| 1 | mod mc; |
| 2 | pub use mc::{H264MC, McBlock}; |
| 3 | #[cfg(target_arch="x86_64")] |
| 4 | use std::arch::asm; |
| 5 | |
| 6 | pub const CHROMA_QUANTS: [u8; 52] = [ |
| 7 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, |
| 8 | 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 29, 30, |
| 9 | 31, 32, 32, 33, 34, 34, 35, 35, 36, 36, 37, 37, 37, 38, 38, 38, |
| 10 | 39, 39, 39, 39 |
| 11 | ]; |
| 12 | |
| 13 | pub const CHROMA_DC_SCAN: [usize; 4] = [ 0, 1, 2, 3]; |
| 14 | pub const ZIGZAG: [usize; 16] = [ |
| 15 | 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15 |
| 16 | ]; |
| 17 | pub const ZIGZAG1: [usize; 15] = [ |
| 18 | 0, 3, 7, 4, 1, 2, 5, 8, 11, 12, 9, 6, 10, 13, 14 |
| 19 | ]; |
| 20 | /*pub const IL_SCAN: [usize; 16] = [ |
| 21 | 0, 4, 1, 8, 12, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 |
| 22 | ];*/ |
| 23 | pub const ZIGZAG8X8: [usize; 64] = [ |
| 24 | 0, 1, 8, 16, 9, 2, 3, 10, |
| 25 | 17, 24, 32, 25, 18, 11, 4, 5, |
| 26 | 12, 19, 26, 33, 40, 48, 41, 34, |
| 27 | 27, 20, 13, 6, 7, 14, 21, 28, |
| 28 | 35, 42, 49, 56, 57, 50, 43, 36, |
| 29 | 29, 22, 15, 23, 30, 37, 44, 51, |
| 30 | 58, 59, 52, 45, 38, 31, 39, 46, |
| 31 | 53, 60, 61, 54, 47, 55, 62, 63 |
| 32 | ]; |
| 33 | |
| 34 | const LEVEL_SCALE: [[i16; 6]; 3] = [ |
| 35 | [ 10, 11, 13, 14, 16, 18 ], |
| 36 | [ 16, 18, 20, 23, 25, 29 ], |
| 37 | [ 13, 14, 16, 18, 20, 23 ] |
| 38 | ]; |
| 39 | |
| 40 | pub fn chroma_dc_transform(blk: &mut [i16; 4], qp: u8) { |
| 41 | let t0 = blk[0] + blk[2]; |
| 42 | let t1 = blk[0] - blk[2]; |
| 43 | let t2 = blk[1] + blk[3]; |
| 44 | let t3 = blk[1] - blk[3]; |
| 45 | blk[0] = t0 + t2; |
| 46 | blk[1] = t0 - t2; |
| 47 | blk[2] = t1 + t3; |
| 48 | blk[3] = t1 - t3; |
| 49 | if qp < 6 { |
| 50 | let mul = LEVEL_SCALE[0][qp as usize]; |
| 51 | for el in blk.iter_mut() { |
| 52 | *el = el.wrapping_mul(mul) >> 1; |
| 53 | } |
| 54 | } else { |
| 55 | let mul = LEVEL_SCALE[0][(qp % 6) as usize]; |
| 56 | let shift = qp / 6 - 1; |
| 57 | for el in blk.iter_mut() { |
| 58 | *el = el.wrapping_mul(mul) << shift; |
| 59 | } |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | macro_rules! transform { |
| 64 | (luma_dc; $a: expr, $b: expr, $c: expr, $d: expr) => ({ |
| 65 | let t0 = $a.wrapping_add($c); |
| 66 | let t1 = $a.wrapping_sub($c); |
| 67 | let t2 = $b.wrapping_add($d); |
| 68 | let t3 = $b.wrapping_sub($d); |
| 69 | $a = t0.wrapping_add(t2); |
| 70 | $b = t1.wrapping_add(t3); |
| 71 | $c = t1.wrapping_sub(t3); |
| 72 | $d = t0.wrapping_sub(t2); |
| 73 | }); |
| 74 | ($a: expr, $b: expr, $c: expr, $d: expr, $shift: expr) => ({ |
| 75 | let t0 = $a.wrapping_add($c); |
| 76 | let t1 = $a.wrapping_sub($c); |
| 77 | let t2 = ($b >> 1).wrapping_sub($d); |
| 78 | let t3 = $b.wrapping_add($d >> 1); |
| 79 | let bias = 1 << $shift >> 1; |
| 80 | $a = t0.wrapping_add(t3).wrapping_add(bias) >> $shift; |
| 81 | $b = t1.wrapping_add(t2).wrapping_add(bias) >> $shift; |
| 82 | $c = t1.wrapping_sub(t2).wrapping_add(bias) >> $shift; |
| 83 | $d = t0.wrapping_sub(t3).wrapping_add(bias) >> $shift; |
| 84 | }); |
| 85 | ($a: expr, $b: expr, $c: expr, $d: expr, $e: expr, $f: expr, $g: expr, $h: expr) => { |
| 86 | let e0 = $a + $e; |
| 87 | let e1 = -$d + $f - $h - ($h >> 1); |
| 88 | let e2 = $a - $e; |
| 89 | let e3 = $b + $h - $d - ($d >> 1); |
| 90 | let e4 = ($c >> 1) - $g; |
| 91 | let e5 = -$b + $h + $f + ($f >> 1); |
| 92 | let e6 = $c + ($g >> 1); |
| 93 | let e7 = $d + $f + $b + ($b >> 1); |
| 94 | |
| 95 | let f0 = e0 + e6; |
| 96 | let f1 = e1 + (e7 >> 2); |
| 97 | let f2 = e2 + e4; |
| 98 | let f3 = e3 + (e5 >> 2); |
| 99 | let f4 = e2 - e4; |
| 100 | let f5 = (e3 >> 2) - e5; |
| 101 | let f6 = e0 - e6; |
| 102 | let f7 = e7 - (e1 >> 2); |
| 103 | |
| 104 | $a = f0 + f7; |
| 105 | $b = f2 + f5; |
| 106 | $c = f4 + f3; |
| 107 | $d = f6 + f1; |
| 108 | $e = f6 - f1; |
| 109 | $f = f4 - f3; |
| 110 | $g = f2 - f5; |
| 111 | $h = f0 - f7; |
| 112 | }; |
| 113 | } |
| 114 | |
| 115 | pub fn idct_luma_dc(blk: &mut [i16; 16], qp: u8) { |
| 116 | if qp < 12 { |
| 117 | let mul = LEVEL_SCALE[0][(qp % 6) as usize]; |
| 118 | let shift = 2 - qp / 6; |
| 119 | let bias = 1 << shift >> 1; |
| 120 | for el in blk.iter_mut() { |
| 121 | *el = el.wrapping_mul(mul).wrapping_add(bias) >> shift; |
| 122 | } |
| 123 | } else { |
| 124 | let mul = LEVEL_SCALE[0][(qp % 6) as usize]; |
| 125 | let shift = qp / 6 - 2; |
| 126 | for el in blk.iter_mut() { |
| 127 | *el = el.wrapping_mul(mul) << shift; |
| 128 | } |
| 129 | } |
| 130 | for i in 0..4 { |
| 131 | transform!(luma_dc; blk[i], blk[i + 4], blk[i + 8], blk[i + 12]); |
| 132 | } |
| 133 | for row in blk.chunks_exact_mut(4) { |
| 134 | transform!(luma_dc; row[0], row[1], row[2], row[3]); |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | pub fn idct_skip_dc(blk: &mut [i16; 16], qp: u8) { |
| 139 | const BLK_INDEX: [usize; 16] = [ |
| 140 | 0, 2, 0, 2, |
| 141 | 2, 1, 2, 1, |
| 142 | 0, 2, 0, 2, |
| 143 | 2, 1, 2, 1 |
| 144 | ]; |
| 145 | let qidx = (qp % 6) as usize; |
| 146 | let shift = qp / 6; |
| 147 | for (el, &idx) in blk.iter_mut().zip(BLK_INDEX.iter()).skip(1) { |
| 148 | *el = (*el * LEVEL_SCALE[idx][qidx]) << shift; |
| 149 | } |
| 150 | for row in blk.chunks_exact_mut(4) { |
| 151 | transform!(row[0], row[1], row[2], row[3], 0); |
| 152 | } |
| 153 | for i in 0..4 { |
| 154 | transform!(blk[i], blk[i + 4], blk[i + 8], blk[i + 12], 6); |
| 155 | } |
| 156 | } |
