]>
Commit | Line | Data |
---|---|---|
1 | use nihav_core::codecs::*; | |
2 | use nihav_core::io::byteio::*; | |
3 | use nihav_codec_support::vq::*; | |
4 | ||
5 | #[derive(Default,Clone,Copy,PartialEq,Debug)] | |
6 | struct YUVCode { | |
7 | y: [u8; 4], | |
8 | u: u8, | |
9 | v: u8, | |
10 | } | |
11 | impl VQElement for YUVCode { | |
12 | fn dist(&self, rval: Self) -> u32 { | |
13 | let mut ysum = 0; | |
14 | for (y0, y1) in self.y.iter().zip(rval.y.iter()) { | |
15 | let yd = i32::from(*y0) - i32::from(*y1); | |
16 | ysum += yd * yd; | |
17 | } | |
18 | let ud = i32::from(self.u) - i32::from(rval.u); | |
19 | let vd = i32::from(self.v) - i32::from(rval.v); | |
20 | (ysum + ud * ud + vd * vd) as u32 | |
21 | } | |
22 | fn min_cw() -> Self { YUVCode { y: [0; 4], u: 0, v: 0 } } | |
23 | fn max_cw() -> Self { YUVCode { y: [255; 4], u: 255, v: 255 } } | |
24 | fn min(&self, rval: Self) -> Self { | |
25 | let mut ycode = YUVCode::default(); | |
26 | for i in 0..4 { | |
27 | ycode.y[i] = self.y[i].min(rval.y[i]); | |
28 | } | |
29 | ycode.u = self.u.min(rval.u); | |
30 | ycode.v = self.v.min(rval.v); | |
31 | ycode | |
32 | } | |
33 | fn max(&self, rval: Self) -> Self { | |
34 | let mut ycode = YUVCode::default(); | |
35 | for i in 0..4 { | |
36 | ycode.y[i] = self.y[i].max(rval.y[i]); | |
37 | } | |
38 | ycode.u = self.u.max(rval.u); | |
39 | ycode.v = self.v.max(rval.v); | |
40 | ycode | |
41 | } | |
42 | fn num_components() -> usize { 6 } | |
43 | fn sort_by_component(arr: &mut [Self], component: usize) { | |
44 | let mut counts = [0; 256]; | |
45 | for entry in arr.iter() { | |
46 | let idx = match component { | |
47 | 0 | 1 | 2 | 3 => entry.y[component], | |
48 | 4 => entry.u, | |
49 | _ => entry.v, | |
50 | } as usize; | |
51 | counts[idx] += 1; | |
52 | } | |
53 | let mut offs = [0; 256]; | |
54 | for i in 0..255 { | |
55 | offs[i + 1] = offs[i] + counts[i]; | |
56 | } | |
57 | let mut dst = vec![YUVCode::default(); arr.len()]; | |
58 | for entry in arr.iter() { | |
59 | let idx = match component { | |
60 | 0 | 1 | 2 | 3 => entry.y[component], | |
61 | 4 => entry.u, | |
62 | _ => entry.v, | |
63 | } as usize; | |
64 | dst[offs[idx]] = *entry; | |
65 | offs[idx] += 1; | |
66 | } | |
67 | arr.copy_from_slice(dst.as_slice()); | |
68 | } | |
69 | fn max_dist_component(min: &Self, max: &Self) -> usize { | |
70 | let mut comp = 0; | |
71 | let mut diff = 0; | |
72 | for i in 0..4 { | |
73 | let d = u32::from(max.y[i]) - u32::from(min.y[i]); | |
74 | if d > diff { | |
75 | diff = d; | |
76 | comp = i; | |
77 | } | |
78 | } | |
79 | let ud = u32::from(max.u) - u32::from(min.u); | |
80 | if ud > diff { | |
81 | diff = ud; | |
82 | comp = 4; | |
83 | } | |
84 | let vd = u32::from(max.v) - u32::from(min.v); | |
85 | if vd > diff { | |
86 | comp = 5; | |
87 | } | |
88 | comp | |
89 | } | |
90 | } | |
91 | ||
92 | #[derive(Default)] | |
93 | struct YUVCodeSum { | |
94 | ysum: [u64; 4], | |
95 | usum: u64, | |
96 | vsum: u64, | |
97 | count: u64, | |
98 | } | |
99 | ||
100 | impl VQElementSum<YUVCode> for YUVCodeSum { | |
101 | fn zero() -> Self { Self::default() } | |
102 | fn add(&mut self, rval: YUVCode, count: u64) { | |
103 | for i in 0..4 { | |
104 | self.ysum[i] += u64::from(rval.y[i]) * count; | |
105 | } | |
106 | self.usum += u64::from(rval.u) * count; | |
107 | self.vsum += u64::from(rval.v) * count; | |
108 | self.count += count; | |
109 | } | |
110 | fn get_centroid(&self) -> YUVCode { | |
111 | if self.count != 0 { | |
112 | let mut ycode = YUVCode::default(); | |
113 | for i in 0..4 { | |
114 | ycode.y[i] = ((self.ysum[i] + self.count / 2) / self.count) as u8; | |
115 | } | |
116 | ycode.u = ((self.usum + self.count / 2) / self.count) as u8; | |
117 | ycode.v = ((self.vsum + self.count / 2) / self.count) as u8; | |
118 | ycode | |
119 | } else { | |
120 | YUVCode::default() | |
121 | } | |
122 | } | |
123 | } | |
124 | ||
125 | struct RNG { | |
126 | seed: u32, | |
127 | } | |
128 | ||
129 | impl RNG { | |
130 | fn new() -> Self { Self { seed: 0x12345678 } } | |
131 | fn next(&mut self) -> u8 { | |
132 | let mut x = self.seed; | |
133 | x ^= x.wrapping_shl(13); | |
134 | x ^= x >> 17; | |
135 | self.seed = x; | |
136 | (self.seed >> 24) as u8 | |
137 | } | |
138 | fn fill_entry(&mut self, entry: &mut YUVCode) { | |
139 | for y in entry.y.iter_mut() { | |
140 | *y = self.next(); | |
141 | } | |
142 | entry.u = self.next(); | |
143 | entry.v = self.next(); | |
144 | } | |
145 | } | |
146 | ||
147 | const GRAY_FORMAT: NAPixelFormaton = NAPixelFormaton { | |
148 | model: ColorModel::YUV(YUVSubmodel::YUVJ), | |
149 | components: 1, | |
150 | comp_info: [Some(NAPixelChromaton{h_ss: 0, v_ss: 0, packed: false, depth: 8, shift: 0, comp_offs: 0, next_elem: 1}), None, None, None, None], | |
151 | elem_size: 1, | |
152 | be: true, | |
153 | alpha: false, | |
154 | palette: false, | |
155 | }; | |
156 | ||
157 | struct MaskWriter { | |
158 | masks: Vec<u32>, | |
159 | mask: u32, | |
160 | pos: u8, | |
161 | } | |
162 | ||
163 | impl MaskWriter { | |
164 | fn new() -> Self { | |
165 | Self { | |
166 | masks: Vec::new(), | |
167 | mask: 0, | |
168 | pos: 0, | |
169 | } | |
170 | } | |
171 | fn reset(&mut self) { | |
172 | self.masks.clear(); | |
173 | self.mask = 0; | |
174 | self.pos = 0; | |
175 | } | |
176 | fn put_v1(&mut self) { | |
177 | self.mask <<= 1; | |
178 | self.pos += 1; | |
179 | if self.pos == 32 { | |
180 | self.flush(); | |
181 | } | |
182 | } | |
183 | fn put_v4(&mut self) { | |
184 | self.mask <<= 1; | |
185 | self.mask |= 1; | |
186 | self.pos += 1; | |
187 | if self.pos == 32 { | |
188 | self.flush(); | |
189 | } | |
190 | } | |
191 | fn put_inter(&mut self, skip: bool) { | |
192 | self.mask <<= 1; | |
193 | self.mask |= !skip as u32; | |
194 | self.pos += 1; | |
195 | if self.pos == 32 { | |
196 | self.flush(); | |
197 | } | |
198 | } | |
199 | fn flush(&mut self) { | |
200 | self.masks.push(self.mask); | |
201 | self.mask = 0; | |
202 | self.pos = 0; | |
203 | } | |
204 | fn end(&mut self) { | |
205 | if self.pos == 0 { return; } | |
206 | while self.pos < 32 { | |
207 | self.mask <<= 1; | |
208 | self.pos += 1; | |
209 | } | |
210 | self.flush(); | |
211 | } | |
212 | } | |
213 | ||
214 | #[derive(Clone,Copy,PartialEq)] | |
215 | enum QuantMode { | |
216 | ELBG, | |
217 | Fast, | |
