FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/dnxhdenc.c
Date: 2024-03-29 01:21:52
Exec Total Coverage
Lines: 683 802 85.2%
Functions: 30 31 96.8%
Branches: 308 419 73.5%

Line Branch Exec Source
1 /*
2 * VC3/DNxHD encoder
3 * Copyright (c) 2007 Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
4 * Copyright (c) 2011 MirriAd Ltd
5 *
6 * VC-3 encoder funded by the British Broadcasting Corporation
7 * 10 bit support added by MirriAd Ltd, Joseph Artsimovich <joseph@mirriad.com>
8 *
9 * This file is part of FFmpeg.
10 *
11 * FFmpeg is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation; either
14 * version 2.1 of the License, or (at your option) any later version.
15 *
16 * FFmpeg is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
20 *
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with FFmpeg; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 */
25
26 #include "libavutil/attributes.h"
27 #include "libavutil/internal.h"
28 #include "libavutil/mem_internal.h"
29 #include "libavutil/opt.h"
30
31 #include "avcodec.h"
32 #include "blockdsp.h"
33 #include "codec_internal.h"
34 #include "encode.h"
35 #include "fdctdsp.h"
36 #include "mathops.h"
37 #include "mpegvideo.h"
38 #include "mpegvideoenc.h"
39 #include "pixblockdsp.h"
40 #include "packet_internal.h"
41 #include "profiles.h"
42 #include "dnxhdenc.h"
43
44 // The largest value that will not lead to overflow for 10-bit samples.
45 #define DNX10BIT_QMAT_SHIFT 18
46 #define RC_VARIANCE 1 // use variance or ssd for fast rc
47 #define LAMBDA_FRAC_BITS 10
48
49 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
50 static const AVOption options[] = {
51 { "nitris_compat", "encode with Avid Nitris compatibility",
52 offsetof(DNXHDEncContext, nitris_compat), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
53 { "ibias", "intra quant bias",
54 offsetof(DNXHDEncContext, intra_quant_bias), AV_OPT_TYPE_INT,
55 { .i64 = 0 }, INT_MIN, INT_MAX, VE },
56 { "profile", NULL, offsetof(DNXHDEncContext, profile), AV_OPT_TYPE_INT,
57 { .i64 = AV_PROFILE_DNXHD },
58 AV_PROFILE_DNXHD, AV_PROFILE_DNXHR_444, VE, .unit = "profile" },
59 { "dnxhd", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHD },
60 0, 0, VE, .unit = "profile" },
61 { "dnxhr_444", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHR_444 },
62 0, 0, VE, .unit = "profile" },
63 { "dnxhr_hqx", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHR_HQX },
64 0, 0, VE, .unit = "profile" },
65 { "dnxhr_hq", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHR_HQ },
66 0, 0, VE, .unit = "profile" },
67 { "dnxhr_sq", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHR_SQ },
68 0, 0, VE, .unit = "profile" },
69 { "dnxhr_lb", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_PROFILE_DNXHR_LB },
70 0, 0, VE, .unit = "profile" },
71 { NULL }
72 };
73
74 static const AVClass dnxhd_class = {
75 .class_name = "dnxhd",
76 .item_name = av_default_item_name,
77 .option = options,
78 .version = LIBAVUTIL_VERSION_INT,
79 };
80
81 97920 static void dnxhd_8bit_get_pixels_8x4_sym(int16_t *restrict block,
82 const uint8_t *pixels,
83 ptrdiff_t line_size)
84 {
85 int i;
86
2/2
✓ Branch 0 taken 391680 times.
✓ Branch 1 taken 97920 times.
489600 for (i = 0; i < 4; i++) {
87 391680 block[0] = pixels[0];
88 391680 block[1] = pixels[1];
89 391680 block[2] = pixels[2];
90 391680 block[3] = pixels[3];
91 391680 block[4] = pixels[4];
92 391680 block[5] = pixels[5];
93 391680 block[6] = pixels[6];
94 391680 block[7] = pixels[7];
95 391680 pixels += line_size;
96 391680 block += 8;
97 }
98 97920 memcpy(block, block - 8, sizeof(*block) * 8);
99 97920 memcpy(block + 8, block - 16, sizeof(*block) * 8);
100 97920 memcpy(block + 16, block - 24, sizeof(*block) * 8);
101 97920 memcpy(block + 24, block - 32, sizeof(*block) * 8);
102 97920 }
103
104 static av_always_inline
105 43200 void dnxhd_10bit_get_pixels_8x4_sym(int16_t *restrict block,
106 const uint8_t *pixels,
107 ptrdiff_t line_size)
108 {
109 43200 memcpy(block + 0 * 8, pixels + 0 * line_size, 8 * sizeof(*block));
110 43200 memcpy(block + 7 * 8, pixels + 0 * line_size, 8 * sizeof(*block));
111 43200 memcpy(block + 1 * 8, pixels + 1 * line_size, 8 * sizeof(*block));
112 43200 memcpy(block + 6 * 8, pixels + 1 * line_size, 8 * sizeof(*block));
113 43200 memcpy(block + 2 * 8, pixels + 2 * line_size, 8 * sizeof(*block));
114 43200 memcpy(block + 5 * 8, pixels + 2 * line_size, 8 * sizeof(*block));
115 43200 memcpy(block + 3 * 8, pixels + 3 * line_size, 8 * sizeof(*block));
116 43200 memcpy(block + 4 * 8, pixels + 3 * line_size, 8 * sizeof(*block));
117 43200 }
118
119 static int dnxhd_10bit_dct_quantize_444(MpegEncContext *ctx, int16_t *block,
120 int n, int qscale, int *overflow)
121 {
122 int i, j, level, last_non_zero, start_i;
123 const int *qmat;
124 const uint8_t *scantable= ctx->intra_scantable.scantable;
125 int bias;
126 int max = 0;
127 unsigned int threshold1, threshold2;
128
129 ctx->fdsp.fdct(block);
130
131 block[0] = (block[0] + 2) >> 2;
132 start_i = 1;
133 last_non_zero = 0;
134 qmat = n < 4 ? ctx->q_intra_matrix[qscale] : ctx->q_chroma_intra_matrix[qscale];
135 bias= ctx->intra_quant_bias * (1 << (16 - 8));
136 threshold1 = (1 << 16) - bias - 1;
137 threshold2 = (threshold1 << 1);
138
139 for (i = 63; i >= start_i; i--) {
140 j = scantable[i];
141 level = block[j] * qmat[j];
142
143 if (((unsigned)(level + threshold1)) > threshold2) {
144 last_non_zero = i;
145 break;
146 } else{
147 block[j]=0;
148 }
149 }
150
151 for (i = start_i; i <= last_non_zero; i++) {
152 j = scantable[i];
153 level = block[j] * qmat[j];
154
155 if (((unsigned)(level + threshold1)) > threshold2) {
156 if (level > 0) {
157 level = (bias + level) >> 16;
158 block[j] = level;
159 } else{
160 level = (bias - level) >> 16;
161 block[j] = -level;
162 }
163 max |= level;
164 } else {
165 block[j] = 0;
166 }
167 }
168 *overflow = ctx->max_qcoeff < max; //overflow might have happened
169
170 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
171 if (ctx->idsp.perm_type != FF_IDCT_PERM_NONE)
172 ff_block_permute(block, ctx->idsp.idct_permutation,
173 scantable, last_non_zero);
174
175 return last_non_zero;
176 }
177
178 3945600 static int dnxhd_10bit_dct_quantize(MpegEncContext *ctx, int16_t *block,
179 int n, int qscale, int *overflow)
180 {
181 3945600 const uint8_t *scantable= ctx->intra_scantable.scantable;
182
2/2
✓ Branch 0 taken 1972800 times.
✓ Branch 1 taken 1972800 times.
3945600 const int *qmat = n<4 ? ctx->q_intra_matrix[qscale] : ctx->q_chroma_intra_matrix[qscale];
183 3945600 int last_non_zero = 0;
184 int i;
185
186 3945600 ctx->fdsp.fdct(block);
187
188 // Divide by 4 with rounding, to compensate scaling of DCT coefficients
189 3945600 block[0] = (block[0] + 2) >> 2;
190
191
2/2
✓ Branch 0 taken 248572800 times.
✓ Branch 1 taken 3945600 times.
252518400 for (i = 1; i < 64; ++i) {
192 248572800 int j = scantable[i];
193 248572800 int sign = FF_SIGNBIT(block[j]);
194 248572800 int level = (block[j] ^ sign) - sign;
195 248572800 level = level * qmat[j] >> DNX10BIT_QMAT_SHIFT;
196 248572800 block[j] = (level ^ sign) - sign;
197
2/2
✓ Branch 0 taken 32824223 times.
✓ Branch 1 taken 215748577 times.
248572800 if (level)
198 32824223 last_non_zero = i;
199 }
200
201 /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
202
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3945600 times.
3945600 if (ctx->idsp.perm_type != FF_IDCT_PERM_NONE)
203 ff_block_permute(block, ctx->idsp.idct_permutation,
204 scantable, last_non_zero);
205
206 3945600 return last_non_zero;
207 }
208
209 73 static av_cold int dnxhd_init_vlc(DNXHDEncContext *ctx)
210 {
211 int i, j, level, run;
212 73 int max_level = 1 << (ctx->bit_depth + 2);
213
214
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 if (!FF_ALLOCZ_TYPED_ARRAY(ctx->orig_vlc_codes, max_level * 4) ||
215
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->orig_vlc_bits, max_level * 4) ||
216
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !(ctx->run_codes = av_mallocz(63 * 2)) ||
217
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 !(ctx->run_bits = av_mallocz(63)))
218 return AVERROR(ENOMEM);
219 73 ctx->vlc_codes = ctx->orig_vlc_codes + max_level * 2;
220 73 ctx->vlc_bits = ctx->orig_vlc_bits + max_level * 2;
221
2/2
✓ Branch 0 taken 192512 times.
