [DST_BUF_SIZE]);
225 13 LOCAL_ALIGNED_32(uint8_t, dst1, [DST_BUF_SIZE]);
226 13 LOCAL_ALIGNED_32(uint8_t, src0, [SRC_BUF_SIZE]);
227 13 LOCAL_ALIGNED_32(uint8_t, src1, [SRC_BUF_SIZE]);
228
229 VVCDSPContext c;
230 13 declare_func(void, uint8_t *dst, ptrdiff_t dststride,
231 const uint8_t *src, ptrdiff_t srcstride, int height,
232 const int8_t *hf, const int8_t *vf, int width);
233
234
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
235 39 ff_vvc_dsp_init(&c, bit_depth);
236
2/2
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 39 times.
 710814 randomize_pixels(src0, src1, SRC_BUF_SIZE);
237
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int i = 0; i < 2; i++) {
238
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int j = 0; j < 2; j++) {
239
2/2
✓ Branch 0 taken 936 times.
✓ Branch 1 taken 156 times.
 1092 for (int h = 4; h <= MAX_CTU_SIZE; h *= 2) {
240
2/2
✓ Branch 0 taken 5616 times.
✓ Branch 1 taken 936 times.
 6552 for (int w = 4; w <= MAX_CTU_SIZE; w *= 2) {
241 5616 const int idx = av_log2(w) - 1;
242 5616 const int mx = rnd() % VVC_INTER_CHROMA_FACTS;
243 5616 const int my = rnd() % VVC_INTER_CHROMA_FACTS;
244 5616 const int8_t *hf = ff_vvc_inter_chroma_filters[rnd() % VVC_INTER_CHROMA_FILTER_TYPES][mx];
245 5616 const int8_t *vf = ff_vvc_inter_chroma_filters[rnd() % VVC_INTER_CHROMA_FILTER_TYPES][my];
246 const char *type;
247
248
4/5
✓ Branch 0 taken 1404 times.
✓ Branch 1 taken 1404 times.
✓ Branch 2 taken 1404 times.
✓ Branch 3 taken 1404 times.
✗ Branch 4 not taken.
 5616 switch ((j << 1) | i) {
249 1404 case 0: type = "put_uni_pixels"; break; // 0 0
250 1404 case 1: type = "put_uni_h"; break; // 0 1
251 1404 case 2: type = "put_uni_v"; break; // 1 0
252 1404 case 3: type = "put_uni_hv"; break; // 1 1
253 }
254
255
2/2
✓ Branch 3 taken 1110 times.
✓ Branch 4 taken 4506 times.
 5616 if (check_func(c.inter.put_uni[CHROMA][idx][j][i], "%s_chroma_%d_%dx%d", type, bit_depth, w, h)) {
256 1110 memset(dst0, 0, DST_BUF_SIZE);
257 1110 memset(dst1, 0, DST_BUF_SIZE);
258 1110 call_ref(dst0, PIXEL_STRIDE, src0 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
259 1110 call_new(dst1, PIXEL_STRIDE, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
260
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1110 times.
 1110 if (memcmp(dst0, dst1, DST_BUF_SIZE))
261 ✗ fail();
262
2/2
✓ Branch 0 taken 185 times.
✓ Branch 1 taken 925 times.
 1110 if (w == h)
263
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 185 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 185 bench_new(dst1, PIXEL_STRIDE, src1 + SRC_OFFSET, PIXEL_STRIDE, h, hf, vf, w);
264 }
265 }
266 }
267 }
268 }
269 }
270 13 report("put_uni_chroma");
271 13 }
272
273 #define AVG_SRC_BUF_SIZE (MAX_CTU_SIZE * MAX_CTU_SIZE)
274 #define AVG_DST_BUF_SIZE (MAX_PB_SIZE * MAX_PB_SIZE * 2)
275
276 13 static void check_avg(void)
277 {
278 13 LOCAL_ALIGNED_32(int16_t, src00, [AVG_SRC_BUF_SIZE]);
279 13 LOCAL_ALIGNED_32(int16_t, src01, [AVG_SRC_BUF_SIZE]);
280 13 LOCAL_ALIGNED_32(int16_t, src10, [AVG_SRC_BUF_SIZE]);
281 13 LOCAL_ALIGNED_32(int16_t, src11, [AVG_SRC_BUF_SIZE]);
282 13 LOCAL_ALIGNED_32(uint8_t, dst0, [AVG_DST_BUF_SIZE]);
283 13 LOCAL_ALIGNED_32(uint8_t, dst1, [AVG_DST_BUF_SIZE]);
284 VVCDSPContext c;
285
286
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
287
2/2
✓ Branch 1 taken 319488 times.
