Directory: | ../../../ffmpeg/ |
---|---|
File: | src/libavcodec/idctdsp.c |
Date: | 2022-07-04 00:18:54 |
Exec | Total | Coverage | |
---|---|---|---|
Lines: | 133 | 173 | 76.9% |
Branches: | 49 | 61 | 80.3% |
Line | Branch | Exec | Source |
---|---|---|---|
1 | /* | ||
2 | * This file is part of FFmpeg. | ||
3 | * | ||
4 | * FFmpeg is free software; you can redistribute it and/or | ||
5 | * modify it under the terms of the GNU Lesser General Public | ||
6 | * License as published by the Free Software Foundation; either | ||
7 | * version 2.1 of the License, or (at your option) any later version. | ||
8 | * | ||
9 | * FFmpeg is distributed in the hope that it will be useful, | ||
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
12 | * Lesser General Public License for more details. | ||
13 | * | ||
14 | * You should have received a copy of the GNU Lesser General Public | ||
15 | * License along with FFmpeg; if not, write to the Free Software | ||
16 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
17 | */ | ||
18 | |||
19 | #include "config.h" | ||
20 | #include "config_components.h" | ||
21 | #include "libavutil/attributes.h" | ||
22 | #include "libavutil/common.h" | ||
23 | #include "avcodec.h" | ||
24 | #include "dct.h" | ||
25 | #include "faanidct.h" | ||
26 | #include "idctdsp.h" | ||
27 | #include "simple_idct.h" | ||
28 | #include "xvididct.h" | ||
29 | |||
30 | 41068 | av_cold void ff_init_scantable(const uint8_t *permutation, ScanTable *st, | |
31 | const uint8_t *src_scantable) | ||
32 | { | ||
33 | int i, end; | ||
34 | |||
35 | 41068 | st->scantable = src_scantable; | |
36 | |||
37 |
2/2✓ Branch 0 taken 2628352 times.
✓ Branch 1 taken 41068 times.
|
2669420 | for (i = 0; i < 64; i++) { |
38 | 2628352 | int j = src_scantable[i]; | |
39 | 2628352 | st->permutated[i] = permutation[j]; | |
40 | } | ||
41 | |||
42 | 41068 | end = -1; | |
43 |
2/2✓ Branch 0 taken 2628352 times.
✓ Branch 1 taken 41068 times.
|
2669420 | for (i = 0; i < 64; i++) { |
44 | 2628352 | int j = st->permutated[i]; | |
45 |
2/2✓ Branch 0 taken 555305 times.
✓ Branch 1 taken 2073047 times.
|
2628352 | if (j > end) |
46 | 555305 | end = j; | |
47 | 2628352 | st->raster_end[i] = end; | |
48 | } | ||
49 | 41068 | } | |
50 | |||
51 | 1890 | av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation, | |
52 | enum idct_permutation_type perm_type) | ||
53 | { | ||
54 | int i; | ||
55 | |||
56 | #if ARCH_X86 | ||
57 |
2/2✓ Branch 1 taken 4 times.
✓ Branch 2 taken 1886 times.
|
1890 | if (ff_init_scantable_permutation_x86(idct_permutation, |
58 | perm_type)) | ||
59 | 4 | return; | |
60 | #endif | ||
61 | |||
62 |
3/5✓ Branch 0 taken 1568 times.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 314 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
1886 | switch (perm_type) { |
63 | 1568 | case FF_IDCT_PERM_NONE: | |
64 |
2/2✓ Branch 0 taken 100352 times.
✓ Branch 1 taken 1568 times.
|
101920 | for (i = 0; i < 64; i++) |
65 | 100352 | idct_permutation[i] = i; | |
66 | 1568 | break; | |
67 | 4 | case FF_IDCT_PERM_LIBMPEG2: | |
68 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 4 times.
|
260 | for (i = 0; i < 64; i++) |
69 | 256 | idct_permutation[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); | |
70 | 4 | break; | |
71 | 314 | case FF_IDCT_PERM_TRANSPOSE: | |
72 |
2/2✓ Branch 0 taken 20096 times.
