FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/tests/checkasm/sw_scale.c
Date: 2024-11-20 23:03:26
Exec Total Coverage
Lines: 143 183 78.1%
Functions: 6 8 75.0%
Branches: 80 122 65.6%
Line Branch Exec Source
1 /*
2 *
3 * This file is part of FFmpeg.
4 *
5 * FFmpeg is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * FFmpeg is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #include <string.h>
21
22 #include "libavutil/common.h"
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/mem.h"
25 #include "libavutil/mem_internal.h"
26
27 #include "libswscale/swscale.h"
28 #include "libswscale/swscale_internal.h"
29
30 #include "checkasm.h"
31
32 #define randomize_buffers(buf, size) \
33 do { \
34 int j; \
35 for (j = 0; j < size; j+=4) \
36 AV_WN32(buf + j, rnd()); \
37 } while (0)
38
39 608 static void yuv2planeX_8_ref(const int16_t *filter, int filterSize,
40 const int16_t **src, uint8_t *dest, int dstW,
41 const uint8_t *dither, int offset)
42 {
43 // This corresponds to the yuv2planeX_8_c function
44 int i;
45
2/2
✓ Branch 0 taken 122112 times.
✓ Branch 1 taken 608 times.
 122720 for (i = 0; i < dstW; i++) {
46 122112 int val = dither[(i + offset) & 7] << 12;
47 int j;
48
2/2
✓ Branch 0 taken 915840 times.
✓ Branch 1 taken 122112 times.
 1037952 for (j = 0; j < filterSize; j++)
49 915840 val += src[j][i] * filter[j];
50
51 122112 dest[i]= av_clip_uint8(val >> 19);
52 }
53 608 }
54
55 752 static int cmp_off_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy)
56 {
57
2/2
✓ Branch 0 taken 337024 times.
✓ Branch 1 taken 752 times.
 337776 for (size_t i = 0; i < n; i++) {
58
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 337024 times.
 337024 if (abs(ref[i] - test[i]) > accuracy)
59 ✗ return 1;
60 }
61 752 return 0;
62 }
63
64 ✗ static void print_data(uint8_t *p, size_t len, size_t offset)
65 {
66 ✗ size_t i = 0;
67 ✗ for (; i < len; i++) {
68 ✗ if (i % 8 == 0) {
69 ✗ printf("0x%04zx: ", i+offset);
70 }
71 ✗ printf("0x%02x ", (uint32_t) p[i]);
72 ✗ if (i % 8 == 7) {
73 ✗ printf("\n");
74 }
75 }
76 ✗ if (i % 8 != 0) {
77 ✗ printf("\n");
78 }
79 ✗ }
80
81 ✗ static size_t show_differences(uint8_t *a, uint8_t *b, size_t len)
82 {
83 ✗ for (size_t i = 0; i < len; i++) {
84 ✗ if (a[i] != b[i]) {
85 ✗ size_t offset_of_mismatch = i;
86 size_t offset;
87 ✗ if (i >= 8) i-=8;
88 ✗ offset = i & (~7);
89 ✗ printf("test a:\n");
90 ✗ print_data(&a[offset], 32, offset);
91 ✗ printf("\ntest b:\n");
92 ✗ print_data(&b[offset], 32, offset);
93 ✗ printf("\n");
94 ✗ return offset_of_mismatch;
95 }
96 }
97 ✗ return len;
98 }
99
100 26 static void check_yuv2yuv1(int accurate)
101 {
102 SwsContext *sws;
103 SwsInternal *c;
104 int osi, isi;
105 int dstW, offset;
106 size_t fail_offset;
107 26 const int input_sizes[] = {8, 24, 128, 144, 256, 512};
108 #define LARGEST_INPUT_SIZE 512
109
110 26 const int offsets[] = {0, 3, 8, 11, 16, 19};
111 26 const int OFFSET_SIZES = sizeof(offsets)/sizeof(offsets[0]);
112
2/2
✓ Branch 0 taken 13 times.
✓ Branch 1 taken 13 times.
