nihav-realmedia/src/codecs/rv30dsp.rs

   1 use nihav_core::frame::{FrameType, NAVideoBuffer};
   2 use nihav_codec_support::codecs::MV;
   3 use nihav_codec_support::codecs::blockdsp::edge_emu;
   4 use super::rv3040::{RV34DSP, RV34MBInfo};
   5
   6 fn clip8(a: i16) -> u8 {
   7     if a < 0 { 0 }
   8     else if a > 255 { 255 }
   9     else { a as u8 }
  10 }
  11
  12 fn rv3_filter_h(dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize, bsize: usize, c1: i16, c2: i16) {
  13     for _ in 0..bsize {
  14         for x in 0..bsize {
  15             dst[didx + x] = clip8((-((src[sidx + x - 1] as i16) + (src[sidx + x + 2] as i16)) + (src[sidx + x + 0] as i16) * c1 + (src[sidx + x + 1] as i16) * c2 + 8) >> 4);
  16         }
  17         sidx += sstride;
  18         didx += dstride;
  19     }
  20 }
  21
  22 fn rv3_filter_v(dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize, bsize: usize, c1: i16, c2: i16) {
  23     for _ in 0..bsize {
  24         for x in 0..bsize {
  25             dst[didx + x] = clip8((-((src[sidx + x - 1 * sstride] as i16) + (src[sidx + x + 2 * sstride] as i16)) + (src[sidx + x + 0 * sstride] as i16) * c1 + (src[sidx + x + 1 * sstride] as i16) * c2 + 8) >> 4);
  26         }
  27         sidx += sstride;
  28         didx += dstride;
  29     }
  30 }
  31
  32 macro_rules! mc_matrix {
  33     ($s: ident, $o: expr, $c1: expr) => (
  34             ($c1 * 6) * ($s[$o] as i32) + ($c1 * 9) * ($s[$o + 1] as i32) + ($c1) * ($s[$o + 2] as i32)
  35         );
  36     ($s: ident, $o: expr, $c1: expr, $d1: expr, $d2: expr) => (
  37             -($c1) * ($s[$o - 1] as i32) + ($c1 * $d1) * ($s[$o] as i32) + ($c1 * $d2) * ($s[$o + 1] as i32) + -($c1) * ($s[$o + 2] as i32)
  38         );
  39     ($s: ident, $o: expr, $ss: expr, $c1: expr, $c2: expr, $d1: expr, $d2: expr) => (
  40         ((mc_matrix!($s, $o -     $ss,  -1, $d1, $d2) +
  41           mc_matrix!($s, $o          , $c1, $d1, $d2) +
  42           mc_matrix!($s, $o +     $ss, $c2, $d1, $d2) +
  43           mc_matrix!($s, $o + 2 * $ss,  -1, $d1, $d2) + 128) >> 8) as i16
  44         );
  45     (m22; $s: ident, $o: expr, $ss: expr) => (
  46         ((mc_matrix!($s, $o + 0 * $ss, 6) +
  47           mc_matrix!($s, $o + 1 * $ss, 9) +
  48           mc_matrix!($s, $o + 2 * $ss, 1) + 128) >> 8) as i16
  49         );
  50 }
  51
  52 macro_rules! mc_func {
  53     (copy; $name: ident, $size: expr) => (
  54         fn $name (dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize) {
  55             for _ in 0..$size {
  56                 let d = &mut dst[didx..][..$size];
  57                 let s = &src[sidx..][..$size];
  58                 for x in 0..$size { d[x] = s[x]; }
  59                 didx += dstride;
  60                 sidx += sstride;
  61             }
  62         }
  63         );
  64     (hor; $name: ident, $c1: expr, $c2: expr, $size: expr) => (
  65         fn $name (dst: &mut [u8], didx: usize, dstride: usize, src: &[u8], sidx: usize, sstride: usize) {
  66             rv3_filter_h(dst, didx, dstride, src, sidx, sstride, $size, $c1, $c2);
  67         }
  68         );
  69     (ver; $name: ident, $c1: expr, $c2: expr, $size: expr) => (
  70         fn $name (dst: &mut [u8], didx: usize, dstride: usize, src: &[u8], sidx: usize, sstride: usize) {
  71             rv3_filter_v(dst, didx, dstride, src, sidx, sstride, $size, $c1, $c2);
  72         }
  73         );
  74     (m11; $name: ident, $size: expr) => (
  75         fn $name (dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize) {
  76             for _ in 0..$size {
  77                 for x in 0..$size {
  78                     dst[didx + x] = clip8(mc_matrix!(src, sidx + x, sstride, 12, 6, 12, 6));
  79                 }
  80                 didx += dstride;
