| 1 | //! Helpers for taking a slice of indices (indices into `PLTE` and/or `trNS` |
| 2 | //! entries) and transforming this into RGB or RGBA output. |
| 3 | //! |
| 4 | //! # Memoization |
| 5 | //! |
| 6 | //! To achieve higher throughput, `create_rgba_palette` combines entries from |
| 7 | //! `PLTE` and `trNS` chunks into a single lookup table. This is based on the |
| 8 | //! ideas explored in <https://crbug.com/706134>. |
| 9 | //! |
| 10 | //! Memoization is a trade-off: |
| 11 | //! * On one hand, memoization requires spending X ns before starting to call |
| 12 | //! `expand_paletted_...` functions. |
| 13 | //! * On the other hand, memoization improves the throughput of the |
| 14 | //! `expand_paletted_...` functions - they take Y ns less to process each byte |
| 15 | //! |
| 16 | //! Based on X and Y, we can try to calculate the breakeven point. It seems |
| 17 | //! that memoization is a net benefit for images bigger than around 13x13 pixels. |
| 18 | |
| 19 | use super::{unpack_bits, TransformFn}; |
| 20 | use crate::{BitDepth, Info}; |
| 21 | |
| 22 | pub fn create_expansion_into_rgb8(info: &Info) -> TransformFn { |
| 23 | let rgba_palette: [[u8; 4]; 256] = create_rgba_palette(info); |
| 24 | |
| 25 | if info.bit_depth == BitDepth::Eight { |
| 26 | Box::new(move |input: &[u8], output: &mut [u8], _info: &Info<'_>| expand_8bit_into_rgb8(input, output, &rgba_palette)) |
| 27 | } else { |
| 28 | Box::new(move |input: &[u8], output: &mut [u8], info: &Info<'_>| expand_into_rgb8(row:input, buffer:output, info, &rgba_palette)) |
| 29 | } |
| 30 | } |
| 31 | |
| 32 | pub fn create_expansion_into_rgba8(info: &Info) -> TransformFn { |
| 33 | let rgba_palette: [[u8; 4]; 256] = create_rgba_palette(info); |
| 34 | Box::new(move |input: &[u8], output: &mut [u8], info: &Info<'_>| { |
| 35 | expand_paletted_into_rgba8(row:input, buffer:output, info, &rgba_palette) |
| 36 | }) |
| 37 | } |
| 38 | |
| 39 | fn create_rgba_palette(info: &Info) -> [[u8; 4]; 256] { |
| 40 | let palette = info.palette.as_deref().expect("Caller should verify" ); |
| 41 | let trns = info.trns.as_deref().unwrap_or(&[]); |
| 42 | |
| 43 | // > The tRNS chunk shall not contain more alpha values than there are palette |
| 44 | // entries, but a tRNS chunk may contain fewer values than there are palette |
| 45 | // entries. In this case, the alpha value for all remaining palette entries is |
| 46 | // assumed to be 255. |
| 47 | // |
| 48 | // It seems, accepted reading is to fully *ignore* an invalid tRNS as if it were |
| 49 | // completely empty / all pixels are non-transparent. |
| 50 | let trns = if trns.len() <= palette.len() / 3 { |
| 51 | trns |
| 52 | } else { |
| 53 | &[] |
| 54 | }; |
| 55 | |
| 56 | // Default to black, opaque entries. |
| 57 | let mut rgba_palette = [[0, 0, 0, 0xFF]; 256]; |
| 58 | |
| 59 | // Copy `palette` (RGB) entries into `rgba_palette`. This may clobber alpha |
| 60 | // values in `rgba_palette` - we need to fix this later. |
| 61 | { |
| 62 | let mut palette_iter = palette; |
| 63 | let mut rgba_iter = &mut rgba_palette[..]; |
| 64 | while palette_iter.len() >= 4 { |
| 65 | // Copying 4 bytes at a time is more efficient than copying 3. |
| 66 | // OTOH, this clobbers the alpha value in `rgba_iter[0][3]` - we |
| 67 | // need to fix this later. |
| 68 | rgba_iter[0].copy_from_slice(&palette_iter[0..4]); |
| 69 | |
| 70 | palette_iter = &palette_iter[3..]; |
