| 1 | use super::{BlockTokens, MAX_DIST, RateControl, calc_diff}; |
| 2 | use super::super::binkviddata::*; |
| 3 | |
| 4 | const BINK_INV_SCAN: [usize; 64] = [ |
| 5 | 0, 1, 4, 5, 8, 9, 12, 13, |
| 6 | 2, 3, 6, 7, 10, 11, 14, 15, |
| 7 | 24, 25, 44, 45, 16, 17, 20, 21, |
| 8 | 26, 27, 46, 47, 18, 19, 22, 23, |
| 9 | 28, 29, 32, 33, 48, 49, 52, 53, |
| 10 | 30, 31, 34, 35, 50, 51, 54, 55, |
| 11 | 36, 37, 40, 41, 56, 57, 60, 61, |
| 12 | 38, 39, 42, 43, 58, 59, 62, 63 |
| 13 | ]; |
| 14 | |
| 15 | trait WriteBit { |
| 16 | fn write_bit(&mut self, val: u16); |
| 17 | fn write_coef(&mut self, aval: u32, sign: bool, bits: u8); |
| 18 | } |
| 19 | |
| 20 | impl WriteBit for BlockTokens { |
| 21 | fn write_bit(&mut self, val: u16) { |
| 22 | self.other.push((val, 1)); |
| 23 | } |
| 24 | fn write_coef(&mut self, aval: u32, sign: bool, bits: u8) { |
| 25 | if bits > 1 { |
| 26 | self.other.push((aval as u16 & ((1 << (bits - 1)) - 1), bits - 1)); |
| 27 | } |
| 28 | self.write_bit(sign as u16); |
| 29 | } |
| 30 | } |
| 31 | |
| 32 | pub struct DSP { |
| 33 | diff: [i16; 64], |
| 34 | dct_i: [u8; 64], |
| 35 | dct_p: [u8; 64], |
| 36 | qmats: QuantMats, |
| 37 | i_start: usize, |
| 38 | i_len: usize, |
| 39 | p_start: usize, |
| 40 | p_len: usize, |
| 41 | } |
| 42 | |
| 43 | impl DSP { |
| 44 | pub fn new() -> Self { |
| 45 | let mut qmats = QuantMats::default(); |
| 46 | qmats.calc_binkb_quants(); |
| 47 | Self { |
| 48 | diff: [0; 64], |
| 49 | dct_i: [0; 64], |
| 50 | dct_p: [0; 64], |
| 51 | qmats, |
| 52 | i_start: 0, |
| 53 | i_len: 16, |
| 54 | p_start: 0, |
| 55 | p_len: 16, |
| 56 | } |
| 57 | } |
| 58 | pub fn get_diff(&mut self, mc_blk: &[u8; 64], cur_blk: &[u8; 64]) { |
| 59 | for (dst, (&prev, &cur)) in self.diff.iter_mut() |
| 60 | .zip(mc_blk.iter().zip(cur_blk.iter())) { |
| 61 | *dst = i16::from(cur) - i16::from(prev); |
| 62 | } |
| 63 | } |
| 64 | pub fn recon_residue(&self, dst: &mut [u8], dstride: usize, mc_blk: &[u8; 64]) { |
| 65 | for (dline, (prow, drow)) in dst.chunks_mut(dstride) |
| 66 | .zip(mc_blk.chunks_exact(8).zip(self.diff.chunks_exact(8))) { |
| 67 | for (dst, (&prev, &diff)) in dline.iter_mut().zip(prow.iter().zip(drow.iter())) { |
| 68 | *dst = (i16::from(prev) + diff) as u8; |
| 69 | } |
| 70 | } |
| 71 | } |
| 72 | pub fn recon_dct_i(&self, dst: &mut [u8], dstride: usize) { |
| 73 | for (dline, srow) in dst.chunks_mut(dstride).zip(self.dct_i.chunks_exact(8)) { |
| 74 | dline[..8].copy_from_slice(srow); |
| 75 | } |
| 76 | } |
| 77 | pub fn recon_dct_p(&self, dst: &mut [u8], dstride: usize) { |
| 78 | for (dline, srow) in dst.chunks_mut(dstride).zip(self.dct_p.chunks_exact(8)) { |
| 79 | dline[..8].copy_from_slice(srow); |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | pub fn try_residue(&self, tokens: &mut BlockTokens) -> u32 { |
| 84 | let mut tree = Tree::new(true); |
| 85 | let mut flat = [0; 64]; |
| 86 | let mut blen = [0; 64]; |
| 87 | |
| 88 | for (&idx, &val) in BINK_INV_SCAN.iter().zip(self.diff.iter()) { |