| 157 | |
| 158 | pub fn idct(blk: &mut [i16; 16], qp: u8) { |
| 159 | const BLK_INDEX: [usize; 16] = [ |
| 160 | 0, 2, 0, 2, |
| 161 | 2, 1, 2, 1, |
| 162 | 0, 2, 0, 2, |
| 163 | 2, 1, 2, 1 |
| 164 | ]; |
| 165 | let qidx = (qp % 6) as usize; |
| 166 | let shift = qp / 6; |
| 167 | for (el, &idx) in blk.iter_mut().zip(BLK_INDEX.iter()) { |
| 168 | *el = (*el * LEVEL_SCALE[idx][qidx]) << shift; |
| 169 | } |
| 170 | for row in blk.chunks_exact_mut(4) { |
| 171 | transform!(row[0], row[1], row[2], row[3], 0); |
| 172 | } |
| 173 | for i in 0..4 { |
| 174 | transform!(blk[i], blk[i + 4], blk[i + 8], blk[i + 12], 6); |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | pub fn idct_dc(blk: &mut [i16; 16], qp: u8, quant_dc: bool) { |
| 179 | let dc = if quant_dc { |
| 180 | (blk[0] * LEVEL_SCALE[0][(qp % 6) as usize]) << (qp / 6) |
| 181 | } else { |
| 182 | blk[0] |
| 183 | }; |
| 184 | *blk = [(dc + 0x20) >> 6; 16]; |
| 185 | } |
| 186 | |
| 187 | const QMAT_8X8: [[u8; 16]; 6] = [ |
| 188 | [ |
| 189 | 20, 19, 25, 19, |
| 190 | 19, 18, 24, 18, |
| 191 | 25, 24, 32, 24, |
| 192 | 19, 18, 24, 18 |
| 193 | ], [ |
| 194 | 22, 21, 28, 21, |
| 195 | 21, 19, 26, 19, |
| 196 | 28, 26, 35, 26, |
| 197 | 21, 19, 26, 19 |
| 198 | ], [ |
| 199 | 26, 24, 33, 24, |
| 200 | 24, 23, 31, 23, |
| 201 | 33, 31, 42, 31, |
| 202 | 24, 23, 31, 23 |
| 203 | ], [ |
| 204 | 28, 26, 35, 26, |
| 205 | 26, 25, 33, 25, |
| 206 | 35, 33, 45, 33, |
| 207 | 26, 25, 33, 25 |
| 208 | ], [ |
| 209 | 32, 30, 40, 30, |
| 210 | 30, 28, 38, 28, |
| 211 | 40, 38, 51, 38, |
| 212 | 30, 28, 38, 28 |
| 213 | ], [ |
| 214 | 36, 34, 46, 34, |
| 215 | 34, 32, 43, 32, |
| 216 | 46, 43, 58, 43, |
| 217 | 34, 32, 43, 32 |
| 218 | ] |
| 219 | ]; |
| 220 | |
| 221 | pub fn dequant8x8(blk: &mut [i16; 64], slist: &[u8; 64]) { |
| 222 | for (el, &scan) in blk.iter_mut().zip(ZIGZAG8X8.iter()) { |
| 223 | if *el != 0 { |
| 224 | *el = el.wrapping_mul(i16::from(slist[scan])); |
| 225 | } |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | pub fn idct8x8(blk: &mut [i16; 64], qp: u8) { |
| 230 | let mut tmp = [0i32; 64]; |
| 231 | let qmat = &QMAT_8X8[(qp % 6) as usize]; |
| 232 | if qp >= 36 { |
| 233 | let shift = qp / 6 - 6; |
| 234 | for (i, (dst, &src)) in tmp.iter_mut().zip(blk.iter()).enumerate() { |
| 235 | let x = i & 7; |
| 236 | let y = i >> 3; |
| 237 | let idx = (x & 3) + (y & 3) * 4; |
| 238 | *dst = i32::from(src).wrapping_mul(i32::from(qmat[idx])) << shift; |
| 239 | } |
| 240 | } else { |
| 241 | let shift = 6 - qp / 6; |
| 242 | let bias = (1 << shift) >> 1; |
| 243 | for (i, (dst, &src)) in tmp.iter_mut().zip(blk.iter()).enumerate() { |
| 244 | let x = i & 7; |
| 245 | let y = i >> 3; |
| 246 | let idx = (x & 3) + (y & 3) * 4; |
| 247 | *dst = i32::from(src).wrapping_mul(i32::from(qmat[idx])).wrapping_add(bias) >> shift; |
| 248 | } |
| 249 | } |
| 250 | for row in tmp.chunks_exact_mut(8) { |
| 251 | transform!(row[0], row[1], row[2], row[3], row[4], row[5], row[6], row[7]); |
| 252 | } |
| 253 | for col in 0..8 { |
| 254 | transform!(tmp[col], tmp[col + 8], tmp[col + 8 * 2], tmp[col + 8 * 3], |
| 255 | tmp[col + 8 * 4], tmp[col + 8 * 5], tmp[col + 8 * 6], tmp[col + 8 * 7]); |
| 256 | } |
| 257 | for (dst, &src) in blk.iter_mut().zip(tmp.iter()) { |
| 258 | *dst = ((src + 0x20) >> 6) as i16; |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | pub fn add_coeffs(dst: &mut [u8], offset: usize, stride: usize, coeffs: &[i16]) { |
| 263 | let out = &mut dst[offset..][..stride * 3 + 4]; |
| 264 | for (line, src) in out.chunks_mut(stride).take(4).zip(coeffs.chunks_exact(4)) { |
| 265 | for (dst, src) in line.iter_mut().take(4).zip(src.iter()) { |
| 266 | *dst = (i32::from(*dst) + i32::from(*src)).max(0).min(255) as u8; |
| 267 | } |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | pub fn add_coeffs8(dst: &mut [u8], offset: usize, stride: usize, coeffs: &[i16; 64]) { |
| 272 | let out = &mut dst[offset..]; |
| 273 | for (line, src) in out.chunks_mut(stride).take(8).zip(coeffs.chunks_exact(8)) { |
| 274 | for (dst, src) in line.iter_mut().take(8).zip(src.iter()) { |
| 275 | *dst = (i32::from(*dst) + i32::from(*src)).max(0).min(255) as u8; |
| 276 | } |
| 277 | } |
| 278 | } |
| 279 | |
| 280 | fn clip8(val: i16) -> u8 { val.max(0).min(255) as u8 } |
| 281 | |
| 282 | fn ipred_dc128(buf: &mut [u8], stride: usize, bsize: usize) { |
| 283 | for row in buf.chunks_mut(stride).take(bsize) { |
| 284 | for el in row[..bsize].iter_mut() { |
| 285 | *el = 128; |
| 286 | } |
| 287 | } |
| 288 | } |
| 289 | fn ipred_ver(buf: &mut [u8], stride: usize, top: &[u8], bsize: usize) { |
| 290 | for row in buf.chunks_mut(stride).take(bsize) { |
| 291 | row[..bsize].copy_from_slice(&top[..bsize]); |
| 292 | } |
| 293 | } |
| 294 | fn ipred_hor(buf: &mut [u8], stride: usize, left: &[u8], bsize: usize) { |
| 295 | for (row, &left) in buf.chunks_mut(stride).zip(left[1..].iter()).take(bsize) { |
| 296 | for el in row[..bsize].iter_mut() { |
| 297 | *el = left; |
| 298 | } |
| 299 | } |
| 300 | } |
| 301 | fn ipred_dc(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], bsize: usize, shift: u8) { |
| 302 | let mut adc: u16 = 0; |
| 303 | for i in 0..bsize { adc += u16::from(top[i]); } |
| 304 | for i in 0..bsize { adc += u16::from(left[i + 1]); } |
| 305 | let dc = ((adc + (1 << (shift - 1))) >> shift) as u8; |
| 306 | |
| 307 | for row in buf.chunks_mut(stride).take(bsize) { |