 26 const char *accurate_str = (accurate) ? "accurate" : "approximate";
113
114 26 declare_func(void,
115 const int16_t *src, uint8_t *dest,
116 int dstW, const uint8_t *dither, int offset);
117
118 26 LOCAL_ALIGNED_16(int16_t, src_pixels, [LARGEST_INPUT_SIZE]);
119 26 LOCAL_ALIGNED_16(uint8_t, dst0, [LARGEST_INPUT_SIZE]);
120 26 LOCAL_ALIGNED_16(uint8_t, dst1, [LARGEST_INPUT_SIZE]);
121 26 LOCAL_ALIGNED_8(uint8_t, dither, [8]);
122
123
2/2
✓ Branch 1 taken 52 times.
✓ Branch 2 taken 26 times.
 78 randomize_buffers((uint8_t*)dither, 8);
124
2/2
✓ Branch 1 taken 6656 times.
✓ Branch 2 taken 26 times.
 6682 randomize_buffers((uint8_t*)src_pixels, LARGEST_INPUT_SIZE * sizeof(int16_t));
125 26 sws = sws_alloc_context();
126 26 c = sws_internal(sws);
127
2/2
✓ Branch 0 taken 13 times.
✓ Branch 1 taken 13 times.
 26 if (accurate)
128 13 c->flags |= SWS_ACCURATE_RND;
129
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 26 times.
 26 if (sws_init_context(sws, NULL, NULL) < 0)
130 ✗ fail();
131
132 26 ff_sws_init_scale(c);
133
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 26 times.
 182 for (isi = 0; isi < FF_ARRAY_ELEMS(input_sizes); ++isi) {
134 156 dstW = input_sizes[isi];
135
2/2
✓ Branch 0 taken 936 times.
✓ Branch 1 taken 156 times.
 1092 for (osi = 0; osi < OFFSET_SIZES; osi++) {
136 936 offset = offsets[osi];
137
2/2
✓ Branch 3 taken 144 times.
✓ Branch 4 taken 792 times.
 936 if (check_func(c->yuv2plane1, "yuv2yuv1_%d_%d_%s", offset, dstW, accurate_str)){
138 144 memset(dst0, 0, LARGEST_INPUT_SIZE * sizeof(dst0[0]));
139 144 memset(dst1, 0, LARGEST_INPUT_SIZE * sizeof(dst1[0]));
140
141 144 call_ref(src_pixels, dst0, dstW, dither, offset);
142 144 call_new(src_pixels, dst1, dstW, dither, offset);
143
3/4
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 108 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 144 times.
 144 if (cmp_off_by_n(dst0, dst1, dstW * sizeof(dst0[0]), accurate ? 0 : 2)) {
144 ✗ fail();
145 ✗ printf("failed: yuv2yuv1_%d_%di_%s\n", offset, dstW, accurate_str);
146 ✗ fail_offset = show_differences(dst0, dst1, LARGEST_INPUT_SIZE * sizeof(dst0[0]));
147 ✗ printf("failing values: src: 0x%04x dither: 0x%02x dst-c: %02x dst-asm: %02x\n",
148 ✗ (int) src_pixels[fail_offset],
149 ✗ (int) dither[(fail_offset + fail_offset) & 7],
150 ✗ (int) dst0[fail_offset],
151 ✗ (int) dst1[fail_offset]);
152 }
153
2/2
✓ Branch 0 taken 24 times.
✓ Branch 1 taken 120 times.
 144 if(dstW == LARGEST_INPUT_SIZE)
154
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 24 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.
 24 bench_new(src_pixels, dst1, dstW, dither, offset);
155 }
156 }
157 }
158 26 sws_freeContext(sws);
159 26 }
160
161 26 static void check_yuv2yuvX(int accurate)
162 {
163 SwsContext *sws;
164 SwsInternal *c;
165 int fsi, osi, isi, i, j;
166 int dstW;
167 #define LARGEST_FILTER 16
168 // ff_yuv2planeX_8_sse2 can't handle odd filter sizes
169 26 const int filter_sizes[] = {2, 4, 8, 16};
170 26 const int FILTER_SIZES = sizeof(filter_sizes)/sizeof(filter_sizes[0]);
171 #define LARGEST_INPUT_SIZE 512
172 static const int input_sizes[] = {8, 24, 128, 144, 256, 512};
173
2/2
✓ Branch 0 taken 13 times.
✓ Branch 1 taken 13 times.
 26 const char *accurate_str = (accurate) ? "accurate" : "approximate";
174
175
2/2
✓ Branch 1 taken 24 times.
✓ Branch 2 taken 2 times.