✓ Branch 1 taken 73 times.
192585 for (level = -max_level; level < max_level; level++) {
222
2/2
✓ Branch 0 taken 385024 times.
✓ Branch 1 taken 192512 times.
577536 for (run = 0; run < 2; run++) {
223 385024 int index = level * (1 << 1) | run;
224 385024 int sign, offset = 0, alevel = level;
225
226 385024 MASK_ABS(sign, alevel);
227
2/2
✓ Branch 0 taken 366190 times.
✓ Branch 1 taken 18834 times.
385024 if (alevel > 64) {
228 366190 offset = (alevel - 1) >> 6;
229 366190 alevel -= offset << 6;
230 }
231
2/2
✓ Branch 0 taken 65136947 times.
✓ Branch 1 taken 73 times.
65137020 for (j = 0; j < 257; j++) {
232
4/4
✓ Branch 0 taken 1197496 times.
✓ Branch 1 taken 63939451 times.
✓ Branch 2 taken 1163352 times.
✓ Branch 3 taken 34144 times.
65136947 if (ctx->cid_table->ac_info[2*j+0] >> 1 == alevel &&
233
5/6
✓ Branch 0 taken 549285 times.
✓ Branch 1 taken 614067 times.
✓ Branch 2 taken 549285 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 390917 times.
✓ Branch 5 taken 192512 times.
1197496 (!offset || (ctx->cid_table->ac_info[2*j+1] & 1) && offset) &&
234
3/4
✓ Branch 0 taken 192439 times.
✓ Branch 1 taken 198478 times.
✓ Branch 2 taken 192439 times.
✗ Branch 3 not taken.
390917 (!run || (ctx->cid_table->ac_info[2*j+1] & 2) && run)) {
235 av_assert1(!ctx->vlc_codes[index]);
236
2/2
✓ Branch 0 taken 384878 times.
✓ Branch 1 taken 73 times.
384951 if (alevel) {
237 384878 ctx->vlc_codes[index] =
238 384878 (ctx->cid_table->ac_codes[j] << 1) | (sign & 1);
239 384878 ctx->vlc_bits[index] = ctx->cid_table->ac_bits[j] + 1;
240 } else {
241 73 ctx->vlc_codes[index] = ctx->cid_table->ac_codes[j];
242 73 ctx->vlc_bits[index] = ctx->cid_table->ac_bits[j];
243 }
244 384951 break;
245 }
246 }
247
3/4
✓ Branch 0 taken 384878 times.
✓ Branch 1 taken 146 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 384878 times.
385024 av_assert0(!alevel || j < 257);
248
2/2
✓ Branch 0 taken 366190 times.
✓ Branch 1 taken 18834 times.
385024 if (offset) {
249 366190 ctx->vlc_codes[index] =
250 366190 (ctx->vlc_codes[index] << ctx->cid_table->index_bits) | offset;
251 366190 ctx->vlc_bits[index] += ctx->cid_table->index_bits;
252 }
253 }
254 }
255
2/2
✓ Branch 0 taken 4526 times.
✓ Branch 1 taken 73 times.
4599 for (i = 0; i < 62; i++) {
256 4526 int run = ctx->cid_table->run[i];
257
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4526 times.
4526 av_assert0(run < 63);
258 4526 ctx->run_codes[run] = ctx->cid_table->run_codes[i];
259 4526 ctx->run_bits[run] = ctx->cid_table->run_bits[i];
260 }
261 73 return 0;
262 }
263
264 73 static av_cold int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
265 {
266 // init first elem to 1 to avoid div by 0 in convert_matrix
267 73 uint16_t weight_matrix[64] = { 1, }; // convert_matrix needs uint16_t*
268 int qscale, i;
269 73 const uint8_t *luma_weight_table = ctx->cid_table->luma_weight;
270 73 const uint8_t *chroma_weight_table = ctx->cid_table->chroma_weight;
271
272
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 if (!FF_ALLOCZ_TYPED_ARRAY(ctx->qmatrix_l, ctx->m.avctx->qmax + 1) ||
273
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->qmatrix_c, ctx->m.avctx->qmax + 1) ||
274
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->qmatrix_l16, ctx->m.avctx->qmax + 1) ||
275
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->qmatrix_c16, ctx->m.avctx->qmax + 1))
276 return AVERROR(ENOMEM);
277
278
2/2
✓ Branch 0 taken 66 times.
✓ Branch 1 taken 7 times.
73 if (ctx->bit_depth == 8) {
279
2/2
✓ Branch 0 taken 4158 times.
✓ Branch 1 taken 66 times.
4224 for (i = 1; i < 64; i++) {
280 4158 int j = ctx->m.idsp.idct_permutation[ff_zigzag_direct[i]];
281 4158 weight_matrix[j] = ctx->cid_table->luma_weight[i];
282 }
283 66 ff_convert_matrix(&ctx->m, ctx->qmatrix_l, ctx->qmatrix_l16,
284 weight_matrix, ctx->intra_quant_bias, 1,
285 66 ctx->m.avctx->qmax, 1);
286
2/2
✓ Branch 0 taken 4158 times.
✓ Branch 1 taken 66 times.
4224 for (i = 1; i < 64; i++) {
287 4158 int j = ctx->m.idsp.idct_permutation[ff_zigzag_direct[i]];
288 4158 weight_matrix[j] = ctx->cid_table->chroma_weight[i];
289 }
290 66 ff_convert_matrix(&ctx->m, ctx->qmatrix_c, ctx->qmatrix_c16,
291 weight_matrix, ctx->intra_quant_bias, 1,
292 66 ctx->m.avctx->qmax, 1);
293
294
2/2
✓ Branch 0 taken 42184 times.
✓ Branch 1 taken 66 times.
42250 for (qscale = 1; qscale <= ctx->m.avctx->qmax; qscale++) {
295
2/2
✓ Branch 0 taken 2699776 times.
✓ Branch 1 taken 42184 times.
2741960 for (i = 0; i < 64; i++) {
296 2699776 ctx->qmatrix_l[qscale][i] <<= 2;
297 2699776 ctx->qmatrix_c[qscale][i] <<= 2;
298 2699776 ctx->qmatrix_l16[qscale][0][i] <<= 2;
299 2699776 ctx->qmatrix_l16[qscale][1][i] <<= 2;
300 2699776 ctx->qmatrix_c16[qscale][0][i] <<= 2;
301 2699776 ctx->qmatrix_c16[qscale][1][i] <<= 2;
302 }
303 }
304 } else {
305 // 10-bit
306
2/2
✓ Branch 0 taken 56 times.
✓ Branch 1 taken 7 times.
63 for (qscale = 1; qscale <= ctx->m.avctx->qmax; qscale++) {
307
2/2
✓ Branch 0 taken 3528 times.
✓ Branch 1 taken 56 times.
3584 for (i = 1; i < 64; i++) {
308 3528 int j = ff_zigzag_direct[i];
309
310 /* The quantization formula from the VC-3 standard is:
311 * quantized = sign(block[i]) * floor(abs(block[i]/s) * p /
312 * (qscale * weight_table[i]))
313 * Where p is 32 for 8-bit samples and 8 for 10-bit ones.
314 * The s factor compensates scaling of DCT coefficients done by
315 * the DCT routines, and therefore is not present in standard.
316 * It's 8 for 8-bit samples and 4 for 10-bit ones.
317 * We want values of ctx->qtmatrix_l and ctx->qtmatrix_r to be:
318 * ((1 << DNX10BIT_QMAT_SHIFT) * (p / s)) /
319 * (qscale * weight_table[i])
320 * For 10-bit samples, p / s == 2 */
321 3528 ctx->qmatrix_l[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) /
322 3528 (qscale * luma_weight_table[i]);
323 3528 ctx->qmatrix_c[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) /
324 3528 (qscale * chroma_weight_table[i]);
325 }
326 }
327 }
328
329 73 ctx->m.q_chroma_intra_matrix16 = ctx->qmatrix_c16;
330 73 ctx->m.q_chroma_intra_matrix = ctx->qmatrix_c;
331 73 ctx->m.q_intra_matrix16 = ctx->qmatrix_l16;
332 73 ctx->m.q_intra_matrix = ctx->qmatrix_l;
333
334 73 return 0;
335 }
336
337 73 static av_cold int dnxhd_init_rc(DNXHDEncContext *ctx)
338 {
339
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 if (!FF_ALLOCZ_TYPED_ARRAY(ctx->mb_rc, (ctx->m.avctx->qmax + 1) * ctx->m.mb_num))
340 return AVERROR(ENOMEM);
341
342
2/2
✓ Branch 0 taken 58 times.
✓ Branch 1 taken 15 times.
73 if (ctx->m.avctx->mb_decision != FF_MB_DECISION_RD) {
343
1/2
✓ Branch 1 taken 58 times.
✗ Branch 2 not taken.
58 if (!FF_ALLOCZ_TYPED_ARRAY(ctx->mb_cmp, ctx->m.mb_num) ||
344
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 58 times.
58 !FF_ALLOCZ_TYPED_ARRAY(ctx->mb_cmp_tmp, ctx->m.mb_num))
345 return AVERROR(ENOMEM);
346 }
347 73 ctx->frame_bits = (ctx->coding_unit_size -
348 73 ctx->data_offset - 4 - ctx->min_padding) * 8;
349 73 ctx->qscale = 1;
350 73 ctx->lambda = 2 << LAMBDA_FRAC_BITS; // qscale 2
351 73 return 0;
352 }
353
354 73 static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
355 {
356 73 DNXHDEncContext *ctx = avctx->priv_data;
357 int i, ret;
358
359
2/3
✓ Branch 0 taken 66 times.
✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
73 switch (avctx->pix_fmt) {
360 66 case AV_PIX_FMT_YUV422P:
361 66 ctx->bit_depth = 8;
362 66 break;
363 7 case AV_PIX_FMT_YUV422P10:
364 case AV_PIX_FMT_YUV444P10:
365 case AV_PIX_FMT_GBRP10:
366 7 ctx->bit_depth = 10;
367 7 break;
368 }
369
370
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
73 if ((ctx->profile == AV_PROFILE_DNXHR_444 && (avctx->pix_fmt != AV_PIX_FMT_YUV444P10 &&
371 avctx->pix_fmt != AV_PIX_FMT_GBRP10)) ||
372
2/4
✓ Branch 0 taken 73 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
✗ Branch 3 not taken.
73 (ctx->profile != AV_PROFILE_DNXHR_444 && (avctx->pix_fmt == AV_PIX_FMT_YUV444P10 ||
373
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 avctx->pix_fmt == AV_PIX_FMT_GBRP10))) {
374 av_log(avctx, AV_LOG_ERROR,
375 "pixel format is incompatible with DNxHD profile\n");
376 return AVERROR(EINVAL);
377 }
378
379
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
73 if (ctx->profile == AV_PROFILE_DNXHR_HQX && avctx->pix_fmt != AV_PIX_FMT_YUV422P10) {
380 av_log(avctx, AV_LOG_ERROR,
381 "pixel format is incompatible with DNxHR HQX profile\n");
382 return AVERROR(EINVAL);
383 }
384
385
2/2
✓ Branch 0 taken 61 times.
✓ Branch 1 taken 12 times.
73 if ((ctx->profile == AV_PROFILE_DNXHR_LB ||
386
2/2
✓ Branch 0 taken 53 times.
✓ Branch 1 taken 8 times.
61 ctx->profile == AV_PROFILE_DNXHR_SQ ||
387
3/4
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 33 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 40 times.
73 ctx->profile == AV_PROFILE_DNXHR_HQ) && avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
388 av_log(avctx, AV_LOG_ERROR,
389 "pixel format is incompatible with DNxHR LB/SQ/HQ profile\n");
390 return AVERROR(EINVAL);
391 }
392
393 73 ctx->is_444 = ctx->profile == AV_PROFILE_DNXHR_444;
394 73 avctx->profile = ctx->profile;
395 73 ctx->cid = ff_dnxhd_find_cid(avctx, ctx->bit_depth);
396
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (!ctx->cid) {
397 av_log(avctx, AV_LOG_ERROR,
398 "video parameters incompatible with DNxHD. Valid DNxHD profiles:\n");
399 ff_dnxhd_print_profiles(avctx, AV_LOG_ERROR);
400 return AVERROR(EINVAL);
401 }
402 73 av_log(avctx, AV_LOG_DEBUG, "cid %d\n", ctx->cid);
403
404
3/4
✓ Branch 0 taken 40 times.
✓ Branch 1 taken 33 times.
✓ Branch 2 taken 40 times.
✗ Branch 3 not taken.
73 if (ctx->cid >= 1270 && ctx->cid <= 1274)
405 40 avctx->codec_tag = MKTAG('A','V','d','h');
406
407
2/4
✓ Branch 0 taken 73 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 73 times.
73 if (avctx->width < 256 || avctx->height < 120) {
408 av_log(avctx, AV_LOG_ERROR,
409 "Input dimensions too small, input must be at least 256x120\n");
410 return AVERROR(EINVAL);
411 }
412
413 73 ctx->cid_table = ff_dnxhd_get_cid_table(ctx->cid);
414
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 av_assert0(ctx->cid_table);
415
416 73 ctx->m.avctx = avctx;
417 73 ctx->m.mb_intra = 1;
418 73 ctx->m.h263_aic = 1;
419
420 73 avctx->bits_per_raw_sample = ctx->bit_depth;
421
422 73 ff_blockdsp_init(&ctx->bdsp);
423 73 ff_fdctdsp_init(&ctx->m.fdsp, avctx);
424 73 ff_mpv_idct_init(&ctx->m);
425 73 ff_mpegvideoencdsp_init(&ctx->m.mpvencdsp, avctx);
426 73 ff_pixblockdsp_init(&ctx->m.pdsp, avctx);
427 73 ff_dct_encode_init(&ctx->m);
428
429
2/2
✓ Branch 0 taken 40 times.
✓ Branch 1 taken 33 times.
73 if (ctx->profile != AV_PROFILE_DNXHD)
430 40 ff_videodsp_init(&ctx->m.vdsp, ctx->bit_depth);
431
432
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (!ctx->m.dct_quantize)
433 ctx->m.dct_quantize = ff_dct_quantize_c;
434
435
2/4
✓ Branch 0 taken 73 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 73 times.
73 if (ctx->is_444 || ctx->profile == AV_PROFILE_DNXHR_HQX) {
436 ctx->m.dct_quantize = dnxhd_10bit_dct_quantize_444;
437 ctx->get_pixels_8x4_sym = dnxhd_10bit_get_pixels_8x4_sym;
438 ctx->block_width_l2 = 4;
439
2/2
✓ Branch 0 taken 7 times.
✓ Branch 1 taken 66 times.
73 } else if (ctx->bit_depth == 10) {
440 7 ctx->m.dct_quantize = dnxhd_10bit_dct_quantize;
441 7 ctx->get_pixels_8x4_sym = dnxhd_10bit_get_pixels_8x4_sym;
442 7 ctx->block_width_l2 = 4;
443 } else {
444 66 ctx->get_pixels_8x4_sym = dnxhd_8bit_get_pixels_8x4_sym;
445 66 ctx->block_width_l2 = 3;
446 }
447
448 73 ff_dnxhdenc_init(ctx);
449
450 73 ctx->m.mb_height = (avctx->height + 15) / 16;
451 73 ctx->m.mb_width = (avctx->width + 15) / 16;
452
453
2/2
✓ Branch 0 taken 11 times.
✓ Branch 1 taken 62 times.
73 if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
454 11 ctx->interlaced = 1;
455 11 ctx->m.mb_height /= 2;
456 }
457
458
3/4
✓ Branch 0 taken 11 times.
✓ Branch 1 taken 62 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 11 times.
73 if (ctx->interlaced && ctx->profile != AV_PROFILE_DNXHD) {
459 av_log(avctx, AV_LOG_ERROR,
460 "Interlaced encoding is not supported for DNxHR profiles.\n");
461 return AVERROR(EINVAL);
462 }
463
464 73 ctx->m.mb_num = ctx->m.mb_height * ctx->m.mb_width;
465
466
2/2
✓ Branch 0 taken 40 times.
✓ Branch 1 taken 33 times.
73 if (ctx->cid_table->frame_size == DNXHD_VARIABLE) {
467 40 ctx->frame_size = ff_dnxhd_get_hr_frame_size(ctx->cid,
468 avctx->width, avctx->height);
469
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 40 times.
40 av_assert0(ctx->frame_size >= 0);
470 40 ctx->coding_unit_size = ctx->frame_size;
471 } else {
472 33 ctx->frame_size = ctx->cid_table->frame_size;
473 33 ctx->coding_unit_size = ctx->cid_table->coding_unit_size;
474 }
475
476
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 61 times.
73 if (ctx->m.mb_height > 68)
477 12 ctx->data_offset = 0x170 + (ctx->m.mb_height << 2);
478 else
479 61 ctx->data_offset = 0x280;
480
481 // XXX tune lbias/cbias
482
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 if ((ret = dnxhd_init_qmat(ctx, ctx->intra_quant_bias, 0)) < 0)
483 return ret;
484
485 /* Avid Nitris hardware decoder requires a minimum amount of padding
486 * in the coding unit payload */
487
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (ctx->nitris_compat)
488 ctx->min_padding = 1600;
489
490
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 if ((ret = dnxhd_init_vlc(ctx)) < 0)
491 return ret;
492
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 if ((ret = dnxhd_init_rc(ctx)) < 0)
493 return ret;
494
495
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 if (!FF_ALLOCZ_TYPED_ARRAY(ctx->slice_size, ctx->m.mb_height) ||
496
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->slice_offs, ctx->m.mb_height) ||
497
1/2
✓ Branch 1 taken 73 times.
✗ Branch 2 not taken.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->mb_bits, ctx->m.mb_num) ||
498
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 73 times.
73 !FF_ALLOCZ_TYPED_ARRAY(ctx->mb_qscale, ctx->m.mb_num))
499 return AVERROR(ENOMEM);
500
501
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (avctx->active_thread_type == FF_THREAD_SLICE) {
502 if (avctx->thread_count > MAX_THREADS) {
503 av_log(avctx, AV_LOG_ERROR, "too many threads\n");
504 return AVERROR(EINVAL);
505 }
506 }
507
508
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (avctx->qmax <= 1) {
509 av_log(avctx, AV_LOG_ERROR, "qmax must be at least 2\n");
510 return AVERROR(EINVAL);
511 }
512
513 73 ctx->thread[0] = ctx;
514
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (avctx->active_thread_type == FF_THREAD_SLICE) {
515 for (i = 1; i < avctx->thread_count; i++) {
516 ctx->thread[i] = av_memdup(ctx, sizeof(DNXHDEncContext));
517 if (!ctx->thread[i])
518 return AVERROR(ENOMEM);
519 }
520 }
521
522 73 return 0;
523 }
524
525 380 static int dnxhd_write_header(AVCodecContext *avctx, uint8_t *buf)
526 {
527 380 DNXHDEncContext *ctx = avctx->priv_data;
528
529 380 memset(buf, 0, ctx->data_offset);
530
531 // * write prefix */
532 380 AV_WB16(buf + 0x02, ctx->data_offset);
533
3/4
✓ Branch 0 taken 200 times.
✓ Branch 1 taken 180 times.
✓ Branch 2 taken 200 times.
✗ Branch 3 not taken.
