| 1 | use nihav_core::frame::*; |
| 2 | use nihav_core::codecs::*; |
| 3 | use nihav_core::io::byteio::read_u16le; |
| 4 | use nihav_core::io::bitreader::*; |
| 5 | use nihav_core::io::codebook::*; |
| 6 | use nihav_core::io::intcode::*; |
| 7 | use nihav_codec_support::dsp::fft::*; |
| 8 | use nihav_codec_support::dsp::mdct::IMDCT; |
| 9 | |
| 10 | use std::str::FromStr; |
| 11 | |
| 12 | use super::on2avcdata::*; |
| 13 | |
| 14 | const COEFFS: usize = 1024; |
| 15 | const MAX_BANDS: usize = 14 * 8; |
| 16 | const QTAB_SIZE: usize = 10000; |
| 17 | const SHORT_WIN_POINT0: usize = 512 - 64; |
| 18 | const SHORT_WIN_POINT1: usize = 512 + 64; |
| 19 | |
| 20 | struct WinInfo { |
| 21 | is_long: bool, |
| 22 | bands: &'static [usize], |
| 23 | } |
| 24 | |
| 25 | impl WinInfo { |
| 26 | fn get_tot_bands(&self) -> usize { |
| 27 | if self.is_long { |
| 28 | self.bands.len() |
| 29 | } else { |
| 30 | self.bands.len() * 8 |
| 31 | } |
| 32 | } |
| 33 | } |
| 34 | |
| 35 | struct Codebooks { |
| 36 | scale_cb: Codebook<i8>, |
| 37 | spec_cb: [Codebook<u16>; 15], |
| 38 | } |
| 39 | |
| 40 | fn scale_map(idx: usize) -> i8 { (idx as i8) - 60 } |
| 41 | fn cb_map1(idx: usize) -> u16 { AVC_SPEC_CB1_SYMS[idx] } |
| 42 | fn cb_map2(idx: usize) -> u16 { AVC_SPEC_CB2_SYMS[idx] } |
| 43 | fn cb_map3(idx: usize) -> u16 { AVC_SPEC_CB3_SYMS[idx] } |
| 44 | fn cb_map4(idx: usize) -> u16 { AVC_SPEC_CB4_SYMS[idx] } |
| 45 | fn cb_map5(idx: usize) -> u16 { AVC_SPEC_CB5_SYMS[idx] } |
| 46 | fn cb_map6(idx: usize) -> u16 { AVC_SPEC_CB6_SYMS[idx] } |
| 47 | fn cb_map7(idx: usize) -> u16 { AVC_SPEC_CB7_SYMS[idx] } |
| 48 | fn cb_map8(idx: usize) -> u16 { AVC_SPEC_CB8_SYMS[idx] } |
| 49 | fn cb_map9(idx: usize) -> u16 { AVC_SPEC_CB9_SYMS[idx] } |
| 50 | fn cb_map10(idx: usize) -> u16 { AVC_SPEC_CB10_SYMS[idx] } |
| 51 | fn cb_map11(idx: usize) -> u16 { AVC_SPEC_CB11_SYMS[idx] } |
| 52 | fn cb_map12(idx: usize) -> u16 { AVC_SPEC_CB12_SYMS[idx] } |
| 53 | fn cb_map13(idx: usize) -> u16 { AVC_SPEC_CB13_SYMS[idx] } |
| 54 | fn cb_map14(idx: usize) -> u16 { AVC_SPEC_CB14_SYMS[idx] } |
| 55 | fn cb_map15(idx: usize) -> u16 { AVC_SPEC_CB15_SYMS[idx] } |
| 56 | |
| 57 | impl Codebooks { |
| 58 | fn new() -> Self { |
| 59 | let mut coderead = TableCodebookDescReader::new(AVC_SCF_CODEBOOK_CODES, AVC_SCF_CODEBOOK_BITS, scale_map); |
| 60 | let scale_cb = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 61 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB1_CODES, AVC_SPEC_CB1_BITS, cb_map1); |
| 62 | let cb1 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 63 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB2_CODES, AVC_SPEC_CB2_BITS, cb_map2); |
| 64 | let cb2 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 65 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB3_CODES, AVC_SPEC_CB3_BITS, cb_map3); |
| 66 | let cb3 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 67 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB4_CODES, AVC_SPEC_CB4_BITS, cb_map4); |
| 68 | let cb4 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 69 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB5_CODES, AVC_SPEC_CB5_BITS, cb_map5); |
| 70 | let cb5 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 71 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB6_CODES, AVC_SPEC_CB6_BITS, cb_map6); |
| 72 | let cb6 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 73 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB7_CODES, AVC_SPEC_CB7_BITS, cb_map7); |
| 74 | let cb7 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 75 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB8_CODES, AVC_SPEC_CB8_BITS, cb_map8); |
| 76 | let cb8 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 77 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB9_CODES, AVC_SPEC_CB9_BITS, cb_map9); |
| 78 | let cb9 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 79 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB10_CODES, AVC_SPEC_CB10_BITS, cb_map10); |
| 80 | let cb10 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 81 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB11_CODES, AVC_SPEC_CB11_BITS, cb_map11); |
| 82 | let cb11 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 83 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB12_CODES, AVC_SPEC_CB12_BITS, cb_map12); |
| 84 | let cb12 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 85 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB13_CODES, AVC_SPEC_CB13_BITS, cb_map13); |
| 86 | let cb13 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 87 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB14_CODES, AVC_SPEC_CB14_BITS, cb_map14); |
| 88 | let cb14 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 89 | let mut coderead = TableCodebookDescReader::new(AVC_SPEC_CB15_CODES, AVC_SPEC_CB15_BITS, cb_map15); |
| 90 | let cb15 = Codebook::new(&mut coderead, CodebookMode::MSB).unwrap(); |
| 91 | let spec_cb = [cb1, cb2, cb3, cb4, cb5, cb6, cb7, cb8, cb9, cb10, cb11, cb12, cb13, cb14, cb15]; |
| 92 | Self { scale_cb, spec_cb } |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | struct SynthDSP { |
| 97 | rdft512: RDFT, |
| 98 | rdft256: RDFT, |
| 99 | irdft128: RDFT, |
| 100 | irdft64: RDFT, |
| 101 | synth_tmp: [FFTComplex; 512], |
| 102 | synth_out: [FFTComplex; 512], |
| 103 | } |
| 104 | |
| 105 | impl SynthDSP { |
| 106 | fn new() -> Self { |
| 107 | let rdft512 = RDFTBuilder::new_rdft(512, true, true); |
| 108 | let rdft256 = RDFTBuilder::new_rdft(256, true, true); |
| 109 | let irdft128 = RDFTBuilder::new_rdft(128, false, false); |
| 110 | let irdft64 = RDFTBuilder::new_rdft(64, false, false); |
| 111 | Self { |
| 112 | rdft512, rdft256, irdft128, irdft64, |
| 113 | synth_tmp: [FFTC_ZERO; 512], |
| 114 | synth_out: [FFTC_ZERO; 512], |
| 115 | } |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | struct AVCDecoder { |
| 120 | chmap: NAChannelMap, |
| 121 | ainfo: NAAudioInfo, |
| 122 | info: NACodecInfoRef, |
| 123 | dsp: SynthDSP, |
| 124 | version: i32, |
| 125 | codebooks: Codebooks, |
| 126 | channels: usize, |
| 127 | is_40khz: bool, |
| 128 | win_long: &'static [f32; 1024], |
| 129 | |
| 130 | prev_win: u8, |