✓ Branch 1 taken 314 times.
|
20410 | for (i = 0; i < 64; i++) |
73 | 20096 | idct_permutation[i] = ((i & 7) << 3) | (i >> 3); | |
74 | 314 | break; | |
75 | ✗ | case FF_IDCT_PERM_PARTTRANS: | |
76 | ✗ | for (i = 0; i < 64; i++) | |
77 | ✗ | idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3); | |
78 | ✗ | break; | |
79 | ✗ | default: | |
80 | ✗ | av_log(NULL, AV_LOG_ERROR, | |
81 | "Internal error, IDCT permutation not set\n"); | ||
82 | } | ||
83 | } | ||
84 | |||
85 | 141453 | void ff_put_pixels_clamped_c(const int16_t *block, uint8_t *av_restrict pixels, | |
86 | ptrdiff_t line_size) | ||
87 | { | ||
88 | int i; | ||
89 | |||
90 | /* read the pixels */ | ||
91 |
2/2✓ Branch 0 taken 1131624 times.
✓ Branch 1 taken 141453 times.
|
1273077 | for (i = 0; i < 8; i++) { |
92 | 1131624 | pixels[0] = av_clip_uint8(block[0]); | |
93 | 1131624 | pixels[1] = av_clip_uint8(block[1]); | |
94 | 1131624 | pixels[2] = av_clip_uint8(block[2]); | |
95 | 1131624 | pixels[3] = av_clip_uint8(block[3]); | |
96 | 1131624 | pixels[4] = av_clip_uint8(block[4]); | |
97 | 1131624 | pixels[5] = av_clip_uint8(block[5]); | |
98 | 1131624 | pixels[6] = av_clip_uint8(block[6]); | |
99 | 1131624 | pixels[7] = av_clip_uint8(block[7]); | |
100 | |||
101 | 1131624 | pixels += line_size; | |
102 | 1131624 | block += 8; | |
103 | } | ||
104 | 141453 | } | |
105 | |||
106 | 14262 | static void put_pixels_clamped4_c(const int16_t *block, uint8_t *av_restrict pixels, | |
107 | int line_size) | ||
108 | { | ||
109 | int i; | ||
110 | |||
111 | /* read the pixels */ | ||
112 |
2/2✓ Branch 0 taken 57048 times.
✓ Branch 1 taken 14262 times.
|
71310 | for(i=0;i<4;i++) { |
113 | 57048 | pixels[0] = av_clip_uint8(block[0]); | |
114 | 57048 | pixels[1] = av_clip_uint8(block[1]); | |
115 | 57048 | pixels[2] = av_clip_uint8(block[2]); | |
116 | 57048 | pixels[3] = av_clip_uint8(block[3]); | |
117 | |||
118 | 57048 | pixels += line_size; | |
119 | 57048 | block += 8; | |
120 | } | ||
121 | 14262 | } | |
122 | |||
123 | ✗ | static void put_pixels_clamped2_c(const int16_t *block, uint8_t *av_restrict pixels, | |
124 | int line_size) | ||
125 | { | ||
126 | int i; | ||
127 | |||
128 | /* read the pixels */ | ||
129 | ✗ | for(i=0;i<2;i++) { | |
130 | ✗ | pixels[0] = av_clip_uint8(block[0]); | |
131 | ✗ | pixels[1] = av_clip_uint8(block[1]); | |
132 | |||
133 | ✗ | pixels += line_size; | |
134 | ✗ | block += 8; | |
135 | } | ||
136 | } | ||
137 | |||
138 | 1027939 | static void put_signed_pixels_clamped_c(const int16_t *block, | |
139 | uint8_t *av_restrict pixels, | ||
140 | ptrdiff_t line_size) | ||
141 | { | ||
142 | int i, j; | ||
143 | |||
144 |
2/2✓ Branch 0 taken 8223512 times.