380 if (ctx->cid >= 1270 && ctx->cid <= 1274)
534 200 buf[4] = 0x03;
535 else
536 180 buf[4] = 0x01;
537
538
2/2
✓ Branch 0 taken 110 times.
✓ Branch 1 taken 270 times.
380 buf[5] = ctx->interlaced ? ctx->cur_field + 2 : 0x01;
539 380 buf[6] = 0x80; // crc flag off
540 380 buf[7] = 0xa0; // reserved
541 380 AV_WB16(buf + 0x18, avctx->height >> ctx->interlaced); // ALPF
542 380 AV_WB16(buf + 0x1a, avctx->width); // SPL
543 380 AV_WB16(buf + 0x1d, avctx->height >> ctx->interlaced); // NAL
544
545
2/2
✓ Branch 0 taken 55 times.
✓ Branch 1 taken 325 times.
380 buf[0x21] = ctx->bit_depth == 10 ? 0x58 : 0x38;
546 380 buf[0x22] = 0x88 + (ctx->interlaced << 2);
547 380 AV_WB32(buf + 0x28, ctx->cid); // CID
548
2/2
✓ Branch 0 taken 270 times.
✓ Branch 1 taken 110 times.
380 buf[0x2c] = (!ctx->interlaced << 7) | (ctx->is_444 << 6) | (avctx->pix_fmt == AV_PIX_FMT_YUV444P10);
549
550 380 buf[0x5f] = 0x01; // UDL
551
552 380 buf[0x167] = 0x02; // reserved
553 380 AV_WB16(buf + 0x16a, ctx->m.mb_height * 4 + 4); // MSIPS
554 380 AV_WB16(buf + 0x16c, ctx->m.mb_height); // Ns
555 380 buf[0x16f] = 0x10; // reserved
556
557 380 ctx->msip = buf + 0x170;
558 380 return 0;
559 }
560
561 26918560 static av_always_inline void dnxhd_encode_dc(DNXHDEncContext *ctx, int diff)
562 {
563 int nbits;
564
2/2
✓ Branch 0 taken 12317800 times.
✓ Branch 1 taken 14600760 times.
26918560 if (diff < 0) {
565 12317800 nbits = av_log2_16bit(-2 * diff);
566 12317800 diff--;
567 } else {
568 14600760 nbits = av_log2_16bit(2 * diff);
569 }
570 26918560 put_bits(&ctx->m.pb, ctx->cid_table->dc_bits[nbits] + nbits,
571 26918560 (ctx->cid_table->dc_codes[nbits] << nbits) +
572 26918560 av_mod_uintp2(diff, nbits));
573 26918560 }
574
575 static av_always_inline
576 26918560 void dnxhd_encode_block(DNXHDEncContext *ctx, int16_t *block,
577 int last_index, int n)
578 {
579 26918560 int last_non_zero = 0;
580 int slevel, i, j;
581
582 26918560 dnxhd_encode_dc(ctx, block[0] - ctx->m.last_dc[n]);
583 26918560 ctx->m.last_dc[n] = block[0];
584
585
2/2
✓ Branch 0 taken 296815535 times.
✓ Branch 1 taken 26918560 times.
323734095 for (i = 1; i <= last_index; i++) {
586 296815535 j = ctx->m.intra_scantable.permutated[i];
587 296815535 slevel = block[j];
588
2/2
✓ Branch 0 taken 111020901 times.
✓ Branch 1 taken 185794634 times.
296815535 if (slevel) {
589 111020901 int run_level = i - last_non_zero - 1;
590 111020901 int rlevel = slevel * (1 << 1) | !!run_level;
591 111020901 put_bits(&ctx->m.pb, ctx->vlc_bits[rlevel], ctx->vlc_codes[rlevel]);
592
2/2
✓ Branch 0 taken 36001883 times.
✓ Branch 1 taken 75019018 times.
111020901 if (run_level)
593 36001883 put_bits(&ctx->m.pb, ctx->run_bits[run_level],
594 36001883 ctx->run_codes[run_level]);
595 111020901 last_non_zero = i;
596 }
597 }
598 26918560 put_bits(&ctx->m.pb, ctx->vlc_bits[0], ctx->vlc_codes[0]); // EOB
599 26918560 }
600
601 static av_always_inline
602 3024000 void dnxhd_unquantize_c(DNXHDEncContext *ctx, int16_t *block, int n,
603 int qscale, int last_index)
604 {
605 const uint8_t *weight_matrix;
606 int level;
607 int i;
608
609
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3024000 times.
3024000 if (ctx->is_444) {
610 weight_matrix = ((n % 6) < 2) ? ctx->cid_table->luma_weight
611 : ctx->cid_table->chroma_weight;
612 } else {
613 3024000 weight_matrix = (n & 2) ? ctx->cid_table->chroma_weight
614
2/2
✓ Branch 0 taken 1512000 times.
✓ Branch 1 taken 1512000 times.
3024000 : ctx->cid_table->luma_weight;
615 }
616
617
2/2
✓ Branch 0 taken 32773315 times.
✓ Branch 1 taken 3024000 times.
35797315 for (i = 1; i <= last_index; i++) {
618 32773315 int j = ctx->m.intra_scantable.permutated[i];
619 32773315 level = block[j];
620
2/2
✓ Branch 0 taken 16774641 times.
✓ Branch 1 taken 15998674 times.
32773315 if (level) {
621
2/2
✓ Branch 0 taken 8458215 times.
✓ Branch 1 taken 8316426 times.
16774641 if (level < 0) {
622 8458215 level = (1 - 2 * level) * qscale * weight_matrix[i];
623
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8458215 times.
8458215 if (ctx->bit_depth == 10) {
624 if (weight_matrix[i] != 8)
625 level += 8;
626 level >>= 4;
627 } else {
628
2/2
✓ Branch 0 taken 6479585 times.
✓ Branch 1 taken 1978630 times.
8458215 if (weight_matrix[i] != 32)
629 6479585 level += 32;
630 8458215 level >>= 6;
631 }
632 8458215 level = -level;
633 } else {
634 8316426 level = (2 * level + 1) * qscale * weight_matrix[i];
635
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8316426 times.
8316426 if (ctx->bit_depth == 10) {
636 if (weight_matrix[i] != 8)
637 level += 8;
638 level >>= 4;
639 } else {
640
2/2
✓ Branch 0 taken 6813970 times.
✓ Branch 1 taken 1502456 times.
8316426 if (weight_matrix[i] != 32)
641 6813970 level += 32;
642 8316426 level >>= 6;
643 }
644 }
645 16774641 block[j] = level;
646 }
647 }
648 3024000 }
649
650 3024000 static av_always_inline int dnxhd_ssd_block(int16_t *qblock, int16_t *block)
651 {
652 3024000 int score = 0;
653 int i;
654
2/2
✓ Branch 0 taken 193536000 times.
✓ Branch 1 taken 3024000 times.
196560000 for (i = 0; i < 64; i++)
655 193536000 score += (block[i] - qblock[i]) * (block[i] - qblock[i]);
656 3024000 return score;
657 }
658
659 static av_always_inline
660 51881432 int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, int16_t *block, int last_index)
661 {
662 51881432 int last_non_zero = 0;
663 51881432 int bits = 0;
664 int i, j, level;
665
2/2
✓ Branch 0 taken 463138205 times.
✓ Branch 1 taken 51881432 times.
515019637 for (i = 1; i <= last_index; i++) {
666 463138205 j = ctx->m.intra_scantable.permutated[i];
667 463138205 level = block[j];
668
2/2
✓ Branch 0 taken 191817206 times.
✓ Branch 1 taken 271320999 times.
463138205 if (level) {
669 191817206 int run_level = i - last_non_zero - 1;
670 191817206 bits += ctx->vlc_bits[level * (1 << 1) |
671 191817206 !!run_level] + ctx->run_bits[run_level];
672 191817206 last_non_zero = i;
673 }
674 }
675 51881432 return bits;
676 }
677
678 static av_always_inline
679 9849999 void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
680 {
681 9849999 const int bs = ctx->block_width_l2;
682 9849999 const int bw = 1 << bs;
683 9849999 int dct_y_offset = ctx->dct_y_offset;
684 9849999 int dct_uv_offset = ctx->dct_uv_offset;
685 9849999 int linesize = ctx->m.linesize;
686 9849999 int uvlinesize = ctx->m.uvlinesize;
687 9849999 const uint8_t *ptr_y = ctx->thread[0]->src[0] +
688 9849999 ((mb_y << 4) * ctx->m.linesize) + (mb_x << bs + 1);
689 9849999 const uint8_t *ptr_u = ctx->thread[0]->src[1] +
690 9849999 ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs + ctx->is_444);
691 9849999 const uint8_t *ptr_v = ctx->thread[0]->src[2] +
692 9849999 ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs + ctx->is_444);
693 9849999 PixblockDSPContext *pdsp = &ctx->m.pdsp;
694 9849999 VideoDSPContext *vdsp = &ctx->m.vdsp;
695
696
6/6
✓ Branch 0 taken 9356799 times.
✓ Branch 1 taken 493200 times.
✓ Branch 2 taken 7207599 times.
✓ Branch 3 taken 2149200 times.
✓ Branch 4 taken 7203152 times.
✓ Branch 5 taken 4447 times.
9849999 if (ctx->bit_depth != 10 && vdsp->emulated_edge_mc && ((mb_x << 4) + 16 > ctx->m.avctx->width ||
697
2/2
✓ Branch 0 taken 18064 times.
✓ Branch 1 taken 7185088 times.