| 131 | cur_win: u8, |
| 132 | winfo: &'static WinInfo, |
| 133 | windows: usize, |
| 134 | win_grp: [bool; 8], |
| 135 | ms_present: bool, |
| 136 | ms_info: [bool; MAX_BANDS], |
| 137 | cbs: [u8; MAX_BANDS], |
| 138 | scales: [u8; MAX_BANDS], |
| 139 | |
| 140 | imdct_long: IMDCT, |
| 141 | imdct_mid: IMDCT, |
| 142 | imdct_short: IMDCT, |
| 143 | coeffs: [[f32; COEFFS]; 2], |
| 144 | delay: [[f32; COEFFS]; 2], |
| 145 | tmp: [f32; COEFFS * 2], |
| 146 | ew_buf: [f32; 1152], |
| 147 | use_generic: bool, |
| 148 | |
| 149 | qtab: [f32; QTAB_SIZE], |
| 150 | scale_tab: [f32; 128], |
| 151 | } |
| 152 | |
| 153 | impl AVCDecoder { |
| 154 | fn new(version: i32) -> Self { |
| 155 | let mut qtab = [0.0f32; QTAB_SIZE]; |
| 156 | for (i, el) in qtab.iter_mut().enumerate() { |
| 157 | *el = (i as f32) * (i as f32).sqrt(); |
| 158 | } |
| 159 | let mut scale_tab = [0.0f32; 128]; |
| 160 | for (i, el) in scale_tab.iter_mut().enumerate() { |
| 161 | let pow = 10.0f32.powf((i as f32) * 0.1); |
| 162 | if i < 20 { |
| 163 | *el = (pow * 16.0).ceil() / 32.0; |
| 164 | } else { |
| 165 | *el = (pow * 0.5).ceil(); |
| 166 | } |
| 167 | *el /= 32768.0; |
| 168 | } |
| 169 | |
| 170 | AVCDecoder { |
| 171 | chmap: NAChannelMap::new(), |
| 172 | ainfo: NAAudioInfo::new(0, 0, SND_F32P_FORMAT, 0), |
| 173 | info: NACodecInfo::new_dummy(), |
| 174 | dsp: SynthDSP::new(), |
| 175 | version, |
| 176 | codebooks: Codebooks::new(), |
| 177 | channels: 0, |
| 178 | is_40khz: false, |
| 179 | win_long: AVC_WIN_LONG_32K, |
| 180 | |
| 181 | prev_win: 0, |
| 182 | cur_win: 0, |
| 183 | winfo: &AVC_WINFO_LONG, |
| 184 | windows: 0, |
| 185 | win_grp: [false; 8], |
| 186 | ms_present: false, |
| 187 | ms_info: [false; MAX_BANDS], |
| 188 | cbs: [0; MAX_BANDS], |
| 189 | scales: [0; MAX_BANDS], |
| 190 | |
| 191 | imdct_long: IMDCT::new(1024 * 2, true), |
| 192 | imdct_mid: IMDCT::new(512 * 2, true), |
| 193 | imdct_short: IMDCT::new(128 * 2, true), |
| 194 | coeffs: [[0.0; COEFFS]; 2], |
| 195 | delay: [[0.0; COEFFS]; 2], |
| 196 | tmp: [0.0; COEFFS * 2], |
| 197 | ew_buf: [0.0; 1152], |
| 198 | use_generic: true, |
| 199 | |
| 200 | qtab, |
| 201 | scale_tab, |
| 202 | } |
| 203 | } |
| 204 | fn decode_channel(&mut self, br: &mut BitReader, chno: usize) -> DecoderResult<()> { |
| 205 | let coeffs = &mut self.coeffs[chno]; |
| 206 | |
| 207 | decode_band_types(br, &mut self.cbs, self.winfo)?; |
| 208 | // band scales |
| 209 | let bands = self.winfo.bands.len(); |
| 210 | let mut cur_band = 0; |
| 211 | let mut scale = 0; |
| 212 | let mut first = true; |
| 213 | for wg in 0..self.windows { |
| 214 | if self.win_grp[wg] { |
| 215 | for _ in 0..bands { |
| 216 | if self.cbs[cur_band] == 0 { |
| 217 | let mut all_zero = true; |
| 218 | let mut band2 = cur_band; |
| 219 | for wg2 in wg + 1..self.windows { |
| 220 | if self.win_grp[wg2] { |
| 221 | break; |
| 222 | } |
| 223 | band2 += bands; |
| 224 | if self.cbs[band2] != 0 { |
| 225 | all_zero = false; |
| 226 | break; |
| 227 | } |
| 228 | } |
| 229 | if all_zero { |
| 230 | self.scales[cur_band] = 0; |
| 231 | cur_band += 1; |
| 232 | continue; |
| 233 | } |
| 234 | } |
| 235 | if first { |
| 236 | scale = br.read(7)? as i16; |
| 237 | first = false; |
| 238 | } else { |
| 239 | scale += i16::from(br.read_cb(&self.codebooks.scale_cb)?); |
| 240 | validate!((scale >= 0) && (scale < 128)); |
| 241 | } |
| 242 | self.scales[cur_band] = scale as u8; |
| 243 | cur_band += 1; |
| 244 | } |
| 245 | } else { |
| 246 | for _ in 0..bands { |
| 247 | self.scales[cur_band] = self.scales[cur_band - bands]; |
| 248 | cur_band += 1; |
| 249 | } |
| 250 | } |
| 251 | } |
| 252 | // coefficients |
| 253 | let mut cur_band = 0; |
| 254 | let mut idx = 0; |
| 255 | *coeffs = [0.0; COEFFS]; |
| 256 | for _ in 0..self.windows { |
| 257 | let mut band_start = 0; |
| 258 | for band_end in self.winfo.bands.iter() { |
| 259 | let band_size = *band_end - band_start; |
| 260 | let cb_idx = self.cbs[cur_band]; |
| 261 | if cb_idx > 0 { |
| 262 | let cb = &self.codebooks.spec_cb[(cb_idx - 1) as usize]; |
| 263 | let scale = self.scale_tab[self.scales[cur_band] as usize]; |
| 264 | let dst = &mut coeffs[idx..][..band_size]; |
| 265 | if cb_idx < 9 { |
| 266 | decode_quads(br, cb, &self.qtab, scale, dst)?; |
| 267 | } else { |
| 268 | decode_pairs(br, cb, &self.qtab, scale, dst, cb_idx == 15)?; |
| 269 | } |
| 270 | } |
| 271 | |
| 272 | band_start = *band_end; |
| 273 | idx += band_size; |
| 274 | cur_band += 1; |
| 275 | } |
| 276 | } |
| 277 | Ok(()) |
| 278 | } |
| 279 | fn decode_frame(&mut self, br: &mut BitReader, adata: &mut NAAudioBuffer<f32>, offset: usize) -> DecoderResult<()> { |
| 280 | let enh_flag = br.read_bool()?; |
| 281 | validate!(!enh_flag); |
| 282 | self.prev_win = self.cur_win; |
| 283 | self.cur_win = br.read(3)? as u8; |
| 284 | self.winfo = match self.cur_win { |
| 285 | 3 => &AVC_WINFO_SHORT, |
| 286 | 4 => if self.is_40khz { &AVC_WINFO_40K_MODE4 } else { &AVC_WINFO_44K_MODE4 }, |
| 287 | 5 => if self.is_40khz { &AVC_WINFO_40K_MODE5 } else { &AVC_WINFO_44K_MODE5 }, |
| 288 | 6 => if self.is_40khz { &AVC_WINFO_40K_MODE6 } else { &AVC_WINFO_44K_MODE6 }, |
| 289 | _ => &AVC_WINFO_LONG, |
| 290 | }; |
| 291 | self.windows = if self.winfo.is_long { 1 } else { 8 }; |
| 292 | let bands = self.winfo.bands.len(); |
| 293 | self.win_grp[0] = true; |
| 294 | for el in self.win_grp.iter_mut().skip(1).take(self.windows - 1) { |
| 295 | *el = !br.read_bool()?; |
| 296 | } |
| 297 | |
| 298 | self.ms_present = br.read_bool()?; |
| 299 | if self.ms_present { |
| 300 | validate!(self.channels == 2); |
| 301 | let mut cur_band = 0; |
| 302 | for wg in self.win_grp.iter().take(self.windows) { |
| 303 | if *wg { |
| 304 | for _ in 0..bands { |
| 305 | self.ms_info[cur_band] = br.read_bool()?; |
| 306 | cur_band += 1; |
| 307 | } |
| 308 | } else { |
| 309 | for _ in 0..bands { |
| 310 | self.ms_info[cur_band] = self.ms_info[cur_band - bands]; |
| 311 | cur_band += 1; |
| 312 | } |
| 313 | } |
| 314 | } |
| 315 | } |
| 316 | for ch in 0..self.channels { |
| 317 | self.decode_channel(br, ch)?; |
| 318 | } |
| 319 | |
| 320 | if (self.channels == 2) && self.ms_present { |
| 321 | let mut idx = 0; |