218 | MedianCut, | |
219 | } | |
220 | ||
221 | impl std::string::ToString for QuantMode { | |
222 | fn to_string(&self) -> String { | |
223 | match *self { | |
224 | QuantMode::ELBG => "elbg".to_string(), | |
225 | QuantMode::Fast => "fast".to_string(), | |
226 | QuantMode::MedianCut => "mediancut".to_string(), | |
227 | } | |
228 | } | |
229 | } | |
230 | ||
231 | #[derive(Clone,Copy,PartialEq)] | |
232 | enum CodingMode { | |
233 | Skip, | |
234 | V1, | |
235 | V4 | |
236 | } | |
237 | ||
238 | struct CinepakEncoder { | |
239 | stream: Option<NAStreamRef>, | |
240 | lastfrm: Option<NAVideoBufferRef<u8>>, | |
241 | pkt: Option<NAPacket>, | |
242 | frmcount: u8, | |
243 | key_int: u8, | |
244 | qmode: QuantMode, | |
245 | quality: u8, | |
246 | refine: bool, | |
247 | nstrips: usize, | |
248 | force_v1: bool, | |
249 | cur_strip: usize, | |
250 | v1_entries: Vec<YUVCode>, | |
251 | v4_entries: Vec<YUVCode>, | |
252 | v1_cb: Vec<[YUVCode; 256]>, | |
253 | v4_cb: Vec<[YUVCode; 256]>, | |
254 | v1_cur_cb: Vec<[YUVCode; 256]>, | |
255 | v4_cur_cb: Vec<[YUVCode; 256]>, | |
256 | v1_len: usize, | |
257 | v4_len: usize, | |
258 | v1_idx: Vec<u8>, | |
259 | v4_idx: Vec<u8>, | |
260 | grayscale: bool, | |
261 | rng: RNG, | |
262 | masks: MaskWriter, | |
263 | skip_dist: Vec<u32>, | |
264 | fst_bins: [Vec<YUVCode>; 4], | |
265 | v1_cand: Vec<YUVCode>, | |
266 | v4_cand: Vec<YUVCode>, | |
267 | cmode: Vec<CodingMode>, | |
268 | } | |
269 | ||
270 | fn avg4(a: u8, b: u8, c: u8, d: u8) -> u8 { | |
271 | ((u16::from(a) + u16::from(b) + u16::from(c) + u16::from(d) + 3) >> 2) as u8 | |
272 | } | |
273 | ||
274 | fn variance(a: u8, mean: u8) -> u32 { | |
275 | if a >= mean { | |
276 | u32::from(a - mean) * u32::from(a - mean) | |
277 | } else { | |
278 | u32::from(mean - a) * u32::from(mean - a) | |
279 | } | |
280 | } | |
281 | ||
282 | fn patch_size(bw: &mut ByteWriter, pos: u64) -> EncoderResult<()> { | |
283 | let size = bw.tell() - pos; | |
284 | bw.seek(SeekFrom::Current(-((size + 3) as i64)))?; | |
285 | bw.write_u24be((size + 4) as u32)?; | |
286 | bw.seek(SeekFrom::End(0))?; | |
287 | Ok(()) | |
288 | } | |
289 | ||
290 | fn elbg_quant(entries: &[YUVCode], codebook: &mut [YUVCode]) -> usize { | |
291 | let cb_len = quantise_median_cut::<YUVCode, YUVCodeSum>(entries, codebook); | |
292 | if cb_len < codebook.len() { | |
293 | cb_len | |
294 | } else { | |
295 | let mut elbg: ELBG<YUVCode, YUVCodeSum> = ELBG::new(codebook); | |
296 | elbg.quantise(entries, codebook) | |
297 | } | |
298 | } | |
299 | ||
300 | fn quant_fast(bins: &mut [Vec<YUVCode>; 4], entries: &[YUVCode], codebook: &mut [YUVCode]) -> usize { | |
301 | for bin in bins.iter_mut() { | |
302 | bin.clear(); | |
303 | } | |
304 | for &entry in entries.iter() { | |
305 | let y_avg = avg4(entry.y[0], entry.y[1], entry.y[2], entry.y[3]); | |
306 | let dist = entry.y.iter().fold(0u32, |acc, &x| acc + variance(x, y_avg)); | |
307 | let ilog = if dist == 0 { 0 } else { 32 - dist.leading_zeros() }; | |
308 | let bin = match ilog { | |
309 | 0..=3 => &mut bins[0], | |
310 | 4..=7 => &mut bins[1], | |
311 | 8..=11 => &mut bins[2], | |
312 | _ => &mut bins[3], | |
313 | }; | |
314 | bin.push(entry); | |
315 | } | |
316 | let mut free_cw = codebook.len(); | |
317 | let mut entries_left = entries.len(); | |
318 | let mut offset = 0; | |
319 | for bin in bins.iter() { | |
320 | if bin.is_empty() { | |
321 | continue; | |
322 | } | |
323 | if free_cw == 0 || entries_left == 0 { | |
324 | break; | |
325 | } | |
326 | let target = (free_cw * bin.len() + entries_left - 1) / entries_left; | |
327 | let cur_len = elbg_quant(bin, &mut codebook[offset..][..target]); | |
328 | offset += cur_len; | |
329 | free_cw -= cur_len; | |
330 | entries_left -= bin.len(); | |
331 | } | |
332 | offset | |
333 | } | |
334 | ||
335 | impl CinepakEncoder { | |
336 | fn new() -> Self { | |
337 | Self { | |
338 | stream: None, | |
339 | pkt: None, | |
340 | lastfrm: None, | |
341 | frmcount: 0, | |
342 | qmode: QuantMode::Fast, | |
343 | key_int: 25, | |
344 | quality: 0, | |
345 | refine: false, | |
346 | nstrips: 2, | |
347 | force_v1: false, | |
348 | cur_strip: 0, | |
349 | v1_entries: Vec::new(), | |
350 | v4_entries: Vec::new(), | |
351 | v1_cb: Vec::with_capacity(2), | |
352 | v4_cb: Vec::with_capacity(2), | |
353 | v1_cur_cb: Vec::with_capacity(2), | |
354 | v4_cur_cb: Vec::with_capacity(2), | |
355 | v1_len: 0, | |
356 | v4_len: 0, | |
357 | grayscale: false, | |
358 | rng: RNG::new(), | |
359 | v1_idx: Vec::new(), | |
360 | v4_idx: Vec::new(), | |
361 | masks: MaskWriter::new(), | |
362 | skip_dist: Vec::new(), | |
363 | fst_bins: [Vec::new(), Vec::new(), Vec::new(), Vec::new()], | |
364 | v1_cand: Vec::new(), | |
365 | v4_cand: Vec::new(), | |
366 | cmode: Vec::new(), | |
367 | } | |
368 | } | |
369 | fn read_strip(&mut self, in_frm: &NAVideoBuffer<u8>, start: usize, end: usize) { | |
370 | let ystride = in_frm.get_stride(0); | |
371 | let mut yoff = in_frm.get_offset(0) + start * ystride; | |
372 | let ustride = in_frm.get_stride(1); | |
373 | let mut uoff = in_frm.get_offset(1) + start / 2 * ustride; | |
374 | let vstride = in_frm.get_stride(2); | |
375 | let mut voff = in_frm.get_offset(2) + start / 2 * vstride; | |
376 | let (width, _) = in_frm.get_dimensions(0); | |
377 | let data = in_frm.get_data(); | |
378 | self.v1_entries.clear(); | |
379 | self.v4_entries.clear(); | |
380 | for _ in (start..end).step_by(4) { | |
381 | for x in (0..width).step_by(4) { | |
382 | let mut yblk = [0; 16]; | |
383 | let mut ublk = [128; 4]; | |
384 | let mut vblk = [128; 4]; | |
385 | for j in 0..4 { | |
386 | for i in 0..4 { | |
387 | yblk[i + j * 4] = data[yoff + x + i + j * ystride]; | |
388 | } | |
389 | } | |
390 | if !self.grayscale { | |
391 | for j in 0..2 { | |
392 | for i in 0..2 { | |
393 | ublk[i + j * 2] = data[uoff + x / 2 + i + j * ustride]; | |
394 | vblk[i + j * 2] = data[voff + x / 2 + i + j * vstride]; | |
395 | } | |
396 | } | |
397 | } | |
398 | self.v1_entries.push(YUVCode { | |
399 | y: [avg4(yblk[ 0], yblk[ 1], yblk[ 4], yblk[ 5]), | |
400 | avg4(yblk[ 2], yblk[ 3], yblk[ 6], yblk[ 7]), | |
401 | avg4(yblk[ 8], yblk[ 9], yblk[12], yblk[13]), | |
402 | avg4(yblk[10], yblk[11], yblk[14], yblk[15])], | |
403 | u: avg4(ublk[0], ublk[1], ublk[2], ublk[3]), | |
404 | v: avg4(vblk[0], vblk[1], vblk[2], vblk[3]), | |
405 | }); | |
406 | for i in 0..4 { | |
407 | let yidx = (i & 1) * 2 + (i & 2) * 4; | |
408 | self.v4_entries.push(YUVCode { | |
409 | y: [ yblk[yidx], yblk[yidx + 1], yblk[yidx + 4], yblk[yidx + 5] ], | |
410 | u: ublk[i], | |
411 | v: vblk[i], | |
412 | }); | |
413 | } | |
414 | } | |
415 | yoff += ystride * 4; | |
416 | uoff += ustride * 2; | |
417 | voff += vstride * 2; | |
418 | } | |
419 | } | |
420 | fn find_nearest(codebook: &[YUVCode], code: YUVCode) -> (u8, u32) { | |
421 | let mut min_dist = std::u32::MAX; | |
422 | let mut idx = 0; | |
423 | for (i, cw) in codebook.iter().enumerate() { | |
424 | let dist = cw.dist(code); | |
425 | if dist < min_dist { | |
426 | min_dist = dist; | |
427 | idx = i; | |
428 | if dist == 0 { | |
429 | break; | |
430 | } | |
431 | } | |
432 | } | |
433 | (idx as u8, min_dist) | |
434 | } | |
435 | fn can_update_cb(new_cb: &[YUVCode], old_cb: &[YUVCode], cb_size: usize) -> bool { | |
436 | let mut skip_count = 0; | |
437 | for (new, old) in new_cb.iter().zip(old_cb.iter()) { | |
438 | if new == old { | |
439 | skip_count += 1; | |
440 | } | |
441 | } | |
442 | let full_size = cb_size * new_cb.len(); | |
443 | let upd_size = cb_size * (new_cb.len() - skip_count) + (new_cb.len() + 31) / 32 * 4; | |
444 | upd_size < full_size | |
445 | } | |
446 | fn write_cb(bw: &mut ByteWriter, mut id: u8, new_cb: &[YUVCode], old_cb: &[YUVCode], grayscale: bool, update: bool, num_elem: usize) -> EncoderResult<()> { | |
447 | if grayscale { | |
448 | id |= 4; | |
449 | } | |
450 | if update { | |
451 | id |= 1; | |
452 | } | |
453 | bw.write_byte(id)?; | |
454 | bw.write_u24be(0)?; | |
455 | let chunk_pos = bw.tell(); | |
456 | if !update { | |
457 | for entry in new_cb.iter().take(num_elem) { | |
458 | bw.write_buf(&entry.y)?; | |
459 | if !grayscale { | |
460 | bw.write_byte(entry.u ^ 0x80)?; | |
461 | bw.write_byte(entry.v ^ 0x80)?; | |
462 | } | |
463 | } | |
464 | } else { | |
465 | let mut end = num_elem; | |
466 | for (i, (ncw, ocw)) in new_cb.iter().zip(old_cb.iter()).enumerate().take(num_elem).rev() { | |
467 | if ncw == ocw { | |
468 | end = i; | |
469 | } else { | |
470 | break; | |
471 | } | |
472 | } | |
473 | for i in (0..end).step_by(32) { | |
474 | let mut mask = 0; | |
475 | for j in 0..32 { | |
476 | mask <<= 1; | |
477 | if new_cb[i + j] != old_cb[i + j] { | |
478 | mask |= 1; | |
479 | } | |
480 | } | |
481 | bw.write_u32be(mask)?; | |
482 | for j in 0..32 { | |
483 | if new_cb[i + j] == old_cb[i + j] { continue; } | |
484 | bw.write_buf(&new_cb[i + j].y)?; | |
485 | if !grayscale { | |
486 | bw.write_byte(new_cb[i + j].u ^ 0x80)?; | |
487 | bw.write_byte(new_cb[i + j].v ^ 0x80)?; | |
488 | } | |
489 | } | |
490 | } | |
491 | } | |
492 | patch_size(bw, chunk_pos)?; | |
493 | Ok(()) | |
494 | } | |
495 | fn render_stripe(&mut self, intra: bool, start: usize, end: usize) { | |
496 | if let Some(ref mut dst_frm) = self.lastfrm { | |
497 | let ystride = dst_frm.get_stride(0); | |
498 | let mut yoff = dst_frm.get_offset(0) + start * ystride; | |
499 | let ustride = dst_frm.get_stride(1); | |
500 | let mut uoff = dst_frm.get_offset(1) + start / 2 * ustride; | |
501 | let vstride = dst_frm.get_stride(2); | |
502 | let mut voff = dst_frm.get_offset(2) + start / 2 * vstride; | |
503 | let (width, _) = dst_frm.get_dimensions(0); | |
504 | let data = dst_frm.get_data_mut().unwrap(); | |
505 | let mut miter = self.masks.masks.iter(); | |
506 | let mut v1_iter = self.v1_idx.iter(); | |
507 | let mut v4_iter = self.v4_idx.iter(); | |
508 | let mut cur_mask = 0; | |
509 | let mut cur_bit = 0; | |
510 | for _ in (start..end).step_by(4) { | |
511 | for x in (0..width).step_by(4) { | |
512 | if cur_bit == 0 { | |
513 | if !intra || !self.v1_idx.is_empty() { | |
514 | cur_mask = *miter.next().unwrap(); | |
515 | } else { | |
516 | cur_mask = 0xFFFFFFFF; | |
517 | } | |
518 | cur_bit = 1 << 31; | |
519 | } | |
520 | if !intra { | |
521 | if (cur_mask & cur_bit) == 0 { | |
522 | cur_bit >>= 1; | |
523 | continue; | |
524 | } | |
525 | cur_bit >>= 1; | |
526 | if cur_bit == 0 { | |
527 | cur_mask = *miter.next().unwrap(); | |
528 | cur_bit = 1 << 31; | |
529 | } | |
530 | } | |
531 | if (cur_mask & cur_bit) == 0 { | |
532 | let idx = *v1_iter.next().unwrap() as usize; | |
533 | let cb = &self.v1_cur_cb[self.cur_strip][idx]; | |
534 | ||
535 | let mut coff = yoff + x; | |
536 | data[coff] = cb.y[0]; data[coff + 1] = cb.y[0]; | |
537 | data[coff + 2] = cb.y[1]; data[coff + 3] = cb.y[1]; | |
538 | coff += ystride; | |
539 | data[coff] = cb.y[0]; data[coff + 1] = cb.y[0]; | |
540 | data[coff + 2] = cb.y[1]; data[coff + 3] = cb.y[1]; | |
541 | coff += ystride; | |
542 | data[coff] = cb.y[2]; data[coff + 1] = cb.y[2]; | |
543 | data[coff + 2] = cb.y[3]; data[coff + 3] = cb.y[3]; | |
544 | coff += ystride; | |
545 | data[coff] = cb.y[2]; data[coff + 1] = cb.y[2]; | |
546 | data[coff + 2] = cb.y[3]; data[coff + 3] = cb.y[3]; | |
547 | ||
548 | if !self.grayscale { | |
549 | let mut coff = uoff + x / 2; | |
550 | data[coff] = cb.u; data[coff + 1] = cb.u; | |
551 | coff += ustride; | |
552 | data[coff] = cb.u; data[coff + 1] = cb.u; | |
553 | ||
554 | let mut coff = voff + x / 2; | |
555 | data[coff] = cb.v; data[coff + 1] = cb.v; | |
556 | coff += vstride; | |
557 | data[coff] = cb.v; data[coff + 1] = cb.v; | |
558 | } | |
559 | } else { | |
560 | let idx0 = *v4_iter.next().unwrap() as usize; | |
561 | let cb0 = &self.v4_cur_cb[self.cur_strip][idx0]; | |
562 | let idx1 = *v4_iter.next().unwrap() as usize; | |
563 | let cb1 = &self.v4_cur_cb[self.cur_strip][idx1]; | |
564 | let idx2 = *v4_iter.next().unwrap() as usize; | |
565 | let cb2 = &self.v4_cur_cb[self.cur_strip][idx2]; | |
566 | let idx3 = *v4_iter.next().unwrap() as usize; | |
567 | let cb3 = &self.v4_cur_cb[self.cur_strip][idx3]; | |
568 | ||
569 | let mut coff = yoff + x; | |
570 | data[coff] = cb0.y[0]; data[coff + 1] = cb0.y[1]; | |
571 | data[coff + 2] = cb1.y[0]; data[coff + 3] = cb1.y[1]; | |
572 | coff += ystride; | |
573 | data[coff] = cb0.y[2]; data[coff + 1] = cb0.y[3]; | |
574 | data[coff + 2] = cb1.y[2]; data[coff + 3] = cb1.y[3]; | |
575 | coff += ystride; | |
576 | data[coff] = cb2.y[0]; data[coff + 1] = cb2.y[1]; | |
577 | data[coff + 2] = cb3.y[0]; data[coff + 3] = cb3.y[1]; | |