  81                 sidx += sstride;
  82             }
  83         }
  84         );
  85     (m12; $name: ident, $size: expr) => (
  86         fn $name (dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize) {
  87             for _ in 0..$size {
  88                 for x in 0..$size {
  89                     dst[didx + x] = clip8(mc_matrix!(src, sidx + x, sstride, 6, 12, 12, 6));
  90                 }
  91                 didx += dstride;
  92                 sidx += sstride;
  93             }
  94         }
  95         );
  96     (m21; $name: ident, $size: expr) => (
  97         fn $name (dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize) {
  98             for _ in 0..$size {
  99                 for x in 0..$size {
 100                     dst[didx + x] = clip8(mc_matrix!(src, sidx + x, sstride, 12, 6, 6, 12));
 101                 }
 102                 didx += dstride;
 103                 sidx += sstride;
 104             }
 105         }
 106         );
 107     (m22; $name: ident, $size: expr) => (
 108         fn $name (dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize) {
 109             for _ in 0..$size {
 110                 for x in 0..$size {
 111                     dst[didx + x] = clip8(mc_matrix!(m22; src, sidx + x, sstride));
 112                 }
 113                 didx += dstride;
 114                 sidx += sstride;
 115             }
 116         }
 117         );
 118 }
 119 mc_func!(copy; copy_16, 16);
 120 mc_func!(copy; copy_8,   8);
 121 mc_func!(hor;  luma_mc_10_16, 12, 6, 16);
 122 mc_func!(hor;  luma_mc_20_16, 6, 12, 16);
 123 mc_func!(hor;  luma_mc_10_8,  12, 6,  8);
 124 mc_func!(hor;  luma_mc_20_8,  6, 12,  8);
 125 mc_func!(ver;  luma_mc_01_16, 12, 6, 16);
 126 mc_func!(ver;  luma_mc_02_16, 6, 12, 16);
 127 mc_func!(ver;  luma_mc_01_8,  12, 6,  8);
 128 mc_func!(ver;  luma_mc_02_8,  6, 12,  8);
 129 mc_func!(m11;  luma_mc_11_16,        16);
 130 mc_func!(m11;  luma_mc_11_8,          8);
 131 mc_func!(m21;  luma_mc_21_16,        16);
 132 mc_func!(m21;  luma_mc_21_8,          8);
 133 mc_func!(m12;  luma_mc_12_16,        16);
 134 mc_func!(m12;  luma_mc_12_8,          8);
 135 mc_func!(m22;  luma_mc_22_16,        16);
 136 mc_func!(m22;  luma_mc_22_8,          8);
 137
 138 const RV30_CHROMA_FRAC1: [u16; 3] = [ 8, 5, 3 ];
 139 const RV30_CHROMA_FRAC2: [u16; 3] = [ 0, 3, 5 ];
 140 fn rv30_chroma_mc(dst: &mut [u8], mut didx: usize, dstride: usize, src: &[u8], mut sidx: usize, sstride: usize, size: usize, x: usize, y: usize) {
 141     if (x == 0) && (y == 0) {
 142         for _ in 0..size {
 143             for x in 0..size { dst[didx + x] = src[sidx + x]; }
 144             didx += dstride;
 145             sidx += sstride;
 146         }
 147         return;
 148     }
 149     let a = RV30_CHROMA_FRAC1[x] * RV30_CHROMA_FRAC1[y];
 150     let b = RV30_CHROMA_FRAC2[x] * RV30_CHROMA_FRAC1[y];
 151     let c = RV30_CHROMA_FRAC1[x] * RV30_CHROMA_FRAC2[y];
 152     let d = RV30_CHROMA_FRAC2[x] * RV30_CHROMA_FRAC2[y];
 153     for _ in 0..size {
 154         for x in 0..size {
 155             dst[didx + x] = ((a * (src[sidx + x] as u16)
 156                             + b * (src[sidx + x + 1] as u16)
 157                             + c * (src[sidx + x + sstride] as u16)
 158                             + d * (src[sidx + x + 1 + sstride] as u16) + 32) >> 6) as u8;
 159         }
 160         didx += dstride;
 161         sidx += sstride;
 162     }
 163 }
 164
 165 pub struct RV30DSP {
 166     luma_mc: [[fn (&mut [u8], usize, usize, &[u8], usize, usize); 9]; 2],
 167 }
 168
 169 impl RV30DSP {
 170     pub fn new() -> Self {
 171         RV30DSP {
 172             luma_mc: [
 173                     [ copy_16,       luma_mc_10_16,  luma_mc_20_16,
 174                       luma_mc_01_16, luma_mc_11_16,  luma_mc_21_16,