| 322 | let mut cur_band = 0; |
| 323 | for _ in 0..self.windows { |
| 324 | let mut band_start = 0; |
| 325 | for band_end in self.winfo.bands.iter() { |
| 326 | let band_size = *band_end - band_start; |
| 327 | if self.ms_info[cur_band] { |
| 328 | for i in 0..band_size { |
| 329 | let l = self.coeffs[0][idx + i]; |
| 330 | let r = self.coeffs[1][idx + i]; |
| 331 | self.coeffs[0][idx + i] = l + r; |
| 332 | self.coeffs[1][idx + i] = l - r; |
| 333 | } |
| 334 | } |
| 335 | cur_band += 1; |
| 336 | idx += band_size; |
| 337 | band_start = *band_end; |
| 338 | } |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | for ch in 0..self.channels { |
| 343 | let off = adata.get_offset(ch) + offset; |
| 344 | let output = adata.get_data_mut().unwrap(); |
| 345 | self.synth_channel(ch, &mut output[off..][..COEFFS]); |
| 346 | } |
| 347 | Ok(()) |
| 348 | } |
| 349 | #[allow(clippy::cyclomatic_complexity)] |
| 350 | fn synth_channel(&mut self, chno: usize, dst: &mut [f32]) { |
| 351 | let coeffs = &mut self.coeffs[chno]; |
| 352 | let delay = &mut self.delay[chno]; |
| 353 | |
| 354 | match self.cur_win { |
| 355 | 0 | 1 => { |
| 356 | self.imdct_long.imdct_half(coeffs, &mut self.tmp); |
| 357 | overlap_half(dst, &self.tmp, delay, self.win_long); |
| 358 | (&mut delay[0..COEFFS/2]).copy_from_slice(&self.tmp[COEFFS/2..COEFFS]); |
| 359 | for i in COEFFS/2..COEFFS { |
| 360 | delay[i] = delay[COEFFS - 1 - i]; |
| 361 | } |
| 362 | }, |
| 363 | 2 | 7 => { |
| 364 | self.imdct_long.imdct(coeffs, &mut self.tmp); |
| 365 | for i in 0..SHORT_WIN_POINT0 { |
| 366 | dst[i] = delay[i]; |
| 367 | } |
| 368 | overlap(&mut dst[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 369 | &self.tmp[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 370 | &delay[SHORT_WIN_POINT0..SHORT_WIN_POINT1], AVC_WIN_SHORT); |
| 371 | for i in SHORT_WIN_POINT1..COEFFS { |
| 372 | dst[i] = self.tmp[i - SHORT_WIN_POINT1 + 128]; |
| 373 | } |
| 374 | delay.copy_from_slice(&self.tmp[COEFFS..]); |
| 375 | }, |
| 376 | 3 => { |
| 377 | for (ain, aout) in coeffs.chunks(128).zip(self.tmp.chunks_mut(256)) { |
| 378 | self.imdct_short.imdct(ain, aout); |
| 379 | } |
| 380 | self.ew_buf = [0.0; 1152]; |
| 381 | for i in 0..128 { |
| 382 | self.ew_buf[i] = self.tmp[i]; |
| 383 | } |
| 384 | for w in 0..7 { |
| 385 | overlap(&mut self.ew_buf[(w + 1) * 128..][..128], |
| 386 | &self.tmp[(w + 1) * 256..][..128], |
| 387 | &self.tmp[w * 256 + 128..][..128], |
| 388 | AVC_WIN_SHORT); |
| 389 | } |
| 390 | for i in 0..128 { |
| 391 | self.ew_buf[1024 + i] = self.tmp[7 * 256 + 128 + i]; |
| 392 | } |
| 393 | for i in 0..SHORT_WIN_POINT0 { |
| 394 | dst[i] = delay[i]; |
| 395 | } |
| 396 | overlap(&mut dst[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 397 | &self.ew_buf[0..128], |
| 398 | &delay[SHORT_WIN_POINT0..SHORT_WIN_POINT1], AVC_WIN_SHORT); |
| 399 | for i in SHORT_WIN_POINT1..COEFFS { |
| 400 | dst[i] = self.ew_buf[i - SHORT_WIN_POINT1 + 128]; |
| 401 | } |
| 402 | for i in 0..SHORT_WIN_POINT1 { |
| 403 | delay[i] = self.ew_buf[SHORT_WIN_POINT1 + i]; |
| 404 | } |
| 405 | for i in SHORT_WIN_POINT1..COEFFS { |
| 406 | delay[i] = 0.0; |
| 407 | } |
| 408 | }, |
| 409 | 4 => { |
| 410 | for i in 0..SHORT_WIN_POINT0 { |
| 411 | dst[i] = delay[i]; |
| 412 | } |
| 413 | if !self.use_generic { |
| 414 | synth1024(&mut self.dsp, coeffs, &mut self.ew_buf, &mut self.tmp, self.is_40khz); |
| 415 | } else { |
| 416 | synth_generic(coeffs, &mut self.ew_buf, &mut self.tmp, self.is_40khz, 1024); |
| 417 | } |
| 418 | overlap_half(&mut dst[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 419 | &self.ew_buf[0..64], |
| 420 | &delay[SHORT_WIN_POINT0..SHORT_WIN_POINT1], AVC_WIN_SHORT); |
| 421 | for i in SHORT_WIN_POINT1..COEFFS { |
| 422 | dst[i] = self.ew_buf[i - SHORT_WIN_POINT1 + 64]; |
| 423 | } |
| 424 | for i in 0..COEFFS/2 { |
| 425 | delay[i] = self.ew_buf[COEFFS/2 + i]; |
| 426 | } |
| 427 | for i in COEFFS/2..COEFFS { |
| 428 | delay[i] = delay[COEFFS - 1 - i]; |
| 429 | } |
| 430 | }, |
| 431 | 5 => { |
| 432 | for i in 0..SHORT_WIN_POINT0 { |
| 433 | dst[i] = delay[i]; |
| 434 | } |
| 435 | if !self.use_generic { |
| 436 | synth512(&mut self.dsp, coeffs, &mut self.ew_buf, &mut self.tmp, self.is_40khz); |
| 437 | } else { |
| 438 | synth_generic(coeffs, &mut self.ew_buf, &mut self.tmp, self.is_40khz, 512); |
| 439 | } |
| 440 | self.imdct_mid.imdct(&coeffs[512..], &mut self.ew_buf[512..]); |
| 441 | overlap(&mut dst[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 442 | &self.ew_buf[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 443 | &delay[SHORT_WIN_POINT0..SHORT_WIN_POINT1], AVC_WIN_SHORT); |
| 444 | for i in SHORT_WIN_POINT1..COEFFS { |
| 445 | dst[i] = self.ew_buf[i]; |
| 446 | } |
| 447 | *delay = [0.0; COEFFS]; |
| 448 | for i in 0..SHORT_WIN_POINT1 { |
| 449 | delay[i] = self.ew_buf[i]; |
| 450 | } |
| 451 | }, |
| 452 | 6 => { |
| 453 | for i in 0..SHORT_WIN_POINT0 { |
| 454 | dst[i] = delay[i]; |
| 455 | } |
| 456 | self.imdct_mid.imdct(coeffs, &mut self.ew_buf); |
| 457 | if !self.use_generic { |
| 458 | synth512(&mut self.dsp, &coeffs[512..], &mut self.ew_buf[512..], &mut self.tmp, self.is_40khz); |
| 459 | } else { |
| 460 | synth_generic(&coeffs[512..], &mut self.ew_buf[512..], &mut self.tmp, self.is_40khz, 512); |
| 461 | } |
| 462 | overlap(&mut dst[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 463 | &self.ew_buf[SHORT_WIN_POINT0..SHORT_WIN_POINT1], |
| 464 | &delay[SHORT_WIN_POINT0..SHORT_WIN_POINT1], AVC_WIN_SHORT); |
| 465 | for i in SHORT_WIN_POINT1..COEFFS { |
| 466 | dst[i] = self.ew_buf[i]; |
| 467 | } |
| 468 | *delay = [0.0; COEFFS]; |
| 469 | for i in 0..SHORT_WIN_POINT1 { |
| 470 | delay[i] = self.ew_buf[i]; |
| 471 | } |
| 472 | }, |
| 473 | _ => unreachable!(), |
| 474 | }; |
| 475 | } |
| 476 | } |
| 477 | |
| 478 | fn overlap(dst: &mut [f32], src: &[f32], delay: &[f32], win: &[f32]) { |
| 479 | for (i, ((out, s), d)) in dst.iter_mut().zip(src.iter()).zip(delay.iter()).enumerate() { |
| 480 | *out = *s * win[i] + *d * win[win.len() - 1 - i]; |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | fn overlap_half(dst: &mut [f32], src: &[f32], delay: &[f32], win: &[f32]) { |