| 89 | flat[idx] = val; |
| 90 | let aval = val.unsigned_abs(); |
| 91 | let mut b = 0u8; |
| 92 | while (1 << b) <= aval { |
| 93 | b += 1; |
| 94 | } |
| 95 | blen[idx] = b.saturating_sub(1); |
| 96 | } |
| 97 | |
| 98 | let mut max_val = 0; |
| 99 | let mut max_bits = 0; |
| 100 | let mut bits = 0; |
| 101 | let mut avals = [0; 64]; |
| 102 | let mut signs = [false; 64]; |
| 103 | for ((aval, sign), (&val, &vlen)) in avals.iter_mut().zip(signs.iter_mut()) |
| 104 | .zip(flat.iter().zip(blen.iter())) { |
| 105 | *aval = val.unsigned_abs(); |
| 106 | *sign = val < 0; |
| 107 | max_val = max_val.max(*aval); |
| 108 | max_bits = max_bits.max(vlen); |
| 109 | bits += aval.count_ones(); |
| 110 | } |
| 111 | |
| 112 | if max_bits > 7 || bits > 127 { |
| 113 | return MAX_DIST; |
| 114 | } |
| 115 | |
| 116 | tokens.nresidues.push(bits as u8); |
| 117 | tokens.other.push((max_bits as u16, 3)); |
| 118 | |
| 119 | let mut nz_cand = Vec::with_capacity(64); |
| 120 | let mut masks_left = bits + 1; |
| 121 | 'tree_loop: for cur_bits in (0..=max_bits).rev() { |
| 122 | let mask = 1 << cur_bits; |
| 123 | for &idx in nz_cand.iter() { |
| 124 | tokens.write_bit(((avals[idx] & mask) != 0) as u16); |
| 125 | if (avals[idx] & mask) != 0 { |
| 126 | masks_left -= 1; |
| 127 | if masks_left == 0 { |
| 128 | break 'tree_loop; |
| 129 | } |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | let mut pos = tree.start; |
| 134 | while pos < tree.end { |
| 135 | if tree.state[pos] == TreeState::None { |
| 136 | pos += 1; |
| 137 | continue; |
| 138 | } |
| 139 | if let TreeState::Candidate(idx) = tree.state[pos] { |
| 140 | let idx = usize::from(idx); |
| 141 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 142 | tree.state[pos] = TreeState::None; |
| 143 | tokens.write_bit(1); |
| 144 | tokens.write_bit(signs[idx] as u16); |
| 145 | nz_cand.push(idx); |
| 146 | masks_left -= 1; |
| 147 | if masks_left == 0 { |
| 148 | break 'tree_loop; |
| 149 | } |
| 150 | } else { |
| 151 | tokens.write_bit(0); |
| 152 | pos += 1; |
| 153 | } |
| 154 | continue; |
| 155 | } |
| 156 | let range = tree.state[pos].get_range(); |
| 157 | let cur_max_bits = blen[range].iter().fold(0u8, |acc, &a| acc.max(a)); |
| 158 | if cur_max_bits == cur_bits { |
| 159 | tokens.write_bit(1); |
| 160 | match tree.state[pos] { |
| 161 | TreeState::Twenty(val) => { |
| 162 | tree.state[pos] = TreeState::Sixteen(val + 4); |
| 163 | for i in 0..4 { |
| 164 | let idx = usize::from(val) + i; |
| 165 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 166 | tokens.write_bit(0); |
| 167 | tokens.write_bit(signs[idx] as u16); |
| 168 | nz_cand.push(idx); |
| 169 | masks_left -= 1; |
| 170 | if masks_left == 0 { |
| 171 | break 'tree_loop; |
| 172 | } |
| 173 | } else { |
| 174 | tokens.write_bit(1); |
| 175 | tree.add_to_head(TreeState::Candidate(idx as u8)); |
| 176 | } |
| 177 | } |
| 178 | }, |
| 179 | TreeState::Sixteen(val) => { |
| 180 | tree.state[pos] = TreeState::Four(val); |