✓ Branch 1 taken 1027939 times.
|
9251451 | for (i = 0; i < 8; i++) { |
145 |
2/2✓ Branch 0 taken 65788096 times.
✓ Branch 1 taken 8223512 times.
|
74011608 | for (j = 0; j < 8; j++) { |
146 |
2/2✓ Branch 0 taken 129 times.
✓ Branch 1 taken 65787967 times.
|
65788096 | if (*block < -128) |
147 | 129 | *pixels = 0; | |
148 |
2/2✓ Branch 0 taken 44010 times.
✓ Branch 1 taken 65743957 times.
|
65787967 | else if (*block > 127) |
149 | 44010 | *pixels = 255; | |
150 | else | ||
151 | 65743957 | *pixels = (uint8_t) (*block + 128); | |
152 | 65788096 | block++; | |
153 | 65788096 | pixels++; | |
154 | } | ||
155 | 8223512 | pixels += (line_size - 8); | |
156 | } | ||
157 | 1027939 | } | |
158 | |||
159 | 451712 | void ff_add_pixels_clamped_c(const int16_t *block, uint8_t *av_restrict pixels, | |
160 | ptrdiff_t line_size) | ||
161 | { | ||
162 | int i; | ||
163 | |||
164 | /* read the pixels */ | ||
165 |
2/2✓ Branch 0 taken 3613696 times.
✓ Branch 1 taken 451712 times.
|
4065408 | for (i = 0; i < 8; i++) { |
166 | 3613696 | pixels[0] = av_clip_uint8(pixels[0] + block[0]); | |
167 | 3613696 | pixels[1] = av_clip_uint8(pixels[1] + block[1]); | |
168 | 3613696 | pixels[2] = av_clip_uint8(pixels[2] + block[2]); | |
169 | 3613696 | pixels[3] = av_clip_uint8(pixels[3] + block[3]); | |
170 | 3613696 | pixels[4] = av_clip_uint8(pixels[4] + block[4]); | |
171 | 3613696 | pixels[5] = av_clip_uint8(pixels[5] + block[5]); | |
172 | 3613696 | pixels[6] = av_clip_uint8(pixels[6] + block[6]); | |
173 | 3613696 | pixels[7] = av_clip_uint8(pixels[7] + block[7]); | |
174 | 3613696 | pixels += line_size; | |
175 | 3613696 | block += 8; | |
176 | } | ||
177 | 451712 | } | |
178 | |||
179 | 73763 | static void add_pixels_clamped4_c(const int16_t *block, uint8_t *av_restrict pixels, | |
180 | int line_size) | ||
181 | { | ||
182 | int i; | ||
183 | |||
184 | /* read the pixels */ | ||
185 |
2/2✓ Branch 0 taken 295052 times.
✓ Branch 1 taken 73763 times.