7203152 (mb_y << 4) + 16 > ctx->m.avctx->height)) {
698 22511 int y_w = ctx->m.avctx->width - (mb_x << 4);
699 22511 int y_h = ctx->m.avctx->height - (mb_y << 4);
700 22511 int uv_w = (y_w + 1) / 2;
701 22511 int uv_h = y_h;
702 22511 linesize = 16;
703 22511 uvlinesize = 8;
704
705 22511 vdsp->emulated_edge_mc(&ctx->edge_buf_y[0], ptr_y,
706 linesize, ctx->m.linesize,
707 linesize, 16,
708 0, 0, y_w, y_h);
709 22511 vdsp->emulated_edge_mc(&ctx->edge_buf_uv[0][0], ptr_u,
710 uvlinesize, ctx->m.uvlinesize,
711 uvlinesize, 16,
712 0, 0, uv_w, uv_h);
713 22511 vdsp->emulated_edge_mc(&ctx->edge_buf_uv[1][0], ptr_v,
714 uvlinesize, ctx->m.uvlinesize,
715 uvlinesize, 16,
716 0, 0, uv_w, uv_h);
717
718 22511 dct_y_offset = bw * linesize;
719 22511 dct_uv_offset = bw * uvlinesize;
720 22511 ptr_y = &ctx->edge_buf_y[0];
721 22511 ptr_u = &ctx->edge_buf_uv[0][0];
722 22511 ptr_v = &ctx->edge_buf_uv[1][0];
723
3/6
✓ Branch 0 taken 493200 times.
✓ Branch 1 taken 9334288 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 493200 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
9827488 } else if (ctx->bit_depth == 10 && vdsp->emulated_edge_mc && ((mb_x << 4) + 16 > ctx->m.avctx->width ||
724 (mb_y << 4) + 16 > ctx->m.avctx->height)) {
725 int y_w = ctx->m.avctx->width - (mb_x << 4);
726 int y_h = ctx->m.avctx->height - (mb_y << 4);
727 int uv_w = ctx->is_444 ? y_w : (y_w + 1) / 2;
728 int uv_h = y_h;
729 linesize = 32;
730 uvlinesize = 16 + 16 * ctx->is_444;
731
732 vdsp->emulated_edge_mc(&ctx->edge_buf_y[0], ptr_y,
733 linesize, ctx->m.linesize,
734 linesize / 2, 16,
735 0, 0, y_w, y_h);
736 vdsp->emulated_edge_mc(&ctx->edge_buf_uv[0][0], ptr_u,
737 uvlinesize, ctx->m.uvlinesize,
738 uvlinesize / 2, 16,
739 0, 0, uv_w, uv_h);
740 vdsp->emulated_edge_mc(&ctx->edge_buf_uv[1][0], ptr_v,
741 uvlinesize, ctx->m.uvlinesize,
742 uvlinesize / 2, 16,
743 0, 0, uv_w, uv_h);
744
745 dct_y_offset = bw * linesize / 2;
746 dct_uv_offset = bw * uvlinesize / 2;
747 ptr_y = &ctx->edge_buf_y[0];
748 ptr_u = &ctx->edge_buf_uv[0][0];
749 ptr_v = &ctx->edge_buf_uv[1][0];
750 }
751
752
1/2
✓ Branch 0 taken 9849999 times.
✗ Branch 1 not taken.
9849999 if (!ctx->is_444) {
753 9849999 pdsp->get_pixels(ctx->blocks[0], ptr_y, linesize);
754 9849999 pdsp->get_pixels(ctx->blocks[1], ptr_y + bw, linesize);
755 9849999 pdsp->get_pixels(ctx->blocks[2], ptr_u, uvlinesize);
756 9849999 pdsp->get_pixels(ctx->blocks[3], ptr_v, uvlinesize);
757
758
4/4
✓ Branch 0 taken 131663 times.
✓ Branch 1 taken 9718336 times.
✓ Branch 2 taken 56560 times.
✓ Branch 3 taken 75103 times.
9849999 if (mb_y + 1 == ctx->m.mb_height && ctx->m.avctx->height == 1080) {
759
2/2
✓ Branch 0 taken 35280 times.
✓ Branch 1 taken 21280 times.
56560 if (ctx->interlaced) {
760 35280 ctx->get_pixels_8x4_sym(ctx->blocks[4],
761 ptr_y + dct_y_offset,
762 linesize);
763 35280 ctx->get_pixels_8x4_sym(ctx->blocks[5],
764 35280 ptr_y + dct_y_offset + bw,
765 linesize);
766 35280 ctx->get_pixels_8x4_sym(ctx->blocks[6],
767 ptr_u + dct_uv_offset,
768 uvlinesize);
769 35280 ctx->get_pixels_8x4_sym(ctx->blocks[7],
770 ptr_v + dct_uv_offset,
771 uvlinesize);
772 } else {
773 21280 ctx->bdsp.clear_block(ctx->blocks[4]);
774 21280 ctx->bdsp.clear_block(ctx->blocks[5]);
775 21280 ctx->bdsp.clear_block(ctx->blocks[6]);
776 21280 ctx->bdsp.clear_block(ctx->blocks[7]);
777 }
778 } else {
779 9793439 pdsp->get_pixels(ctx->blocks[4],
780 ptr_y + dct_y_offset, linesize);
781 9793439 pdsp->get_pixels(ctx->blocks[5],
782 9793439 ptr_y + dct_y_offset + bw, linesize);
783 9793439 pdsp->get_pixels(ctx->blocks[6],
784 ptr_u + dct_uv_offset, uvlinesize);
785 9793439 pdsp->get_pixels(ctx->blocks[7],
786 ptr_v + dct_uv_offset, uvlinesize);
787 }
788 } else {
789 pdsp->get_pixels(ctx->blocks[0], ptr_y, linesize);
790 pdsp->get_pixels(ctx->blocks[1], ptr_y + bw, linesize);
791 pdsp->get_pixels(ctx->blocks[6], ptr_y + dct_y_offset, linesize);
792 pdsp->get_pixels(ctx->blocks[7], ptr_y + dct_y_offset + bw, linesize);
793
794 pdsp->get_pixels(ctx->blocks[2], ptr_u, uvlinesize);
795 pdsp->get_pixels(ctx->blocks[3], ptr_u + bw, uvlinesize);
796 pdsp->get_pixels(ctx->blocks[8], ptr_u + dct_uv_offset, uvlinesize);
797 pdsp->get_pixels(ctx->blocks[9], ptr_u + dct_uv_offset + bw, uvlinesize);
798
799 pdsp->get_pixels(ctx->blocks[4], ptr_v, uvlinesize);
800 pdsp->get_pixels(ctx->blocks[5], ptr_v + bw, uvlinesize);
801 pdsp->get_pixels(ctx->blocks[10], ptr_v + dct_uv_offset, uvlinesize);
802 pdsp->get_pixels(ctx->blocks[11], ptr_v + dct_uv_offset + bw, uvlinesize);
803 }
804 9849999 }
805
806 static av_always_inline
807 78799992 int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
808 {
809 int x;
810
811
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 78799992 times.
78799992 if (ctx->is_444) {
812 x = (i >> 1) % 3;
813 } else {
814 const static uint8_t component[8]={0,0,1,2,0,0,1,2};
815 78799992 x = component[i];
816 }
817 78799992 return x;
818 }
819
820 41557 static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg,
821 int jobnr, int threadnr)
822 {
823 41557 DNXHDEncContext *ctx = avctx->priv_data;
824 41557 int mb_y = jobnr, mb_x;
825 41557 int qscale = ctx->qscale;
826 41557 LOCAL_ALIGNED_16(int16_t, block, [64]);
827 41557 ctx = ctx->thread[threadnr];
828
829 41557 ctx->m.last_dc[0] =
830 41557 ctx->m.last_dc[1] =
831 41557 ctx->m.last_dc[2] = 1 << (ctx->bit_depth + 2);
832
833
2/2
✓ Branch 0 taken 6485179 times.
✓ Branch 1 taken 41557 times.
6526736 for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
834 6485179 unsigned mb = mb_y * ctx->m.mb_width + mb_x;
835 6485179 int ssd = 0;
836 6485179 int ac_bits = 0;
837 6485179 int dc_bits = 0;
838 int i;
839
840 6485179 dnxhd_get_blocks(ctx, mb_x, mb_y);
841
842
2/2
✓ Branch 0 taken 51881432 times.
✓ Branch 1 taken 6485179 times.
58366611 for (i = 0; i < 8 + 4 * ctx->is_444; i++) {
843 51881432 int16_t *src_block = ctx->blocks[i];
844 int overflow, nbits, diff, last_index;
845 51881432 int n = dnxhd_switch_matrix(ctx, i);
846
847 51881432 memcpy(block, src_block, 64 * sizeof(*block));
848 51881432 last_index = ctx->m.dct_quantize(&ctx->m, block,
849
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 51881432 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
51881432 ctx->is_444 ? 4 * (n > 0): 4 & (2*i),
850 qscale, &overflow);
851 51881432 ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
852
853 51881432 diff = block[0] - ctx->m.last_dc[n];
854
2/2
✓ Branch 0 taken 23766663 times.
✓ Branch 1 taken 28114769 times.
51881432 if (diff < 0)
855 23766663 nbits = av_log2_16bit(-2 * diff);
856 else
857 28114769 nbits = av_log2_16bit(2 * diff);
858
859 av_assert1(nbits < ctx->bit_depth + 4);
860 51881432 dc_bits += ctx->cid_table->dc_bits[nbits] + nbits;
861
862 51881432 ctx->m.last_dc[n] = block[0];
863
864
2/2
✓ Branch 0 taken 3024000 times.
✓ Branch 1 taken 48857432 times.