| 181 | for i in 1u8..4 { |
| 182 | tree.add_to_tail(TreeState::Four(val + i * 4)); |
| 183 | } |
| 184 | }, |
| 185 | TreeState::Four(val) => { |
| 186 | tree.state[pos] = TreeState::None; |
| 187 | for i in 0..4 { |
| 188 | let idx = usize::from(val) + i; |
| 189 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 190 | tokens.write_bit(0); |
| 191 | tokens.write_bit(signs[idx] as u16); |
| 192 | nz_cand.push(idx); |
| 193 | masks_left -= 1; |
| 194 | if masks_left == 0 { |
| 195 | break 'tree_loop; |
| 196 | } |
| 197 | } else { |
| 198 | tokens.write_bit(1); |
| 199 | tree.add_to_head(TreeState::Candidate(idx as u8)); |
| 200 | } |
| 201 | } |
| 202 | }, |
| 203 | _ => unreachable!(), |
| 204 | }; |
| 205 | } else { |
| 206 | tokens.write_bit(0); |
| 207 | pos += 1; |
| 208 | } |
| 209 | } |
| 210 | } |
| 211 | |
| 212 | 0 |
| 213 | } |
| 214 | |
| 215 | pub fn set_quant_ranges(&mut self, ranges: [u8; 4]) { |
| 216 | self.i_start = usize::from(ranges[0]); |
| 217 | self.i_len = usize::from(ranges[1]); |
| 218 | self.p_start = usize::from(ranges[2]); |
| 219 | self.p_len = usize::from(ranges[3]); |
| 220 | } |
| 221 | pub fn try_dct_intra(&mut self, blk: &[u8; 64], tokens: &mut BlockTokens, tmp_tok: &mut BlockTokens, is_b: bool, rc: &RateControl, mut best_dist: u32) -> u32 { |
| 222 | tokens.clear(); |
| 223 | if self.i_len == 0 { |
| 224 | return MAX_DIST; |
| 225 | } |
| 226 | |
| 227 | let mut ref_coeffs = [0i32; 64]; |
| 228 | for (dst, &src) in ref_coeffs.iter_mut().zip(blk.iter()) { |
| 229 | *dst = i32::from(src); |
| 230 | } |
| 231 | dct(&mut ref_coeffs); |
| 232 | |
| 233 | let mut dct_out = [0u8; 64]; |
| 234 | let qmats = if is_b { &self.qmats.intra_qmat } else { BINK_INTRA_QUANT }; |
| 235 | for (qidx, qmat) in qmats.iter().enumerate().skip(self.i_start).take(self.i_len) { |
| 236 | let mut coeffs = ref_coeffs; |
| 237 | for (idx, el) in coeffs.iter_mut().enumerate() { |
| 238 | *el /= qmat[BINK_INV_SCAN[idx]]; |
| 239 | } |
| 240 | |
| 241 | if coeffs[0] >= 2048 { |
| 242 | continue; |
| 243 | } |
| 244 | |
| 245 | tmp_tok.clear(); |
| 246 | tmp_tok.intradc.push(coeffs[0] as u16); |
| 247 | Self::code_dct_coeffs(&coeffs, tmp_tok); |
| 248 | if is_b { |
| 249 | tmp_tok.intraq.push(qidx as u8); |
| 250 | } else { |
| 251 | tmp_tok.other.push((qidx as u16, 4)); |
| 252 | } |
| 253 | let bits = tmp_tok.bits(is_b); |
| 254 | if rc.metric(0, bits) >= best_dist { |
| 255 | continue; |
| 256 | } |
| 257 | |
| 258 | for (idx, el) in coeffs.iter_mut().enumerate() { |
| 259 | if *el != 0 { |
| 260 | *el = (*el * qmat[BINK_INV_SCAN[idx]]) >> 11; |
| 261 | } |
| 262 | } |
| 263 | idct(&mut coeffs); |
| 264 | for (dst, &src) in dct_out.iter_mut().zip(coeffs.iter()) { |
| 265 | *dst = src as u8; |
| 266 | } |
| 267 | let diff = calc_diff(&dct_out, 8, blk, 8); |
| 268 | let dist = rc.metric(diff, bits); |
| 269 | if dist < best_dist { |
| 270 | best_dist = dist; |
| 271 | std::mem::swap(tokens, tmp_tok); |