| 308 | for el in row[..bsize].iter_mut() { |
| 309 | *el = dc; |
| 310 | } |
| 311 | } |
| 312 | } |
| 313 | fn ipred_left_dc(buf: &mut [u8], stride: usize, left: &[u8], bsize: usize, shift: u8) { |
| 314 | let mut adc: u16 = 0; |
| 315 | for i in 0..bsize { adc += u16::from(left[i + 1]); } |
| 316 | let dc = ((adc + (1 << (shift - 1))) >> shift) as u8; |
| 317 | |
| 318 | for row in buf.chunks_mut(stride).take(bsize) { |
| 319 | for el in row[..bsize].iter_mut() { |
| 320 | *el = dc; |
| 321 | } |
| 322 | } |
| 323 | } |
| 324 | fn ipred_top_dc(buf: &mut [u8], stride: usize, top: &[u8], bsize: usize, shift: u8) { |
| 325 | let mut adc: u16 = 0; |
| 326 | for i in 0..bsize { adc += u16::from(top[i]); } |
| 327 | let dc = ((adc + (1 << (shift - 1))) >> shift) as u8; |
| 328 | |
| 329 | for row in buf.chunks_mut(stride).take(bsize) { |
| 330 | for el in row[..bsize].iter_mut() { |
| 331 | *el = dc; |
| 332 | } |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | fn load(dst: &mut [u16], src: &[u8]) { |
| 337 | for (dst, &src) in dst.iter_mut().zip(src.iter()) { |
| 338 | *dst = u16::from(src); |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | fn ipred_4x4_ver(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8], _tr: &[u8]) { |
| 343 | ipred_ver(buf, stride, top, 4); |
| 344 | } |
| 345 | fn ipred_4x4_hor(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8], _tr: &[u8]) { |
| 346 | ipred_hor(buf, stride, left, 4); |
| 347 | } |
| 348 | fn ipred_4x4_diag_down_left(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8], tr: &[u8]) { |
| 349 | let mut t: [u16; 9] = [0; 9]; |
| 350 | load(&mut t[..4], top); |
| 351 | load(&mut t[4..8], tr); |
| 352 | t[8] = t[7]; |
| 353 | |
| 354 | for i in 0..4 { |
| 355 | buf[i] = ((t[i] + 2 * t[i + 1] + t[i + 2] + 2) >> 2) as u8; |
| 356 | } |
| 357 | let dst = &mut buf[stride..]; |
| 358 | for i in 0..4 { |
| 359 | dst[i] = ((t[i + 1] + 2 * t[i + 2] + t[i + 3] + 2) >> 2) as u8; |
| 360 | } |
| 361 | let dst = &mut buf[stride * 2..]; |
| 362 | for i in 0..4 { |
| 363 | dst[i] = ((t[i + 2] + 2 * t[i + 3] + t[i + 4] + 2) >> 2) as u8; |
| 364 | } |
| 365 | let dst = &mut buf[stride * 3..]; |
| 366 | for i in 0..4 { |
| 367 | dst[i] = ((t[i + 3] + 2 * t[i + 4] + t[i + 5] + 2) >> 2) as u8; |
| 368 | } |
| 369 | } |
| 370 | fn ipred_4x4_diag_down_right(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], _tr: &[u8]) { |
| 371 | let mut t: [u16; 5] = [0; 5]; |
| 372 | t[0] = u16::from(left[0]); |
| 373 | load(&mut t[1..], top); |
| 374 | let mut l: [u16; 5] = [0; 5]; |
| 375 | load(&mut l, left); |
| 376 | let dst = buf; |
| 377 | |
| 378 | for j in 0..4 { |
| 379 | for i in 0..j { |
| 380 | dst[i + j * stride] = ((l[j - i - 1] + 2 * l[j - i] + l[j - i + 1] + 2) >> 2) as u8; |
| 381 | } |
| 382 | dst[j + j * stride] = ((l[1] + 2 * l[0] + t[1] + 2) >> 2) as u8; |
| 383 | for i in (j+1)..4 { |
| 384 | dst[i + j * stride] = ((t[i - j - 1] + 2 * t[i - j] + t[i - j + 1] + 2) >> 2) as u8; |
| 385 | } |
| 386 | } |
| 387 | } |
| 388 | fn ipred_4x4_ver_right(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], _tr: &[u8]) { |
| 389 | let mut t: [u16; 5] = [0; 5]; |
| 390 | t[0] = u16::from(left[0]); |
| 391 | load(&mut t[1..], top); |
| 392 | let mut l: [u16; 5] = [0; 5]; |
| 393 | load(&mut l, left); |
| 394 | let dst = buf; |
| 395 | |
| 396 | for j in 0..4 { |
| 397 | for i in 0..4 { |
| 398 | let zvr = ((2 * i) as i8) - (j as i8); |
| 399 | let pix; |
| 400 | if zvr >= 0 { |
| 401 | if (zvr & 1) == 0 { |
| 402 | pix = (t[i - (j >> 1)] + t[i - (j >> 1) + 1] + 1) >> 1; |
| 403 | } else { |
| 404 | pix = (t[i - (j >> 1) - 1] + 2 * t[i - (j >> 1)] + t[i - (j >> 1) + 1] + 2) >> 2; |
| 405 | } |
| 406 | } else { |
| 407 | if zvr == -1 { |
| 408 | pix = (l[1] + 2 * l[0] + t[1] + 2) >> 2; |
| 409 | } else { |
| 410 | pix = (l[j] + 2 * l[j - 1] + l[j - 2] + 2) >> 2; |
| 411 | } |
| 412 | } |
| 413 | dst[i + j * stride] = pix as u8; |
| 414 | } |
| 415 | } |
| 416 | } |
| 417 | fn ipred_4x4_ver_left(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8], tr: &[u8]) { |
| 418 | let mut t: [u16; 8] = [0; 8]; |
| 419 | load(&mut t[..4], top); |
| 420 | load(&mut t[4..], tr); |
| 421 | let dst = buf; |
| 422 | |
| 423 | dst[0 + 0 * stride] = ((t[0] + t[1] + 1) >> 1) as u8; |
| 424 | let pix = ((t[1] + t[2] + 1) >> 1) as u8; |
| 425 | dst[1 + 0 * stride] = pix; |
| 426 | dst[0 + 2 * stride] = pix; |
| 427 | let pix = ((t[2] + t[3] + 1) >> 1) as u8; |
| 428 | dst[2 + 0 * stride] = pix; |
| 429 | dst[1 + 2 * stride] = pix; |
| 430 | let pix = ((t[3] + t[4] + 1) >> 1) as u8; |
| 431 | dst[3 + 0 * stride] = pix; |
| 432 | dst[2 + 2 * stride] = pix; |
| 433 | dst[3 + 2 * stride] = ((t[4] + t[5] + 1) >> 1) as u8; |
| 434 | dst[0 + 1 * stride] = ((t[0] + 2*t[1] + t[2] + 2) >> 2) as u8; |
| 435 | let pix = ((t[1] + 2*t[2] + t[3] + 2) >> 2) as u8; |
| 436 | dst[1 + 1 * stride] = pix; |
| 437 | dst[0 + 3 * stride] = pix; |
| 438 | let pix = ((t[2] + 2*t[3] + t[4] + 2) >> 2) as u8; |
| 439 | dst[2 + 1 * stride] = pix; |
| 440 | dst[1 + 3 * stride] = pix; |
| 441 | let pix = ((t[3] + 2*t[4] + t[5] + 2) >> 2) as u8; |
| 442 | dst[3 + 1 * stride] = pix; |
| 443 | dst[2 + 3 * stride] = pix; |
| 444 | dst[3 + 3 * stride] = ((t[4] + 2*t[5] + t[6] + 2) >> 2) as u8; |
| 445 | } |
| 446 | fn ipred_4x4_hor_down(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], _tr: &[u8]) { |
| 447 | let mut t: [u16; 5] = [0; 5]; |
| 448 | t[0] = u16::from(left[0]); |