✓ Branch 2 taken 39 times.
 319527 randomize_avg_src((uint8_t*)src00, (uint8_t*)src10, AVG_SRC_BUF_SIZE * sizeof(int16_t));
288
2/2
✓ Branch 1 taken 319488 times.
✓ Branch 2 taken 39 times.
 319527 randomize_avg_src((uint8_t*)src01, (uint8_t*)src11, AVG_SRC_BUF_SIZE * sizeof(int16_t));
289 39 ff_vvc_dsp_init(&c, bit_depth);
290
2/2
✓ Branch 0 taken 273 times.
✓ Branch 1 taken 39 times.
 312 for (int h = 2; h <= MAX_CTU_SIZE; h *= 2) {
291
2/2
✓ Branch 0 taken 1911 times.
✓ Branch 1 taken 273 times.
 2184 for (int w = 2; w <= MAX_CTU_SIZE; w *= 2) {
292 {
293 1911 declare_func(void, uint8_t *dst, ptrdiff_t dst_stride,
294 const int16_t *src0, const int16_t *src1, int width, int height);
295
2/2
✓ Branch 3 taken 294 times.
✓ Branch 4 taken 1617 times.
 1911 if (check_func(c.inter.avg, "avg_%d_%dx%d", bit_depth, w, h)) {
296 294 memset(dst0, 0, AVG_DST_BUF_SIZE);
297 294 memset(dst1, 0, AVG_DST_BUF_SIZE);
298 294 call_ref(dst0, MAX_CTU_SIZE * SIZEOF_PIXEL, src00, src01, w, h);
299 294 call_new(dst1, MAX_CTU_SIZE * SIZEOF_PIXEL, src10, src11, w, h);
300
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 294 times.
 294 if (memcmp(dst0, dst1, DST_BUF_SIZE))
301 ✗ fail();
302
2/2
✓ Branch 0 taken 42 times.
✓ Branch 1 taken 252 times.
 294 if (w == h)
303
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 42 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 42 bench_new(dst0, MAX_CTU_SIZE * SIZEOF_PIXEL, src00, src01, w, h);
304 }
305 }
306 {
307 1911 declare_func(void, uint8_t *dst, ptrdiff_t dst_stride,
308 const int16_t *src0, const int16_t *src1, int width, int height,
309 int denom, int w0, int w1, int o0, int o1);
310 {
311 1911 const int denom = rnd() % 8;
312 1911 const int w0 = rnd() % 256 - 128;
313 1911 const int w1 = rnd() % 256 - 128;
314 1911 const int o0 = rnd() % 256 - 128;
315 1911 const int o1 = rnd() % 256 - 128;
316
2/2
✓ Branch 3 taken 294 times.
✓ Branch 4 taken 1617 times.
 1911 if (check_func(c.inter.w_avg, "w_avg_%d_%dx%d", bit_depth, w, h)) {
317 294 memset(dst0, 0, AVG_DST_BUF_SIZE);
318 294 memset(dst1, 0, AVG_DST_BUF_SIZE);
319
320 294 call_ref(dst0, MAX_CTU_SIZE * SIZEOF_PIXEL, src00, src01, w, h, denom, w0, w1, o0, o1);
321 294 call_new(dst1, MAX_CTU_SIZE * SIZEOF_PIXEL, src10, src11, w, h, denom, w0, w1, o0, o1);
322
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 294 times.
 294 if (memcmp(dst0, dst1, DST_BUF_SIZE))
323 ✗ fail();
324
2/2
✓ Branch 0 taken 42 times.
✓ Branch 1 taken 252 times.
 294 if (w == h)
325
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 42 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 42 bench_new(dst0, MAX_CTU_SIZE * SIZEOF_PIXEL, src00, src01, w, h, denom, w0, w1, o0, o1);
326 }
327 }
328 }
329 }
330 }
331 }
332 13 report("avg");
333 13 }
334
335 #define SR_RANGE 2
336 13 static void check_dmvr(void)
337 {
338 13 LOCAL_ALIGNED_32(uint16_t, dst0, [DST_BUF_SIZE]);
339 13 LOCAL_ALIGNED_32(uint16_t, dst1, [DST_BUF_SIZE]);
340 13 LOCAL_ALIGNED_32(uint8_t, src0, [SRC_BUF_SIZE]);
341 13 LOCAL_ALIGNED_32(uint8_t, src1, [SRC_BUF_SIZE]);
342 13 const int dst_stride = MAX_PB_SIZE * sizeof(int16_t);
343
344 VVCDSPContext c;
345 13 declare_func(void, int16_t *dst, const uint8_t *src, ptrdiff_t src_stride, int height,
346 intptr_t mx, intptr_t my, int width);
347
348
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
349 39 ff_vvc_dsp_init(&c, bit_depth);
350
2/2
✓ Branch 1 taken 710775 times.