578 | coff += ystride; | |
579 | data[coff] = cb2.y[2]; data[coff + 1] = cb2.y[3]; | |
580 | data[coff + 2] = cb3.y[2]; data[coff + 3] = cb3.y[3]; | |
581 | ||
582 | if !self.grayscale { | |
583 | let mut coff = uoff + x / 2; | |
584 | data[coff] = cb0.u; data[coff + 1] = cb1.u; | |
585 | coff += ustride; | |
586 | data[coff] = cb2.u; data[coff + 1] = cb3.u; | |
587 | ||
588 | let mut coff = voff + x / 2; | |
589 | data[coff] = cb0.v; data[coff + 1] = cb1.v; | |
590 | coff += vstride; | |
591 | data[coff] = cb2.v; data[coff + 1] = cb3.v; | |
592 | } | |
593 | } | |
594 | cur_bit >>= 1; | |
595 | } | |
596 | yoff += ystride * 4; | |
597 | uoff += ustride * 2; | |
598 | voff += vstride * 2; | |
599 | } | |
600 | } else { | |
601 | unreachable!(); | |
602 | } | |
603 | } | |
604 | fn calc_skip_dist(&mut self, in_frm: &NAVideoBuffer<u8>, start: usize, end: usize) { | |
605 | self.skip_dist.clear(); | |
606 | if let Some(ref ref_frm) = self.lastfrm { | |
607 | let rystride = ref_frm.get_stride(0); | |
608 | let mut ryoff = ref_frm.get_offset(0) + start * rystride; | |
609 | let rustride = ref_frm.get_stride(1); | |
610 | let mut ruoff = ref_frm.get_offset(1) + start / 2 * rustride; | |
611 | let rvstride = ref_frm.get_stride(2); | |
612 | let mut rvoff = ref_frm.get_offset(2) + start / 2 * rvstride; | |
613 | let (width, _) = ref_frm.get_dimensions(0); | |
614 | let rdata = ref_frm.get_data(); | |
615 | ||
616 | let iystride = in_frm.get_stride(0); | |
617 | let mut iyoff = in_frm.get_offset(0) + start * iystride; | |
618 | let iustride = in_frm.get_stride(1); | |
619 | let mut iuoff = in_frm.get_offset(1) + start / 2 * iustride; | |
620 | let ivstride = in_frm.get_stride(2); | |
621 | let mut ivoff = in_frm.get_offset(2) + start / 2 * ivstride; | |
622 | let idata = in_frm.get_data(); | |
623 | ||
624 | for _ in (start..end).step_by(4) { | |
625 | for x in (0..width).step_by(4) { | |
626 | let mut dist = 0; | |
627 | let mut roff = ryoff + x; | |
628 | let mut ioff = iyoff + x; | |
629 | for _ in 0..4 { | |
630 | for i in 0..4 { | |
631 | let d = i32::from(rdata[roff + i]) - i32::from(idata[ioff + i]); | |
632 | dist += d * d; | |
633 | } | |
634 | roff += rystride; | |
635 | ioff += iystride; | |
636 | } | |
637 | if !self.grayscale { | |
638 | let mut roff = ruoff + x / 2; | |
639 | let mut ioff = iuoff + x / 2; | |
640 | let ud = i32::from(rdata[roff]) - i32::from(idata[ioff]); | |
641 | dist += ud * ud; | |
642 | let ud = i32::from(rdata[roff + 1]) - i32::from(idata[ioff + 1]); | |
643 | dist += ud * ud; | |
644 | roff += rustride; ioff += iustride; | |
645 | let ud = i32::from(rdata[roff]) - i32::from(idata[ioff]); | |
646 | dist += ud * ud; | |
647 | let ud = i32::from(rdata[roff + 1]) - i32::from(idata[ioff + 1]); | |
648 | dist += ud * ud; | |
649 | ||
650 | let mut roff = rvoff + x / 2; | |
651 | let mut ioff = ivoff + x / 2; | |
652 | let vd = i32::from(rdata[roff]) - i32::from(idata[ioff]); | |
653 | dist += vd * vd; | |
654 | let vd = i32::from(rdata[roff + 1]) - i32::from(idata[ioff + 1]); | |
655 | dist += vd * vd; | |
656 | roff += rvstride; ioff += ivstride; | |
657 | let vd = i32::from(rdata[roff]) - i32::from(idata[ioff]); | |
658 | dist += vd * vd; | |
659 | let vd = i32::from(rdata[roff + 1]) - i32::from(idata[ioff + 1]); | |
660 | dist += vd * vd; | |
661 | } | |
662 | self.skip_dist.push(dist as u32); | |
663 | } | |
664 | ||
665 | iyoff += iystride * 4; | |
666 | iuoff += iustride * 2; | |
667 | ivoff += ivstride * 2; | |
668 | ryoff += rystride * 4; | |
669 | ruoff += rustride * 2; | |
670 | rvoff += rvstride * 2; | |
671 | } | |
672 | } else { | |
673 | unreachable!(); | |
674 | } | |
675 | } | |
676 | fn quant_vectors(&mut self) { | |
677 | match self.qmode { | |
678 | QuantMode::ELBG => { | |
679 | self.v1_len = elbg_quant(&self.v1_entries, &mut self.v1_cur_cb[self.cur_strip]); | |
680 | self.v4_len = if !self.force_v1 { | |
681 | elbg_quant(&self.v4_entries, &mut self.v4_cur_cb[self.cur_strip]) | |
682 | } else { | |
683 | 0 | |
684 | }; | |
685 | }, | |
686 | QuantMode::Fast => { | |
687 | for bin in self.fst_bins.iter_mut() { | |
688 | bin.clear(); | |
689 | } | |
690 | self.v1_len = quant_fast(&mut self.fst_bins, &self.v1_entries, &mut self.v1_cur_cb[self.cur_strip]); | |
691 | self.v4_len = if !self.force_v1 { | |
692 | quant_fast(&mut self.fst_bins, &self.v4_entries, &mut self.v4_cur_cb[self.cur_strip]) | |
693 | } else { | |
694 | 0 | |
695 | }; | |
696 | }, | |
697 | QuantMode::MedianCut => { | |
698 | self.v1_len = quantise_median_cut::<YUVCode, YUVCodeSum>(&self.v1_entries, &mut self.v1_cur_cb[self.cur_strip]); | |
699 | if !self.force_v1 { | |
700 | self.v4_len = quantise_median_cut::<YUVCode, YUVCodeSum>(&self.v4_entries, &mut self.v4_cur_cb[self.cur_strip]); | |
701 | } else { | |
702 | self.v4_len = 0; | |
703 | } | |
704 | }, | |
705 | }; | |
706 | ||
707 | for e in self.v1_cur_cb[self.cur_strip].iter_mut().skip(self.v1_len) { *e = YUVCode::default(); } | |
708 | for e in self.v4_cur_cb[self.cur_strip].iter_mut().skip(self.v4_len) { *e = YUVCode::default(); } | |
709 | } | |
710 | fn refine_vectors(&mut self) { | |
711 | match self.qmode { | |
712 | QuantMode::ELBG => { | |
713 | self.v1_len = if !self.v1_cand.is_empty() { | |
714 | elbg_quant(&self.v1_cand, &mut self.v1_cur_cb[self.cur_strip]) | |
715 | } else { | |
716 | 0 | |
717 | }; | |
718 | self.v4_len = if !self.force_v1 && !self.v4_cand.is_empty() { | |
719 | elbg_quant(&self.v4_cand, &mut self.v4_cur_cb[self.cur_strip]) | |
720 | } else { | |
721 | 0 | |
722 | }; | |
723 | }, | |
724 | QuantMode::Fast => { | |
725 | for bin in self.fst_bins.iter_mut() { | |
726 | bin.clear(); | |
727 | } | |
728 | self.v1_len = if !self.v1_cand.is_empty() { | |
729 | quant_fast(&mut self.fst_bins, &self.v1_cand, &mut self.v1_cur_cb[self.cur_strip]) | |
730 | } else { | |
731 | 0 | |
732 | }; | |
733 | self.v4_len = if !self.force_v1 && !self.v4_cand.is_empty() { | |
734 | quant_fast(&mut self.fst_bins, &self.v4_cand, &mut self.v4_cur_cb[self.cur_strip]) | |
735 | } else { | |
736 | 0 | |
737 | }; | |
738 | }, | |
739 | QuantMode::MedianCut => { | |
740 | self.v1_len = if !self.v1_cand.is_empty() { | |
741 | quantise_median_cut::<YUVCode, YUVCodeSum>(&self.v1_cand, &mut self.v1_cur_cb[self.cur_strip]) | |