 26 declare_func_emms(AV_CPU_FLAG_MMX, void, const int16_t *filter,
176 int filterSize, const int16_t **src, uint8_t *dest,
177 int dstW, const uint8_t *dither, int offset);
178
179 const int16_t **src;
180 26 LOCAL_ALIGNED_16(int16_t, src_pixels, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
181 26 LOCAL_ALIGNED_16(int16_t, filter_coeff, [LARGEST_FILTER]);
182 26 LOCAL_ALIGNED_16(uint8_t, dst0, [LARGEST_INPUT_SIZE]);
183 26 LOCAL_ALIGNED_16(uint8_t, dst1, [LARGEST_INPUT_SIZE]);
184 26 LOCAL_ALIGNED_16(uint8_t, dither, [LARGEST_INPUT_SIZE]);
185 union VFilterData{
186 const int16_t *src;
187 uint16_t coeff[8];
188 } *vFilterData;
189 26 uint8_t d_val = rnd();
190 26 memset(dither, d_val, LARGEST_INPUT_SIZE);
191
2/2
✓ Branch 1 taken 106496 times.
✓ Branch 2 taken 26 times.
 106522 randomize_buffers((uint8_t*)src_pixels, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int16_t));
192 26 sws = sws_alloc_context();
193 26 c = sws_internal(sws);
194
2/2
✓ Branch 0 taken 13 times.
✓ Branch 1 taken 13 times.
 26 if (accurate)
195 13 c->flags |= SWS_ACCURATE_RND;
196
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 26 times.
 26 if (sws_init_context(sws, NULL, NULL) < 0)
197 ✗ fail();
198
199 26 ff_sws_init_scale(c);
200
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 26 times.
 182 for(isi = 0; isi < FF_ARRAY_ELEMS(input_sizes); ++isi){
201 156 dstW = input_sizes[isi];
202
2/2
✓ Branch 0 taken 624 times.
✓ Branch 1 taken 156 times.
 780 for(osi = 0; osi < 64; osi += 16){
203
2/2
✓ Branch 0 taken 130 times.
✓ Branch 1 taken 494 times.
 624 if (dstW <= osi)
204 130 continue;
205
2/2
✓ Branch 0 taken 1976 times.
✓ Branch 1 taken 494 times.
 2470 for (fsi = 0; fsi < FILTER_SIZES; ++fsi) {
206 // Generate filter coefficients for the given filter size,
207 // with some properties:
208 // - The coefficients add up to the intended sum (4096, 1<<12)
209 // - The coefficients contain negative values
210 // - The filter intermediates don't overflow for worst case
211 // inputs (all positive coefficients are coupled with
212 // input_max and all negative coefficients with input_min,
213 // or vice versa).
214 // Produce a filter with all coefficients set to
215 // -((1<<12)/(filter_size-1)) except for one (randomly chosen)
216 // which is set to ((1<<13)-1).
217
2/2
✓ Branch 0 taken 14820 times.
✓ Branch 1 taken 1976 times.
 16796 for (i = 0; i < filter_sizes[fsi]; ++i)
218 14820 filter_coeff[i] = -((1 << 12) / (filter_sizes[fsi] - 1));
219 1976 filter_coeff[rnd() % filter_sizes[fsi]] = (1 << 13) - 1;
220
221 1976 src = av_malloc(sizeof(int16_t*) * filter_sizes[fsi]);
222 1976 vFilterData = av_malloc((filter_sizes[fsi] + 2) * sizeof(union VFilterData));
223 1976 memset(vFilterData, 0, (filter_sizes[fsi] + 2) * sizeof(union VFilterData));
224
2/2
✓ Branch 0 taken 14820 times.
✓ Branch 1 taken 1976 times.
 16796 for (i = 0; i < filter_sizes[fsi]; ++i) {
225 14820 src[i] = &src_pixels[i * LARGEST_INPUT_SIZE];
226 14820 vFilterData[i].src = src[i] - osi;
227
2/2
✓ Branch 0 taken 59280 times.
✓ Branch 1 taken 14820 times.
 74100 for(j = 0; j < 4; ++j)
228 59280 vFilterData[i].coeff[j + 4] = filter_coeff[i];
229 }
230
2/2
✓ Branch 3 taken 608 times.
✓ Branch 4 taken 1368 times.