| 485 | let n = win.len(); |
| 486 | for i in 0..n/2 { |
| 487 | let ii = n - 1 - i; |
| 488 | let c = src [n/2 - 1 - i]; |
| 489 | let d = delay[i]; |
| 490 | let w0 = win[i]; |
| 491 | let w1 = win[ii]; |
| 492 | dst[i] = d * w1 - c * w0; |
| 493 | dst[ii] = d * w0 + c * w1; |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | #[derive(Clone,Copy)] |
| 498 | struct SynthParams { |
| 499 | p0: usize, |
| 500 | p1: usize, |
| 501 | idx: usize |
| 502 | } |
| 503 | |
| 504 | macro_rules! synth_step0_template { |
| 505 | ($name:ident, $tab: ident, $size: expr) => { |
| 506 | fn $name(src: &[f32], dst: &mut [f32], step: usize, off: usize, sp: &SynthParams) { |
| 507 | for i in 0..step { |
| 508 | for j in 0..sp.p0 { |
| 509 | dst[i] += (f64::from(src[j]) * $tab[sp.idx][j][i]) as f32; |
| 510 | } |
| 511 | } |
| 512 | for i in 0..step { |
| 513 | for j in 0..sp.p1 { |
| 514 | dst[$size - step + i] += (f64::from(src[sp.p0 + off + j]) * $tab[sp.idx][sp.p0 + j][i]) as f32; |
| 515 | } |
| 516 | } |
| 517 | } |
| 518 | } |
| 519 | } |
| 520 | |
| 521 | synth_step0_template!(synth_step0_10, AVC_SYNTH_TABS10, 10); |
| 522 | synth_step0_template!(synth_step0_20, AVC_SYNTH_TABS20, 20); |
| 523 | synth_step0_template!(synth_step0_40, AVC_SYNTH_TABS40, 40); |
| 524 | synth_step0_template!(synth_step0_84, AVC_SYNTH_TABS84, 84); |
| 525 | |
| 526 | fn synth_step1(src: &[f32], dst: &mut [f32], size: usize, stride: usize, step: usize, off: usize, mut p0: usize, win: &[f64]) |
| 527 | { |
| 528 | let mut pos = step - 1; |
| 529 | for _ in 0..off { |
| 530 | let scale = f64::from(src[p0]); |
| 531 | p0 += 1; |
| 532 | pos &= size - 1; |
| 533 | for i in 0..pos.min(step) { |
| 534 | dst[pos - i] += (scale * win[i]) as f32; |
| 535 | } |
| 536 | pos += stride; |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | fn zero_some(buf: &mut [FFTComplex], p0: usize, p1: usize) |
| 541 | { |
| 542 | for el in buf.iter_mut().take(p0/2) { |
| 543 | *el = FFTC_ZERO; |
| 544 | } |
| 545 | if (p0 & 1) != 0 { |
| 546 | buf[p0 / 2].re = 0.0; |
| 547 | } |
| 548 | let len = buf.len(); |
| 549 | let dst = &mut buf[len - p1/2..]; |
| 550 | for el in dst.iter_mut() { |
| 551 | *el = FFTC_ZERO; |
| 552 | } |
| 553 | if (p1 & 1) != 0 { |
| 554 | buf[len - p1/2 - 1].im = 0.0; |
| 555 | } |
| 556 | } |
| 557 | |
| 558 | fn merge_bands(src: &[FFTComplex], dst: &mut [FFTComplex], size: usize) { |
| 559 | let step = if size == 512 { 2 } else { 4 }; |
| 560 | let (s01, s23) = src.split_at(size / 2); |
| 561 | let (src0, src1) = s01.split_at(size / 4); |
| 562 | let (src2, src3) = s23.split_at(size / 4); |
| 563 | let (t00, t01) = AVC_SYNTH_TAB0.split_at(512); |
| 564 | let (t10, t11) = AVC_SYNTH_TAB1.split_at(512); |
| 565 | let (t20, t21) = AVC_SYNTH_TAB2.split_at(512); |
| 566 | let (t30, t31) = AVC_SYNTH_TAB3.split_at(512); |
| 567 | let hsize = size / 2; |
| 568 | let qsize = size / 4; |
| 569 | let osize = size / 8; |
| 570 | dst[0].re = src0[0].re * t00[0] + src1[0].re * t10[0] + src2[0].re * t20[0] + src3[0].re * t30[0]; |
| 571 | dst[0].im = src0[0].re * t00[1] + src1[0].re * t10[1] + src2[0].re * t20[1] + src3[0].re * t30[1]; |
| 572 | dst[qsize].re = src0[0].im * t00[256] + src1[0].im * t10[256] + |
| 573 | src2[0].im * t20[256] + src3[0].im * t30[256]; |
| 574 | dst[qsize].im = src0[0].im * t00[257] + src1[0].im * t10[257] + |
| 575 | src2[0].im * t20[257] + src3[0].im * t30[257]; |
| 576 | for i in 1..qsize { |
| 577 | let s0 = src0[i]; |
| 578 | let s1 = src1[i]; |
| 579 | let s2 = src2[i]; |
| 580 | let s3 = src3[i]; |
| 581 | let t0 = s0 * FFTComplex { re: t00[i * step + 0], im: t00[i * step + 1] } + |
| 582 | s1 * FFTComplex { re: t10[i * step + 0], im: t10[i * step + 1] } + |
| 583 | s2 * FFTComplex { re: t20[i * step + 0], im: t20[i * step + 1] } + |
| 584 | s3 * FFTComplex { re: t30[i * step + 0], im: t30[i * step + 1] }; |
| 585 | let t1 = s0 * FFTComplex { re: t01[i * step + 0], im: t01[i * step + 1] } + |
| 586 | s1 * FFTComplex { re: t11[i * step + 0], im: t11[i * step + 1] } + |
| 587 | s2 * FFTComplex { re: t21[i * step + 0], im: t21[i * step + 1] } + |
| 588 | s3 * FFTComplex { re: t31[i * step + 0], im: t31[i * step + 1] }; |
| 589 | dst[i] = t0; |
| 590 | dst[qsize + i] = t1; |
| 591 | } |
| 592 | |
| 593 | dst[hsize].re = src0[0].im * t01[256] + src1[0].im * t11[256] + |
| 594 | src2[0].im * t21[256] + src3[0].im * t31[256]; |
| 595 | dst[hsize].im = src0[0].im * t01[257] + src1[0].im * t11[257] + |
| 596 | src2[0].im * t21[257] + src3[0].im * t31[257]; |
| 597 | dst[hsize + qsize].re = src0[0].re * t01[256] + src1[0].re * t11[256] + |
| 598 | src2[0].re * t21[256] + src3[0].re * t31[256]; |
| 599 | dst[hsize + qsize].im = src0[0].re * t01[257] + src1[0].re * t11[257] + |
| 600 | src2[0].re * t21[257] + src3[0].re * t31[257]; |
| 601 | for i in 1..osize { |
| 602 | let s0 = !src0[qsize - i]; |
| 603 | let s1 = !src1[qsize - i]; |
| 604 | let s2 = !src2[qsize - i]; |
| 605 | let s3 = !src3[qsize - i]; |
| 606 | let t0 = s0 * FFTComplex { re: t00[256 + i * step + 0], im: t00[256 + i * step + 1] } + |
| 607 | s1 * FFTComplex { re: t10[256 + i * step + 0], im: t10[256 + i * step + 1] } + |
| 608 | s2 * FFTComplex { re: t20[256 + i * step + 0], im: t20[256 + i * step + 1] } + |
| 609 | s3 * FFTComplex { re: t30[256 + i * step + 0], im: t30[256 + i * step + 1] }; |
| 610 | let t1 = s0 * FFTComplex { re: t01[256 + i * step + 0], im: t01[256 + i * step + 1] } + |
| 611 | s1 * FFTComplex { re: t11[256 + i * step + 0], im: t11[256 + i * step + 1] } + |
| 612 | s2 * FFTComplex { re: t21[256 + i * step + 0], im: t21[256 + i * step + 1] } + |
| 613 | s3 * FFTComplex { re: t31[256 + i * step + 0], im: t31[256 + i * step + 1] }; |
| 614 | dst[hsize + i] = !t0; |
| 615 | dst[hsize + qsize + i] = !t1; |
| 616 | } |
| 617 | } |
| 618 | |
| 619 | const SPARAMS10: [SynthParams; 2] = [ |
| 620 | SynthParams { p0: 1, p1: 3, idx: 0 }, |
| 621 | SynthParams { p0: 3, p1: 1, idx: 1 } ]; |
| 622 | const SPARAMS20: [SynthParams; 2] = [ |
| 623 | SynthParams { p0: 5, p1: 4, idx: 0 }, |
| 624 | SynthParams { p0: 4, p1: 5, idx: 1 } ]; |
| 625 | const SPARAMS40: [SynthParams; 2] = [ |
| 626 | SynthParams { p0: 11, p1: 8, idx: 0 }, |