| 449 | load(&mut t[1..], top); |
| 450 | let mut l: [u16; 5] = [0; 5]; |
| 451 | load(&mut l, left); |
| 452 | let dst = buf; |
| 453 | |
| 454 | for j in 0..4 { |
| 455 | for i in 0..4 { |
| 456 | let zhd = ((2 * j) as i8) - (i as i8); |
| 457 | let pix; |
| 458 | if zhd >= 0 { |
| 459 | if (zhd & 1) == 0 { |
| 460 | pix = (l[j - (i >> 1)] + l[j - (i >> 1) + 1] + 1) >> 1; |
| 461 | } else { |
| 462 | pix = (l[j - (i >> 1) - 1] + 2 * l[j - (i >> 1)] + l[j - (i >> 1) + 1] + 2) >> 2; |
| 463 | } |
| 464 | } else { |
| 465 | if zhd == -1 { |
| 466 | pix = (l[1] + 2 * l[0] + t[1] + 2) >> 2; |
| 467 | } else { |
| 468 | pix = (t[i - 2] + 2 * t[i - 1] + t[i] + 2) >> 2; |
| 469 | } |
| 470 | } |
| 471 | dst[i + j * stride] = pix as u8; |
| 472 | } |
| 473 | } |
| 474 | } |
| 475 | fn ipred_4x4_hor_up(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8], _tr: &[u8]) { |
| 476 | let mut l: [u16; 8] = [0; 8]; |
| 477 | load(&mut l, &left[1..]); |
| 478 | let dst = buf; |
| 479 | |
| 480 | dst[0 + 0 * stride] = ((l[0] + l[1] + 1) >> 1) as u8; |
| 481 | dst[1 + 0 * stride] = ((l[0] + 2*l[1] + l[2] + 2) >> 2) as u8; |
| 482 | let pix = ((l[1] + l[2] + 1) >> 1) as u8; |
| 483 | dst[2 + 0 * stride] = pix; |
| 484 | dst[0 + 1 * stride] = pix; |
| 485 | let pix = ((l[1] + 2*l[2] + l[3] + 2) >> 2) as u8; |
| 486 | dst[3 + 0 * stride] = pix; |
| 487 | dst[1 + 1 * stride] = pix; |
| 488 | let pix = ((l[2] + l[3] + 1) >> 1) as u8; |
| 489 | dst[2 + 1 * stride] = pix; |
| 490 | dst[0 + 2 * stride] = pix; |
| 491 | let pix = ((l[2] + 3*l[3] + 2) >> 2) as u8; |
| 492 | dst[3 + 1 * stride] = pix; |
| 493 | dst[1 + 2 * stride] = pix; |
| 494 | dst[3 + 2 * stride] = l[3] as u8; |
| 495 | dst[1 + 3 * stride] = l[3] as u8; |
| 496 | dst[0 + 3 * stride] = l[3] as u8; |
| 497 | dst[2 + 2 * stride] = l[3] as u8; |
| 498 | dst[2 + 3 * stride] = l[3] as u8; |
| 499 | dst[3 + 3 * stride] = l[3] as u8; |
| 500 | } |
| 501 | fn ipred_4x4_dc(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], _tr: &[u8]) { |
| 502 | ipred_dc(buf, stride, top, left, 4, 3); |
| 503 | } |
| 504 | fn ipred_4x4_left_dc(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8], _tr: &[u8]) { |
| 505 | ipred_left_dc(buf, stride, left, 4, 2); |
| 506 | } |
| 507 | fn ipred_4x4_top_dc(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8], _tr: &[u8]) { |
| 508 | ipred_top_dc(buf, stride, top, 4, 2); |
| 509 | } |
| 510 | fn ipred_4x4_dc128(buf: &mut [u8], stride: usize, _top: &[u8], _left: &[u8], _tr: &[u8]) { |
| 511 | ipred_dc128(buf, stride, 4); |
| 512 | } |
| 513 | |
| 514 | pub struct IPred8Context { |
| 515 | pub t: [u8; 16], |
| 516 | pub l: [u8; 8], |
| 517 | pub tl: u8, |
| 518 | } |
| 519 | |
| 520 | impl IPred8Context { |
| 521 | pub fn new() -> Self { |
| 522 | Self { |
| 523 | t: [128; 16], |
| 524 | l: [128; 8], |
| 525 | tl: 128, |
| 526 | } |
| 527 | } |
| 528 | pub fn fill(&mut self, top: &[u8], left: &[u8], has_t: bool, has_tr: bool, has_l: bool, has_tl: bool) { |
| 529 | let mut t = [0x80u8; 19]; |
| 530 | let mut l = [0x80u8; 11]; |
| 531 | if has_t { |
| 532 | t[1..8 + 1].copy_from_slice(&top[..8]); |
| 533 | } |
| 534 | if has_tr { |
| 535 | t[8 + 1..16 + 1].copy_from_slice(&top[8..][..8]); |
| 536 | t[16 + 1] = t[15 + 1]; |
| 537 | t[17 + 1] = t[15 + 1]; |
| 538 | } else { |
| 539 | let (t0, t1) = t.split_at_mut(8 + 1); |
| 540 | for el in t1.iter_mut() { |
| 541 | *el = t0[7 + 1]; |
| 542 | } |
| 543 | } |
| 544 | if has_l { |
| 545 | l[1..9].copy_from_slice(&left[1..9]); |
| 546 | l[8 + 1] = l[7 + 1]; |
| 547 | l[9 + 1] = l[7 + 1]; |
| 548 | } |
| 549 | if has_tl { |
| 550 | t[0] = left[0]; |
| 551 | l[0] = left[0]; |
| 552 | } else { |
| 553 | t[0] = t[1]; |
| 554 | l[0] = l[1]; |
| 555 | } |
| 556 | |
| 557 | for i in 0..16 { |
| 558 | self.t[i] = ((u16::from(t[i]) + 2 * u16::from(t[i + 1]) + u16::from(t[i + 2]) + 2) >> 2) as u8; |
| 559 | } |
| 560 | for i in 0..8 { |
| 561 | self.l[i] = ((u16::from(l[i]) + 2 * u16::from(l[i + 1]) + u16::from(l[i + 2]) + 2) >> 2) as u8; |
| 562 | } |
| 563 | self.tl = if has_t && has_l { |
| 564 | ((u16::from(t[1]) + 2 * u16::from(t[0]) + u16::from(l[1]) + 2) >> 2) as u8 |
| 565 | } else if has_t { |
| 566 | ((3 * u16::from(t[0]) + u16::from(t[1]) + 2) >> 2) as u8 |
| 567 | } else if has_l { |
| 568 | ((3 * u16::from(l[0]) + u16::from(l[1]) + 2) >> 2) as u8 |
| 569 | } else { |
| 570 | t[0] |
| 571 | }; |
| 572 | } |
| 573 | } |
| 574 | |
| 575 | fn ipred_y_8x8_ver(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 576 | for row in buf.chunks_mut(stride).take(8) { |
| 577 | row[..8].copy_from_slice(&ctx.t[..8]); |
| 578 | } |
| 579 | } |
| 580 | fn ipred_y_8x8_hor(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 581 | for (row, &l) in buf.chunks_mut(stride).zip(ctx.l.iter()).take(8) { |
| 582 | row[..8].copy_from_slice(&[l; 8]); |
| 583 | } |
| 584 | } |
| 585 | fn ipred_y_8x8_diag_down_left(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 586 | let mut t = [0u16; 16]; |
| 587 | load(&mut t, &ctx.t); |
| 588 | |
| 589 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 590 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 591 | *pix = ((if (x != 7) || (y != 7) { |
| 592 | t[x + y] + 2 * t[x + y + 1] + t[x + y + 2] |
| 593 | } else { |
| 594 | t[14] + 3 * t[15] |
| 595 | } + 2) >> 2) as u8; |
| 596 | } |
| 597 | } |
| 598 | } |
| 599 | fn ipred_y_8x8_diag_down_right(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 600 | let mut t = [0u16; 9]; |