| 272 | self.dct_i.copy_from_slice(&dct_out); |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | best_dist |
| 277 | } |
| 278 | pub fn try_dct_inter(&mut self, ref_blk: &[u8; 64], cur_blk: &[u8; 64], tokens: &mut BlockTokens, tmp_tok: &mut BlockTokens, is_b: bool, rc: &RateControl, mut best_dist: u32) -> u32 { |
| 279 | let mv_x = tokens.xoff[0]; |
| 280 | let mv_y = tokens.yoff[0]; |
| 281 | |
| 282 | let mut ref_coeffs = [0i32; 64]; |
| 283 | for (dst, &src) in ref_coeffs.iter_mut().zip(self.diff.iter()) { |
| 284 | *dst = i32::from(src); |
| 285 | } |
| 286 | dct(&mut ref_coeffs); |
| 287 | |
| 288 | let mut dct_out = [0u8; 64]; |
| 289 | let qmats = if is_b { &self.qmats.inter_qmat } else { BINK_INTER_QUANT }; |
| 290 | for (qidx, qmat) in qmats.iter().enumerate().skip(self.p_start).take(self.p_len) { |
| 291 | let mut coeffs = ref_coeffs; |
| 292 | |
| 293 | for (idx, el) in coeffs.iter_mut().enumerate() { |
| 294 | *el /= qmat[BINK_INV_SCAN[idx]]; |
| 295 | } |
| 296 | |
| 297 | if coeffs[0].unsigned_abs() >= 1024 { |
| 298 | continue; |
| 299 | } |
| 300 | |
| 301 | tmp_tok.clear(); |
| 302 | tmp_tok.interdc.push(coeffs[0] as i16); |
| 303 | tmp_tok.xoff.push(mv_x); |
| 304 | tmp_tok.yoff.push(mv_y); |
| 305 | Self::code_dct_coeffs(&coeffs, tmp_tok); |
| 306 | if is_b { |
| 307 | tmp_tok.interq.push(qidx as u8); |
| 308 | } else { |
| 309 | tmp_tok.other.push((qidx as u16, 4)); |
| 310 | } |
| 311 | let bits = tmp_tok.bits(is_b); |
| 312 | if rc.metric(0, bits) >= best_dist { |
| 313 | continue; |
| 314 | } |
| 315 | |
| 316 | for (idx, el) in coeffs.iter_mut().enumerate() { |
| 317 | if *el != 0 { |
| 318 | *el = (*el * qmat[BINK_INV_SCAN[idx]]) >> 11; |
| 319 | } |
| 320 | } |
| 321 | idct(&mut coeffs); |
| 322 | for (dst, (&prev, &diff)) in dct_out.iter_mut().zip(ref_blk.iter().zip(coeffs.iter())) { |
| 323 | *dst = (i32::from(prev) + diff) as u8; |
| 324 | } |
| 325 | let diff = calc_diff(&dct_out, 8, cur_blk, 8); |
| 326 | let dist = rc.metric(diff, bits); |
| 327 | if dist < best_dist { |
| 328 | best_dist = dist; |
| 329 | std::mem::swap(tokens, tmp_tok); |
| 330 | self.dct_p.copy_from_slice(&dct_out); |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | best_dist |
| 335 | } |
| 336 | |
| 337 | fn code_dct_coeffs(coeffs: &[i32; 64], tokens: &mut BlockTokens) { |
| 338 | let mut tree = Tree::new(false); |
| 339 | let mut flat = [0; 64]; |
| 340 | let mut blen = [0; 64]; |
| 341 | |
| 342 | for (&idx, &val) in BINK_INV_SCAN.iter().zip(coeffs.iter()).skip(1) { |
| 343 | flat[idx] = val; |
| 344 | let aval = val.unsigned_abs(); |
| 345 | let mut b = 0u8; |
| 346 | while (1 << b) <= aval { |
| 347 | b += 1; |
| 348 | } |
| 349 | blen[idx] = b; |
| 350 | } |
| 351 | |
| 352 | let mut max_val = 0; |
| 353 | let mut max_bits = 0; |
| 354 | let mut avals = [0; 64]; |
| 355 | let mut signs = [false; 64]; |
| 356 | for ((aval, sign), (&val, &vlen)) in avals.iter_mut().zip(signs.iter_mut()) |
| 357 | .zip(flat.iter().zip(blen.iter())) { |
