Implement RGBA output for images without alpha, as well as a rewind() function.

Also add a fuzzer, and fix a few of fuzzer-found crashes / infinite loops / unexpected slowness, and fix a case of UB-under-stacked-borrows due to using .as_ptr() instead .as_mut_ptr().

What's Changed

• Add jxl-perfhistory benchmark CI stage by @hjanuschka in #540
• Added extra workflow for benchmark comments by @zond in #542
• Fix (or silence) clippy warnings/errors in jxl_transforms by @tirr-c in #545
• Added efb.jxl from #512 by @zond in #534
• Split out the AUTHORS check so we can make it optional. by @veluca93 in #554
• Edit existing benchmark comments to reduce noise by @FooIbar in #551
• slice_from_cachelines_mut: fix invalid *const T to *mut T cast by @Mrmaxmeier in #557
• Add simple fuzzers via cargo-fuzz (decode, decode_header) by @Mrmaxmeier in #527
• Fix various bugs discovered by fuzzing. by @veluca93 in #559
• Fix more fuzzer-found issues. by @veluca93 in #560
• Fix issues with parsing of permuted TOCs by @veluca93 in #561
• Another batch of fuzzer fixes. by @veluca93 in #562
• Make section handling non-quadratic. by @veluca93 in #564
• Fix computation of output rectangle in case of non0 origin. by @veluca93 in #563
• