FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/dxvenc.c
Date: 2024-11-20 23:03:26
Exec Total Coverage
Lines: 126 144 87.5%
Functions: 7 7 100.0%
Branches: 75 98 76.5%
Line Branch Exec Source
1 /*
2 * Resolume DXV encoder
3 * Copyright (C) 2024 Connor Worley <connorbworley@gmail.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (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 GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include <stdint.h>
23
24 #include "libavutil/crc.h"
25 #include "libavutil/imgutils.h"
26 #include "libavutil/mem.h"
27 #include "libavutil/opt.h"
28
29 #include "bytestream.h"
30 #include "codec_internal.h"
31 #include "dxv.h"
32 #include "encode.h"
33 #include "texturedsp.h"
34
35 #define DXV_HEADER_LENGTH 12
36
37 /*
38 * DXV uses LZ-like back-references to avoid copying words that have already
39 * appeared in the decompressed stream. Using a simple hash table (HT)
40 * significantly speeds up the lookback process while encoding.
41 */
42 #define LOOKBACK_HT_ELEMS 0x40000
43 #define LOOKBACK_WORDS 0x20202
44
45 typedef struct HTEntry {
46 uint32_t key;
47 uint32_t pos;
48 } HTEntry;
49
50 2 static void ht_init(HTEntry *ht)
51 {
52
2/2
✓ Branch 0 taken 524288 times.
✓ Branch 1 taken 2 times.
 524290 for (size_t i = 0; i < LOOKBACK_HT_ELEMS; i++) {
53 524288 ht[i].pos = -1;
54 }
55 2 }
56
57 135986 static uint32_t ht_lookup_and_upsert(HTEntry *ht, const AVCRC *hash_ctx,
58 uint32_t key, uint32_t pos)
59 {
60 135986 uint32_t ret = -1;
61 135986 size_t hash = av_crc(hash_ctx, 0, (uint8_t*)&key, 4) % LOOKBACK_HT_ELEMS;
62
1/2
✓ Branch 0 taken 136009 times.
✗ Branch 1 not taken.
 136009 for (size_t i = hash; i < hash + LOOKBACK_HT_ELEMS; i++) {
63 136009 size_t wrapped_index = i % LOOKBACK_HT_ELEMS;
64 136009 HTEntry *entry = &ht[wrapped_index];
65
4/4
✓ Branch 0 taken 5904 times.
✓ Branch 1 taken 130105 times.
✓ Branch 2 taken 5881 times.
✓ Branch 3 taken 23 times.
 136009 if (entry->key == key || entry->pos == -1) {
66 135986 ret = entry->pos;
67 135986 entry->key = key;
68 135986 entry->pos = pos;
69 135986 break;
70 }
71 }
72 135986 return ret;
73 }
74
75 127614 static void ht_delete(HTEntry *ht, const AVCRC *hash_ctx,
76 uint32_t key, uint32_t pos)
77 {
78 127614 HTEntry *removed_entry = NULL;
79 size_t removed_hash;
80 127614 size_t hash = av_crc(hash_ctx, 0, (uint8_t*)&key, 4) % LOOKBACK_HT_ELEMS;
81
82
1/2
✓ Branch 0 taken 128361 times.
✗ Branch 1 not taken.
 128361 for (size_t i = hash; i < hash + LOOKBACK_HT_ELEMS; i++) {
83 128361 size_t wrapped_index = i % LOOKBACK_HT_ELEMS;
84 128361 HTEntry *entry = &ht[wrapped_index];
85
2/2
✓ Branch 0 taken 796 times.
✓ Branch 1 taken 127565 times.
 128361 if (entry->pos == -1)
86 796 return;
87
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 127560 times.
 127565 if (removed_entry) {
88 5 size_t candidate_hash = av_crc(hash_ctx, 0, (uint8_t*)&entry->key, 4) % LOOKBACK_HT_ELEMS;
89
4/8
✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 5 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 5 times.
 5 if ((wrapped_index > removed_hash && (candidate_hash <= removed_hash || candidate_hash > wrapped_index)) ||
90 ✗ (wrapped_index < removed_hash && (candidate_hash <= removed_hash && candidate_hash > wrapped_index))) {
91 ✗ *removed_entry = *entry;
92 ✗ entry->pos = -1;
93 ✗ removed_entry = entry;
94 ✗ removed_hash = wrapped_index;
95 }
96
1/2
✓ Branch 0 taken 127560 times.
✗ Branch 1 not taken.
 127560 } else if (entry->key == key) {
97
2/2
✓ Branch 0 taken 742 times.
✓ Branch 1 taken 126818 times.