| 601 | t[0] = u16::from(ctx.tl); |
| 602 | load(&mut t[1..], &ctx.t); |
| 603 | let mut l = [0u16; 9]; |
| 604 | l[0] = u16::from(ctx.tl); |
| 605 | load(&mut l[1..], &ctx.l); |
| 606 | let diag = t[1] + 2 * t[0] + l[1]; |
| 607 | |
| 608 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 609 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 610 | *pix = ((if x > y { |
| 611 | t[x - y - 1] + 2 * t[x - y] + t[x - y + 1] |
| 612 | } else if x < y { |
| 613 | l[y - x - 1] + 2 * l[y - x] + l[y - x + 1] |
| 614 | } else { |
| 615 | diag |
| 616 | } + 2) >> 2) as u8; |
| 617 | } |
| 618 | } |
| 619 | } |
| 620 | fn ipred_y_8x8_ver_right(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 621 | let mut t = [0u16; 9]; |
| 622 | t[0] = u16::from(ctx.tl); |
| 623 | load(&mut t[1..], &ctx.t); |
| 624 | let mut l = [0u16; 9]; |
| 625 | l[0] = u16::from(ctx.tl); |
| 626 | load(&mut l[1..], &ctx.l); |
| 627 | |
| 628 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 629 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 630 | let zvr = 2 * (x as i8) - (y as i8); |
| 631 | *pix = if zvr >= 0 { |
| 632 | let ix = x - (y >> 1); |
| 633 | if (zvr & 1) == 0 { |
| 634 | (t[ix] + t[ix + 1] + 1) >> 1 |
| 635 | } else { |
| 636 | (t[ix - 1] + 2 * t[ix] + t[ix + 1] + 2) >> 2 |
| 637 | } |
| 638 | } else if zvr == -1 { |
| 639 | (l[1] + 2 * l[0] + t[1] + 2) >> 2 |
| 640 | } else { |
| 641 | let ix = y - 2 * x; |
| 642 | (l[ix] + 2 * l[ix - 1] + l[ix - 2] + 2) >> 2 |
| 643 | } as u8; |
| 644 | } |
| 645 | } |
| 646 | } |
| 647 | fn ipred_y_8x8_ver_left(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 648 | let mut t = [0u16; 16]; |
| 649 | load(&mut t, &ctx.t); |
| 650 | |
| 651 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 652 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 653 | let ix = x + (y >> 1); |
| 654 | *pix = if (y & 1) == 0 { |
| 655 | (t[ix] + t[ix + 1] + 1) >> 1 |
| 656 | } else { |
| 657 | (t[ix] + 2 * t[ix + 1] + t[ix + 2] + 2) >> 2 |
| 658 | } as u8; |
| 659 | } |
| 660 | } |
| 661 | |
| 662 | } |
| 663 | fn ipred_y_8x8_hor_down(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 664 | let mut t = [0u16; 9]; |
| 665 | t[0] = u16::from(ctx.tl); |
| 666 | load(&mut t[1..], &ctx.t); |
| 667 | let mut l = [0u16; 9]; |
| 668 | l[0] = u16::from(ctx.tl); |
| 669 | load(&mut l[1..], &ctx.l); |
| 670 | |
| 671 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 672 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 673 | let zhd = 2 * (y as i8) - (x as i8); |
| 674 | *pix = if zhd >= 0 { |
| 675 | let ix = y - (x >> 1); |
| 676 | if (zhd & 1) == 0 { |
| 677 | (l[ix] + l[ix + 1] + 1) >> 1 |
| 678 | } else { |
| 679 | (l[ix - 1] + 2 * l[ix] + l[ix + 1] + 2) >> 2 |
| 680 | } |
| 681 | } else if zhd == -1 { |
| 682 | (l[1] + 2 * l[0] + t[1] + 2) >> 2 |
| 683 | } else { |
| 684 | let ix = x - 2 * y; |
| 685 | (t[ix] + 2 * t[ix - 1] + t[ix - 2] + 2) >> 2 |
| 686 | } as u8; |
| 687 | } |
| 688 | } |
| 689 | } |
| 690 | fn ipred_y_8x8_hor_up(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 691 | let mut l = [0u16; 8]; |
| 692 | load(&mut l, &ctx.l); |
| 693 | |
| 694 | for (y, row) in buf.chunks_mut(stride).take(8).enumerate() { |
| 695 | for (x, pix) in row.iter_mut().take(8).enumerate() { |
| 696 | let zhu = x + 2 * y; |
| 697 | let ix = y + (x >> 1); |
| 698 | *pix = if zhu > 13 { |
| 699 | l[7] |
| 700 | } else if zhu == 13 { |
| 701 | (l[6] + 3 * l[7] + 2) >> 2 |
| 702 | } else if (zhu & 1) != 0 { |
| 703 | (l[ix] + 2 * l[ix + 1] + l[ix + 2] + 2) >> 2 |
| 704 | } else { |
| 705 | (l[ix] + l[ix + 1] + 1) >> 1 |
| 706 | } as u8; |
| 707 | } |
| 708 | } |
| 709 | } |
| 710 | fn ipred_y_8x8_dc(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 711 | let mut sum = 0u16; |
| 712 | for &t in ctx.t[..8].iter() { |
| 713 | sum += u16::from(t); |
| 714 | } |
| 715 | for &l in ctx.l[..8].iter() { |
| 716 | sum += u16::from(l); |
| 717 | } |
| 718 | let dc = ((sum + 8) >> 4) as u8; |
| 719 | for row in buf.chunks_mut(stride).take(8) { |
| 720 | for pix in row.iter_mut().take(8) { |
| 721 | *pix = dc; |
| 722 | } |
| 723 | } |
| 724 | } |
| 725 | fn ipred_y_8x8_left_dc(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 726 | let mut sum = 0u16; |
| 727 | for &l in ctx.l[..8].iter() { |
| 728 | sum += u16::from(l); |
| 729 | } |
| 730 | let dc = ((sum + 4) >> 3) as u8; |
| 731 | for row in buf.chunks_mut(stride).take(8) { |
| 732 | for pix in row.iter_mut().take(8) { |
| 733 | *pix = dc; |
| 734 | } |
| 735 | } |
| 736 | } |
| 737 | fn ipred_y_8x8_top_dc(buf: &mut [u8], stride: usize, ctx: &IPred8Context) { |
| 738 | let mut sum = 0u16; |
| 739 | for &t in ctx.t[..8].iter() { |
| 740 | sum += u16::from(t); |
| 741 | } |
| 742 | let dc = ((sum + 4) >> 3) as u8; |
| 743 | for row in buf.chunks_mut(stride).take(8) { |
| 744 | for pix in row.iter_mut().take(8) { |
| 745 | *pix = dc; |
| 746 | } |
| 747 | } |
| 748 | } |
| 749 | fn ipred_y_8x8_dc128(buf: &mut [u8], stride: usize, _ctx: &IPred8Context) { |
| 750 | ipred_dc128(buf, stride, 8); |
| 751 | } |
| 752 | |
| 753 | fn ipred_8x8_ver(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8]) { |
| 754 | ipred_ver(buf, stride, top, 8); |
| 755 | } |
| 756 | fn ipred_8x8_hor(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8]) { |
| 757 | ipred_hor(buf, stride, left, 8); |
| 758 | } |
| 759 | fn ipred_8x8_dc(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8]) { |