| 358 | *aval = val.unsigned_abs(); |
| 359 | *sign = val < 0; |
| 360 | max_val = max_val.max(*aval); |
| 361 | max_bits = max_bits.max(vlen); |
| 362 | } |
| 363 | |
| 364 | tokens.other.push((u16::from(max_bits), 4)); |
| 365 | for cur_bits in (1..=max_bits).rev() { |
| 366 | let mut pos = tree.start; |
| 367 | while pos < tree.end { |
| 368 | if tree.state[pos] == TreeState::None { |
| 369 | pos += 1; |
| 370 | continue; |
| 371 | } |
| 372 | if let TreeState::Candidate(idx) = tree.state[pos] { |
| 373 | let idx = usize::from(idx); |
| 374 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 375 | tree.state[pos] = TreeState::None; |
| 376 | tokens.write_bit(1); |
| 377 | tokens.write_coef(avals[idx], signs[idx], cur_bits); |
| 378 | } else { |
| 379 | tokens.write_bit(0); |
| 380 | pos += 1; |
| 381 | } |
| 382 | continue; |
| 383 | } |
| 384 | let range = tree.state[pos].get_range(); |
| 385 | let cur_max_bits = blen[range].iter().fold(0u8, |acc, &a| acc.max(a)); |
| 386 | if cur_max_bits == cur_bits { |
| 387 | tokens.write_bit(1); |
| 388 | match tree.state[pos] { |
| 389 | TreeState::Twenty(val) => { |
| 390 | tree.state[pos] = TreeState::Sixteen(val + 4); |
| 391 | for i in 0..4 { |
| 392 | let idx = usize::from(val) + i; |
| 393 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 394 | tokens.write_bit(0); |
| 395 | tokens.write_coef(avals[idx], signs[idx], cur_bits); |
| 396 | } else { |
| 397 | tokens.write_bit(1); |
| 398 | tree.add_to_head(TreeState::Candidate(idx as u8)); |
| 399 | } |
| 400 | } |
| 401 | }, |
| 402 | TreeState::Sixteen(val) => { |
| 403 | tree.state[pos] = TreeState::Four(val); |
| 404 | for i in 1u8..4 { |
| 405 | tree.add_to_tail(TreeState::Four(val + i * 4)); |
| 406 | } |
| 407 | }, |
| 408 | TreeState::Four(val) => { |
| 409 | tree.state[pos] = TreeState::None; |
| 410 | for i in 0..4 { |
| 411 | let idx = usize::from(val) + i; |
| 412 | if blen[idx] == cur_bits && flat[idx] != 0 { |
| 413 | tokens.write_bit(0); |
| 414 | tokens.write_coef(avals[idx], signs[idx], cur_bits); |
| 415 | } else { |
| 416 | tokens.write_bit(1); |
| 417 | tree.add_to_head(TreeState::Candidate(idx as u8)); |
| 418 | } |
| 419 | } |
| 420 | }, |
| 421 | _ => unreachable!(), |
| 422 | }; |
| 423 | } else { |
| 424 | tokens.write_bit(0); |
| 425 | pos += 1; |
| 426 | } |
| 427 | } |
| 428 | } |
| 429 | } |
| 430 | } |
| 431 | |
| 432 | #[derive(Clone,Copy,Debug,PartialEq)] |
| 433 | enum TreeState { |
| 434 | None, |
| 435 | Twenty(u8), |
| 436 | Sixteen(u8), |
| 437 | Four(u8), |
| 438 | Candidate(u8), |
| 439 | } |
| 440 | |
| 441 | impl TreeState { |
| 442 | fn get_range(self) -> std::ops::Range<usize> { |
| 443 | let (base, len) = match self { |
| 444 | TreeState::None => (0, 0), |
| 445 | TreeState::Twenty(val) => (val, 20), |
| 446 | TreeState::Sixteen(val) => (val, 16), |
| 447 | TreeState::Four(val) => (val, 4), |
| 448 | TreeState::Candidate(val) => (val, 1), |
| 449 | }; |
| 450 | usize::from(base)..usize::from(base + len) |