 127560 if (entry->pos <= pos) {
98 742 entry->pos = -1;
99 742 removed_entry = entry;
100 742 removed_hash = wrapped_index;
101 } else {
102 126818 return;
103 }
104 }
105 }
106 }
107
108 typedef struct DXVEncContext {
109 AVClass *class;
110
111 PutByteContext pbc;
112
113 uint8_t *tex_data; // Compressed texture
114 int64_t tex_size; // Texture size
115
116 /* Optimal number of slices for parallel decoding */
117 int slice_count;
118
119 TextureDSPThreadContext enc;
120
121 DXVTextureFormat tex_fmt;
122 int (*compress_tex)(AVCodecContext *avctx);
123
124 const AVCRC *crc_ctx;
125
126 HTEntry color_lookback_ht[LOOKBACK_HT_ELEMS];
127 HTEntry lut_lookback_ht[LOOKBACK_HT_ELEMS];
128 } DXVEncContext;
129
130 /* Converts an index offset value to a 2-bit opcode and pushes it to a stream.
131 * Inverse of CHECKPOINT in dxv.c. */
132 #define PUSH_OP(x) \
133 do { \
134 if (state == 16) { \
135 if (bytestream2_get_bytes_left_p(pbc) < 4) { \
136 return AVERROR_INVALIDDATA; \
137 } \
138 value = pbc->buffer; \
139 bytestream2_put_le32(pbc, 0); \
140 state = 0; \
141 } \
142 if (idx >= 0x102 * x) { \
143 op = 3; \
144 bytestream2_put_le16(pbc, (idx / x) - 0x102); \
145 } else if (idx >= 2 * x) { \
146 op = 2; \
147 bytestream2_put_byte(pbc, (idx / x) - 2); \
148 } else if (idx == x) { \
149 op = 1; \
150 } else { \
151 op = 0; \
152 } \
153 AV_WL32(value, AV_RL32(value) | (op << (state * 2))); \
154 state++; \
155 } while (0)
156
157 1 static int dxv_compress_dxt1(AVCodecContext *avctx)
158 {
159 1 DXVEncContext *ctx = avctx->priv_data;
160 1 PutByteContext *pbc = &ctx->pbc;
161 void *value;
162 1 uint32_t color, lut, idx, color_idx, lut_idx, prev_pos, state = 16, pos = 2, op = 0;
163
164 1 ht_init(ctx->color_lookback_ht);
165 1 ht_init(ctx->lut_lookback_ht);
166
167 1 bytestream2_put_le32(pbc, AV_RL32(ctx->tex_data));
168 1 bytestream2_put_le32(pbc, AV_RL32(ctx->tex_data + 4));
169
170 1 ht_lookup_and_upsert(ctx->color_lookback_ht, ctx->crc_ctx, AV_RL32(ctx->tex_data), 0);
171 1 ht_lookup_and_upsert(ctx->lut_lookback_ht, ctx->crc_ctx, AV_RL32(ctx->tex_data + 4), 1);
172
173
2/2
✓ Branch 0 taken 129599 times.
✓ Branch 1 taken 1 times.
 129600 while (pos + 2 <= ctx->tex_size / 4) {
174 129599 idx = 0;
175
176 129599 color = AV_RL32(ctx->tex_data + pos * 4);
177 129599 prev_pos = ht_lookup_and_upsert(ctx->color_lookback_ht, ctx->crc_ctx, color, pos);
178
2/2
✓ Branch 0 taken 127924 times.
✓ Branch 1 taken 1675 times.
 129599 color_idx = prev_pos != -1 ? pos - prev_pos : 0;
179
2/2
✓ Branch 0 taken 63807 times.
✓ Branch 1 taken 65792 times.
 129599 if (pos >= LOOKBACK_WORDS) {
180 63807 uint32_t old_pos = pos - LOOKBACK_WORDS;
181 63807 uint32_t old_color = AV_RL32(ctx->tex_data + old_pos * 4);
182 63807 ht_delete(ctx->color_lookback_ht, ctx->crc_ctx, old_color, old_pos);
183 }
184 129599 pos++;
185
186 129599 lut = AV_RL32(ctx->tex_data + pos * 4);
187
4/4
✓ Branch 0 taken 127924 times.
✓ Branch 1 taken 1675 times.
✓ Branch 2 taken 123214 times.
✓ Branch 3 taken 4710 times.
 129599 if (color_idx && lut == AV_RL32(ctx->tex_data + (pos - color_idx) * 4)) {
188 123214 idx = color_idx;
189 } else {
190 6385 idx = 0;
191 6385 prev_pos = ht_lookup_and_upsert(ctx->lut_lookback_ht, ctx->crc_ctx, lut, pos);
192
2/2
✓ Branch 0 taken 2179 times.