742 | } else { | |
743 | 0 | |
744 | }; | |
745 | if !self.force_v1 && !self.v4_cand.is_empty() { | |
746 | self.v4_len = quantise_median_cut::<YUVCode, YUVCodeSum>(&self.v4_cand, &mut self.v4_cur_cb[self.cur_strip]); | |
747 | } else { | |
748 | self.v4_len = 0; | |
749 | } | |
750 | }, | |
751 | }; | |
752 | ||
753 | for e in self.v1_cur_cb[self.cur_strip].iter_mut().skip(self.v1_len) { *e = YUVCode::default(); } | |
754 | for e in self.v4_cur_cb[self.cur_strip].iter_mut().skip(self.v4_len) { *e = YUVCode::default(); } | |
755 | } | |
756 | fn encode_intra(&mut self, bw: &mut ByteWriter, in_frm: &NAVideoBuffer<u8>) -> EncoderResult<bool> { | |
757 | let (width, height) = in_frm.get_dimensions(0); | |
758 | let mut strip_h = (height / self.nstrips + 3) & !3; | |
759 | if strip_h == 0 { | |
760 | self.nstrips = 1; | |
761 | strip_h = height; | |
762 | } | |
763 | let mut start_line = 0; | |
764 | let mut end_line = strip_h; | |
765 | ||
766 | bw.write_byte(0)?; // intra flag | |
767 | bw.write_u24be(0)?; // frame size | |
768 | let frame_data_pos = bw.tell(); | |
769 | bw.write_u16be(width as u16)?; | |
770 | bw.write_u16be(height as u16)?; | |
771 | bw.write_u16be(self.nstrips as u16)?; | |
772 | ||
773 | self.cur_strip = 0; | |
774 | for entry in self.v1_cb[self.cur_strip].iter_mut() { | |
775 | self.rng.fill_entry(entry); | |
776 | } | |
777 | for entry in self.v4_cb[self.cur_strip].iter_mut() { | |
778 | self.rng.fill_entry(entry); | |
779 | } | |
780 | while start_line < height { | |
781 | self.read_strip(in_frm, start_line, end_line); | |
782 | ||
783 | if self.cur_strip > 0 { | |
784 | self.v1_cb[self.cur_strip] = self.v1_cb[self.cur_strip - 1]; | |
785 | self.v4_cb[self.cur_strip] = self.v4_cb[self.cur_strip - 1]; | |
786 | } | |
787 | self.quant_vectors(); | |
788 | if self.grayscale { | |
789 | for cw in self.v1_cur_cb[self.cur_strip].iter_mut() { | |
790 | cw.u = 128; | |
791 | cw.v = 128; | |
792 | } | |
793 | for cw in self.v4_cur_cb[self.cur_strip].iter_mut() { | |
794 | cw.u = 128; | |
795 | cw.v = 128; | |
796 | } | |
797 | } | |
798 | ||
799 | self.v1_idx.clear(); | |
800 | self.v4_idx.clear(); | |
801 | self.masks.reset(); | |
802 | ||
803 | self.cmode.clear(); | |
804 | self.v1_cand.clear(); | |
805 | self.v4_cand.clear(); | |
806 | for (v1_entry, v4_entries) in self.v1_entries.iter().zip(self.v4_entries.chunks(4)) { | |
807 | let (v1_idx, v1_dist) = Self::find_nearest(&self.v1_cur_cb[self.cur_strip][..self.v1_len], *v1_entry); | |
808 | if v1_dist == 0 || self.force_v1 { | |
809 | if !self.refine { | |
810 | self.masks.put_v1(); | |
811 | self.v1_idx.push(v1_idx); | |
812 | } else { | |
813 | self.cmode.push(CodingMode::V1); | |
814 | self.v1_cand.push(*v1_entry); | |
815 | } | |
816 | continue; | |
817 | } | |
818 | let (v40_idx, v40_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[0]); | |
819 | let (v41_idx, v41_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[1]); | |
820 | let (v42_idx, v42_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[2]); | |
821 | let (v43_idx, v43_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[3]); | |
822 | if v40_dist + v41_dist + v42_dist + v43_dist > v1_dist { | |
823 | if !self.refine { | |
824 | self.masks.put_v4(); | |
825 | self.v4_idx.push(v40_idx); | |
826 | self.v4_idx.push(v41_idx); | |
827 | self.v4_idx.push(v42_idx); | |
828 | self.v4_idx.push(v43_idx); | |
829 | } else { | |
830 | self.cmode.push(CodingMode::V4); | |
831 | self.v4_cand.extend_from_slice(v4_entries); | |
832 | } | |
833 | } else { | |
834 | if !self.refine { | |
835 | self.masks.put_v1(); | |
836 | self.v1_idx.push(v1_idx); | |
837 | } else { | |
838 | self.cmode.push(CodingMode::V1); | |
839 | self.v1_cand.push(*v1_entry); | |
840 | } | |
841 | } | |
842 | } | |
843 | if self.refine { | |
844 | self.refine_vectors(); | |
845 | let mut v1_src = self.v1_cand.iter(); | |
846 | let mut v4_src = self.v4_cand.chunks_exact(4); | |
847 | for &cmode in self.cmode.iter() { | |
848 | match cmode { | |
849 | CodingMode::Skip => unreachable!(), | |
850 | CodingMode::V1 => { | |
851 | let v1_entry = v1_src.next().unwrap(); | |
852 | let (v1_idx, _) = Self::find_nearest(&self.v1_cur_cb[self.cur_strip][..self.v1_len], *v1_entry); | |
853 | self.masks.put_v1(); | |
854 | self.v1_idx.push(v1_idx); | |
855 | }, | |
856 | CodingMode::V4 => { | |
857 | let v4_entries = v4_src.next().unwrap(); | |
858 | let (v40_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[0]); | |
859 | let (v41_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[1]); | |
860 | let (v42_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[2]); | |
861 | let (v43_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[3]); | |
862 | ||
863 | self.masks.put_v4(); | |
864 | self.v4_idx.push(v40_idx); | |
865 | self.v4_idx.push(v41_idx); | |
866 | self.v4_idx.push(v42_idx); | |
867 | self.v4_idx.push(v43_idx); | |
868 | }, | |
869 | }; | |
870 | } | |
871 | } | |
872 | self.masks.end(); | |
873 | ||
874 | let mut is_intra_strip = start_line == 0; | |
875 | let (upd_v1, upd_v4) = if !is_intra_strip { | |
876 | let cb_size = if self.grayscale { 4 } else { 6 }; | |
877 | (Self::can_update_cb(&self.v1_cur_cb[self.cur_strip][..self.v1_len], &self.v1_cb[self.cur_strip][..self.v1_len], cb_size), | |
878 | Self::can_update_cb(&self.v4_cur_cb[self.cur_strip][..self.v4_len], &self.v4_cb[self.cur_strip][..self.v4_len], cb_size)) | |
879 | } else { | |
880 | (false, false) | |
881 | }; | |
882 | if !is_intra_strip && !upd_v1 && !upd_v4 { | |
883 | is_intra_strip = true; | |
884 | } | |
885 | bw.write_byte(if is_intra_strip { 0x10 } else { 0x11 })?; | |
886 | bw.write_u24be(0)?; // strip size | |
887 | let strip_data_pos = bw.tell(); | |
888 | bw.write_u16be(0)?; // yoff | |
889 | bw.write_u16be(0)?; // xoff | |
890 | bw.write_u16be((end_line - start_line) as u16)?; | |
891 | bw.write_u16be(width as u16)?; | |
892 | ||
893 | Self::write_cb(bw, 0x20, &self.v4_cur_cb[self.cur_strip], &self.v4_cb[self.cur_strip], self.grayscale, upd_v4, self.v4_len)?; | |
894 | Self::write_cb(bw, 0x22, &self.v1_cur_cb[self.cur_strip], &self.v1_cb[self.cur_strip], self.grayscale, upd_v1, self.v1_len)?; | |
895 | ||
896 | self.render_stripe(true, start_line, end_line); | |
897 | ||
898 | if self.v4_idx.is_empty() { | |