| 760 | let mut l = [0; 8]; |
| 761 | load(&mut l, &left[1..]); |
| 762 | let mut t = [0; 8]; |
| 763 | load(&mut t, top); |
| 764 | |
| 765 | let dc0 = ((t[0] + t[1] + t[2] + t[3] + l[0] + l[1] + l[2] + l[3] + 4) >> 3) as u8; |
| 766 | let sum1 = t[4] + t[5] + t[6] + t[7]; |
| 767 | let dc1 = ((sum1 + 2) >> 2) as u8; |
| 768 | let sum2 = l[4] + l[5] + l[6] + l[7]; |
| 769 | let dc2 = ((sum2 + 2) >> 2) as u8; |
| 770 | let dc3 = ((sum1 + sum2 + 4) >> 3) as u8; |
| 771 | |
| 772 | for row in buf.chunks_mut(stride).take(4) { |
| 773 | row[..4].copy_from_slice(&[dc0; 4]); |
| 774 | row[4..8].copy_from_slice(&[dc1; 4]); |
| 775 | } |
| 776 | for row in buf.chunks_mut(stride).skip(4).take(4) { |
| 777 | row[..4].copy_from_slice(&[dc2; 4]); |
| 778 | row[4..8].copy_from_slice(&[dc3; 4]); |
| 779 | } |
| 780 | } |
| 781 | fn ipred_8x8_left_dc(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8]) { |
| 782 | let mut left_dc0 = 0; |
| 783 | let mut left_dc1 = 0; |
| 784 | for &el in left[1..].iter().take(4) { |
| 785 | left_dc0 += u16::from(el); |
| 786 | } |
| 787 | for &el in left[1..].iter().skip(4).take(4) { |
| 788 | left_dc1 += u16::from(el); |
| 789 | } |
| 790 | let dc0 = ((left_dc0 + 2) >> 2) as u8; |
| 791 | let dc2 = ((left_dc1 + 2) >> 2) as u8; |
| 792 | for row in buf.chunks_mut(stride).take(4) { |
| 793 | row[..8].copy_from_slice(&[dc0; 8]); |
| 794 | } |
| 795 | for row in buf.chunks_mut(stride).skip(4).take(4) { |
| 796 | row[..8].copy_from_slice(&[dc2; 8]); |
| 797 | } |
| 798 | } |
| 799 | fn ipred_8x8_top_dc(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8]) { |
| 800 | ipred_top_dc(buf, stride, top, 4, 2); |
| 801 | ipred_top_dc(&mut buf[4..], stride, &top[4..], 4, 2); |
| 802 | let mut top = [0; 8]; |
| 803 | top.copy_from_slice(&buf[stride * 3..][..8]); |
| 804 | ipred_top_dc(&mut buf[4 * stride..], stride, &top, 4, 2); |
| 805 | ipred_top_dc(&mut buf[4 + 4 * stride..], stride, &top[4..], 4, 2); |
| 806 | } |
| 807 | fn ipred_8x8_dc128(buf: &mut [u8], stride: usize, _top: &[u8], _left: &[u8]) { |
| 808 | ipred_dc128(buf, stride, 8); |
| 809 | } |
| 810 | fn ipred_8x8_plane(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8]) { |
| 811 | let mut h: i32 = 4 * (i32::from(top[7]) - i32::from(left[0])); |
| 812 | let mut v: i32 = 4 * (i32::from(left[8]) - i32::from(left[0])); |
| 813 | for i in 0..3 { |
| 814 | let i1 = (i + 1) as i32; |
| 815 | h += i1 * (i32::from(top[4 + i]) - i32::from(top[2 - i])); |
| 816 | v += i1 * (i32::from(left[5 + i]) - i32::from(left[3 - i])); |
| 817 | } |
| 818 | let b = (17 * h + 16) >> 5; |
| 819 | let c = (17 * v + 16) >> 5; |
| 820 | let mut a = 16 * (i32::from(left[8]) + i32::from(top[7])) - 3 * (b + c) + 16; |
| 821 | for line in buf.chunks_mut(stride).take(8) { |
| 822 | let mut acc = a; |
| 823 | for el in line.iter_mut().take(8) { |
| 824 | *el = clip8((acc >> 5) as i16); |
| 825 | acc += b; |
| 826 | } |
| 827 | a += c; |
| 828 | } |
| 829 | } |
| 830 | |
| 831 | fn ipred_16x16_ver(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8]) { |
| 832 | ipred_ver(buf, stride, top, 16); |
| 833 | } |
| 834 | fn ipred_16x16_hor(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8]) { |
| 835 | ipred_hor(buf, stride, left, 16); |
| 836 | } |
| 837 | fn ipred_16x16_dc(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8]) { |
| 838 | ipred_dc(buf, stride, top, left, 16, 5); |
| 839 | } |
| 840 | fn ipred_16x16_left_dc(buf: &mut [u8], stride: usize, _top: &[u8], left: &[u8]) { |
| 841 | ipred_left_dc(buf, stride, left, 16, 4); |
| 842 | } |
| 843 | fn ipred_16x16_top_dc(buf: &mut [u8], stride: usize, top: &[u8], _left: &[u8]) { |
| 844 | ipred_top_dc(buf, stride, top, 16, 4); |
| 845 | } |
| 846 | fn ipred_16x16_dc128(buf: &mut [u8], stride: usize, _top: &[u8], _left: &[u8]) { |
| 847 | ipred_dc128(buf, stride, 16); |
| 848 | } |
| 849 | fn ipred_16x16_plane(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8]) { |
| 850 | let mut h = 8 * (i32::from(top[15]) - i32::from(left[0])); |
| 851 | let mut v = 8 * (i32::from(left[16]) - i32::from(left[0])); |
| 852 | for k in 0..7 { |
| 853 | h += ((k as i32) + 1) * (i32::from(top[8 + k]) - i32::from(top[6 - k])); |
| 854 | v += ((k as i32) + 1) * (i32::from(left[9 + k]) - i32::from(left[7 - k])); |
| 855 | } |
| 856 | |
| 857 | h = (5 * h + 32) >> 6; |
| 858 | v = (5 * v + 32) >> 6; |
| 859 | |
| 860 | let mut a = 16 * (i32::from(left[16]) + i32::from(top[15]) + 1) - 7 * (v + h); |
| 861 | |
| 862 | for row in buf.chunks_mut(stride).take(16) { |
| 863 | let mut b = a; |
| 864 | a += v; |
| 865 | |
| 866 | for dst in row.chunks_exact_mut(4).take(4) { |
| 867 | dst[0] = clip8(((b ) >> 5) as i16); |
| 868 | dst[1] = clip8(((b + h) >> 5) as i16); |
| 869 | dst[2] = clip8(((b + 2*h) >> 5) as i16); |
| 870 | dst[3] = clip8(((b + 3*h) >> 5) as i16); |
| 871 | b += h * 4; |
| 872 | } |
| 873 | } |
| 874 | } |
| 875 | |
| 876 | pub type IPred4x4Func = fn(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8], tr: &[u8]); |
| 877 | pub type IPred8x8Func = fn(buf: &mut [u8], stride: usize, top: &[u8], left: &[u8]); |
| 878 | pub type IPred8x8LumaFunc = fn(buf: &mut [u8], stride: usize, ctx: &IPred8Context); |
| 879 | |
| 880 | pub const IPRED4_DC128: usize = 11; |
| 881 | pub const IPRED4_DC_TOP: usize = 10; |
| 882 | pub const IPRED4_DC_LEFT: usize = 9; |
| 883 | pub const IPRED8_DC128: usize = 6; |