51881432 if (avctx->mb_decision == FF_MB_DECISION_RD || !RC_VARIANCE) {
865 3024000 dnxhd_unquantize_c(ctx, block, i, qscale, last_index);
866 3024000 ctx->m.idsp.idct(block);
867 3024000 ssd += dnxhd_ssd_block(block, src_block);
868 }
869 }
870 6485179 ctx->mb_rc[(qscale * ctx->m.mb_num) + mb].ssd = ssd;
871 6485179 ctx->mb_rc[(qscale * ctx->m.mb_num) + mb].bits = ac_bits + dc_bits + 12 +
872 6485179 (1 + ctx->is_444) * 8 * ctx->vlc_bits[0];
873 }
874 41557 return 0;
875 }
876
877 20765 static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg,
878 int jobnr, int threadnr)
879 {
880 20765 DNXHDEncContext *ctx = avctx->priv_data;
881 20765 int mb_y = jobnr, mb_x;
882 20765 ctx = ctx->thread[threadnr];
883 20765 init_put_bits(&ctx->m.pb, (uint8_t *)arg + ctx->data_offset + ctx->slice_offs[jobnr],
884 20765 ctx->slice_size[jobnr]);
885
886 20765 ctx->m.last_dc[0] =
887 20765 ctx->m.last_dc[1] =
888 20765 ctx->m.last_dc[2] = 1 << (ctx->bit_depth + 2);
889
2/2
✓ Branch 0 taken 3364820 times.
✓ Branch 1 taken 20765 times.
3385585 for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
890 3364820 unsigned mb = mb_y * ctx->m.mb_width + mb_x;
891 3364820 int qscale = ctx->mb_qscale[mb];
892 int i;
893
894 3364820 put_bits(&ctx->m.pb, 11, qscale);
895 3364820 put_bits(&ctx->m.pb, 1, avctx->pix_fmt == AV_PIX_FMT_YUV444P10);
896
897 3364820 dnxhd_get_blocks(ctx, mb_x, mb_y);
898
899
2/2
✓ Branch 0 taken 26918560 times.
✓ Branch 1 taken 3364820 times.
30283380 for (i = 0; i < 8 + 4 * ctx->is_444; i++) {
900 26918560 int16_t *block = ctx->blocks[i];
901 26918560 int overflow, n = dnxhd_switch_matrix(ctx, i);
902 26918560 int last_index = ctx->m.dct_quantize(&ctx->m, block,
903
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 26918560 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
26918560 ctx->is_444 ? (((i >> 1) % 3) < 1 ? 0 : 4): 4 & (2*i),
904 qscale, &overflow);
905
906 26918560 dnxhd_encode_block(ctx, block, last_index, n);
907 }
908 }
909
2/2
✓ Branch 1 taken 20141 times.
✓ Branch 2 taken 624 times.
20765 if (put_bits_count(&ctx->m.pb) & 31)
910 20141 put_bits(&ctx->m.pb, 32 - (put_bits_count(&ctx->m.pb) & 31), 0);
911 20765 flush_put_bits(&ctx->m.pb);
912 20765 memset(put_bits_ptr(&ctx->m.pb), 0, put_bytes_left(&ctx->m.pb, 0));
913 20765 return 0;
914 }
915
916 380 static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx)
917 {
918 int mb_y, mb_x;
919 380 int offset = 0;
920
2/2
✓ Branch 0 taken 20765 times.
✓ Branch 1 taken 380 times.
21145 for (mb_y = 0; mb_y < ctx->m.mb_height; mb_y++) {
921 int thread_size;
922 20765 ctx->slice_offs[mb_y] = offset;
923 20765 ctx->slice_size[mb_y] = 0;
924
2/2
✓ Branch 0 taken 3364820 times.
✓ Branch 1 taken 20765 times.
3385585 for (mb_x = 0; mb_x < ctx->m.mb_width; mb_x++) {
925 3364820 unsigned mb = mb_y * ctx->m.mb_width + mb_x;
926 3364820 ctx->slice_size[mb_y] += ctx->mb_bits[mb];
927 }
928 20765 ctx->slice_size[mb_y] = (ctx->slice_size[mb_y] + 31U) & ~31U;
929 20765 ctx->slice_size[mb_y] >>= 3;
930 20765 thread_size = ctx->slice_size[mb_y];
931 20765 offset += thread_size;
932 }
933 380 }
934
935 11520 static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg,
936 int jobnr, int threadnr)
937 {
938 11520 DNXHDEncContext *ctx = avctx->priv_data;
939 11520 int mb_y = jobnr, mb_x, x, y;
940
2/2
✓ Branch 0 taken 220 times.
✓ Branch 1 taken 11300 times.
11740 int partial_last_row = (mb_y == ctx->m.mb_height - 1) &&
941
2/2
✓ Branch 0 taken 135 times.
✓ Branch 1 taken 85 times.
220 ((avctx->height >> ctx->interlaced) & 0xF);
942
943 11520 ctx = ctx->thread[threadnr];
944
2/2
✓ Branch 0 taken 10955 times.
✓ Branch 1 taken 565 times.
11520 if (ctx->bit_depth == 8) {
945 10955 const uint8_t *pix = ctx->thread[0]->src[0] + ((mb_y << 4) * ctx->m.linesize);
946
2/2
✓ Branch 0 taken 1919700 times.
✓ Branch 1 taken 10955 times.
1930655 for (mb_x = 0; mb_x < ctx->m.mb_width; ++mb_x, pix += 16) {
947 1919700 unsigned mb = mb_y * ctx->m.mb_width + mb_x;
948 int sum;
949 int varc;
950
951
6/6
✓ Branch 0 taken 1908265 times.
✓ Branch 1 taken 11435 times.
✓ Branch 2 taken 1907880 times.
✓ Branch 3 taken 385 times.
✓ Branch 4 taken 1901720 times.
✓ Branch 5 taken 6160 times.
1919700 if (!partial_last_row && mb_x * 16 <= avctx->width - 16 && (avctx->width % 16) == 0) {
952 1901720 sum = ctx->m.mpvencdsp.pix_sum(pix, ctx->m.linesize);
953 1901720 varc = ctx->m.mpvencdsp.pix_norm1(pix, ctx->m.linesize);
954 } else {
955 17980 int bw = FFMIN(avctx->width - 16 * mb_x, 16);
956 17980 int bh = FFMIN((avctx->height >> ctx->interlaced) - 16 * mb_y, 16);
957 17980 sum = varc = 0;
958
2/2
✓ Branch 0 taken 226195 times.
✓ Branch 1 taken 17980 times.
244175 for (y = 0; y < bh; y++) {
959
2/2
✓ Branch 0 taken 3573835 times.
✓ Branch 1 taken 226195 times.
3800030 for (x = 0; x < bw; x++) {
960 3573835 uint8_t val = pix[x + y * ctx->m.linesize];
961 3573835 sum += val;
962 3573835 varc += val * val;
963 }
964 }
965 }
966 1919700 varc = (varc - (((unsigned) sum * sum) >> 8) + 128) >> 8;
967
968 1919700 ctx->mb_cmp[mb].value = varc;
969 1919700 ctx->mb_cmp[mb].mb = mb;
970 }
971 } else { // 10-bit
972 565 const int linesize = ctx->m.linesize >> 1;
973
2/2
✓ Branch 0 taken 58800 times.
✓ Branch 1 taken 565 times.
59365 for (mb_x = 0; mb_x < ctx->m.mb_width; ++mb_x) {
974 58800 const uint16_t *pix = (const uint16_t *)ctx->thread[0]->src[0] +
975 58800 ((mb_y << 4) * linesize) + (mb_x << 4);
976 58800 unsigned mb = mb_y * ctx->m.mb_width + mb_x;
977 58800 int sum = 0;
978 58800 int sqsum = 0;
979 58800 int bw = FFMIN(avctx->width - 16 * mb_x, 16);
980 58800 int bh = FFMIN((avctx->height >> ctx->interlaced) - 16 * mb_y, 16);
981 int mean, sqmean;
982 int i, j;
983 // Macroblocks are 16x16 pixels, unlike DCT blocks which are 8x8.
984
2/2
✓ Branch 0 taken 936000 times.
✓ Branch 1 taken 58800 times.
994800 for (i = 0; i < bh; ++i) {
985
2/2
✓ Branch 0 taken 14976000 times.
✓ Branch 1 taken 936000 times.
15912000 for (j = 0; j < bw; ++j) {
986 // Turn 16-bit pixels into 10-bit ones.
987 14976000 const int sample = (unsigned) pix[j] >> 6;
988 14976000 sum += sample;
989 14976000 sqsum += sample * sample;
990 // 2^10 * 2^10 * 16 * 16 = 2^28, which is less than INT_MAX
991 }
992 936000 pix += linesize;
993 }
994 58800 mean = sum >> 8; // 16*16 == 2^8
995 58800 sqmean = sqsum >> 8;
996 58800 ctx->mb_cmp[mb].value = sqmean - mean * mean;
997 58800 ctx->mb_cmp[mb].mb = mb;
998 }
999 }
1000 11520 return 0;
1001 }
1002
1003 15 static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
1004 {
1005 int lambda, up_step, down_step;
1006 15 int last_lower = INT_MAX, last_higher = 0;
1007 int x, y, q;
1008
1009
2/2
✓ Branch 0 taken 105 times.
✓ Branch 1 taken 15 times.
120 for (q = 1; q < avctx->qmax; q++) {
1010 105 ctx->qscale = q;
1011 105 avctx->execute2(avctx, dnxhd_calc_bits_thread,
1012 NULL, NULL, ctx->m.mb_height);
1013 }
1014 15 up_step = down_step = 2 << LAMBDA_FRAC_BITS;
1015 15 lambda = ctx->lambda;
1016
1017 73 for (;;) {
1018 88 int bits = 0;
1019 88 int end = 0;
1020
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 83 times.
88 if (lambda == last_higher) {
1021 5 lambda++;
1022 5 end = 1; // need to set final qscales/bits
1023 }
1024
2/2
✓ Branch 0 taken 3948 times.
✓ Branch 1 taken 61 times.
4009 for (y = 0; y < ctx->m.mb_height; y++) {
1025
2/2
✓ Branch 0 taken 315840 times.
✓ Branch 1 taken 3948 times.