✓ Branch 2 taken 39 times.
 710814 randomize_pixels(src0, src1, SRC_BUF_SIZE);
351
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int i = 0; i < 2; i++) {
352
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int j = 0; j < 2; j++) {
353
2/2
✓ Branch 0 taken 312 times.
✓ Branch 1 taken 156 times.
 468 for (int h = 8; h <= 16; h *= 2) {
354
2/2
✓ Branch 0 taken 624 times.
✓ Branch 1 taken 312 times.
 936 for (int w = 8; w <= 16; w *= 2) {
355 624 const int pred_w = w + 2 * SR_RANGE;
356 624 const int pred_h = h + 2 * SR_RANGE;
357 624 const int mx = rnd() % VVC_INTER_LUMA_DMVR_FACTS;
358 624 const int my = rnd() % VVC_INTER_LUMA_DMVR_FACTS;
359 const char *type;
360
361
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 468 times.
 624 if (w * h < 128)
362 156 continue;
363
364
4/5
✓ Branch 0 taken 117 times.
✓ Branch 1 taken 117 times.
✓ Branch 2 taken 117 times.
✓ Branch 3 taken 117 times.
✗ Branch 4 not taken.
 468 switch ((j << 1) | i) {
365 117 case 0: type = "dmvr"; break; // 0 0
366 117 case 1: type = "dmvr_h"; break; // 0 1
367 117 case 2: type = "dmvr_v"; break; // 1 0
368 117 case 3: type = "dmvr_hv"; break; // 1 1
369 }
370
371
2/2
✓ Branch 3 taken 72 times.
✓ Branch 4 taken 396 times.
 468 if (check_func(c.inter.dmvr[j][i], "%s_%d_%dx%d", type, bit_depth, pred_w, pred_h)) {
372 72 memset(dst0, 0, DST_BUF_SIZE);
373 72 memset(dst1, 0, DST_BUF_SIZE);
374 72 call_ref(dst0, src0 + SRC_OFFSET, PIXEL_STRIDE, pred_h, mx, my, pred_w);
375 72 call_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, pred_h, mx, my, pred_w);
376
2/2
✓ Branch 0 taken 1248 times.
✓ Branch 1 taken 72 times.
 1320 for (int k = 0; k < pred_h; k++) {
377
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1248 times.
 1248 if (memcmp(dst0 + k * dst_stride, dst1 + k * dst_stride, pred_w * sizeof(int16_t))) {
378 ✗ fail();
379 ✗ break;
380 }
381 }
382
383
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 72 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 72 bench_new(dst1, src1 + SRC_OFFSET, PIXEL_STRIDE, pred_h, mx, my, pred_w);
384 }
385 }
386 }
387 }
388 }
389 }
390 13 report("dmvr");
391 13 }
392
393 #define BDOF_BLOCK_SIZE 16
394 #define BDOF_SRC_SIZE (MAX_PB_SIZE* (BDOF_BLOCK_SIZE + 2))
395 #define BDOF_SRC_OFFSET (MAX_PB_SIZE + 1)
396 #define BDOF_DST_SIZE (BDOF_BLOCK_SIZE * BDOF_BLOCK_SIZE * 2)
397 13 static void check_bdof(void)
398 {
399 13 LOCAL_ALIGNED_32(uint8_t, dst0, [BDOF_DST_SIZE]);
400 13 LOCAL_ALIGNED_32(uint8_t, dst1, [BDOF_DST_SIZE]);
401 13 LOCAL_ALIGNED_32(uint16_t, src00, [BDOF_SRC_SIZE]);
402 13 LOCAL_ALIGNED_32(uint16_t, src01, [BDOF_SRC_SIZE]);
403 13 LOCAL_ALIGNED_32(uint16_t, src10, [BDOF_SRC_SIZE]);
404 13 LOCAL_ALIGNED_32(uint16_t, src11, [BDOF_SRC_SIZE]);
405
406 VVCDSPContext c;
407 13 declare_func(void, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1, int block_w, int block_h);
408
409
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
 52 for (int bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
410 39 const int dst_stride = BDOF_BLOCK_SIZE * SIZEOF_PIXEL;
411
412 39 ff_vvc_dsp_init(&c, bit_depth);
413
2/2
✓ Branch 1 taken 44928 times.