 175                       luma_mc_02_16, luma_mc_12_16,  luma_mc_22_16 ],
 176                     [ copy_8,        luma_mc_10_8,   luma_mc_20_8,
 177                       luma_mc_01_8,  luma_mc_11_8,   luma_mc_21_8,
 178                       luma_mc_02_8,  luma_mc_12_8,   luma_mc_22_8  ] ],
 179         }
 180     }
 181 }
 182
 183 macro_rules! el {
 184     ($src: ident, $o: expr) => ($src[$o] as i16);
 185 }
 186
 187 fn clip_symm(a: i16, lim: i16) -> i16 {
 188     if a < -lim {
 189         -lim
 190     } else if a > lim {
 191         lim
 192     } else {
 193         a
 194     }
 195 }
 196
 197 const RV30_LOOP_FILTER_STRENGTH: [i16; 32] = [
 198     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
 199     2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5
 200 ];
 201
 202 macro_rules! test_bit {
 203     ($pat: expr, $x: expr) => ( (($pat >> $x) & 1) != 0 )
 204 }
 205
 206 fn rv30_loop_filter4(pix: &mut [u8], mut off: usize, step: usize, stride: usize, lim: i16) {
 207     for _ in 0..4 {
 208         let a = el!(pix, off - 2*step);
 209         let b = el!(pix, off -   step);
 210         let c = el!(pix, off);
 211         let d = el!(pix, off +   step);
 212         let diff0 = ((a - d) - (b - c) * 4) >> 3;
 213         let diff = clip_symm(diff0, lim);
 214         pix[off - step] = clip8(b + diff);
 215         pix[off       ] = clip8(c - diff);
 216         off += stride;
 217     }
 218 }
 219
 220 fn rv30_div_mv(mv: i16) -> (i16, usize) {
 221     let i = mv / 3;
 222     let f = mv - i * 3;
 223     if f < 0 {
 224         (i - 1, (f + 3) as usize)
 225     } else {
 226         (i, f as usize)
 227     }
 228 }
 229
 230 fn check_pos(x: usize, y: usize, size: usize, w: usize, h: usize, dx: i16, dy: i16, e0: isize, e1: isize, e2: isize, e3: isize) -> bool {
 231     let xn = (x as isize) + (dx as isize);
 232     let yn = (y as isize) + (dy as isize);
 233
 234     (xn - e0 >= 0) && (xn + (size as isize) + e1 <= (w as isize)) && (yn - e2 >= 0) && (yn + (size as isize) + e3 <= (h as isize))
 235 }
 236
 237 const RV30_EDGE1: [isize; 3] = [ 0, 1, 1 ];
 238 const RV30_EDGE2: [isize; 3] = [ 0, 2, 2 ];
 239
 240 impl RV34DSP for RV30DSP {
 241     fn loop_filter(&self, frame: &mut NAVideoBuffer<u8>, _ftype: FrameType, mbinfo: &[RV34MBInfo], mb_w: usize, _mb_h: usize, row: usize) {
 242         let mut offs:   [usize; 3] = [0; 3];
 243         let mut stride: [usize; 3] = [0; 3];
 244
 245         for comp in 0..3 {
 246             stride[comp] = frame.get_stride(comp);
 247             let start = if comp == 0 { row * 16 } else { row * 8 };
 248             offs[comp] = frame.get_offset(comp) + start * stride[comp];
 249         }
 250
 251         let data = frame.get_data_mut().unwrap();
 252         let dst: &mut [u8] = data.as_mut_slice();
 253
 254         // vertical filter
 255         let mut left_cbp = 0;
 256         let mut left_lim = 0;
 257         let mut left_dbk = 0;
 258         let mut mb_pos: usize = row * mb_w;
 259         for mb_x in 0..mb_w {
 260             let cur_lim = RV30_LOOP_FILTER_STRENGTH[mbinfo[mb_pos].q as usize];
 261             let cur_dbk = mbinfo[mb_pos].deblock;
 262             let cur_cbp = mbinfo[mb_pos].cbp_c;
 263             let xstart = if mb_x == 0 { 1 } else { 0 };
 264             for y in 0..4 {
 265                 let yoff = offs[0] + mb_x * 16 + y * 4 * stride[0];
 266                 for x in xstart..4 {
 267                     let cs = x + y*4;
 268                     let loc_lim;
 269
 270                     if test_bit!(cur_dbk, cs) {
 271                         loc_lim = cur_lim;
 272                     } else if (x == 0) && test_bit!(left_dbk, cs + 3) {
 273                         loc_lim = left_lim;
 274                     } else if (x != 0) && test_bit!(cur_dbk,  cs - 1) {