| 627 | SynthParams { p0: 8, p1: 11, idx: 1 } ]; |
| 628 | const SPARAMS84: [SynthParams; 4] = [ |
| 629 | SynthParams { p0: 16, p1: 4, idx: 0 }, |
| 630 | SynthParams { p0: 16, p1: 4, idx: 1 }, |
| 631 | SynthParams { p0: 13, p1: 7, idx: 2 }, |
| 632 | SynthParams { p0: 15, p1: 5, idx: 3 } ]; |
| 633 | |
| 634 | const MERGE_ORDER_44K: &[u8] = &[ |
| 635 | 4, 4, 2, 2, 0, 0, 2, 0, 0, 2, 2, 0, 0, 2, 0, 0, 2, 0, 0, |
| 636 | 2, 0, 0, 4, 0, 0, 0, 0, 2, 0, 0, 0 |
| 637 | ]; |
| 638 | const MERGE_ORDER_40K: &[u8] = &[ |
| 639 | 4, 4, 2, 2, 0, 0, 2, 0, 0, 2, 2, 0, 0, 2, 0, 0, 2, 0, 0, |
| 640 | 2, 0, 0, 4, 2, 0, 0, 2, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0 |
| 641 | ]; |
| 642 | |
| 643 | fn synth_filter(src: &[f32], dst: &mut [f32], len: usize, step: usize, bands: usize, idx: usize) |
| 644 | { |
| 645 | let off = (len - step) / bands + 1; |
| 646 | let params = match step { |
| 647 | 10 => &SPARAMS10[idx], |
| 648 | 20 => &SPARAMS20[idx], |
| 649 | 40 => &SPARAMS40[idx], |
| 650 | _ => &SPARAMS84[idx], |
| 651 | }; |
| 652 | |
| 653 | if step == 10 { |
| 654 | synth_step0_10(src, dst, step, off, params); |
| 655 | synth_step1(src, dst, len, bands, step, off, params.p0, &AVC_TAB10[idx]); |
| 656 | } else if step == 20 { |
| 657 | synth_step0_20(src, dst, step, off, params); |
| 658 | synth_step1(src, dst, len, bands, step, off, params.p0, &AVC_TAB20[idx]); |
| 659 | } else if step == 40 { |
| 660 | synth_step0_40(src, dst, step, off, params); |
| 661 | synth_step1(src, dst, len, bands, step, off, params.p0, &AVC_TAB40[idx]); |
| 662 | } else { |
| 663 | synth_step0_84(src, dst, step, off, params); |
| 664 | synth_step1(src, dst, len, bands, step, off, params.p0, &AVC_TAB84[idx]); |
| 665 | } |
| 666 | } |
| 667 | |
| 668 | fn synth_recursive(dst: &mut [f32], tmp: &mut [f32], size: usize, order: &[u8], order_idx: &mut usize, dir: bool) { |
| 669 | let bands = order[*order_idx] as usize; |
| 670 | *order_idx += 1; |
| 671 | if bands == 0 { return; } |
| 672 | let sub_size = size / bands; |
| 673 | let mut sub_dir = false; |
| 674 | for (dst, tmp) in dst.chunks_mut(sub_size).take(bands).zip(tmp.chunks_mut(sub_size)) { |
| 675 | synth_recursive(dst, tmp, sub_size, order, order_idx, sub_dir); |
| 676 | sub_dir = !sub_dir; |
| 677 | } |
| 678 | for el in tmp.iter_mut().take(size) { *el = 0.0; } |
| 679 | let step = if bands == 2 { |
| 680 | if sub_size <= 20 { |
| 681 | 10 |
| 682 | } else if sub_size <= 40 { |
| 683 | 20 |
| 684 | } else { |
| 685 | 40 |
| 686 | } |
| 687 | } else { |
| 688 | 84 |
| 689 | }; |
| 690 | for (i, src) in dst.chunks_mut(sub_size).take(bands).enumerate() { |
| 691 | let idx = if !dir { i } else { bands - 1 - i }; |
| 692 | synth_filter(src, tmp, size, step, bands, idx); |
| 693 | } |
| 694 | (&mut dst[..size]).copy_from_slice(&tmp[..size]); |
| 695 | } |
| 696 | |
| 697 | fn synth_generic(src: &[f32], dst: &mut [f32], tmpbuf: &mut [f32; COEFFS * 2], is_40khz: bool, size: usize) { |
| 698 | let order = if is_40khz { MERGE_ORDER_40K } else { MERGE_ORDER_44K }; |
| 699 | (&mut dst[..size]).copy_from_slice(&src[..size]); |
| 700 | let mut order_idx = 0; |
| 701 | synth_recursive(dst, tmpbuf, size, order, &mut order_idx, false); |
| 702 | for el in dst.iter_mut().take(COEFFS) { *el *= 0.125; } |
| 703 | } |
| 704 | |
| 705 | fn synth1024(dsp: &mut SynthDSP, src: &[f32], dst: &mut [f32], tmpbuf: &mut [f32; COEFFS * 2], is_40khz: bool) { |
| 706 | for el in tmpbuf.iter_mut() { |
| 707 | *el = 0.0; |
| 708 | } |
| 709 | let (tmp, tmp2) = tmpbuf.split_at_mut(COEFFS); |
| 710 | for el in dsp.synth_tmp.iter_mut() { *el = FFTC_ZERO; } |
| 711 | for el in dsp.synth_out.iter_mut() { *el = FFTC_ZERO; } |
| 712 | |
| 713 | let (size, step, stride) = (32, 20, 2); |
| 714 | let off = (size - step) / stride + 1; |
| 715 | let mut sparam0 = SynthParams { p0: 5, p1: 4, idx: 0 }; |
| 716 | let mut sparam1 = SynthParams { p0: 4, p1: 5, idx: 1 }; |
| 717 | for (i, dst) in tmp.chunks_mut(size).take(8).enumerate() { |
| 718 | let coeffs = &src[i * 32..]; |
| 719 | synth_step0_20(coeffs, dst, step, off, &sparam0); |
| 720 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB20[sparam0.idx]); |
| 721 | let coeffs = &src[i * 32 + 16..]; |
| 722 | synth_step0_20(coeffs, dst, step, off, &sparam1); |
| 723 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB20[sparam1.idx]); |
| 724 | std::mem::swap(&mut sparam0, &mut sparam1); |
| 725 | } |
| 726 | |
| 727 | let (size, step, stride) = (64, 40, 2); |
| 728 | let off = (size - step) / stride + 1; |
| 729 | let mut sparam0 = SynthParams { p0: 11, p1: 8, idx: 0 }; |
| 730 | let mut sparam1 = SynthParams { p0: 8, p1: 11, idx: 1 }; |
| 731 | let nchunks = if !is_40khz { 8 } else { 12 }; |
| 732 | for (i, dst) in tmp2.chunks_mut(size).take(nchunks).enumerate() { |
| 733 | let coeffs = if i < 4 { &tmp[i * 64..] } else { &src[i * 64 + 32..] }; |
| 734 | synth_step0_40(coeffs, dst, step, off, &sparam0); |
| 735 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB40[sparam0.idx]); |
| 736 | let coeffs = if i < 4 { &tmp[i * 64 + 32..] } else { &src[i * 64 + 32..] }; |
| 737 | synth_step0_40(coeffs, dst, step, off, &sparam1); |
| 738 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB40[sparam1.idx]); |
| 739 | std::mem::swap(&mut sparam0, &mut sparam1); |
| 740 | } |
| 741 | |
| 742 | for el in tmp.iter_mut().take(128) { |
| 743 | *el = 0.0; |
| 744 | } |
| 745 | let (size, step, stride) = (256, 84, 4); |
| 746 | let off = (size - step) / stride + 1; |
| 747 | for i in 0..4 { |
| 748 | let coeffs = &tmp2[i * 64..]; |
| 749 | synth_step0_84(coeffs, tmp, step, off, &SPARAMS84[i]); |
| 750 | synth_step1(coeffs, tmp, size, stride, step, off, SPARAMS84[i].p0, &AVC_TAB84[i]); |
| 751 | } |
| 752 | for i in 0..4 { |
| 753 | let coeffs = if i < 2 && is_40khz { &tmp2[256 + i * 64..] } else { &src[256 + i * 64..] }; |
| 754 | let dst = &mut tmp[256..]; |
| 755 | synth_step0_84(coeffs, dst, step, off, &SPARAMS84[3 - i]); |
| 756 | synth_step1(coeffs, dst, size, stride, step, off, SPARAMS84[3 - i].p0, &AVC_TAB84[3 - i]); |
| 757 | } |
| 758 | |
| 759 | if !is_40khz { |
| 760 | let (size, step, stride) = (256, 40, 2); |