|
368815 | for(i=0;i<4;i++) { |
186 | 295052 | pixels[0] = av_clip_uint8(pixels[0] + block[0]); | |
187 | 295052 | pixels[1] = av_clip_uint8(pixels[1] + block[1]); | |
188 | 295052 | pixels[2] = av_clip_uint8(pixels[2] + block[2]); | |
189 | 295052 | pixels[3] = av_clip_uint8(pixels[3] + block[3]); | |
190 | 295052 | pixels += line_size; | |
191 | 295052 | block += 8; | |
192 | } | ||
193 | 73763 | } | |
194 | |||
195 | ✗ | static void add_pixels_clamped2_c(const int16_t *block, uint8_t *av_restrict pixels, | |
196 | int line_size) | ||
197 | { | ||
198 | int i; | ||
199 | |||
200 | /* read the pixels */ | ||
201 | ✗ | for(i=0;i<2;i++) { | |
202 | ✗ | pixels[0] = av_clip_uint8(pixels[0] + block[0]); | |
203 | ✗ | pixels[1] = av_clip_uint8(pixels[1] + block[1]); | |
204 | ✗ | pixels += line_size; | |
205 | ✗ | block += 8; | |
206 | } | ||
207 | } | ||
208 | |||
209 | 14262 | static void ff_jref_idct4_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
210 | { | ||
211 | 14262 | ff_j_rev_dct4 (block); | |
212 | 14262 | put_pixels_clamped4_c(block, dest, line_size); | |
213 | 14262 | } | |
214 | 73763 | static void ff_jref_idct4_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
215 | { | ||
216 | 73763 | ff_j_rev_dct4 (block); | |
217 | 73763 | add_pixels_clamped4_c(block, dest, line_size); | |
218 | 73763 | } | |
219 | |||
220 | ✗ | static void ff_jref_idct2_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
221 | { | ||
222 | ✗ | ff_j_rev_dct2 (block); | |
223 | ✗ | put_pixels_clamped2_c(block, dest, line_size); | |
224 | } | ||
225 | ✗ | static void ff_jref_idct2_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
226 | { | ||
227 | ✗ | ff_j_rev_dct2 (block); | |
228 | ✗ | add_pixels_clamped2_c(block, dest, line_size); | |
229 | } | ||
230 | |||
231 | ✗ | static void ff_jref_idct1_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
232 | { | ||
233 | ✗ | dest[0] = av_clip_uint8((block[0] + 4)>>3); | |
234 | } | ||
235 | ✗ | static void ff_jref_idct1_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block) | |
236 | { | ||
237 | ✗ | dest[0] = av_clip_uint8(dest[0] + ((block[0] + 4)>>3)); | |
238 | } | ||
239 | |||
240 | 1665 | av_cold void ff_idctdsp_init(IDCTDSPContext *c, AVCodecContext *avctx) | |
241 | { | ||
242 | 1665 | av_unused const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8; | |
243 | |||
244 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1664 times.
|
1665 | if (avctx->lowres==1) { |
245 | 1 | c->idct_put = ff_jref_idct4_put; | |
246 | 1 | c->idct_add = ff_jref_idct4_add; | |
247 | 1 | c->idct = ff_j_rev_dct4; | |
248 | 1 | c->perm_type = FF_IDCT_PERM_NONE; | |
249 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1664 times.
|
1664 | } else if (avctx->lowres==2) { |
250 | ✗ | c->idct_put = ff_jref_idct2_put; | |
251 | ✗ | c->idct_add = ff_jref_idct2_add; | |
252 | ✗ | c->idct = ff_j_rev_dct2; | |
253 | ✗ | c->perm_type = FF_IDCT_PERM_NONE; | |
254 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1664 times.
|
1664 | } else if (avctx->lowres==3) { |
255 | ✗ | c->idct_put = ff_jref_idct1_put; | |
256 | ✗ | c->idct_add = ff_jref_idct1_add; | |
257 | ✗ | c->idct = ff_j_rev_dct1; | |
258 | ✗ | c->perm_type = FF_IDCT_PERM_NONE; | |
259 | } else { | ||
260 |
3/4✓ Branch 0 taken 1639 times.
✓ Branch 1 taken 25 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 1639 times.
|
1664 | if (avctx->bits_per_raw_sample == 10 || avctx->bits_per_raw_sample == 9) { |
261 | /* 10-bit MPEG-4 Simple Studio Profile requires a higher precision IDCT | ||
262 | However, it only uses idct_put */ | ||
263 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 24 times.