| 451 | } |
| 452 | } |
| 453 | |
| 454 | struct Tree { |
| 455 | state: [TreeState; 128], |
| 456 | start: usize, |
| 457 | end: usize, |
| 458 | } |
| 459 | |
| 460 | impl Tree { |
| 461 | fn new(is_res: bool) -> Self { |
| 462 | let mut state = [TreeState::None; 128]; |
| 463 | let start = 64; |
| 464 | let mut end = start; |
| 465 | |
| 466 | state[end] = TreeState::Twenty(4); |
| 467 | end += 1; |
| 468 | state[end] = TreeState::Twenty(24); |
| 469 | end += 1; |
| 470 | state[end] = TreeState::Twenty(44); |
| 471 | end += 1; |
| 472 | if is_res { |
| 473 | state[end] = TreeState::Four(0); |
| 474 | end += 1; |
| 475 | } else { |
| 476 | for i in 1..4 { |
| 477 | state[end] = TreeState::Candidate(i); |
| 478 | end += 1; |
| 479 | } |
| 480 | } |
| 481 | Self { state, start, end } |
| 482 | } |
| 483 | fn add_to_tail(&mut self, ts: TreeState) { |
| 484 | self.state[self.end] = ts; |
| 485 | self.end += 1; |
| 486 | } |
| 487 | fn add_to_head(&mut self, ts: TreeState) { |
| 488 | self.start -= 1; |
| 489 | self.state[self.start] = ts; |
| 490 | } |
| 491 | } |
| 492 | |
| 493 | const A1: i32 = 2896; |
| 494 | const A2: i32 = 2217; |
| 495 | const A3: i32 = 3784; |
| 496 | const A4: i32 = -5352; |
| 497 | |
| 498 | macro_rules! idct { |
| 499 | ($src: expr, $sstep: expr, $dst: expr, $dstep: expr, $off: expr, $bias: expr, $shift: expr) => { |
| 500 | let a0 = $src[$off + 0 * $sstep] + $src[$off + 4 * $sstep]; |
| 501 | let a1 = $src[$off + 0 * $sstep] - $src[$off + 4 * $sstep]; |
| 502 | let a2 = $src[$off + 2 * $sstep] + $src[$off + 6 * $sstep]; |
| 503 | let a3 = A1.wrapping_mul($src[$off + 2 * $sstep] - $src[$off + 6 * $sstep]) >> 11; |
| 504 | let a4 = $src[$off + 5 * $sstep] + $src[$off + 3 * $sstep]; |
| 505 | let a5 = $src[$off + 5 * $sstep] - $src[$off + 3 * $sstep]; |
| 506 | let a6 = $src[$off + 1 * $sstep] + $src[$off + 7 * $sstep]; |
| 507 | let a7 = $src[$off + 1 * $sstep] - $src[$off + 7 * $sstep]; |
| 508 | let b0 = a4 + a6; |
| 509 | let b1 = A3.wrapping_mul(a5 + a7) >> 11; |
| 510 | let b2 = (A4.wrapping_mul(a5) >> 11) - b0 + b1; |
| 511 | let b3 = (A1.wrapping_mul(a6 - a4) >> 11) - b2; |
| 512 | let b4 = (A2.wrapping_mul(a7) >> 11) + b3 - b1; |
| 513 | let c0 = a0 + a2; |
| 514 | let c1 = a0 - a2; |
| 515 | let c2 = a1 + (a3 - a2); |
| 516 | let c3 = a1 - (a3 - a2); |
| 517 | |
| 518 | $dst[$off + 0 * $dstep] = (c0 + b0 + $bias) >> $shift; |
| 519 | $dst[$off + 1 * $dstep] = (c2 + b2 + $bias) >> $shift; |
| 520 | $dst[$off + 2 * $dstep] = (c3 + b3 + $bias) >> $shift; |
| 521 | $dst[$off + 3 * $dstep] = (c1 - b4 + $bias) >> $shift; |
| 522 | $dst[$off + 4 * $dstep] = (c1 + b4 + $bias) >> $shift; |
| 523 | $dst[$off + 5 * $dstep] = (c3 - b3 + $bias) >> $shift; |
| 524 | $dst[$off + 6 * $dstep] = (c2 - b2 + $bias) >> $shift; |
| 525 | $dst[$off + 7 * $dstep] = (c0 - b0 + $bias) >> $shift; |
| 526 | }; |
| 527 | } |
| 528 | |
| 529 | fn idct(coeffs: &mut [i32; 64]) { |
| 530 | let mut tmp: [i32; 64] = [0; 64]; |