 275                         loc_lim = cur_lim;
 276                     } else {
 277                         loc_lim = 0;
 278                     }
 279                     if loc_lim != 0 {
 280                         rv30_loop_filter4(dst, yoff + x * 4, 1, stride[0], loc_lim);
 281                     }
 282                 }
 283             }
 284
 285             for comp in 1..3 {
 286                 for y in 0..2 {
 287                     let coff = offs[comp] + mb_x * 8 + y * 4 * stride[comp];
 288                     for x in xstart..2 {
 289                         let cs = x + y * 2 + (comp - 1) * 4;
 290                         let loc_lim;
 291
 292                         if test_bit!(cur_cbp, cs) {
 293                             loc_lim = cur_lim;
 294                         } else if (x == 0) && test_bit!(left_cbp, cs + 1) {
 295                             loc_lim = left_lim;
 296                         } else if (x != 0) && test_bit!(cur_cbp,  cs - 1) {
 297                             loc_lim = cur_lim;
 298                         } else {
 299                             loc_lim = 0;
 300                         }
 301                         if loc_lim != 0 {
 302                             rv30_loop_filter4(dst, coff + x * 4, 1, stride[comp], loc_lim);
 303                         }
 304                     }
 305                 }
 306             }
 307
 308             left_lim = cur_lim;
 309             left_dbk = cur_dbk;
 310             left_cbp = cur_cbp;
 311             mb_pos += 1;
 312         }
 313
 314         // horizontal filter
 315         let mut mb_pos: usize = row * mb_w;
 316         for mb_x in 0..mb_w {
 317             let cur_lim = RV30_LOOP_FILTER_STRENGTH[mbinfo[mb_pos].q as usize];
 318             let cur_dbk = mbinfo[mb_pos].deblock;
 319             let cur_cbp = mbinfo[mb_pos].cbp_c;
 320             let ystart = if row == 0 { 1 } else { 0 };
 321             let top_lim;
 322             let top_dbk;
 323             let top_cbp;
 324             if row > 0 {
 325                 top_lim = RV30_LOOP_FILTER_STRENGTH[mbinfo[mb_pos - mb_w].q as usize];
 326                 top_dbk = mbinfo[mb_pos - mb_w].deblock;
 327                 top_cbp = mbinfo[mb_pos - mb_w].cbp_c;
 328             } else {
 329                 top_lim = 0;
 330                 top_dbk = 0;
 331                 top_cbp = 0;
 332             }
 333             for y in ystart..4 {
 334                 let yoff = offs[0] + mb_x * 16 + y * 4 * stride[0];
 335                 for x in 0..4 {
 336                     let cs = x + y*4;
 337                     let loc_lim;
 338
 339                     if test_bit!(cur_dbk, cs) {
 340                         loc_lim = cur_lim;
 341                     } else if (y == 0) && test_bit!(top_dbk, cs + 12) {
 342                         loc_lim = top_lim;
 343                     } else if (y != 0) && test_bit!(cur_dbk, cs - 4) {
 344                         loc_lim = cur_lim;
 345                     } else {
 346                         loc_lim = 0;
 347                     }
 348                     if loc_lim != 0 {
 349                         rv30_loop_filter4(dst, yoff + x * 4, stride[0], 1, loc_lim);
 350                     }
 351                 }
 352             }
 353
 354             for comp in 1..3 {
 355                 for y in ystart..2 {
 356                     let coff = offs[comp] + mb_x * 8 + y * 4 * stride[comp];
 357                     for x in 0..2 {
 358                         let cs = x + y * 2 + (comp - 1) * 4;
 359                         let loc_lim;
 360
 361                         if test_bit!(cur_cbp, cs) {
 362                             loc_lim = cur_lim;
 363                         } else if (y == 0) && test_bit!(top_cbp, cs + 2) {