| 761 | let off = (size - step) / stride + 1; |
| 762 | let sparams = [ SynthParams { p0: 11, p1: 8, idx: 0 }, |
| 763 | SynthParams { p0: 8, p1: 11, idx: 1 } ]; |
| 764 | for i in 0..2 { |
| 765 | let coeffs = &src[512 + i * 128..]; |
| 766 | let dst = &mut tmp[512..]; |
| 767 | synth_step0_40(coeffs, dst, step, off, &sparams[i]); |
| 768 | synth_step1(coeffs, dst, size, stride, step, off, sparams[i].p0, &AVC_TAB40[i]); |
| 769 | } |
| 770 | } else { |
| 771 | let (size, step, stride) = (256, 84, 4); |
| 772 | let off = (size - step) / stride + 1; |
| 773 | for i in 0..4 { |
| 774 | let coeffs = &src[512 + i * 64..]; |
| 775 | let dst = &mut tmp[512..]; |
| 776 | synth_step0_84(coeffs, dst, step, off, &SPARAMS84[i]); |
| 777 | synth_step1(coeffs, dst, size, stride, step, off, SPARAMS84[i].p0, &AVC_TAB84[i]); |
| 778 | } |
| 779 | } |
| 780 | |
| 781 | for (i, s) in tmp.chunks(2).take(768/2).enumerate() { |
| 782 | dsp.synth_tmp[i] = FFTComplex { re: s[0], im: s[1] }; |
| 783 | } |
| 784 | for i in (768..1024).step_by(2) { |
| 785 | dsp.synth_tmp[i >> 1] = FFTComplex { re: src[i], im: src[i + 1] }; |
| 786 | } |
| 787 | { |
| 788 | let dst = &mut tmp[768..1024]; |
| 789 | dst.copy_from_slice(&src[768..1024]); |
| 790 | } |
| 791 | for (i, chunk) in dsp.synth_tmp.chunks_exact_mut(128).enumerate() { |
| 792 | zero_some(chunk, SPARAMS84[i].p0, SPARAMS84[i].p1); |
| 793 | } |
| 794 | for chunk in dsp.synth_tmp.chunks_exact_mut(128) { |
| 795 | dsp.irdft128.do_rdft_inplace(chunk); |
| 796 | for el in chunk.iter_mut() { *el = el.scale(1.0/128.0f32); } |
| 797 | } |
| 798 | merge_bands(&dsp.synth_tmp, &mut dsp.synth_out, 512); |
| 799 | dsp.rdft512.do_rdft_inplace(&mut dsp.synth_out); |
| 800 | for (out, src) in dst.chunks_exact_mut(2).zip(dsp.synth_out.iter()) { |
| 801 | out[0] = src.re * (1.0 / 1024.0f32.sqrt()); |
| 802 | out[1] = src.im * (1.0 / 1024.0f32.sqrt()); |
| 803 | } |
| 804 | let (size, step, stride) = (1024, 84, 4); |
| 805 | let off = (size - step) / stride + 1; |
| 806 | for i in 0..4 { |
| 807 | let src = &tmp[256 * i..]; |
| 808 | synth_step0_84(src, dst, step, off, &SPARAMS84[i]); |
| 809 | } |
| 810 | } |
| 811 | fn synth512(dsp: &mut SynthDSP, src: &[f32], dst: &mut [f32], tmpbuf: &mut [f32; COEFFS * 2], is_40khz: bool) { |
| 812 | for el in tmpbuf.iter_mut() { |
| 813 | *el = 0.0; |
| 814 | } |
| 815 | let (tmp, tmp2) = tmpbuf.split_at_mut(COEFFS); |
| 816 | |
| 817 | let (size, step, stride) = (16, 10, 2); |
| 818 | let off = (size - step) / stride + 1; |
| 819 | let mut sparam0 = SynthParams { p0: 1, p1: 3, idx: 0 }; |
| 820 | let mut sparam1 = SynthParams { p0: 3, p1: 1, idx: 1 }; |
| 821 | let mut tab_ptr0: &[f64] = AVC_TAB10_0; |
| 822 | let mut tab_ptr1: &[f64] = AVC_TAB10_1; |
| 823 | for (i, dst) in tmp.chunks_mut(size).take(8).enumerate() { |
| 824 | let coeffs = &src[i * 16..]; |
| 825 | synth_step0_10(coeffs, dst, step, off, &sparam0); |
| 826 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, tab_ptr0); |
| 827 | let coeffs = &src[i * 16 + 8..]; |
| 828 | synth_step0_10(coeffs, dst, step, off, &sparam1); |
| 829 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, tab_ptr1); |
| 830 | std::mem::swap(&mut sparam0, &mut sparam1); |
| 831 | std::mem::swap(&mut tab_ptr0, &mut tab_ptr1); |
| 832 | } |
| 833 | |
| 834 | let (size, step, stride) = (32, 20, 2); |
| 835 | let off = (size - step) / stride + 1; |
| 836 | let mut sparam0 = SynthParams { p0: 5, p1: 4, idx: 0 }; |
| 837 | let mut sparam1 = SynthParams { p0: 4, p1: 5, idx: 1 }; |
| 838 | let nchunks = if !is_40khz { 8 } else { 12 }; |
| 839 | for (i, dst) in tmp2.chunks_mut(size).take(nchunks).enumerate() { |
| 840 | let coeffs = if i < 4 { &tmp[i * 32..] } else { &src[i * 32..] }; |
| 841 | synth_step0_20(coeffs, dst, step, off, &sparam0); |
| 842 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB20[sparam0.idx]); |
| 843 | let coeffs = if i < 4 { &tmp[i * 32 + 16..] } else { &src[i * 32 + 16..] }; |
| 844 | synth_step0_20(coeffs, dst, step, off, &sparam1); |
| 845 | synth_step1(coeffs, dst, size, stride, step, off, sparam0.p0, &AVC_TAB20[sparam1.idx]); |
| 846 | std::mem::swap(&mut sparam0, &mut sparam1); |
| 847 | } |
| 848 | |
| 849 | for el in tmp.iter_mut().take(64) { |
| 850 | *el = 0.0; |
| 851 | } |
| 852 | let (size, step, stride) = (128, 84, 4); |
| 853 | let off = (size - step) / stride + 1; |
| 854 | for i in 0..4 { |
| 855 | let coeffs = &tmp2[i * 32..]; |
| 856 | synth_step0_84(coeffs, tmp, step, off, &SPARAMS84[i]); |
| 857 | synth_step1(coeffs, tmp, size, stride, step, off, SPARAMS84[i].p0, &AVC_TAB84[i]); |
| 858 | } |
| 859 | for i in 0..4 { |
| 860 | let coeffs = if i < 2 && is_40khz { &tmp2[128 + i * 32..] } else { &src[128 + i * 32..] }; |
| 861 | let dst = &mut tmp[128..]; |
| 862 | synth_step0_84(coeffs, dst, step, off, &SPARAMS84[3 - i]); |
| 863 | synth_step1(coeffs, dst, size, stride, step, off, SPARAMS84[3 - i].p0, &AVC_TAB84[3 - i]); |
| 864 | } |
| 865 | |
| 866 | if !is_40khz { |
| 867 | let (size, step, stride) = (128, 40, 2); |
| 868 | let off = (size - step) / stride + 1; |
| 869 | let sparams = [ SynthParams { p0: 11, p1: 8, idx: 0 }, |
| 870 | SynthParams { p0: 8, p1: 11, idx: 1 } ]; |
| 871 | for i in 0..2 { |
| 872 | let coeffs = &src[256 + i * 64..]; |
| 873 | let dst = &mut tmp[256..]; |
| 874 | synth_step0_40(coeffs, dst, step, off, &sparams[i]); |
| 875 | synth_step1(coeffs, dst, size, stride, step, off, sparams[i].p0, &AVC_TAB40[i]); |
| 876 | } |
| 877 | } else { |
| 878 | let (size, step, stride) = (128, 84, 4); |
| 879 | let off = (size - step) / stride + 1; |
| 880 | for i in 0..4 { |
| 881 | let coeffs = &src[256 + i * 32..]; |
| 882 | let dst = &mut tmp[256..]; |
| 883 | synth_step0_84(coeffs, dst, step, off, &SPARAMS84[i]); |
| 884 | synth_step1(coeffs, dst, size, stride, step, off, SPARAMS84[i].p0, &AVC_TAB84[i]); |
| 885 | } |
| 886 | } |
| 887 | |
| 888 | for (i, s) in tmp.chunks(2).take(384/2).enumerate() { |
| 889 | dsp.synth_tmp[i] = FFTComplex { re: s[0], im: s[1] }; |
| 890 | } |
| 891 | for i in (384..512).step_by(2) { |
| 892 | dsp.synth_tmp[i >> 1] = FFTComplex { re: src[i], im: src[i + 1] }; |
| 893 | } |
| 894 | { |
| 895 | let dst = &mut tmp[384..512]; |