| 531 | let mut row: [i32; 8] = [0; 8]; |
| 532 | for i in 0..8 { |
| 533 | idct!(coeffs, 8, tmp, 8, i, 0, 0); |
| 534 | } |
| 535 | for (drow, srow) in coeffs.chunks_exact_mut(8).zip(tmp.chunks_exact(8)) { |
| 536 | idct!(srow, 1, row, 1, 0, 0x7F, 8); |
| 537 | drow.copy_from_slice(&row); |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | const B1: i32 = 2896; |
| 542 | const B2: i32 = 3789; |
| 543 | const B3: i32 = 1569; |
| 544 | const B4: i32 = 4464; |
| 545 | const B5: i32 = 6679; |
| 546 | const B6: i32 = 1327; |
| 547 | const B7: i32 = 888; |
| 548 | macro_rules! dct { |
| 549 | ($src: expr, $sstep: expr, $dst: expr, $dstep: expr, $off: expr, $bias: expr, $shift: expr) => { |
| 550 | let a0 = $src[$off + 0 * $sstep] + $src[$off + 7 * $sstep]; |
| 551 | let a1 = $src[$off + 0 * $sstep] - $src[$off + 7 * $sstep]; |
| 552 | let a2 = $src[$off + 1 * $sstep] + $src[$off + 6 * $sstep]; |
| 553 | let a3 = $src[$off + 1 * $sstep] - $src[$off + 6 * $sstep]; |
| 554 | let a4 = $src[$off + 2 * $sstep] + $src[$off + 5 * $sstep]; |
| 555 | let a5 = $src[$off + 2 * $sstep] - $src[$off + 5 * $sstep]; |
| 556 | let a6 = $src[$off + 3 * $sstep] + $src[$off + 4 * $sstep]; |
| 557 | let a7 = $src[$off + 3 * $sstep] - $src[$off + 4 * $sstep]; |
| 558 | |
| 559 | let b0 = (a0 + a4) << 7; |
| 560 | let b1 = (a0 - a4) << 7; |
| 561 | let b2 = (a2 + a6) << 7; |
| 562 | let b3 = (a2 - a6) << 7; |
| 563 | |
| 564 | $dst[$off + 0 * $dstep] = (b0 + b2 + $bias) >> $shift; |
| 565 | $dst[$off + 4 * $dstep] = (b1 - b3 + $bias) >> $shift; |
| 566 | |
| 567 | let c0 = (a0 - a6) << 7; |
| 568 | let c1 = B1.wrapping_mul((b1 + b3) >> 7) >> 5; |
| 569 | $dst[$off + 2 * $dstep] = (c0 + c1 + $bias) >> $shift; |
| 570 | $dst[$off + 6 * $dstep] = (c0 - c1 + $bias) >> $shift; |
| 571 | |
| 572 | let d0 = B2.wrapping_mul(a1) + B3.wrapping_mul(a7); |
| 573 | let d1 = ( d0 + B4.wrapping_mul(a5) + B5.wrapping_mul(a3) + (1 << 4)) >> 5; |
| 574 | let d2 = (-d0 + B6.wrapping_mul(a5) - B7.wrapping_mul(a3) + (1 << 4)) >> 5; |
| 575 | $dst[$off + 1 * $dstep] = ((a1 << 7) + d1 + $bias) >> $shift; |
| 576 | $dst[$off + 7 * $dstep] = ((a1 << 7) + d2 + $bias) >> $shift; |
| 577 | |
| 578 | let e0 = B3.wrapping_mul(a1) - B2.wrapping_mul(a7); |
| 579 | let e1 = ( e0 - B6.wrapping_mul(a3) - B5.wrapping_mul(a5) + (1 << 4)) >> 5; |
| 580 | let e2 = (-e0 - B4.wrapping_mul(a3) + B7.wrapping_mul(a5) + (1 << 4)) >> 5; |
| 581 | |
| 582 | $dst[$off + 3 * $dstep] = ((a1 << 7) + e1 + $bias) >> $shift; |
| 583 | $dst[$off + 5 * $dstep] = ((a1 << 7) + e2 + $bias) >> $shift; |
| 584 | }; |
| 585 | } |
| 586 | |
| 587 | fn dct(coeffs: &mut [i32; 64]) { |
| 588 | let mut tmp: [i32; 64] = [0; 64]; |
| 589 | let mut row: [i32; 8] = [0; 8]; |
| 590 | for i in 0..8 { |
| 591 | dct!(coeffs, 8, tmp, 8, i, 1, 1); |
| 592 | } |
| 593 | for (drow, srow) in coeffs.chunks_exact_mut(8).zip(tmp.chunks_exact(8)) { |
| 594 | dct!(srow, 1, row, 1, 0, 0, 0); |
| 595 | drow.copy_from_slice(&row); |
| 596 | } |
| 597 | } |
| 598 | |