899 | bw.write_byte(0x32)?; | |
900 | bw.write_u24be((self.v1_idx.len() + 4) as u32)?; | |
901 | bw.write_buf(self.v1_idx.as_slice())?; | |
902 | } else { | |
903 | bw.write_byte(0x30)?; | |
904 | bw.write_u24be(0)?; | |
905 | let chunk_pos = bw.tell(); | |
906 | let mut v1_pos = 0; | |
907 | let mut v4_pos = 0; | |
908 | for _ in 0..32 { | |
909 | self.v1_idx.push(0); | |
910 | self.v4_idx.push(0); | |
911 | self.v4_idx.push(0); | |
912 | self.v4_idx.push(0); | |
913 | self.v4_idx.push(0); | |
914 | } | |
915 | for mask in self.masks.masks.iter() { | |
916 | bw.write_u32be(*mask)?; | |
917 | for j in (0..32).rev() { | |
918 | if (mask & (1 << j)) == 0 { | |
919 | bw.write_byte(self.v1_idx[v1_pos])?; | |
920 | v1_pos += 1; | |
921 | } else { | |
922 | bw.write_byte(self.v4_idx[v4_pos])?; | |
923 | bw.write_byte(self.v4_idx[v4_pos + 1])?; | |
924 | bw.write_byte(self.v4_idx[v4_pos + 2])?; | |
925 | bw.write_byte(self.v4_idx[v4_pos + 3])?; | |
926 | v4_pos += 4; | |
927 | } | |
928 | } | |
929 | } | |
930 | patch_size(bw, chunk_pos)?; | |
931 | } | |
932 | ||
933 | patch_size(bw, strip_data_pos)?; | |
934 | ||
935 | self.v1_cb[self.cur_strip].copy_from_slice(&self.v1_cur_cb[self.cur_strip]); | |
936 | self.v4_cb[self.cur_strip].copy_from_slice(&self.v4_cur_cb[self.cur_strip]); | |
937 | start_line = end_line; | |
938 | end_line = (end_line + strip_h).min(height); | |
939 | ||
940 | self.cur_strip += 1; | |
941 | } | |
942 | patch_size(bw, frame_data_pos)?; | |
943 | Ok(true) | |
944 | } | |
945 | fn encode_inter(&mut self, bw: &mut ByteWriter, in_frm: &NAVideoBuffer<u8>) -> EncoderResult<bool> { | |
946 | let (width, height) = in_frm.get_dimensions(0); | |
947 | let mut strip_h = (height / self.nstrips + 3) & !3; | |
948 | if strip_h == 0 { | |
949 | self.nstrips = 1; | |
950 | strip_h = height; | |
951 | } | |
952 | let mut start_line = 0; | |
953 | let mut end_line = strip_h; | |
954 | ||
955 | bw.write_byte(1)?; // intra flag | |
956 | bw.write_u24be(0)?; // frame size | |
957 | let frame_data_pos = bw.tell(); | |
958 | bw.write_u16be(width as u16)?; | |
959 | bw.write_u16be(height as u16)?; | |
960 | bw.write_u16be(self.nstrips as u16)?; | |
961 | ||
962 | self.cur_strip = 0; | |
963 | while start_line < height { | |
964 | self.read_strip(in_frm, start_line, end_line); | |
965 | self.calc_skip_dist(in_frm, start_line, end_line); | |
966 | ||
967 | self.quant_vectors(); | |
968 | if self.grayscale { | |
969 | for cw in self.v1_cur_cb[self.cur_strip].iter_mut() { | |
970 | cw.u = 128; | |
971 | cw.v = 128; | |
972 | } | |
973 | for cw in self.v4_cur_cb[self.cur_strip].iter_mut() { | |
974 | cw.u = 128; | |
975 | cw.v = 128; | |
976 | } | |
977 | } | |
978 | ||
979 | self.v1_idx.clear(); | |
980 | self.v4_idx.clear(); | |
981 | self.masks.reset(); | |
982 | ||
983 | self.cmode.clear(); | |
984 | self.v1_cand.clear(); | |
985 | self.v4_cand.clear(); | |
986 | ||
987 | let mut skip_iter = self.skip_dist.iter(); | |
988 | for (v1_entry, v4_entries) in self.v1_entries.iter().zip(self.v4_entries.chunks(4)) { | |
989 | let skip_dist = *skip_iter.next().unwrap(); | |
990 | if skip_dist == 0 { | |
991 | if !self.refine { | |
992 | self.masks.put_inter(true); | |
993 | } else { | |
994 | self.cmode.push(CodingMode::Skip); | |
995 | } | |
996 | continue; | |
997 | } | |
998 | let (v1_idx, v1_dist) = Self::find_nearest(&self.v1_cur_cb[self.cur_strip][..self.v1_len], *v1_entry); | |
999 | if skip_dist < v1_dist { | |
1000 | if !self.refine { | |
1001 | self.masks.put_inter(true); | |
1002 | } else { | |
1003 | self.cmode.push(CodingMode::Skip); | |
1004 | } | |
1005 | continue; | |
1006 | } else if !self.refine { | |
1007 | self.masks.put_inter(false); | |
1008 | } | |
1009 | if v1_dist == 0 || self.force_v1 { | |
1010 | if !self.refine { | |
1011 | self.masks.put_v1(); | |
1012 | self.v1_idx.push(v1_idx); | |
1013 | } else { | |
1014 | self.cmode.push(CodingMode::V1); | |
1015 | self.v1_cand.push(*v1_entry); | |
1016 | } | |
1017 | continue; | |
1018 | } | |
1019 | let (v40_idx, v40_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[0]); | |
1020 | let (v41_idx, v41_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[1]); | |
1021 | let (v42_idx, v42_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[2]); | |
1022 | let (v43_idx, v43_dist) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[3]); | |
1023 | if v40_dist + v41_dist + v42_dist + v43_dist > v1_dist { | |
1024 | if !self.refine { | |
1025 | self.masks.put_v4(); | |
1026 | self.v4_idx.push(v40_idx); | |
1027 | self.v4_idx.push(v41_idx); | |
1028 | self.v4_idx.push(v42_idx); | |
1029 | self.v4_idx.push(v43_idx); | |
1030 | } else { | |
1031 | self.cmode.push(CodingMode::V4); | |
1032 | self.v4_cand.extend_from_slice(v4_entries); | |
1033 | } | |
1034 | } else { | |
1035 | if !self.refine { | |
1036 | self.masks.put_v1(); | |
1037 | self.v1_idx.push(v1_idx); | |
1038 | } else { | |
1039 | self.cmode.push(CodingMode::V1); | |
1040 | self.v1_cand.push(*v1_entry); | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | if self.refine { | |
1045 | self.refine_vectors(); | |
1046 | let mut v1_src = self.v1_cand.iter(); | |
1047 | let mut v4_src = self.v4_cand.chunks_exact(4); | |
1048 | for &cmode in self.cmode.iter() { | |
1049 | match cmode { | |
1050 | CodingMode::Skip => { | |
1051 | self.masks.put_inter(true); | |
1052 | }, | |
1053 | CodingMode::V1 => { | |
1054 | let v1_entry = v1_src.next().unwrap(); | |
1055 | let (v1_idx, _) = Self::find_nearest(&self.v1_cur_cb[self.cur_strip][..self.v1_len], *v1_entry); | |
1056 | self.masks.put_inter(false); | |
1057 | self.masks.put_v1(); | |
1058 | self.v1_idx.push(v1_idx); | |
1059 | }, | |
1060 | CodingMode::V4 => { | |
1061 | let v4_entries = v4_src.next().unwrap(); | |
1062 | let (v40_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[0]); | |
1063 | let (v41_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[1]); | |
1064 | let (v42_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[2]); | |
1065 | let (v43_idx, _) = Self::find_nearest(&self.v4_cur_cb[self.cur_strip][..self.v4_len], v4_entries[3]); | |
1066 | ||
1067 | self.masks.put_inter(false); | |
1068 | self.masks.put_v4(); | |
1069 | self.v4_idx.push(v40_idx); | |
1070 | self.v4_idx.push(v41_idx); | |
1071 | self.v4_idx.push(v42_idx); | |