✓ Branch 2 taken 39 times.
 44967 randomize_prof_src(src00, src10, BDOF_SRC_SIZE);
414
2/2
✓ Branch 1 taken 44928 times.
✓ Branch 2 taken 39 times.
 44967 randomize_prof_src(src01, src11, BDOF_SRC_SIZE);
415
2/2
✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
 117 for (int h = 8; h <= 16; h *= 2) {
416
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
 234 for (int w = 8; w <= 16; w *= 2) {
417
2/2
✓ Branch 0 taken 39 times.
✓ Branch 1 taken 117 times.
 156 if (w * h < 128)
418 39 continue;
419
2/2
✓ Branch 3 taken 18 times.
✓ Branch 4 taken 99 times.
 117 if (check_func(c.inter.apply_bdof, "apply_bdof_%d_%dx%d", bit_depth, w, h)) {
420 18 memset(dst0, 0, BDOF_DST_SIZE);
421 18 memset(dst1, 0, BDOF_DST_SIZE);
422 18 call_ref(dst0, dst_stride, src00 + BDOF_SRC_OFFSET, src01 + BDOF_SRC_OFFSET, w, h);
423 18 call_new(dst1, dst_stride, src10 + BDOF_SRC_OFFSET, src11 + BDOF_SRC_OFFSET, w, h);
424
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 if (memcmp(dst0, dst1, BDOF_DST_SIZE))
425 ✗ fail();
426
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 18 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 18 bench_new(dst0, dst_stride, src00 + BDOF_SRC_OFFSET, src01 + BDOF_SRC_OFFSET, w, h);
427 }
428 }
429 }
430 }
431 13 report("apply_bdof");
432 13 }
433
434 13 static void check_vvc_sad(void)
435 {
436 13 const int bit_depth = 10;
437 VVCDSPContext c;
438 13 LOCAL_ALIGNED_32(uint16_t, src0, [MAX_CTU_SIZE * MAX_CTU_SIZE * 4]);
439 13 LOCAL_ALIGNED_32(uint16_t, src1, [MAX_CTU_SIZE * MAX_CTU_SIZE * 4]);
440 13 declare_func(int, const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
441
442 13 ff_vvc_dsp_init(&c, bit_depth);
443
2/2
✓ Branch 1 taken 425984 times.
✓ Branch 2 taken 13 times.
 425997 randomize_pixels(src0, src1, MAX_CTU_SIZE * MAX_CTU_SIZE * 4);
444
2/2
✓ Branch 0 taken 26 times.
✓ Branch 1 taken 13 times.
 39 for (int h = 8; h <= 16; h *= 2) {
445
2/2
✓ Branch 0 taken 52 times.
✓ Branch 1 taken 26 times.
 78 for (int w = 8; w <= 16; w *= 2) {
446
2/2
✓ Branch 0 taken 260 times.
✓ Branch 1 taken 52 times.
 312 for(int offy = 0; offy <= 4; offy++) {
447
2/2
✓ Branch 0 taken 1300 times.
✓ Branch 1 taken 260 times.
 1560 for(int offx = 0; offx <= 4; offx++) {
448
2/2
✓ Branch 0 taken 325 times.
✓ Branch 1 taken 975 times.
 1300 if (w * h < 128)
449 325 continue;
450
451
2/2
✓ Branch 3 taken 6 times.
✓ Branch 4 taken 969 times.
 975 if (check_func(c.inter.sad, "sad_%dx%d", w, h)) {
452 int result0;
453 int result1;
454
455 6 result0 = call_ref(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
456 6 result1 = call_new(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
457
458
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
 6 if (result1 != result0)
459 ✗ fail();
460
2/4
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
 6 if(offx == 0 && offy == 0)
461
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
 6 bench_new(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
462 }
463 }
464 }
465 }
466 }
467
468 13 report("sad");
469 13 }
470
471 13 void checkasm_check_vvc_mc(void)
472 {
473 13 check_dmvr();
474 13 check_bdof();
475 13 check_vvc_sad();
476 13 check_put_vvc_luma();
477 13 check_put_vvc_luma_uni();
478 13 check_put_vvc_chroma();
479 13 check_put_vvc_chroma_uni();
480 13 check_avg();
481 13 }
482