| 71 | rgba_iter = &mut rgba_iter[1..]; |
| 72 | } |
| 73 | if !palette_iter.is_empty() { |
| 74 | rgba_iter[0][0..3].copy_from_slice(&palette_iter[0..3]); |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | // Copy `trns` (alpha) entries into `rgba_palette`. `trns.len()` may be |
| 79 | // smaller than `palette.len()` and therefore this is not sufficient to fix |
| 80 | // all the clobbered alpha values. |
| 81 | for (alpha, rgba) in trns.iter().copied().zip(rgba_palette.iter_mut()) { |
| 82 | rgba[3] = alpha; |
| 83 | } |
| 84 | |
| 85 | // Unclobber the remaining alpha values. |
| 86 | for rgba in rgba_palette[trns.len()..(palette.len() / 3)].iter_mut() { |
| 87 | rgba[3] = 0xFF; |
| 88 | } |
| 89 | |
| 90 | rgba_palette |
| 91 | } |
| 92 | |
| 93 | fn expand_8bit_into_rgb8(mut input: &[u8], mut output: &mut [u8], rgba_palette: &[[u8; 4]; 256]) { |
| 94 | while output.len() >= 4 { |
| 95 | // Copying 4 bytes at a time is more efficient than 3. |
| 96 | let rgba: &[u8; 4] = &rgba_palette[input[0] as usize]; |
| 97 | output[0..4].copy_from_slice(src:rgba); |
| 98 | |
| 99 | input = &input[1..]; |
| 100 | output = &mut output[3..]; |
| 101 | } |
| 102 | if !output.is_empty() { |
| 103 | let rgba: &[u8; 4] = &rgba_palette[input[0] as usize]; |
| 104 | output[0..3].copy_from_slice(&rgba[0..3]); |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | fn expand_into_rgb8(row: &[u8], buffer: &mut [u8], info: &Info, rgba_palette: &[[u8; 4]; 256]) { |
| 109 | unpack_bits(input:row, output:buffer, channels:3, info.bit_depth as u8, |i: u8, chunk: &mut [u8]| { |
| 110 | let rgba: &[u8; 4] = &rgba_palette[i as usize]; |
| 111 | chunk[0] = rgba[0]; |
| 112 | chunk[1] = rgba[1]; |
| 113 | chunk[2] = rgba[2]; |
| 114 | }) |
| 115 | } |
| 116 | |
| 117 | fn expand_paletted_into_rgba8( |
| 118 | row: &[u8], |
| 119 | buffer: &mut [u8], |
| 120 | info: &Info, |
| 121 | rgba_palette: &[[u8; 4]; 256], |
| 122 | ) { |
| 123 | unpack_bits(input:row, output:buffer, channels:4, info.bit_depth as u8, |i: u8, chunk: &mut [u8]| { |
| 124 | chunk.copy_from_slice(&rgba_palette[i as usize]); |
| 125 | }); |
| 126 | } |
| 127 | |
| 128 | #[cfg (test)] |
| 129 | mod test { |
| 130 | use crate::{BitDepth, ColorType, Info, Transformations}; |
| 131 | |
| 132 | /// Old, non-memoized version of the code is used as a test oracle. |
| 133 | fn oracle_expand_paletted_into_rgb8(row: &[u8], buffer: &mut [u8], info: &Info) { |
| 134 | let palette = info.palette.as_deref().expect("Caller should verify" ); |
| 135 | let black = [0, 0, 0]; |
| 136 | |
| 137 | super::unpack_bits(row, buffer, 3, info.bit_depth as u8, |i, chunk| { |
| 138 | let rgb = palette |
| 139 | .get(3 * i as usize..3 * i as usize + 3) |
| 140 | .unwrap_or(&black); |
| 141 | chunk[0] = rgb[0]; |
| 142 | chunk[1] = rgb[1]; |
| 143 | chunk[2] = rgb[2]; |
| 144 | }) |
| 145 | } |
| 146 | |
| 147 | /// Old, non-memoized version of the code is used as a test oracle. |
| 148 | fn oracle_expand_paletted_into_rgba8(row: &[u8], buffer: &mut [u8], info: &Info) { |
| 149 | let palette = info.palette.as_deref().expect("Caller should verify" ); |
| 150 | let trns = info.trns.as_deref().unwrap_or(&[]); |
| 151 | let black = [0, 0, 0]; |
| 152 | |
| 153 | // > The tRNS chunk shall not contain more alpha values than there are palette |
| 154 | // entries, but a tRNS chunk may contain fewer values than there are palette |