1072 | self.v4_idx.push(v43_idx); | |
1073 | }, | |
1074 | }; | |
1075 | } | |
1076 | } | |
1077 | self.masks.end(); | |
1078 | ||
1079 | let (upd_v1, upd_v4) = { | |
1080 | let cb_size = if self.grayscale { 4 } else { 6 }; | |
1081 | (Self::can_update_cb(&self.v1_cur_cb[self.cur_strip][..self.v1_len], &self.v1_cb[self.cur_strip][..self.v1_len], cb_size), | |
1082 | Self::can_update_cb(&self.v4_cur_cb[self.cur_strip][..self.v4_len], &self.v4_cb[self.cur_strip][..self.v4_len], cb_size)) | |
1083 | }; | |
1084 | bw.write_byte(0x11)?; | |
1085 | bw.write_u24be(0)?; // strip size | |
1086 | let strip_data_pos = bw.tell(); | |
1087 | bw.write_u16be(0)?; // yoff | |
1088 | bw.write_u16be(0)?; // xoff | |
1089 | bw.write_u16be((end_line - start_line) as u16)?; | |
1090 | bw.write_u16be(width as u16)?; | |
1091 | ||
1092 | Self::write_cb(bw, 0x20, &self.v4_cur_cb[self.cur_strip], &self.v4_cb[self.cur_strip], self.grayscale, upd_v4, self.v4_len)?; | |
1093 | Self::write_cb(bw, 0x22, &self.v1_cur_cb[self.cur_strip], &self.v1_cb[self.cur_strip], self.grayscale, upd_v1, self.v1_len)?; | |
1094 | ||
1095 | self.render_stripe(false, start_line, end_line); | |
1096 | ||
1097 | bw.write_byte(0x31)?; | |
1098 | bw.write_u24be(0)?; | |
1099 | let chunk_pos = bw.tell(); | |
1100 | let mut v1_pos = 0; | |
1101 | let mut v4_pos = 0; | |
1102 | for _ in 0..32 { | |
1103 | self.v1_idx.push(0); | |
1104 | self.v4_idx.push(0); | |
1105 | self.v4_idx.push(0); | |
1106 | self.v4_idx.push(0); | |
1107 | self.v4_idx.push(0); | |
1108 | } | |
1109 | let mut skip = true; | |
1110 | for mask in self.masks.masks.iter() { | |
1111 | bw.write_u32be(*mask)?; | |
1112 | if *mask == 0 && skip { continue; } | |
1113 | let mut bit = 1 << 31; | |
1114 | while bit > 0 { | |
1115 | if skip { | |
1116 | skip = (mask & bit) == 0; | |
1117 | bit >>= 1; | |
1118 | } else { | |
1119 | if (mask & bit) == 0 { | |
1120 | bw.write_byte(self.v1_idx[v1_pos])?; | |
1121 | v1_pos += 1; | |
1122 | } else { | |
1123 | bw.write_byte(self.v4_idx[v4_pos])?; | |
1124 | bw.write_byte(self.v4_idx[v4_pos + 1])?; | |
1125 | bw.write_byte(self.v4_idx[v4_pos + 2])?; | |
1126 | bw.write_byte(self.v4_idx[v4_pos + 3])?; | |
1127 | v4_pos += 4; | |
1128 | } | |
1129 | bit >>= 1; | |
1130 | skip = true; | |
1131 | } | |
1132 | } | |
1133 | } | |
1134 | patch_size(bw, chunk_pos)?; | |
1135 | ||
1136 | patch_size(bw, strip_data_pos)?; | |
1137 | ||
1138 | self.v1_cb[self.cur_strip].copy_from_slice(&self.v1_cur_cb[self.cur_strip]); | |
1139 | self.v4_cb[self.cur_strip].copy_from_slice(&self.v4_cur_cb[self.cur_strip]); | |
1140 | start_line = end_line; | |
1141 | end_line = (end_line + strip_h).min(height); | |
1142 | ||
1143 | self.cur_strip += 1; | |
1144 | } | |
1145 | patch_size(bw, frame_data_pos)?; | |
1146 | Ok(true) | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | impl NAEncoder for CinepakEncoder { | |
1151 | fn negotiate_format(&self, encinfo: &EncodeParameters) -> EncoderResult<EncodeParameters> { | |
1152 | match encinfo.format { | |
1153 | NACodecTypeInfo::None => { | |
1154 | Ok(EncodeParameters { | |
1155 | format: NACodecTypeInfo::Video(NAVideoInfo::new(0, 0, true, YUV420_FORMAT)), | |
1156 | ..Default::default() | |
1157 | }) | |
1158 | }, | |
1159 | NACodecTypeInfo::Audio(_) => Err(EncoderError::FormatError), | |
1160 | NACodecTypeInfo::Video(vinfo) => { | |
1161 | let pix_fmt = if vinfo.format == GRAY_FORMAT { GRAY_FORMAT } else { YUV420_FORMAT }; | |
1162 | let outinfo = NAVideoInfo::new((vinfo.width + 3) & !3, (vinfo.height + 3) & !3, false, pix_fmt); | |
1163 | let mut ofmt = *encinfo; | |
1164 | ofmt.format = NACodecTypeInfo::Video(outinfo); | |
1165 | Ok(ofmt) | |
1166 | } | |
1167 | } | |
1168 | } | |
1169 | fn get_capabilities(&self) -> u64 { 0 } | |
1170 | fn init(&mut self, stream_id: u32, encinfo: EncodeParameters) -> EncoderResult<NAStreamRef> { | |
1171 | match encinfo.format { | |
1172 | NACodecTypeInfo::None => Err(EncoderError::FormatError), | |
1173 | NACodecTypeInfo::Audio(_) => Err(EncoderError::FormatError), | |
1174 | NACodecTypeInfo::Video(vinfo) => { | |
1175 | if vinfo.format != YUV420_FORMAT && vinfo.format != GRAY_FORMAT { | |
1176 | return Err(EncoderError::FormatError); | |
1177 | } | |
1178 | if ((vinfo.width | vinfo.height) & 3) != 0 { | |
1179 | return Err(EncoderError::FormatError); | |
1180 | } | |
1181 | if (vinfo.width | vinfo.height) >= (1 << 16) { | |
1182 | return Err(EncoderError::FormatError); | |
1183 | } | |
1184 | ||
1185 | let out_info = NAVideoInfo::new(vinfo.width, vinfo.height, false, vinfo.format); | |
1186 | let info = NACodecInfo::new("cinepak", NACodecTypeInfo::Video(out_info), None); | |
1187 | let mut stream = NAStream::new(StreamType::Video, stream_id, info, encinfo.tb_num, encinfo.tb_den, 0); | |
1188 | stream.set_num(stream_id as usize); | |
1189 | let stream = stream.into_ref(); | |
1190 | ||
1191 | self.stream = Some(stream.clone()); | |
1192 | self.quality = encinfo.quality; | |
1193 | self.grayscale = vinfo.format != YUV420_FORMAT; | |
1194 | let num_blocks = vinfo.width / 2 * vinfo.height / 2; | |
1195 | self.v1_entries = Vec::with_capacity(num_blocks); | |
1196 | self.v4_entries = Vec::with_capacity(num_blocks * 4); | |
1197 | self.v1_idx = Vec::with_capacity(num_blocks); | |
1198 | self.v4_idx = Vec::with_capacity(num_blocks * 4); | |
1199 | self.skip_dist = Vec::with_capacity(vinfo.width / 4 * vinfo.height / 4); | |
1200 | ||
1201 | let buf = alloc_video_buffer(out_info, 2)?; | |
1202 | self.lastfrm = Some(buf.get_vbuf().unwrap()); | |
1203 | ||
1204 | Ok(stream) | |
1205 | }, | |
1206 | } | |
1207 | } | |
1208 | fn encode(&mut self, frm: &NAFrame) -> EncoderResult<()> { | |
1209 | let buf = frm.get_buffer(); | |
1210 | if let Some(ref vbuf) = buf.get_vbuf() { | |
1211 | if self.nstrips == 0 { | |
1212 | let (w, h) = vbuf.get_dimensions(0); | |
1213 | self.nstrips = ((((w * h) >> 4) + 1200) / 2400).max(1).min(3); | |
1214 | let strip_h = ((h + self.nstrips - 1) / self.nstrips + 3) & !3; | |
1215 | self.nstrips = (h + strip_h - 1) / strip_h; | |
1216 | } | |
1217 | let cur_strips = self.v1_cb.len(); | |
1218 | if cur_strips != self.nstrips { | |
1219 | self.frmcount = 0; | |
1220 | } | |
1221 | if cur_strips < self.nstrips { | |
1222 | for _ in cur_strips..self.nstrips { | |