| 896 | dst.copy_from_slice(&src[384..512]); |
| 897 | } |
| 898 | for (i, chunk) in dsp.synth_tmp.chunks_exact_mut(128).enumerate() { |
| 899 | zero_some(chunk, SPARAMS84[i].p0, SPARAMS84[i].p1); |
| 900 | } |
| 901 | for chunk in dsp.synth_tmp.chunks_exact_mut(64) { |
| 902 | dsp.irdft64.do_rdft_inplace(chunk); |
| 903 | } |
| 904 | merge_bands(&dsp.synth_tmp, &mut dsp.synth_out, 256); |
| 905 | dsp.rdft256.do_rdft_inplace(&mut dsp.synth_out); |
| 906 | for (out, src) in dst.chunks_exact_mut(2).zip(dsp.synth_out.iter()).take(256) { |
| 907 | out[0] = src.re * (1.0 / 512.0f32 / 1024.0f32); |
| 908 | out[1] = src.im * (1.0 / 512.0f32 / 1024.0f32); |
| 909 | } |
| 910 | let (size, step, stride) = (512, 84, 4); |
| 911 | let off = (size - step) / stride + 1; |
| 912 | for i in 0..4 { |
| 913 | let src = &tmp[128 * i..]; |
| 914 | synth_step0_84(src, dst, step, off, &SPARAMS84[i]); |
| 915 | } |
| 916 | } |
| 917 | |
| 918 | fn decode_band_types(br: &mut BitReader, cbs: &mut [u8; MAX_BANDS], winfo: &WinInfo) -> DecoderResult<()> { |
| 919 | let bits_per_sect = if winfo.is_long { 5 } else { 3 }; |
| 920 | let esc_val = (1 << bits_per_sect) - 1; |
| 921 | let tot_bands = winfo.get_tot_bands(); |
| 922 | |
| 923 | let mut cur_band = 0; |
| 924 | while cur_band < tot_bands { |
| 925 | let codebook = br.read(4)? as u8; |
| 926 | let mut run = 1; |
| 927 | loop { |
| 928 | let run_add = br.read(bits_per_sect)? as usize; |
| 929 | run += run_add; |
| 930 | if run_add != esc_val { break; } |
| 931 | } |
| 932 | validate!(cur_band + run <= tot_bands); |
| 933 | for _ in 0..run { |
| 934 | cbs[cur_band] = codebook; |
| 935 | cur_band += 1; |
| 936 | } |
| 937 | } |
| 938 | Ok(()) |
| 939 | } |
| 940 | |
| 941 | fn dequant(val: i16, qtab: &[f32; QTAB_SIZE], scale: f32) -> f32 { |
| 942 | if val >= 0 { |
| 943 | qtab[val as usize] * scale |
| 944 | } else { |
| 945 | -qtab[val.abs() as usize] * scale |
| 946 | } |
| 947 | } |
| 948 | |
| 949 | fn decode_quads(br: &mut BitReader, cb: &Codebook<u16>, qtab: &[f32; QTAB_SIZE], scale: f32, dst: &mut [f32]) -> DecoderResult<()> { |
| 950 | for quad in dst.chunks_exact_mut(4) { |
| 951 | let mut idx = br.read_cb(cb)? as i16; |
| 952 | for el in quad.iter_mut() { |
| 953 | *el = dequant(idx >> 12, qtab, scale); |
| 954 | idx <<= 4; |
| 955 | } |
| 956 | } |
| 957 | Ok(()) |
| 958 | } |
| 959 | |
| 960 | fn decode_esc_val(br: &mut BitReader, qtab: &[f32; QTAB_SIZE], scale: f32, sign: bool) -> DecoderResult<f32> { |
| 961 | let pfx = br.read_code(UintCodeType::LimitedOnes(24))? + 4; |
| 962 | let add = br.read(pfx as u8)? as usize; |
| 963 | let val = (1 << pfx) + add; |
| 964 | let fval = scale * if val < qtab.len() { |
| 965 | qtab[val] |
| 966 | } else { |
| 967 | (val as f32) * (val as f32).sqrt() |
| 968 | }; |
| 969 | if !sign { |
| 970 | Ok(fval) |
| 971 | } else { |
| 972 | Ok(-fval) |
| 973 | } |
| 974 | } |
| 975 | |
| 976 | fn decode_pairs(br: &mut BitReader, cb: &Codebook<u16>, qtab: &[f32; QTAB_SIZE], scale: f32, dst: &mut [f32], is_esc: bool) -> DecoderResult<()> { |
| 977 | for pair in dst.chunks_exact_mut(2) { |
| 978 | let idx = br.read_cb(cb)? as i16; |
| 979 | let idx0 = idx >> 8; |
| 980 | let idx1 = (idx << 8) >> 8; |
| 981 | if !is_esc || (idx0.abs() < 16) { |
| 982 | pair[0] = dequant(idx0, qtab, scale); |
| 983 | } else { |
| 984 | pair[0] = decode_esc_val(br, qtab, scale, idx0 < 0)?; |
| 985 | } |
| 986 | if !is_esc || (idx1.abs() < 16) { |
| 987 | pair[1] = dequant(idx1, qtab, scale); |
| 988 | } else { |
| 989 | pair[1] = decode_esc_val(br, qtab, scale, idx1 < 0)?; |
| 990 | } |
| 991 | } |
| 992 | Ok(()) |
| 993 | } |
| 994 | |
| 995 | impl NADecoder for AVCDecoder { |
| 996 | fn init(&mut self, _supp: &mut NADecoderSupport, info: NACodecInfoRef) -> DecoderResult<()> { |
| 997 | if let NACodecTypeInfo::Audio(ainfo) = info.get_properties() { |
| 998 | self.chmap = match ainfo.get_channels() { |
| 999 | 1 => { NAChannelMap::from_str("C").unwrap() }, |
| 1000 | 2 => { NAChannelMap::from_str("L,R").unwrap() }, |
| 1001 | _ => { return Err(DecoderError::InvalidData); }, |
| 1002 | }; |
| 1003 | let srate = ainfo.get_sample_rate(); |
| 1004 | self.ainfo = NAAudioInfo::new(srate, ainfo.get_channels(), |
| 1005 | SND_F32P_FORMAT, COEFFS); |
| 1006 | self.info = info.replace_info(NACodecTypeInfo::Audio(self.ainfo)); |
| 1007 | self.channels = ainfo.get_channels() as usize; |
| 1008 | self.is_40khz = srate <= 40000; |
| 1009 | |
| 1010 | self.win_long = if (srate > 24000) && ((srate > 32000) || (self.channels == 2)) { |
| 1011 | AVC_WIN_LONG_32K |
| 1012 | } else { |
| 1013 | AVC_WIN_LONG_24K |
| 1014 | }; |
| 1015 | |
| 1016 | Ok(()) |
| 1017 | } else { |
| 1018 | Err(DecoderError::InvalidData) |
| 1019 | } |
| 1020 | } |
| 1021 | fn decode(&mut self, _supp: &mut NADecoderSupport, pkt: &NAPacket) -> DecoderResult<NAFrameRef> { |
| 1022 | let info = pkt.get_stream().get_info(); |
| 1023 | validate!(info.get_properties().is_audio()); |
| 1024 | let src = pkt.get_buffer(); |
| 1025 | |
| 1026 | let abuf; |
| 1027 | |
| 1028 | if self.version == 500 { |
| 1029 | abuf = alloc_audio_buffer(self.ainfo, COEFFS, self.chmap.clone())?; |
| 1030 | let mut adata = abuf.get_abuf_f32().unwrap(); |
| 1031 | let mut br = BitReader::new(src.as_slice(), BitReaderMode::BE); |
| 1032 | self.decode_frame(&mut br, &mut adata, 0)?; |
| 1033 | } else { |
| 1034 | let mut offsets: Vec<usize> = Vec::new(); |
| 1035 | let mut sizes: Vec<usize> = Vec::new(); |
| 1036 | |
| 1037 | let mut cur_off = 0; |
| 1038 | while cur_off + 2 < src.len() { |
| 1039 | let sz = read_u16le(&src[cur_off..])? as usize; |
| 1040 | validate!(sz > 0); |
| 1041 | validate!(cur_off + sz + 2 <= src.len()); |
| 1042 | offsets.push(cur_off + 2); |
| 1043 | sizes.push(sz); |
| 1044 | cur_off += sz + 2; |
| 1045 | } |
| 1046 | |
| 1047 | validate!(!sizes.is_empty()); |
| 1048 | |
| 1049 | abuf = alloc_audio_buffer(self.ainfo, COEFFS * sizes.len(), self.chmap.clone())?; |
| 1050 | let mut adata = abuf.get_abuf_f32().unwrap(); |
| 1051 | let mut aoffset = 0; |
| 1052 | for (o, s) in offsets.iter().zip(sizes.iter()) { |
| 1053 | let mut br = BitReader::new(&src[*o..][..*s], BitReaderMode::BE); |