 364                             loc_lim = top_lim;
 365                         } else if (y != 0) && test_bit!(cur_cbp, cs - 2) {
 366                             loc_lim = cur_lim;
 367                         } else {
 368                             loc_lim = 0;
 369                         }
 370                         if loc_lim != 0 {
 371                             rv30_loop_filter4(dst, coff + x * 4, stride[comp], 1, loc_lim);
 372                         }
 373                     }
 374                 }
 375             }
 376
 377             mb_pos += 1;
 378         }
 379     }
 380     fn do_luma_mc(&self, frame: &mut NAVideoBuffer<u8>, prev_frame: &NAVideoBuffer<u8>, x: usize, y: usize, mv: MV, use16: bool, avg: bool) {
 381         let size: usize = if use16 { 16 } else { 8 };
 382         let dstride = frame.get_stride(0);
 383         let doffset = frame.get_offset(0) + (if !avg { x + y * dstride } else { 0 });
 384         let data = frame.get_data_mut().unwrap();
 385         let dst: &mut [u8] = data.as_mut_slice();
 386
 387         let (w_, h_) = prev_frame.get_dimensions(0);
 388         let w = (w_ + 15) & !15;
 389         let h = (h_ + 15) & !15;
 390
 391         let (dx, cx) = rv30_div_mv(mv.x);
 392         let (dy, cy) = rv30_div_mv(mv.y);
 393         let mode = cx + cy * 3;
 394
 395         if check_pos(x, y, size, w, h, dx, dy, RV30_EDGE1[cx], RV30_EDGE2[cx], RV30_EDGE1[cy], RV30_EDGE2[cy]) {
 396             let sstride = prev_frame.get_stride(0);
 397             let mut soffset = prev_frame.get_offset(0) + x + y * sstride;
 398             let data = prev_frame.get_data();
 399             let src: &[u8] = data.as_slice();
 400             soffset = ((soffset as isize) + (dx as isize) + (dy as isize) * (sstride as isize)) as usize;
 401             self.luma_mc[if use16 { 0 } else { 1 }][mode](dst, doffset, dstride, src, soffset, sstride);
 402         } else {
 403             let mut ebuf: [u8; 32*20] = [0; 32*20];
 404             edge_emu(prev_frame, (x as isize) + (dx as isize) - 1, (y as isize) + (dy as isize) - 1, 16+3, 16+3, &mut ebuf, 32, 0, 4);
 405             self.luma_mc[if use16 { 0 } else { 1 }][mode](dst, doffset, dstride, &ebuf, 32 + 1, 32);
 406         }
 407     }
 408     fn do_chroma_mc(&self, frame: &mut NAVideoBuffer<u8>, prev_frame: &NAVideoBuffer<u8>, x: usize, y: usize, comp: usize, mv: MV, use8: bool, avg: bool) {
 409         let size: usize = if use8 { 8 } else { 4 };
 410         let dstride = frame.get_stride(comp);
 411         let doffset = frame.get_offset(comp) + (if !avg { x + y * dstride } else { 0 });
 412         let data = frame.get_data_mut().unwrap();
 413         let dst: &mut [u8] = data.as_mut_slice();
 414
 415         let (w_, h_) = prev_frame.get_dimensions(comp);
 416         let w = (w_ + 7) & !7;
 417         let h = (h_ + 7) & !7;
 418
 419         let (dx, cx) = rv30_div_mv(mv.x / 2);
 420         let (dy, cy) = rv30_div_mv(mv.y / 2);
 421
 422         if check_pos(x, y, size, w, h, dx, dy, 0, 1, 0, 1) {
 423             let sstride = prev_frame.get_stride(comp);
 424             let mut soffset = prev_frame.get_offset(comp) + x + y * sstride;
 425             let data = prev_frame.get_data();
 426             let src: &[u8] = data.as_slice();
 427             soffset = ((soffset as isize) + (dx as isize) + (dy as isize) * (sstride as isize)) as usize;
 428             rv30_chroma_mc(dst, doffset, dstride, src, soffset, sstride, size, cx, cy);
 429         } else {
 430             let mut ebuf: [u8; 16*10] = [0; 16*10];
 431             edge_emu(prev_frame, (x as isize) + (dx as isize), (y as isize) + (dy as isize), 8+1, 8+1, &mut ebuf, 16, comp, 4);
 432             rv30_chroma_mc(dst, doffset, dstride, &ebuf, 0, 16, size, cx, cy);
 433         }
 434     }
 435 }