| 884 | pub const IPRED8_DC_TOP: usize = 5; |
| 885 | pub const IPRED8_DC_LEFT: usize = 4; |
| 886 | |
| 887 | pub const IPRED_FUNCS4X4: [IPred4x4Func; 12] = [ |
| 888 | ipred_4x4_ver, ipred_4x4_hor, ipred_4x4_dc, |
| 889 | ipred_4x4_diag_down_left, ipred_4x4_diag_down_right, |
| 890 | ipred_4x4_ver_right, ipred_4x4_hor_down, ipred_4x4_ver_left, ipred_4x4_hor_up, |
| 891 | ipred_4x4_left_dc, ipred_4x4_top_dc, ipred_4x4_dc128 |
| 892 | ]; |
| 893 | |
| 894 | pub const IPRED_FUNCS8X8_LUMA: [IPred8x8LumaFunc; 12] = [ |
| 895 | ipred_y_8x8_ver, ipred_y_8x8_hor, ipred_y_8x8_dc, |
| 896 | ipred_y_8x8_diag_down_left, ipred_y_8x8_diag_down_right, |
| 897 | ipred_y_8x8_ver_right, ipred_y_8x8_hor_down, |
| 898 | ipred_y_8x8_ver_left, ipred_y_8x8_hor_up, |
| 899 | ipred_y_8x8_left_dc, ipred_y_8x8_top_dc, ipred_y_8x8_dc128 |
| 900 | ]; |
| 901 | |
| 902 | pub const IPRED_FUNCS8X8_CHROMA: [IPred8x8Func; 7] = [ |
| 903 | ipred_8x8_dc, ipred_8x8_hor, ipred_8x8_ver, ipred_8x8_plane, |
| 904 | ipred_8x8_left_dc, ipred_8x8_top_dc, ipred_8x8_dc128 |
| 905 | ]; |
| 906 | |
| 907 | pub const IPRED_FUNCS16X16: [IPred8x8Func; 7] = [ |
| 908 | ipred_16x16_ver, ipred_16x16_hor, ipred_16x16_dc, ipred_16x16_plane, |
| 909 | ipred_16x16_left_dc, ipred_16x16_top_dc, ipred_16x16_dc128 |
| 910 | ]; |
| 911 | |
| 912 | macro_rules! loop_filter { |
| 913 | (lumaedge; $buf: expr, $off: expr, $step: expr, $alpha: expr, $beta: expr) => { |
| 914 | let p2 = i16::from($buf[$off - $step * 3]); |
| 915 | let p1 = i16::from($buf[$off - $step * 2]); |
| 916 | let p0 = i16::from($buf[$off - $step]); |
| 917 | let q0 = i16::from($buf[$off]); |
| 918 | let q1 = i16::from($buf[$off + $step]); |
| 919 | let q2 = i16::from($buf[$off + $step * 2]); |
| 920 | let a_p = (p2 - p0).abs() < $beta; |
| 921 | let a_q = (q2 - q0).abs() < $beta; |
| 922 | if a_p && (p0 - q0).abs() < (($alpha >> 2) + 2) { |
| 923 | let p3 = i16::from($buf[$off - $step * 4]); |
| 924 | $buf[$off - $step * 3] = ((2 * p3 + 3 * p2 + p1 + p0 + q0 + 4) >> 3) as u8; |
| 925 | $buf[$off - $step * 2] = ((p2 + p1 + p0 + q0 + 2) >> 2) as u8; |
| 926 | $buf[$off - $step] = ((p2 + 2 * p1 + 2 * p0 + 2 * q0 + q1 + 4) >> 3) as u8; |
| 927 | } else { |
| 928 | $buf[$off - $step] = ((2 * p1 + p0 + q1 + 2) >> 2) as u8; |
| 929 | } |
| 930 | if a_q && (p0 - q0).abs() < (($alpha >> 2) + 2) { |
| 931 | let q3 = i16::from($buf[$off + $step * 3]); |
| 932 | $buf[$off] = ((p1 + 2 * p0 + 2 * q0 + 2 * q1 + q2 + 4) >> 3) as u8; |
| 933 | $buf[$off + $step] = ((p0 + q0 + q1 + q2 + 2) >> 2) as u8; |
| 934 | $buf[$off + $step * 2] = ((2 * q3 + 3 * q2 + q1 + q0 + p0 + 4) >> 3) as u8; |
| 935 | } else { |
| 936 | $buf[$off] = ((2 * q1 + q0 + p1 + 2) >> 2) as u8; |
| 937 | } |
| 938 | }; |
| 939 | (chromaedge; $buf: expr, $off: expr, $step: expr) => { |
| 940 | let p1 = i16::from($buf[$off - $step * 2]); |
| 941 | let p0 = i16::from($buf[$off - $step]); |
| 942 | let q0 = i16::from($buf[$off]); |
| 943 | let q1 = i16::from($buf[$off + $step]); |
| 944 | $buf[$off - $step] = ((2 * p1 + p0 + q1 + 2) >> 2) as u8; |
| 945 | $buf[$off] = ((2 * q1 + q0 + p1 + 2) >> 2) as u8; |
| 946 | }; |
| 947 | (lumanormal; $buf: expr, $off: expr, $step: expr, $tc0: expr, $beta: expr) => { |
| 948 | let p2 = i16::from($buf[$off - $step * 3]); |
| 949 | let p1 = i16::from($buf[$off - $step * 2]); |
| 950 | let p0 = i16::from($buf[$off - $step]); |
| 951 | let q0 = i16::from($buf[$off]); |
| 952 | let q1 = i16::from($buf[$off + $step]); |
| 953 | let q2 = i16::from($buf[$off + $step * 2]); |
| 954 | let a_p = (p2 - p0).abs() < $beta; |
| 955 | let a_q = (q2 - q0).abs() < $beta; |
| 956 | let tc = $tc0 + (a_p as i16) + (a_q as i16); |
| 957 | let delta = (((q0 - p0) * 4 + (p1 - q1) + 4) >> 3).max(-tc).min(tc); |
| 958 | if a_p && ($tc0 > 0) { |
| 959 | $buf[$off - $step * 2] = clip8(p1 + ((p2 + ((p0 + q0 + 1) >> 1) - p1 * 2) >> 1).max(-$tc0).min($tc0)); |
| 960 | } |
| 961 | $buf[$off - $step] = clip8(p0 + delta); |
| 962 | $buf[$off] = clip8(q0 - delta); |
| 963 | if a_q && ($tc0 > 0) { |
| 964 | $buf[$off + $step] = clip8(q1 + ((q2 + ((p0 + q0 + 1) >> 1) - q1 * 2) >> 1).max(-$tc0).min($tc0)); |
| 965 | } |
| 966 | }; |
| 967 | (chromanormal; $buf: expr, $off: expr, $step: expr, $tc0: expr) => { |
| 968 | let p1 = i16::from($buf[$off - $step * 2]); |
| 969 | let p0 = i16::from($buf[$off - $step]); |
| 970 | let q0 = i16::from($buf[$off]); |
| 971 | let q1 = i16::from($buf[$off + $step]); |
| 972 | let tc = $tc0 + 1; |
| 973 | let delta = (((q0 - p0) * 4 + (p1 - q1) + 4) >> 3).max(-tc).min(tc); |
| 974 | $buf[$off - $step] = clip8(p0 + delta); |
| 975 | $buf[$off] = clip8(q0 - delta); |
| 976 | } |
| 977 | } |
| 978 | |
| 979 | fn check_filter(buf: &[u8], off: usize, step: usize, alpha: i16, beta: i16) -> bool { |
| 980 | let p1 = i16::from(buf[off - step * 2]); |
| 981 | let p0 = i16::from(buf[off - step]); |
| 982 | let q0 = i16::from(buf[off]); |
| 983 | let q1 = i16::from(buf[off + step]); |
| 984 | (p0 - q0).abs() < alpha && (p1 - p0).abs() < beta && (q1 - q0).abs() < beta |
| 985 | } |
| 986 | |
| 987 | #[cfg(not(target_arch="x86_64"))] |
| 988 | fn check_filter4(buf: &[u8], mut off: usize, step: usize, stride: usize, alpha: i16, beta: i16) -> [bool; 4] { |
| 989 | let mut flags = [false; 4]; |
| 990 | for flag in flags.iter_mut() { |
| 991 | let p1 = i16::from(buf[off - step * 2]); |
| 992 | let p0 = i16::from(buf[off - step]); |
| 993 | let q0 = i16::from(buf[off]); |