|
25 | if (c->mpeg4_studio_profile) { |
264 | 1 | c->idct_put = ff_simple_idct_put_int32_10bit; | |
265 | 1 | c->idct_add = NULL; | |
266 | 1 | c->idct = NULL; | |
267 | } else { | ||
268 | 24 | c->idct_put = ff_simple_idct_put_int16_10bit; | |
269 | 24 | c->idct_add = ff_simple_idct_add_int16_10bit; | |
270 | 24 | c->idct = ff_simple_idct_int16_10bit; | |
271 | } | ||
272 | 25 | c->perm_type = FF_IDCT_PERM_NONE; | |
273 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1635 times.
|
1639 | } else if (avctx->bits_per_raw_sample == 12) { |
274 | 4 | c->idct_put = ff_simple_idct_put_int16_12bit; | |
275 | 4 | c->idct_add = ff_simple_idct_add_int16_12bit; | |
276 | 4 | c->idct = ff_simple_idct_int16_12bit; | |
277 | 4 | c->perm_type = FF_IDCT_PERM_NONE; | |
278 | } else { | ||
279 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1631 times.
|
1635 | if (avctx->idct_algo == FF_IDCT_INT) { |
280 | 4 | c->idct_put = ff_jref_idct_put; | |
281 | 4 | c->idct_add = ff_jref_idct_add; | |
282 | 4 | c->idct = ff_j_rev_dct; | |
283 | 4 | c->perm_type = FF_IDCT_PERM_LIBMPEG2; | |
284 | #if CONFIG_FAANIDCT | ||
285 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1631 times.
|
1631 | } else if (avctx->idct_algo == FF_IDCT_FAAN) { |
286 | ✗ | c->idct_put = ff_faanidct_put; | |
287 | ✗ | c->idct_add = ff_faanidct_add; | |
288 | ✗ | c->idct = ff_faanidct; | |
289 | ✗ | c->perm_type = FF_IDCT_PERM_NONE; | |
290 | #endif /* CONFIG_FAANIDCT */ | ||
291 | } else { // accurate/default | ||
292 | 1631 | c->idct_put = ff_simple_idct_put_int16_8bit; | |
293 | 1631 | c->idct_add = ff_simple_idct_add_int16_8bit; | |
294 | 1631 | c->idct = ff_simple_idct_int16_8bit; | |
295 | 1631 | c->perm_type = FF_IDCT_PERM_NONE; | |
296 | } | ||
297 | } | ||
298 | } | ||
299 | |||
300 | 1665 | c->put_pixels_clamped = ff_put_pixels_clamped_c; | |
301 | 1665 | c->put_signed_pixels_clamped = put_signed_pixels_clamped_c; | |
302 | 1665 | c->add_pixels_clamped = ff_add_pixels_clamped_c; | |
303 | |||
304 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1661 times.
|
1665 | if (CONFIG_MPEG4_DECODER && avctx->idct_algo == FF_IDCT_XVID) |
305 | 4 | ff_xvid_idct_init(c, avctx); | |
306 | |||
307 | #if ARCH_AARCH64 | ||
308 | ff_idctdsp_init_aarch64(c, avctx, high_bit_depth); | ||
309 | #elif ARCH_ALPHA | ||
310 | ff_idctdsp_init_alpha(c, avctx, high_bit_depth); | ||
311 | #elif ARCH_ARM | ||
312 | ff_idctdsp_init_arm(c, avctx, high_bit_depth); | ||
313 | #elif ARCH_PPC | ||
314 | ff_idctdsp_init_ppc(c, avctx, high_bit_depth); | ||
315 | #elif ARCH_X86 | ||
316 | 1665 | ff_idctdsp_init_x86(c, avctx, high_bit_depth); | |
317 | #elif ARCH_MIPS | ||
318 | ff_idctdsp_init_mips(c, avctx, high_bit_depth); | ||
319 | #elif ARCH_LOONGARCH | ||
320 | ff_idctdsp_init_loongarch(c, avctx, high_bit_depth); | ||
321 | #endif | ||
322 | |||
323 | 1665 | ff_init_scantable_permutation(c->idct_permutation, | |
324 | c->perm_type); | ||
325 | 1665 | } | |
326 |