✓ Branch 1 taken 4206 times.
 6385 lut_idx = prev_pos != -1 ? pos - prev_pos : 0;
193 }
194
2/2
✓ Branch 0 taken 63807 times.
✓ Branch 1 taken 65792 times.
 129599 if (pos >= LOOKBACK_WORDS) {
195 63807 uint32_t old_pos = pos - LOOKBACK_WORDS;
196 63807 uint32_t old_lut = AV_RL32(ctx->tex_data + old_pos * 4);
197 63807 ht_delete(ctx->lut_lookback_ht, ctx->crc_ctx, old_lut, old_pos);
198 }
199 129599 pos++;
200
201
9/10
✓ Branch 0 taken 8111 times.
✓ Branch 1 taken 121488 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 8111 times.
✓ Branch 6 taken 2535 times.
✓ Branch 7 taken 127064 times.
✓ Branch 9 taken 798 times.
✓ Branch 10 taken 126266 times.
✓ Branch 12 taken 119881 times.
✓ Branch 13 taken 6385 times.
 129599 PUSH_OP(2);
202
203
2/2
✓ Branch 0 taken 6385 times.
✓ Branch 1 taken 123214 times.
 129599 if (!idx) {
204 6385 idx = color_idx;
205
9/10
✓ Branch 0 taken 385 times.
✓ Branch 1 taken 6000 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 385 times.
✓ Branch 6 taken 3456 times.
✓ Branch 7 taken 2929 times.
✓ Branch 9 taken 730 times.
✓ Branch 10 taken 2199 times.
✓ Branch 12 taken 524 times.
✓ Branch 13 taken 1675 times.
 6385 PUSH_OP(2);
206
2/2
✓ Branch 0 taken 1675 times.
✓ Branch 1 taken 4710 times.
 6385 if (!idx)
207 1675 bytestream2_put_le32(pbc, color);
208
209 6385 idx = lut_idx;
210
9/10
✓ Branch 0 taken 403 times.
✓ Branch 1 taken 5982 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 403 times.
✓ Branch 6 taken 1797 times.
✓ Branch 7 taken 4588 times.
✓ Branch 9 taken 277 times.
✓ Branch 10 taken 4311 times.
✓ Branch 12 taken 105 times.
✓ Branch 13 taken 4206 times.
 6385 PUSH_OP(2);
211
2/2
✓ Branch 0 taken 4206 times.
✓ Branch 1 taken 2179 times.
 6385 if (!idx)
212 4206 bytestream2_put_le32(pbc, lut);
213 }
214 }
215
216 1 return 0;
217 }
218
219 1 static int dxv_encode(AVCodecContext *avctx, AVPacket *pkt,
220 const AVFrame *frame, int *got_packet)
221 {
222 1 DXVEncContext *ctx = avctx->priv_data;
223 1 PutByteContext *pbc = &ctx->pbc;
224 int ret;
225
226 /* unimplemented: needs to depend on compression ratio of tex format */
227 /* under DXT1, we need 3 words to encode load ops for 32 words.
228 * the first 2 words of the texture do not need load ops. */
229 1 ret = ff_alloc_packet(avctx, pkt, DXV_HEADER_LENGTH + ctx->tex_size + AV_CEIL_RSHIFT(ctx->tex_size - 8, 7) * 12);
230
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (ret < 0)
231 ✗ return ret;
232
233
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
 1 if (ctx->enc.tex_funct) {
234 1 ctx->enc.tex_data.out = ctx->tex_data;
235 1 ctx->enc.frame_data.in = frame->data[0];
236 1 ctx->enc.stride = frame->linesize[0];
237 1 ctx->enc.width = avctx->width;
238 1 ctx->enc.height = avctx->height;
239 1 ff_texturedsp_exec_compress_threads(avctx, &ctx->enc);
240 } else {
241 /* unimplemented: YCoCg formats */
242 ✗ return AVERROR_INVALIDDATA;
243 }
244
245 1 bytestream2_init_writer(pbc, pkt->data, pkt->size);
246
247 1 bytestream2_put_le32(pbc, ctx->tex_fmt);
248 1 bytestream2_put_byte(pbc, 4);
249 1 bytestream2_put_byte(pbc, 0);
250 1 bytestream2_put_byte(pbc, 0);
251 1 bytestream2_put_byte(pbc, 0);
252 /* Fill in compressed size later */
253 1 bytestream2_skip_p(pbc, 4);
254
255 1 ret = ctx->compress_tex(avctx);
256
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (ret < 0)
257 ✗ return ret;
258
259 1 AV_WL32(pkt->data + 8, bytestream2_tell_p(pbc) - DXV_HEADER_LENGTH);
260 1 av_shrink_packet(pkt, bytestream2_tell_p(pbc));
261
262 1 *got_packet = 1;
263 1 return 0;
264 }
265
266 9 static av_cold int dxv_init(AVCodecContext *avctx)
267 {
268 9 DXVEncContext *ctx = avctx->priv_data;
269 TextureDSPEncContext texdsp;
270 9 int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx);
271
272
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (ret < 0) {
273 ✗ av_log(avctx, AV_LOG_ERROR, "Invalid image size %dx%d.\n",
274 avctx->width, avctx->height);
275 ✗ return ret;
276 }
277
278
2/4
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 9 times.