319788 for (x = 0; x < ctx->m.mb_width; x++) {
1026 315840 unsigned min = UINT_MAX;
1027 315840 int qscale = 1;
1028 315840 int mb = y * ctx->m.mb_width + x;
1029 315840 int rc = 0;
1030
2/2
✓ Branch 0 taken 2210880 times.
✓ Branch 1 taken 315840 times.
2526720 for (q = 1; q < avctx->qmax; q++) {
1031 2210880 int i = (q*ctx->m.mb_num) + mb;
1032 2210880 unsigned score = ctx->mb_rc[i].bits * lambda +
1033 2210880 ((unsigned) ctx->mb_rc[i].ssd << LAMBDA_FRAC_BITS);
1034
2/2
✓ Branch 0 taken 532068 times.
✓ Branch 1 taken 1678812 times.
2210880 if (score < min) {
1035 532068 min = score;
1036 532068 qscale = q;
1037 532068 rc = i;
1038 }
1039 }
1040 315840 bits += ctx->mb_rc[rc].bits;
1041 315840 ctx->mb_qscale[mb] = qscale;
1042 315840 ctx->mb_bits[mb] = ctx->mb_rc[rc].bits;
1043 }
1044 3948 bits = (bits + 31) & ~31; // padding
1045
2/2
✓ Branch 0 taken 27 times.
✓ Branch 1 taken 3921 times.
3948 if (bits > ctx->frame_bits)
1046 27 break;
1047 }
1048
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 83 times.
88 if (end) {
1049
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 5 times.
5 if (bits > ctx->frame_bits)
1050 return AVERROR(EINVAL);
1051 5 break;
1052 }
1053
2/2
✓ Branch 0 taken 56 times.
✓ Branch 1 taken 27 times.
83 if (bits < ctx->frame_bits) {
1054 56 last_lower = FFMIN(lambda, last_lower);
1055
2/2
✓ Branch 0 taken 34 times.
✓ Branch 1 taken 22 times.
56 if (last_higher != 0)
1056 34 lambda = (lambda+last_higher)>>1;
1057 else
1058 22 lambda -= down_step;
1059 56 down_step = FFMIN((int64_t)down_step*5, INT_MAX);
1060 56 up_step = 1<<LAMBDA_FRAC_BITS;
1061 56 lambda = FFMAX(1, lambda);
1062
2/2
✓ Branch 0 taken 10 times.
✓ Branch 1 taken 46 times.
56 if (lambda == last_lower)
1063 10 break;
1064 } else {
1065 27 last_higher = FFMAX(lambda, last_higher);
1066
2/2
✓ Branch 0 taken 26 times.
✓ Branch 1 taken 1 times.
27 if (last_lower != INT_MAX)
1067 26 lambda = (lambda+last_lower)>>1;
1068
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 else if ((int64_t)lambda + up_step > INT_MAX)
1069 return AVERROR(EINVAL);
1070 else
1071 1 lambda += up_step;
1072 27 up_step = FFMIN((int64_t)up_step*5, INT_MAX);
1073 27 down_step = 1<<LAMBDA_FRAC_BITS;
1074 }
1075 }
1076 15 ctx->lambda = lambda;
1077 15 return 0;
1078 }
1079
1080 365 static int dnxhd_find_qscale(DNXHDEncContext *ctx)
1081 {
1082 365 int bits = 0;
1083 365 int up_step = 1;
1084 365 int down_step = 1;
1085 365 int last_higher = 0;
1086 365 int last_lower = INT_MAX;
1087 int qscale;
1088 int x, y;
1089
1090 365 qscale = ctx->qscale;
1091 for (;;) {
1092 691 bits = 0;
1093 691 ctx->qscale = qscale;
1094 // XXX avoid recalculating bits
1095 691 ctx->m.avctx->execute2(ctx->m.avctx, dnxhd_calc_bits_thread,
1096 NULL, NULL, ctx->m.mb_height);
1097
2/2
✓ Branch 0 taken 34838 times.
✓ Branch 1 taken 393 times.
35231 for (y = 0; y < ctx->m.mb_height; y++) {
1098
2/2
✓ Branch 0 taken 5760121 times.
✓ Branch 1 taken 34838 times.
5794959 for (x = 0; x < ctx->m.mb_width; x++)
1099 5760121 bits += ctx->mb_rc[(qscale*ctx->m.mb_num) + (y*ctx->m.mb_width+x)].bits;
1100 34838 bits = (bits+31)&~31; // padding
1101
2/2
✓ Branch 0 taken 298 times.
✓ Branch 1 taken 34540 times.
34838 if (bits > ctx->frame_bits)
1102 298 break;
1103 }
1104
2/2
✓ Branch 0 taken 393 times.
✓ Branch 1 taken 298 times.
691 if (bits < ctx->frame_bits) {
1105
2/2
✓ Branch 0 taken 145 times.
✓ Branch 1 taken 248 times.
393 if (qscale == 1)
1106 145 return 1;
1107
2/2
✓ Branch 0 taken 195 times.
✓ Branch 1 taken 53 times.
248 if (last_higher == qscale - 1) {
1108 195 qscale = last_higher;
1109 195 break;
1110 }
1111 53 last_lower = FFMIN(qscale, last_lower);
1112
2/2
✓ Branch 0 taken 38 times.
✓ Branch 1 taken 15 times.
53 if (last_higher != 0)
1113 38 qscale = (qscale + last_higher) >> 1;
1114 else
1115 15 qscale -= down_step++;
1116
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 53 times.
53 if (qscale < 1)
1117 qscale = 1;
1118 53 up_step = 1;
1119 } else {
1120
2/2
✓ Branch 0 taken 25 times.
✓ Branch 1 taken 273 times.
298 if (last_lower == qscale + 1)
1121 25 break;
1122 273 last_higher = FFMAX(qscale, last_higher);
1123
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 268 times.
273 if (last_lower != INT_MAX)
1124 5 qscale = (qscale + last_lower) >> 1;
1125 else
1126 268 qscale += up_step++;
1127 273 down_step = 1;
1128
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 273 times.
273 if (qscale >= ctx->m.avctx->qmax)
1129 return AVERROR(EINVAL);
1130 }
1131 }
1132 220 ctx->qscale = qscale;
1133 220 return 0;
1134 }
1135
1136 #define BUCKET_BITS 8
1137 #define RADIX_PASSES 4
1138 #define NBUCKETS (1 << BUCKET_BITS)
1139
1140 11871000 static inline int get_bucket(int value, int shift)
1141 {
1142 11871000 value >>= shift;
1143 11871000 value &= NBUCKETS - 1;
1144 11871000 return NBUCKETS - 1 - value;
1145 }
1146
1147 220 static void radix_count(const RCCMPEntry *data, int size,
1148 int buckets[RADIX_PASSES][NBUCKETS])
1149 {
1150 int i, j;
1151 220 memset(buckets, 0, sizeof(buckets[0][0]) * RADIX_PASSES * NBUCKETS);
1152
2/2
✓ Branch 0 taken 1978500 times.
✓ Branch 1 taken 220 times.
1978720 for (i = 0; i < size; i++) {
1153 1978500 int v = data[i].value;
1154
2/2
✓ Branch 0 taken 7914000 times.
✓ Branch 1 taken 1978500 times.
9892500 for (j = 0; j < RADIX_PASSES; j++) {
1155 7914000 buckets[j][get_bucket(v, 0)]++;
1156 7914000 v >>= BUCKET_BITS;
1157 }
1158 av_assert1(!v);
1159 }
1160
2/2
✓ Branch 0 taken 880 times.
✓ Branch 1 taken 220 times.
1100 for (j = 0; j < RADIX_PASSES; j++) {
1161 880 int offset = size;
1162
2/2
✓ Branch 0 taken 225280 times.
✓ Branch 1 taken 880 times.
226160 for (i = NBUCKETS - 1; i >= 0; i--)
1163 225280 buckets[j][i] = offset -= buckets[j][i];
1164 av_assert1(!buckets[j][0]);
1165 }
1166 220 }
1167
1168 440 static void radix_sort_pass(RCCMPEntry *dst, const RCCMPEntry *data,
1169 int size, int buckets[NBUCKETS], int pass)
1170 {
1171 440 int shift = pass * BUCKET_BITS;
1172 int i;
1173
2/2
✓ Branch 0 taken 3957000 times.
✓ Branch 1 taken 440 times.
3957440 for (i = 0; i < size; i++) {
1174 3957000 int v = get_bucket(data[i].value, shift);
1175 3957000 int pos = buckets[v]++;
1176 3957000 dst[pos] = data[i];
1177 }
1178 440 }
1179
1180 220 static void radix_sort(RCCMPEntry *data, RCCMPEntry *tmp, int size)
1181 {
1182 int buckets[RADIX_PASSES][NBUCKETS];
1183 220 radix_count(data, size, buckets);
1184 220 radix_sort_pass(tmp, data, size, buckets[0], 0);
1185 220 radix_sort_pass(data, tmp, size, buckets[1], 1);
1186
2/4
✓ Branch 0 taken 220 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 220 times.
220 if (buckets[2][NBUCKETS - 1] || buckets[3][NBUCKETS - 1]) {
1187 radix_sort_pass(tmp, data, size, buckets[2], 2);
1188 radix_sort_pass(data, tmp, size, buckets[3], 3);
1189 }
1190 220 }
1191
1192 365 static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
1193 {
1194 365 int max_bits = 0;
1195 int ret, x, y;
1196
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 365 times.
365 if ((ret = dnxhd_find_qscale(ctx)) < 0)
1197 return ret;
1198
2/2
✓ Branch 0 taken 20090 times.
✓ Branch 1 taken 365 times.
20455 for (y = 0; y < ctx->m.mb_height; y++) {
1199
2/2
✓ Branch 0 taken 3310820 times.
✓ Branch 1 taken 20090 times.