| 1054 | self.decode_frame(&mut br, &mut adata, aoffset)?; |
| 1055 | aoffset += COEFFS; |
| 1056 | } |
| 1057 | } |
| 1058 | |
| 1059 | let mut frm = NAFrame::new_from_pkt(pkt, self.info.clone(), abuf); |
| 1060 | frm.set_keyframe(true); |
| 1061 | Ok(frm.into_ref()) |
| 1062 | } |
| 1063 | fn flush(&mut self) { |
| 1064 | } |
| 1065 | } |
| 1066 | |
| 1067 | impl NAOptionHandler for AVCDecoder { |
| 1068 | fn get_supported_options(&self) -> &[NAOptionDefinition] { &[] } |
| 1069 | fn set_options(&mut self, _options: &[NAOption]) { } |
| 1070 | fn query_option_value(&self, _name: &str) -> Option<NAValue> { None } |
| 1071 | } |
| 1072 | |
| 1073 | pub fn get_decoder_500() -> Box<dyn NADecoder + Send> { |
| 1074 | Box::new(AVCDecoder::new(500)) |
| 1075 | } |
| 1076 | |
| 1077 | pub fn get_decoder_501() -> Box<dyn NADecoder + Send> { |
| 1078 | Box::new(AVCDecoder::new(501)) |
| 1079 | } |
| 1080 | |
| 1081 | #[cfg(test)] |
| 1082 | mod test { |
| 1083 | use nihav_core::codecs::RegisteredDecoders; |
| 1084 | use nihav_core::demuxers::RegisteredDemuxers; |
| 1085 | use nihav_codec_support::test::dec_video::*; |
| 1086 | use crate::duck_register_all_decoders; |
| 1087 | use nihav_commonfmt::generic_register_all_demuxers; |
| 1088 | #[test] |
| 1089 | fn test_avc() { |
| 1090 | let mut dmx_reg = RegisteredDemuxers::new(); |
| 1091 | generic_register_all_demuxers(&mut dmx_reg); |
| 1092 | let mut dec_reg = RegisteredDecoders::new(); |
| 1093 | duck_register_all_decoders(&mut dec_reg); |
| 1094 | |
| 1095 | //let file = "assets/Duck/Cell-140.vp5"; |
| 1096 | //let file = "assets/Duck/Chocolat-500.vp5"; |
| 1097 | let file = "assets/Duck/potter-500.vp7"; |
| 1098 | test_decode_audio("avi", file, Some(1500), None/*Some("avc")*/, &dmx_reg, &dec_reg); |
| 1099 | } |
| 1100 | } |
| 1101 | |
| 1102 | const AVC_WINFO_LONG: WinInfo = WinInfo { |
| 1103 | is_long: true, |
| 1104 | bands: &[ |
| 1105 | 4, 8, 12, 16, 20, 24, 28, 32, |
| 1106 | 36, 40, 48, 56, 64, 72, 80, 88, |
| 1107 | 96, 108, 120, 132, 144, 156, 172, 188, |
| 1108 | 204, 224, 244, 264, 288, 312, 340, 368, |
| 1109 | 400, 432, 464, 496, 528, 560, 592, 624, |
| 1110 | 656, 688, 720, 752, 784, 816, 848, 880, 1024 |
| 1111 | ], |
| 1112 | }; |
| 1113 | |
| 1114 | const AVC_WINFO_SHORT: WinInfo = WinInfo { |
| 1115 | is_long: false, |
| 1116 | bands: &[ 4, 8, 12, 16, 20, 24, 32, 40, 48, 56, 68, 80, 108, 128 ], |
| 1117 | }; |
| 1118 | |
| 1119 | const AVC_WINFO_40K_MODE4: WinInfo = WinInfo { |
| 1120 | is_long: true, |
| 1121 | bands: &[ |
| 1122 | 8, 16, 24, 32, 40, 48, 56, 64, |
| 1123 | 72, 80, 88, 96, 104, 112, 120, 128, |
| 1124 | 144, 160, 176, 192, 208, 224, 240, 256, |
| 1125 | 264, 272, 280, 288, 296, 304, 312, 320, |
| 1126 | 328, 336, 344, 352, 360, 368, 376, 384, |
| 1127 | 400, 416, 432, 448, 464, 480, 496, 512, |
| 1128 | 520, 528, 536, 544, 552, 560, 568, 576, |
| 1129 | 584, 592, 600, 608, 616, 624, 632, 640, |
| 1130 | 648, 656, 664, 672, 680, 688, 696, 704, |
| 1131 | 712, 720, 728, 736, 744, 752, 760, 768, |
| 1132 | 800, 832, 864, 896, 928, 960, 992, 1024 |
| 1133 | ], |
| 1134 | }; |
| 1135 | const AVC_WINFO_40K_MODE5: WinInfo = WinInfo { |
| 1136 | is_long: true, |
| 1137 | bands: &[ |
| 1138 | 8, 16, 24, 32, 40, 48, 56, 64, |
| 1139 | 80, 96, 112, 128, 136, 144, 152, 160, |
| 1140 | 168, 176, 184, 192, 208, 224, 240, 256, |
| 1141 | 264, 272, 280, 288, 296, 304, 312, 320, |
| 1142 | 328, 336, 344, 352, 360, 368, 376, 384, |
| 1143 | 416, 448, 480, 512, 516, 520, 524, 528, |
| 1144 | 532, 536, 540, 548, 556, 568, 580, 592, |
| 1145 | 608, 624, 640, 656, 672, 688, 704, 720, |
| 1146 | 736, 752, 768, 784, 800, 816, 832, 848, |
| 1147 | 864, 880, 896, 912, 928, 944, 1024 |
| 1148 | ], |
| 1149 | }; |
| 1150 | const AVC_WINFO_40K_MODE6: WinInfo = WinInfo { |
| 1151 | is_long: true, |
| 1152 | bands: &[ |
| 1153 | 4, 8, 12, 16, 20, 24, 28, 36, |
| 1154 | 44, 56, 68, 80, 96, 112, 128, 144, |
| 1155 | 160, 176, 192, 208, 224, 240, 256, 272, |
| 1156 | 288, 304, 320, 336, 352, 368, 384, 400, |
| 1157 | 416, 432, 512, 520, 528, 536, 544, 552, |
| 1158 | 560, 568, 576, 592, 608, 624, 640, 648, |
| 1159 | 656, 664, 672, 680, 688, 696, 704, 720, |
| 1160 | 736, 752, 768, 776, 784, 792, 800, 808, |
| 1161 | 816, 824, 832, 840, 848, 856, 864, 872, |
| 1162 | 880, 888, 896, 928, 960, 992, 1024 |
| 1163 | ], |
| 1164 | }; |
| 1165 | |
| 1166 | const AVC_WINFO_44K_MODE4: WinInfo = WinInfo { |
| 1167 | is_long: true, |
| 1168 | bands: &[ |
| 1169 | 8, 16, 24, 32, 40, 48, 56, 64, |
| 1170 | 72, 80, 88, 96, 104, 112, 120, 128, |
| 1171 | 136, 144, 152, 160, 168, 176, 184, 192, |
| 1172 | 200, 208, 216, 224, 232, 240, 248, 256, |
| 1173 | 264, 272, 280, 288, 296, 304, 312, 320, |
| 1174 | 328, 336, 344, 352, 360, 368, 376, 384, |
| 1175 | 392, 400, 408, 416, 424, 432, 440, 448, |
| 1176 | 456, 464, 472, 480, 488, 496, 504, 512, |
| 1177 | 528, 544, 560, 576, 592, 608, 624, 640, |
| 1178 | 672, 704, 736, 768, 832, 896, 960, 1024 |
| 1179 | ], |
| 1180 | }; |
| 1181 | const AVC_WINFO_44K_MODE5: WinInfo = WinInfo { |
| 1182 | is_long: true, |
| 1183 | bands: &[ |
| 1184 | 8, 16, 24, 32, 40, 48, 56, 64, |
| 1185 | 72, 80, 88, 96, 104, 112, 120, 128, |
| 1186 | 136, 144, 152, 160, 168, 176, 184, 192, |
| 1187 | 200, 208, 216, 224, 232, 240, 248, 256, |
| 1188 | 272, 288, 304, 320, 352, 384, 448, 512, |
| 1189 | 516, 520, 524, 528, 532, 536, 540, 548, |
| 1190 | 556, 568, 580, 592, 608, 624, 640, 656, |
| 1191 | 672, 688, 704, 720, 736, 752, 768, 784, |
| 1192 | 800, 816, 832, 848, 864, 880, 896, 912, |
| 1193 | 928, 944, 1024 |
| 1194 | ], |
| 1195 | }; |
| 1196 | const AVC_WINFO_44K_MODE6: WinInfo = WinInfo { |
| 1197 | is_long: true, |
| 1198 | bands: &[ |
| 1199 | 4, 8, 12, 16, 20, 24, 28, 36, |
| 1200 | 44, 56, 68, 80, 96, 112, 128, 144, |
| 1201 | 160, 176, 192, 208, 224, 240, 256, 272, |
| 1202 | 288, 304, 320, 336, 352, 368, 384, 400, |
| 1203 | 416, 432, 512, 520, 528, 536, 544, 552, |
| 1204 | 560, 568, 576, 584, 592, 600, 608, 616, |
| 1205 | 624, 632, 640, 648, 656, 664, 672, 680, |
| 1206 | 688, 696, 704, 712, 720, 728, 736, 744, |
| 1207 | 752, 760, 768, 784, 800, 816, 832, 864, |
| 1208 | 896, 960, 1024 |
| 1209 | ], |
| 1210 | }; |