| 994 | let q1 = i16::from(buf[off + step]); |
| 995 | *flag = (p0 - q0).abs() < alpha && (p1 - p0).abs() < beta && (q1 - q0).abs() < beta; |
| 996 | off += stride; |
| 997 | } |
| 998 | flags |
| 999 | } |
| 1000 | |
| 1001 | #[cfg(target_arch="x86_64")] |
| 1002 | fn check_filter4(buf: &[u8], off: usize, step: usize, stride: usize, alpha: i16, beta: i16) -> [bool; 4] { |
| 1003 | unsafe { |
| 1004 | let mut flags = [false; 4]; |
| 1005 | let src = buf[off - step * 2..].as_ptr(); |
| 1006 | let load_stride = step.max(stride); |
| 1007 | let fptr = flags.as_mut_ptr(); |
| 1008 | let tflag = u32::from(step == 1); |
| 1009 | asm! { |
| 1010 | // load block |
| 1011 | "pxor xmm4, xmm4", |
| 1012 | "movd xmm0, dword ptr [{src}]", |
| 1013 | "lea {tmp}, [{src} + {stride} * 2]", |
| 1014 | "movd xmm1, dword ptr [{src} + {stride}]", |
| 1015 | "movd xmm2, dword ptr [{tmp}]", |
| 1016 | "movd xmm3, dword ptr [{tmp} + {stride}]", |
| 1017 | "punpcklbw xmm0, xmm4", |
| 1018 | "punpcklbw xmm1, xmm4", |
| 1019 | "punpcklbw xmm2, xmm4", |
| 1020 | "punpcklbw xmm3, xmm4", |
| 1021 | |
| 1022 | // transpose block if necessary so it's always processed by rows |
| 1023 | "test {tflag:e}, {tflag:e}", |
| 1024 | "jz 1f", |
| 1025 | "punpcklwd xmm0, xmm1", |
| 1026 | "movhlps xmm4, xmm0", |
| 1027 | "punpcklwd xmm2, xmm3", |
| 1028 | "movhlps xmm1, xmm2", |
| 1029 | "punpckldq xmm0, xmm2", |
| 1030 | "punpckldq xmm4, xmm1", |
| 1031 | "movhlps xmm1, xmm0", |
| 1032 | "movhlps xmm3, xmm4", |
| 1033 | "movaps xmm2, xmm4", |
| 1034 | "1:", |
| 1035 | |
| 1036 | // calculate deltas and flags |
| 1037 | "movd xmm4, {alpha:r}", |
| 1038 | "movd xmm5, {beta:r}", |
| 1039 | "psubw xmm0, xmm1", |
| 1040 | "psubw xmm1, xmm2", |
| 1041 | "psubw xmm3, xmm2", |
| 1042 | "pshuflw xmm4, xmm4, 0", |
| 1043 | "pshuflw xmm5, xmm5, 0", |
| 1044 | "pabsw xmm0, xmm0", // |p1 - p0| |
| 1045 | "pabsw xmm1, xmm1", // |p0 - q0| |
| 1046 | "pabsw xmm2, xmm3", // |q1 - q0| |
| 1047 | "movaps xmm3, xmm5", |
| 1048 | "pcmpgtw xmm4, xmm1", |
| 1049 | "pcmpgtw xmm5, xmm0", |
| 1050 | "pcmpgtw xmm3, xmm2", |
| 1051 | "pand xmm4, xmm5", |
| 1052 | "pand xmm4, xmm3", |
| 1053 | "packsswb xmm4, xmm4", |
| 1054 | "movd [{flags}], xmm4", |
| 1055 | tmp = out(reg) _, |
| 1056 | src = in(reg) src, |
| 1057 | stride = in(reg) load_stride, |
| 1058 | alpha = in(reg) alpha, |
| 1059 | beta = in(reg) beta, |
| 1060 | flags = in(reg) fptr, |
| 1061 | tflag = in(reg) tflag, |
| 1062 | out("xmm0") _, |
| 1063 | out("xmm1") _, |
| 1064 | out("xmm2") _, |
| 1065 | out("xmm3") _, |
| 1066 | out("xmm4") _, |
| 1067 | out("xmm5") _, |
| 1068 | } |
| 1069 | flags |
| 1070 | } |
| 1071 | } |
| 1072 | |
| 1073 | pub fn loop_filter_lumaedge_v(dst: &mut [u8], mut off: usize, stride: usize, alpha: i16, beta: i16) { |
| 1074 | let flags = check_filter4(dst, off, 1, stride, alpha, beta); |
| 1075 | for &flag in flags.iter() { |
| 1076 | if flag { |
| 1077 | loop_filter!(lumaedge; dst, off, 1, alpha, beta); |
| 1078 | } |
| 1079 | off += stride; |
| 1080 | } |
| 1081 | } |
| 1082 | pub fn loop_filter_lumaedge_h(dst: &mut [u8], off: usize, stride: usize, alpha: i16, beta: i16) { |
| 1083 | let flags = check_filter4(dst, off, stride, 1, alpha, beta); |
| 1084 | for (x, &flag) in flags.iter().enumerate() { |
| 1085 | if flag { |
| 1086 | loop_filter!(lumaedge; dst, off + x, stride, alpha, beta); |
| 1087 | } |
| 1088 | } |
| 1089 | } |
| 1090 | pub fn loop_filter_lumanormal_v(dst: &mut [u8], mut off: usize, stride: usize, alpha: i16, beta: i16, tc0: i16) { |
| 1091 | let flags = check_filter4(dst, off, 1, stride, alpha, beta); |
| 1092 | for &flag in flags.iter() { |
| 1093 | if flag { |
| 1094 | loop_filter!(lumanormal; dst, off, 1, tc0, beta); |
| 1095 | } |
| 1096 | off += stride; |
| 1097 | } |
| 1098 | } |
| 1099 | pub fn loop_filter_lumanormal_h(dst: &mut [u8], off: usize, stride: usize, alpha: i16, beta: i16, tc0: i16) { |
| 1100 | let flags = check_filter4(dst, off, stride, 1, alpha, beta); |
| 1101 | for (x, &flag) in flags.iter().enumerate() { |
| 1102 | if flag { |
| 1103 | loop_filter!(lumanormal; dst, off + x, stride, tc0, beta); |
| 1104 | } |
| 1105 | } |
| 1106 | } |
| 1107 | pub fn loop_filter_chromaedge_v(dst: &mut [u8], mut off: usize, stride: usize, alpha: i16, beta: i16) { |
| 1108 | for _ in 0..2 { |
| 1109 | if check_filter(dst, off, 1, alpha, beta) { |
| 1110 | loop_filter!(chromaedge; dst, off, 1); |
| 1111 | } |
| 1112 | off += stride; |
| 1113 | } |
| 1114 | } |
| 1115 | pub fn loop_filter_chromaedge_h(dst: &mut [u8], off: usize, stride: usize, alpha: i16, beta: i16) { |
| 1116 | for x in 0..2 { |
| 1117 | if check_filter(dst, off + x, stride, alpha, beta) { |
| 1118 | loop_filter!(chromaedge; dst, off + x, stride); |
| 1119 | } |
| 1120 | } |
| 1121 | } |
| 1122 | pub fn loop_filter_chromanormal_v(dst: &mut [u8], mut off: usize, stride: usize, alpha: i16, beta: i16, tc0: i16) { |
| 1123 | for _ in 0..2 { |
| 1124 | if check_filter(dst, off, 1, alpha, beta) { |
| 1125 | loop_filter!(chromanormal; dst, off, 1, tc0); |
| 1126 | } |
| 1127 | off += stride; |
| 1128 | } |
| 1129 | } |
| 1130 | pub fn loop_filter_chromanormal_h(dst: &mut [u8], off: usize, stride: usize, alpha: i16, beta: i16, tc0: i16) { |
| 1131 | for x in 0..2 { |
| 1132 | if check_filter(dst, off + x, stride, alpha, beta) { |
| 1133 | loop_filter!(chromanormal; dst, off + x, stride, tc0); |
| 1134 | } |
| 1135 | } |
| 1136 | } |