3330910 for (x = 0; x < ctx->m.mb_width; x++) {
1200 3310820 int mb = y * ctx->m.mb_width + x;
1201 3310820 int rc = (ctx->qscale * ctx->m.mb_num ) + mb;
1202 int delta_bits;
1203 3310820 ctx->mb_qscale[mb] = ctx->qscale;
1204 3310820 ctx->mb_bits[mb] = ctx->mb_rc[rc].bits;
1205 3310820 max_bits += ctx->mb_rc[rc].bits;
1206 if (!RC_VARIANCE) {
1207 delta_bits = ctx->mb_rc[rc].bits -
1208 ctx->mb_rc[rc + ctx->m.mb_num].bits;
1209 ctx->mb_cmp[mb].mb = mb;
1210 ctx->mb_cmp[mb].value =
1211 delta_bits ? ((ctx->mb_rc[rc].ssd -
1212 ctx->mb_rc[rc + ctx->m.mb_num].ssd) * 100) /
1213 delta_bits
1214 : INT_MIN; // avoid increasing qscale
1215 }
1216 }
1217 20090 max_bits += 31; // worst padding
1218 }
1219
2/2
✓ Branch 0 taken 220 times.
✓ Branch 1 taken 145 times.
365 if (!ret) {
1220 if (RC_VARIANCE)
1221 220 avctx->execute2(avctx, dnxhd_mb_var_thread,
1222 NULL, NULL, ctx->m.mb_height);
1223 220 radix_sort(ctx->mb_cmp, ctx->mb_cmp_tmp, ctx->m.mb_num);
1224 222 retry:
1225
4/4
✓ Branch 0 taken 647231 times.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 647011 times.
✓ Branch 3 taken 220 times.
647233 for (x = 0; x < ctx->m.mb_num && max_bits > ctx->frame_bits; x++) {
1226 647011 int mb = ctx->mb_cmp[x].mb;
1227 647011 int rc = (ctx->qscale * ctx->m.mb_num ) + mb;
1228 647011 max_bits -= ctx->mb_rc[rc].bits -
1229 647011 ctx->mb_rc[rc + ctx->m.mb_num].bits;
1230
1/2
✓ Branch 0 taken 647011 times.
✗ Branch 1 not taken.
647011 if (ctx->mb_qscale[mb] < 255)
1231 647011 ctx->mb_qscale[mb]++;
1232 647011 ctx->mb_bits[mb] = ctx->mb_rc[rc + ctx->m.mb_num].bits;
1233 }
1234
1235
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 220 times.
222 if (max_bits > ctx->frame_bits)
1236 2 goto retry;
1237 }
1238 365 return 0;
1239 }
1240
1241 325 static void dnxhd_load_picture(DNXHDEncContext *ctx, const AVFrame *frame)
1242 {
1243 int i;
1244
1245
2/2
✓ Branch 0 taken 325 times.
✓ Branch 1 taken 325 times.
650 for (i = 0; i < ctx->m.avctx->thread_count; i++) {
1246 325 ctx->thread[i]->m.linesize = frame->linesize[0] << ctx->interlaced;
1247 325 ctx->thread[i]->m.uvlinesize = frame->linesize[1] << ctx->interlaced;
1248 325 ctx->thread[i]->dct_y_offset = ctx->m.linesize *8;
1249 325 ctx->thread[i]->dct_uv_offset = ctx->m.uvlinesize*8;
1250 }
1251
1252
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 325 times.
325 ctx->cur_field = (frame->flags & AV_FRAME_FLAG_INTERLACED) &&
1253 !(frame->flags & AV_FRAME_FLAG_TOP_FIELD_FIRST);
1254 325 }
1255
1256 325 static int dnxhd_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
1257 const AVFrame *frame, int *got_packet)
1258 {
1259 325 DNXHDEncContext *ctx = avctx->priv_data;
1260 325 int first_field = 1;
1261 int offset, i, ret;
1262 uint8_t *buf;
1263
1264
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 325 times.
325 if ((ret = ff_get_encode_buffer(avctx, pkt, ctx->frame_size, 0)) < 0)
1265 return ret;
1266 325 buf = pkt->data;
1267
1268 325 dnxhd_load_picture(ctx, frame);
1269
1270 380 encode_coding_unit:
1271
2/2
✓ Branch 0 taken 1140 times.
✓ Branch 1 taken 380 times.
1520 for (i = 0; i < 3; i++) {
1272 1140 ctx->src[i] = frame->data[i];
1273
4/4
✓ Branch 0 taken 330 times.
✓ Branch 1 taken 810 times.
✓ Branch 2 taken 165 times.
✓ Branch 3 taken 165 times.
1140 if (ctx->interlaced && ctx->cur_field)
1274 165 ctx->src[i] += frame->linesize[i];
1275 }
1276
1277 380 dnxhd_write_header(avctx, buf);
1278
1279
2/2
✓ Branch 0 taken 15 times.
✓ Branch 1 taken 365 times.
380 if (avctx->mb_decision == FF_MB_DECISION_RD)
1280 15 ret = dnxhd_encode_rdo(avctx, ctx);
1281 else
1282 365 ret = dnxhd_encode_fast(avctx, ctx);
1283
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 380 times.
380 if (ret < 0) {
1284 av_log(avctx, AV_LOG_ERROR,
1285 "picture could not fit ratecontrol constraints, increase qmax\n");
1286 return ret;
1287 }
1288
1289 380 dnxhd_setup_threads_slices(ctx);
1290
1291 380 offset = 0;
1292
2/2
✓ Branch 0 taken 20765 times.
✓ Branch 1 taken 380 times.
21145 for (i = 0; i < ctx->m.mb_height; i++) {
1293 20765 AV_WB32(ctx->msip + i * 4, offset);
1294 20765 offset += ctx->slice_size[i];
1295 av_assert1(!(ctx->slice_size[i] & 3));
1296 }
1297
1298 380 avctx->execute2(avctx, dnxhd_encode_thread, buf, NULL, ctx->m.mb_height);
1299
1300 av_assert1(ctx->data_offset + offset + 4 <= ctx->coding_unit_size);
1301 380 memset(buf + ctx->data_offset + offset, 0,
1302 380 ctx->coding_unit_size - 4 - offset - ctx->data_offset);
1303
1304 380 AV_WB32(buf + ctx->coding_unit_size - 4, 0x600DC0DE); // EOF
1305
1306
4/4
✓ Branch 0 taken 110 times.
✓ Branch 1 taken 270 times.
✓ Branch 2 taken 55 times.
✓ Branch 3 taken 55 times.
380 if (ctx->interlaced && first_field) {
1307 55 first_field = 0;
1308 55 ctx->cur_field ^= 1;
1309 55 buf += ctx->coding_unit_size;
1310 55 goto encode_coding_unit;
1311 }
1312
1313 325 ff_side_data_set_encoder_stats(pkt, ctx->qscale * FF_QP2LAMBDA, NULL, 0, AV_PICTURE_TYPE_I);
1314
1315 325 *got_packet = 1;
1316 325 return 0;
1317 }
1318
1319 73 static av_cold int dnxhd_encode_end(AVCodecContext *avctx)
1320 {
1321 73 DNXHDEncContext *ctx = avctx->priv_data;
1322 int i;
1323
1324 73 av_freep(&ctx->orig_vlc_codes);
1325 73 av_freep(&ctx->orig_vlc_bits);
1326 73 av_freep(&ctx->run_codes);
1327 73 av_freep(&ctx->run_bits);
1328
1329 73 av_freep(&ctx->mb_bits);
1330 73 av_freep(&ctx->mb_qscale);
1331 73 av_freep(&ctx->mb_rc);
1332 73 av_freep(&ctx->mb_cmp);
1333 73 av_freep(&ctx->mb_cmp_tmp);
1334 73 av_freep(&ctx->slice_size);
1335 73 av_freep(&ctx->slice_offs);
1336
1337 73 av_freep(&ctx->qmatrix_c);
1338 73 av_freep(&ctx->qmatrix_l);
1339 73 av_freep(&ctx->qmatrix_c16);
1340 73 av_freep(&ctx->qmatrix_l16);
1341
1342
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 73 times.
73 if (ctx->thread[1]) {
1343 for (i = 1; i < avctx->thread_count; i++)
1344 av_freep(&ctx->thread[i]);
1345 }
1346
1347 73 return 0;
1348 }
1349
1350 static const FFCodecDefault dnxhd_defaults[] = {
1351 { "qmax", "1024" }, /* Maximum quantization scale factor allowed for VC-3 */
1352 { NULL },
1353 };
1354
1355 const FFCodec ff_dnxhd_encoder = {
1356 .p.name = "dnxhd",
1357 CODEC_LONG_NAME("VC3/DNxHD"),
1358 .p.type = AVMEDIA_TYPE_VIDEO,
1359 .p.id = AV_CODEC_ID_DNXHD,
1360 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
1361 AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE,
1362 .priv_data_size = sizeof(DNXHDEncContext),
1363 .init = dnxhd_encode_init,
1364 FF_CODEC_ENCODE_CB(dnxhd_encode_picture),
1365 .close = dnxhd_encode_end,
1366 .p.pix_fmts = (const enum AVPixelFormat[]) {
1367 AV_PIX_FMT_YUV422P,
1368 AV_PIX_FMT_YUV422P10,
1369 AV_PIX_FMT_YUV444P10,
1370 AV_PIX_FMT_GBRP10,
1371 AV_PIX_FMT_NONE
1372 },
1373 .p.priv_class = &dnxhd_class,
1374 .defaults = dnxhd_defaults,
1375 .p.profiles = NULL_IF_CONFIG_SMALL(ff_dnxhd_profiles),
1376 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1377 };
1378
1379 73 void ff_dnxhdenc_init(DNXHDEncContext *ctx)
1380 {
1381 #if ARCH_X86
1382 73 ff_dnxhdenc_init_x86(ctx);
1383 #endif
1384 73 }
1385