1223 | self.v1_cb.push([YUVCode::default(); 256]); | |
1224 | self.v4_cb.push([YUVCode::default(); 256]); | |
1225 | self.v1_cur_cb.push([YUVCode::default(); 256]); | |
1226 | self.v4_cur_cb.push([YUVCode::default(); 256]); | |
1227 | } | |
1228 | } | |
1229 | ||
1230 | let mut dbuf = Vec::with_capacity(4); | |
1231 | let mut gw = GrowableMemoryWriter::new_write(&mut dbuf); | |
1232 | let mut bw = ByteWriter::new(&mut gw); | |
1233 | let is_intra = if self.frmcount == 0 { | |
1234 | self.encode_intra(&mut bw, vbuf)? | |
1235 | } else { | |
1236 | self.encode_inter(&mut bw, vbuf)? | |
1237 | }; | |
1238 | self.pkt = Some(NAPacket::new(self.stream.clone().unwrap(), frm.ts, is_intra, dbuf)); | |
1239 | self.frmcount += 1; | |
1240 | if self.frmcount == self.key_int { | |
1241 | self.frmcount = 0; | |
1242 | } | |
1243 | Ok(()) | |
1244 | } else { | |
1245 | Err(EncoderError::InvalidParameters) | |
1246 | } | |
1247 | } | |
1248 | fn get_packet(&mut self) -> EncoderResult<Option<NAPacket>> { | |
1249 | let mut npkt = None; | |
1250 | std::mem::swap(&mut self.pkt, &mut npkt); | |
1251 | Ok(npkt) | |
1252 | } | |
1253 | fn flush(&mut self) -> EncoderResult<()> { | |
1254 | self.frmcount = 0; | |
1255 | Ok(()) | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | const ENCODER_OPTS: &[NAOptionDefinition] = &[ | |
1260 | NAOptionDefinition { | |
1261 | name: KEYFRAME_OPTION, description: KEYFRAME_OPTION_DESC, | |
1262 | opt_type: NAOptionDefinitionType::Int(Some(0), Some(128)) }, | |
1263 | NAOptionDefinition { | |
1264 | name: "nstrips", description: "Number of strips per frame (0 - automatic)", | |
1265 | opt_type: NAOptionDefinitionType::Int(Some(0), Some(16)) }, | |
1266 | NAOptionDefinition { | |
1267 | name: "quant_mode", description: "Quantisation mode", | |
1268 | opt_type: NAOptionDefinitionType::String(Some(&["elbg", "fast", "mediancut"])) }, | |
1269 | NAOptionDefinition { | |
1270 | name: "force_v1", description: "Force coarse (V1-only) mode", | |
1271 | opt_type: NAOptionDefinitionType::Bool }, | |
1272 | NAOptionDefinition { | |
1273 | name: "refine", description: "Try to improve coded picture", | |
1274 | opt_type: NAOptionDefinitionType::Bool }, | |
1275 | ]; | |
1276 | ||
1277 | impl NAOptionHandler for CinepakEncoder { | |
1278 | fn get_supported_options(&self) -> &[NAOptionDefinition] { ENCODER_OPTS } | |
1279 | fn set_options(&mut self, options: &[NAOption]) { | |
1280 | for option in options.iter() { | |
1281 | for opt_def in ENCODER_OPTS.iter() { | |
1282 | if opt_def.check(option).is_ok() { | |
1283 | match option.name { | |
1284 | KEYFRAME_OPTION => { | |
1285 | if let NAValue::Int(intval) = option.value { | |
1286 | self.key_int = intval as u8; | |
1287 | } | |
1288 | }, | |
1289 | "nstrips" => { | |
1290 | if let NAValue::Int(intval) = option.value { | |
1291 | self.nstrips = intval as usize; | |
1292 | } | |
1293 | }, | |
1294 | "quant_mode" => { | |
1295 | if let NAValue::String(ref strval) = option.value { | |
1296 | match strval.as_str() { | |
1297 | "elbg" => self.qmode = QuantMode::ELBG, | |
1298 | "fast" => self.qmode = QuantMode::Fast, | |
1299 | "mediancut" => self.qmode = QuantMode::MedianCut, | |
1300 | _ => {}, | |
1301 | }; | |
1302 | } | |
1303 | }, | |
1304 | "force_v1" => { | |
1305 | if let NAValue::Bool(val) = option.value { | |
1306 | self.force_v1 = val; | |
1307 | } | |
1308 | }, | |
1309 | "refine" => { | |
1310 | if let NAValue::Bool(val) = option.value { | |
1311 | self.refine = val; | |
1312 | } | |
1313 | }, | |
1314 | _ => {}, | |
1315 | }; | |
1316 | } | |
1317 | } | |
1318 | } | |
1319 | } | |
1320 | fn query_option_value(&self, name: &str) -> Option<NAValue> { | |
1321 | match name { | |
1322 | KEYFRAME_OPTION => Some(NAValue::Int(i64::from(self.key_int))), | |
1323 | "nstrips" => Some(NAValue::Int(self.nstrips as i64)), | |
1324 | "quant_mode" => Some(NAValue::String(self.qmode.to_string())), | |
1325 | "force_v1" => Some(NAValue::Bool(self.force_v1)), | |
1326 | "refine" => Some(NAValue::Bool(self.refine)), | |
1327 | _ => None, | |
1328 | } | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | pub fn get_encoder() -> Box<dyn NAEncoder + Send> { | |
1333 | Box::new(CinepakEncoder::new()) | |
1334 | } | |
1335 | ||
1336 | #[cfg(test)] | |
1337 | mod test { | |
1338 | use nihav_core::codecs::*; | |
1339 | use nihav_core::demuxers::*; | |
1340 | use nihav_core::muxers::*; | |
1341 | use crate::*; | |
1342 | use nihav_codec_support::test::enc_video::*; | |
1343 | ||
1344 | #[test] | |
1345 | fn test_cinepak_encoder() { | |
1346 | let mut dmx_reg = RegisteredDemuxers::new(); | |
1347 | generic_register_all_demuxers(&mut dmx_reg); | |
1348 | let mut dec_reg = RegisteredDecoders::new(); | |
1349 | generic_register_all_decoders(&mut dec_reg); | |
1350 | let mut mux_reg = RegisteredMuxers::new(); | |
1351 | generic_register_all_muxers(&mut mux_reg); | |
1352 | let mut enc_reg = RegisteredEncoders::new(); | |
1353 | generic_register_all_encoders(&mut enc_reg); | |
1354 | ||
1355 | // sample: https://samples.mplayerhq.hu/V-codecs/UCOD/TalkingHead_352x288.avi | |
1356 | let dec_config = DecoderTestParams { | |
1357 | demuxer: "avi", | |
1358 | in_name: "assets/Misc/TalkingHead_352x288.avi", | |
1359 | stream_type: StreamType::Video, | |
1360 | limit: Some(2), | |
1361 | dmx_reg, dec_reg, | |
1362 | }; | |
1363 | let enc_config = EncoderTestParams { | |
1364 | muxer: "avi", | |
1365 | enc_name: "cinepak", | |
1366 | out_name: "cinepak.avi", | |
1367 | mux_reg, enc_reg, | |
1368 | }; | |
1369 | let dst_vinfo = NAVideoInfo { | |
1370 | width: 0, | |
1371 | height: 0, | |
1372 | format: YUV420_FORMAT, | |
1373 | flipped: true, | |
1374 | bits: 12, | |
1375 | }; | |
1376 | let enc_params = EncodeParameters { | |
1377 | format: NACodecTypeInfo::Video(dst_vinfo), | |
1378 | quality: 0, | |
1379 | bitrate: 0, | |
1380 | tb_num: 0, | |
1381 | tb_den: 0, | |
1382 | flags: 0, | |
1383 | }; | |
1384 | let enc_options = &[ | |
1385 | NAOption { name: "quant_mode", value: NAValue::String("mediancut".to_string()) }, | |
1386 | ]; | |
1387 | //test_encoding_to_file(&dec_config, &enc_config, enc_params, enc_options); | |
1388 | test_encoding_md5(&dec_config, &enc_config, enc_params, enc_options, | |
1389 | &[0x1d4690c8, 0x3b15b4b3, 0xc2df3c7b, 0x1a25b159]); | |
1390 | } | |
1391 | } |