 9 if (avctx->width % TEXTURE_BLOCK_W || avctx->height % TEXTURE_BLOCK_H) {
279 ✗ av_log(avctx,
280 AV_LOG_ERROR,
281 "Video size %dx%d is not multiple of "AV_STRINGIFY(TEXTURE_BLOCK_W)"x"AV_STRINGIFY(TEXTURE_BLOCK_H)".\n",
282 avctx->width, avctx->height);
283 ✗ return AVERROR_INVALIDDATA;
284 }
285
286 9 ff_texturedspenc_init(&texdsp);
287
288
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 switch (ctx->tex_fmt) {
289 9 case DXV_FMT_DXT1:
290 9 ctx->compress_tex = dxv_compress_dxt1;
291 9 ctx->enc.tex_funct = texdsp.dxt1_block;
292 9 ctx->enc.tex_ratio = 8;
293 9 break;
294 ✗ default:
295 ✗ av_log(avctx, AV_LOG_ERROR, "Invalid format %08X\n", ctx->tex_fmt);
296 ✗ return AVERROR_INVALIDDATA;
297 }
298 9 ctx->enc.raw_ratio = 16;
299 9 ctx->tex_size = avctx->width / TEXTURE_BLOCK_W *
300 9 avctx->height / TEXTURE_BLOCK_H *
301 9 ctx->enc.tex_ratio;
302 9 ctx->enc.slice_count = av_clip(avctx->thread_count, 1, avctx->height / TEXTURE_BLOCK_H);
303
304 9 ctx->tex_data = av_malloc(ctx->tex_size);
305
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (!ctx->tex_data) {
306 ✗ return AVERROR(ENOMEM);
307 }
308
309 9 ctx->crc_ctx = av_crc_get_table(AV_CRC_32_IEEE);
310
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (!ctx->crc_ctx) {
311 ✗ av_log(avctx, AV_LOG_ERROR, "Could not initialize CRC table.\n");
312 ✗ return AVERROR_BUG;
313 }
314
315 9 return 0;
316 }
317
318 9 static av_cold int dxv_close(AVCodecContext *avctx)
319 {
320 9 DXVEncContext *ctx = avctx->priv_data;
321
322 9 av_freep(&ctx->tex_data);
323
324 9 return 0;
325 }
326
327 #define OFFSET(x) offsetof(DXVEncContext, x)
328 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
329 static const AVOption options[] = {
330 { "format", NULL, OFFSET(tex_fmt), AV_OPT_TYPE_INT, { .i64 = DXV_FMT_DXT1 }, DXV_FMT_DXT1, DXV_FMT_DXT1, FLAGS, .unit = "format" },
331 { "dxt1", "DXT1 (Normal Quality, No Alpha)", 0, AV_OPT_TYPE_CONST, { .i64 = DXV_FMT_DXT1 }, 0, 0, FLAGS, .unit = "format" },
332 { NULL },
333 };
334
335 static const AVClass dxvenc_class = {
336 .class_name = "DXV encoder",
337 .option = options,
338 .version = LIBAVUTIL_VERSION_INT,
339 };
340
341 const FFCodec ff_dxv_encoder = {
342 .p.name = "dxv",
343 CODEC_LONG_NAME("Resolume DXV"),
344 .p.type = AVMEDIA_TYPE_VIDEO,
345 .p.id = AV_CODEC_ID_DXV,
346 .init = dxv_init,
347 FF_CODEC_ENCODE_CB(dxv_encode),
348 .close = dxv_close,
349 .priv_data_size = sizeof(DXVEncContext),
350 .p.capabilities = AV_CODEC_CAP_DR1 |
351 AV_CODEC_CAP_SLICE_THREADS |
352 AV_CODEC_CAP_FRAME_THREADS,
353 .p.priv_class = &dxvenc_class,
354 .p.pix_fmts = (const enum AVPixelFormat[]) {
355 AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE,
356 },
357 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
358 };
359