| 155 | // entries. In this case, the alpha value for all remaining palette entries is |
| 156 | // assumed to be 255. |
| 157 | // |
| 158 | // It seems, accepted reading is to fully *ignore* an invalid tRNS as if it were |
| 159 | // completely empty / all pixels are non-transparent. |
| 160 | let trns = if trns.len() <= palette.len() / 3 { |
| 161 | trns |
| 162 | } else { |
| 163 | &[] |
| 164 | }; |
| 165 | |
| 166 | super::unpack_bits(row, buffer, 4, info.bit_depth as u8, |i, chunk| { |
| 167 | let (rgb, a) = ( |
| 168 | palette |
| 169 | .get(3 * i as usize..3 * i as usize + 3) |
| 170 | .unwrap_or(&black), |
| 171 | *trns.get(i as usize).unwrap_or(&0xFF), |
| 172 | ); |
| 173 | chunk[0] = rgb[0]; |
| 174 | chunk[1] = rgb[1]; |
| 175 | chunk[2] = rgb[2]; |
| 176 | chunk[3] = a; |
| 177 | }); |
| 178 | } |
| 179 | |
| 180 | fn create_info<'a>(src_bit_depth: u8, palette: &'a [u8], trns: Option<&'a [u8]>) -> Info<'a> { |
| 181 | Info { |
| 182 | color_type: ColorType::Indexed, |
| 183 | bit_depth: BitDepth::from_u8(src_bit_depth).unwrap(), |
| 184 | palette: Some(palette.into()), |
| 185 | trns: trns.map(Into::into), |
| 186 | ..Info::default() |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | fn expand_paletted( |
| 191 | src: &[u8], |
| 192 | src_bit_depth: u8, |
| 193 | palette: &[u8], |
| 194 | trns: Option<&[u8]>, |
| 195 | ) -> Vec<u8> { |
| 196 | let info = create_info(src_bit_depth, palette, trns); |
| 197 | let output_bytes_per_input_sample = match trns { |
| 198 | None => 3, |
| 199 | Some(_) => 4, |
| 200 | }; |
| 201 | let samples_count_per_byte = (8 / src_bit_depth) as usize; |
| 202 | let samples_count = src.len() * samples_count_per_byte; |
| 203 | |
| 204 | let mut dst = vec![0; samples_count * output_bytes_per_input_sample]; |
| 205 | let transform_fn = |
| 206 | super::super::create_transform_fn(&info, Transformations::EXPAND).unwrap(); |
| 207 | transform_fn(src, dst.as_mut_slice(), &info); |
| 208 | |
| 209 | { |
| 210 | // Compare the memoization-based calculations with the old, non-memoized code. |
| 211 | let mut simple_dst = vec![0; samples_count * output_bytes_per_input_sample]; |
| 212 | if trns.is_none() { |
| 213 | oracle_expand_paletted_into_rgb8(src, &mut simple_dst, &info) |
| 214 | } else { |
| 215 | oracle_expand_paletted_into_rgba8(src, &mut simple_dst, &info) |
| 216 | } |
| 217 | assert_eq!(&dst, &simple_dst); |
| 218 | } |
| 219 | |
| 220 | dst |
| 221 | } |
| 222 | |
| 223 | #[test ] |
| 224 | fn test_expand_paletted_rgba_8bit() { |
| 225 | let actual = expand_paletted( |
| 226 | &[0, 1, 2, 3], // src |
| 227 | 8, // src_bit_depth |
| 228 | &[ |
| 229 | // palette |
| 230 | 0, 1, 2, // entry #0 |
| 231 | 4, 5, 6, // entry #1 |
| 232 | 8, 9, 10, // entry #2 |
| 233 | 12, 13, 14, // entry #3 |
| 234 | ], |
| 235 | Some(&[3, 7, 11, 15]), // trns |
| 236 | ); |
| 237 | assert_eq!(actual, (0..16).collect::<Vec<u8>>()); |
| 238 | } |
| 239 | |
| 240 | #[test ] |
| 241 | fn test_expand_paletted_rgb_8bit() { |
| 242 | let actual = expand_paletted( |
| 243 | &[0, 1, 2, 3], // src |
| 244 | 8, // src_bit_depth |
| 245 | &[ |
| 246 | // palette |
| 247 | 0, 1, 2, // entry #0 |
| 248 | 3, 4, 5, // entry #1 |
| 249 | 6, 7, 8, // entry #2 |
| 250 | 9, 10, 11, // entry #3 |
| 251 | ], |
| 252 | None, // trns |
| 253 | ); |
| 254 | assert_eq!(actual, (0..12).collect::<Vec<u8>>()); |
| 255 | } |
| 256 | |
| 257 | #[test ] |