 1976 if (check_func(c->yuv2planeX, "yuv2yuvX_%d_%d_%d_%s", filter_sizes[fsi], osi, dstW, accurate_str)){
231 // use vFilterData for the mmx function
232
2/2
✓ Branch 0 taken 228 times.
✓ Branch 1 taken 380 times.
 608 const int16_t *filter = c->use_mmx_vfilter ? (const int16_t*)vFilterData : &filter_coeff[0];
233 608 memset(dst0, 0, LARGEST_INPUT_SIZE * sizeof(dst0[0]));
234 608 memset(dst1, 0, LARGEST_INPUT_SIZE * sizeof(dst1[0]));
235
236 // We can't use call_ref here, because we don't know if use_mmx_vfilter was set for that
237 // function or not, so we can't pass it the parameters correctly.
238 608 yuv2planeX_8_ref(&filter_coeff[0], filter_sizes[fsi], src, dst0, dstW - osi, dither, osi);
239
240 608 call_new(filter, filter_sizes[fsi], src, dst1, dstW - osi, dither, osi);
241
3/4
✓ Branch 0 taken 304 times.
✓ Branch 1 taken 304 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 608 times.
 608 if (cmp_off_by_n(dst0, dst1, LARGEST_INPUT_SIZE * sizeof(dst0[0]), accurate ? 0 : 2)) {
242 ✗ fail();
243 ✗ printf("failed: yuv2yuvX_%d_%d_%d_%s\n", filter_sizes[fsi], osi, dstW, accurate_str);
244 ✗ show_differences(dst0, dst1, LARGEST_INPUT_SIZE * sizeof(dst0[0]));
245 }
246
2/2
✓ Branch 0 taken 128 times.
✓ Branch 1 taken 480 times.
 608 if(dstW == LARGEST_INPUT_SIZE)
247
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 128 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.
 128 bench_new((const int16_t*)vFilterData, filter_sizes[fsi], src, dst1, dstW - osi, dither, osi);
248
249 }
250 1976 av_freep(&src);
251 1976 av_freep(&vFilterData);
252 }
253 }
254 }
255 26 sws_freeContext(sws);
256 #undef FILTER_SIZES
257 26 }
258
259 #undef SRC_PIXELS
260 #define SRC_PIXELS 512
261
262 13 static void check_hscale(void)
263 {
264 #define MAX_FILTER_WIDTH 40
265 #define FILTER_SIZES 6
266 static const int filter_sizes[FILTER_SIZES] = { 4, 8, 12, 16, 32, 40 };
267
268 #define HSCALE_PAIRS 2
269 static const int hscale_pairs[HSCALE_PAIRS][2] = {
270 { 8, 14 },
271 { 8, 18 },
272 };
273
274 #define LARGEST_INPUT_SIZE 512
275 static const int input_sizes[] = {8, 24, 128, 144, 256, 512};
276
277 int i, j, fsi, hpi, width, dstWi;
278 SwsContext *sws;
279 SwsInternal *c;
280
281 // padded
282 13 LOCAL_ALIGNED_32(uint8_t, src, [FFALIGN(SRC_PIXELS + MAX_FILTER_WIDTH - 1, 4)]);
283 13 LOCAL_ALIGNED_32(uint32_t, dst0, [SRC_PIXELS]);
284 13 LOCAL_ALIGNED_32(uint32_t, dst1, [SRC_PIXELS]);
285
286 // padded
287 13 LOCAL_ALIGNED_32(int16_t, filter, [SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH]);
288 13 LOCAL_ALIGNED_32(int32_t, filterPos, [SRC_PIXELS]);
289 13 LOCAL_ALIGNED_32(int16_t, filterAvx2, [SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH]);
290 13 LOCAL_ALIGNED_32(int32_t, filterPosAvx, [SRC_PIXELS]);
291
292 // The dst parameter here is either int16_t or int32_t but we use void* to
293 // just cover both cases.
294 13 declare_func(void, void *c, void *dst, int dstW,
295 const uint8_t *src, const int16_t *filter,
296 const int32_t *filterPos, int filterSize);
297
298 13 sws = sws_alloc_context();
299
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 13 times.
 13 if (sws_init_context(sws, NULL, NULL) < 0)
300 ✗ fail();
301
302 13 c = sws_internal(sws);
303
2/2
✓ Branch 1 taken 1794 times.
✓ Branch 2 taken 13 times.