| 258 | fn test_expand_paletted_rgba_4bit() { |
| 259 | let actual = expand_paletted( |
| 260 | &[0x01, 0x23], // src |
| 261 | 4, // src_bit_depth |
| 262 | &[ |
| 263 | // palette |
| 264 | 0, 1, 2, // entry #0 |
| 265 | 4, 5, 6, // entry #1 |
| 266 | 8, 9, 10, // entry #2 |
| 267 | 12, 13, 14, // entry #3 |
| 268 | ], |
| 269 | Some(&[3, 7, 11, 15]), // trns |
| 270 | ); |
| 271 | assert_eq!(actual, (0..16).collect::<Vec<u8>>()); |
| 272 | } |
| 273 | |
| 274 | #[test ] |
| 275 | fn test_expand_paletted_rgb_4bit() { |
| 276 | let actual = expand_paletted( |
| 277 | &[0x01, 0x23], // src |
| 278 | 4, // src_bit_depth |
| 279 | &[ |
| 280 | // palette |
| 281 | 0, 1, 2, // entry #0 |
| 282 | 3, 4, 5, // entry #1 |
| 283 | 6, 7, 8, // entry #2 |
| 284 | 9, 10, 11, // entry #3 |
| 285 | ], |
| 286 | None, // trns |
| 287 | ); |
| 288 | assert_eq!(actual, (0..12).collect::<Vec<u8>>()); |
| 289 | } |
| 290 | |
| 291 | #[test ] |
| 292 | fn test_expand_paletted_rgba_8bit_more_trns_entries_than_palette_entries() { |
| 293 | let actual = expand_paletted( |
| 294 | &[0, 1, 2, 3], // src |
| 295 | 8, // src_bit_depth |
| 296 | &[ |
| 297 | // palette |
| 298 | 0, 1, 2, // entry #0 |
| 299 | 4, 5, 6, // entry #1 |
| 300 | 8, 9, 10, // entry #2 |
| 301 | 12, 13, 14, // entry #3 |
| 302 | ], |
| 303 | Some(&[123; 5]), // trns |
| 304 | ); |
| 305 | |
| 306 | // Invalid (too-long) `trns` means that we'll use 0xFF / opaque alpha everywhere. |
| 307 | assert_eq!( |
| 308 | actual, |
| 309 | vec![0, 1, 2, 0xFF, 4, 5, 6, 0xFF, 8, 9, 10, 0xFF, 12, 13, 14, 0xFF], |
| 310 | ); |
| 311 | } |
| 312 | |
| 313 | #[test ] |
| 314 | fn test_expand_paletted_rgba_8bit_less_trns_entries_than_palette_entries() { |
| 315 | let actual = expand_paletted( |
| 316 | &[0, 1, 2, 3], // src |
| 317 | 8, // src_bit_depth |
| 318 | &[ |
| 319 | // palette |
| 320 | 0, 1, 2, // entry #0 |
| 321 | 4, 5, 6, // entry #1 |
| 322 | 8, 9, 10, // entry #2 |
| 323 | 12, 13, 14, // entry #3 |
| 324 | ], |
| 325 | Some(&[3, 7]), // trns |
| 326 | ); |
| 327 | |
| 328 | // Too-short `trns` is treated differently from too-long - only missing entries are |
| 329 | // replaced with 0XFF / opaque. |
| 330 | assert_eq!( |
| 331 | actual, |
| 332 | vec![0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 0xFF, 12, 13, 14, 0xFF], |
| 333 | ); |
| 334 | } |
| 335 | |
| 336 | #[test ] |
| 337 | fn test_create_rgba_palette() { |
| 338 | fn create_expected_rgba_palette(plte: &[u8], trns: &[u8]) -> [[u8; 4]; 256] { |
| 339 | let mut rgba = [[1, 2, 3, 4]; 256]; |
| 340 | for (i, rgba) in rgba.iter_mut().enumerate() { |
| 341 | rgba[0] = plte.get(i * 3 + 0).map(|&r| r).unwrap_or(0); |
| 342 | rgba[1] = plte.get(i * 3 + 1).map(|&g| g).unwrap_or(0); |
| 343 | rgba[2] = plte.get(i * 3 + 2).map(|&b| b).unwrap_or(0); |
| 344 | rgba[3] = trns.get(i * 1 + 0).map(|&a| a).unwrap_or(0xFF); |
| 345 | } |
| 346 | rgba |
| 347 | } |
| 348 | |
| 349 | for plte_len in 1..=32 { |
| 350 | for trns_len in 0..=plte_len { |
| 351 | let plte: Vec<u8> = (0..plte_len * 3).collect(); |
| 352 | let trns: Vec<u8> = (0..trns_len).map(|alpha| alpha + 200).collect(); |
| 353 | |
| 354 | let info = create_info(8, &plte, Some(&trns)); |
| 355 | let expected = create_expected_rgba_palette(&plte, &trns); |
| 356 | let actual = super::create_rgba_palette(&info); |
| 357 | assert_eq!(actual, expected); |
| 358 | } |
| 359 | } |
| 360 | } |
| 361 | } |
| 362 | |