 1807 randomize_buffers(src, SRC_PIXELS + MAX_FILTER_WIDTH - 1);
304
305
2/2
✓ Branch 0 taken 26 times.
✓ Branch 1 taken 13 times.
 39 for (hpi = 0; hpi < HSCALE_PAIRS; hpi++) {
306
2/2
✓ Branch 0 taken 156 times.
✓ Branch 1 taken 26 times.
 182 for (fsi = 0; fsi < FILTER_SIZES; fsi++) {
307
2/2
✓ Branch 0 taken 936 times.
✓ Branch 1 taken 156 times.
 1092 for (dstWi = 0; dstWi < FF_ARRAY_ELEMS(input_sizes); dstWi++) {
308 936 width = filter_sizes[fsi];
309
310 936 c->srcBpc = hscale_pairs[hpi][0];
311 936 c->dstBpc = hscale_pairs[hpi][1];
312 936 c->hLumFilterSize = c->hChrFilterSize = width;
313
314
2/2
✓ Branch 0 taken 479232 times.
✓ Branch 1 taken 936 times.
 480168 for (i = 0; i < SRC_PIXELS; i++) {
315 479232 filterPos[i] = i;
316 479232 filterPosAvx[i] = i;
317
318 // These filter cofficients are chosen to try break two corner
319 // cases, namely:
320 //
321 // - Negative filter coefficients. The filters output signed
322 // values, and it should be possible to end up with negative
323 // output values.
324 //
325 // - Positive clipping. The hscale filter function has clipping
326 // at (1<<15) - 1
327 //
328 // The coefficients sum to the 1.0 point for the hscale
329 // functions (1 << 14).
330
331
2/2
✓ Branch 0 taken 8945664 times.
✓ Branch 1 taken 479232 times.
 9424896 for (j = 0; j < width; j++) {
332 8945664 filter[i * width + j] = -((1 << 14) / (width - 1));
333 }
334 479232 filter[i * width + (rnd() % width)] = ((1 << 15) - 1);
335 }
336
337
2/2
✓ Branch 0 taken 37440 times.
✓ Branch 1 taken 936 times.
 38376 for (i = 0; i < MAX_FILTER_WIDTH; i++) {
338 // These values should be unused in SIMD implementations but
339 // may still be read, random coefficients here should help show
340 // issues where they are used in error.
341
342 37440 filter[SRC_PIXELS * width + i] = rnd();
343 }
344 936 c->dstW = c->chrDstW = input_sizes[dstWi];
345 936 ff_sws_init_scale(c);
346 936 memcpy(filterAvx2, filter, sizeof(uint16_t) * (SRC_PIXELS * MAX_FILTER_WIDTH + MAX_FILTER_WIDTH));
347 936 ff_shuffle_filter_coefficients(c, filterPosAvx, width, filterAvx2, c->dstW);
348
349
2/2
✓ Branch 3 taken 288 times.
✓ Branch 4 taken 648 times.
 936 if (check_func(c->hcScale, "hscale_%d_to_%d__fs_%d_dstW_%d", c->srcBpc, c->dstBpc + 1, width, c->dstW)) {
350 288 memset(dst0, 0, SRC_PIXELS * sizeof(dst0[0]));
351 288 memset(dst1, 0, SRC_PIXELS * sizeof(dst1[0]));
352
353 288 call_ref(NULL, dst0, c->dstW, src, filter, filterPos, width);
354 288 call_new(NULL, dst1, c->dstW, src, filterAvx2, filterPosAvx, width);
355
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 288 times.
 288 if (memcmp(dst0, dst1, c->dstW * sizeof(dst0[0])))
356 ✗ fail();
357
1/8
✗ Branch 1 not taken.
✓ Branch 2 taken 288 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.
 288 bench_new(NULL, dst0, c->dstW, src, filter, filterPosAvx, width);
358 }
359 }
360 }
361 }
362 13 sws_freeContext(sws);
363 13 }
364
365 13 void checkasm_check_sw_scale(void)
366 {
367 13 check_hscale();
368 13 report("hscale");
369 13 check_yuv2yuv1(0);
370 13 check_yuv2yuv1(1);
371 13 report("yuv2yuv1");
372 13 check_yuv2yuvX(0);
373 13 check_yuv2yuvX(1);
374 13 report("yuv2yuvX");
375 13 }
376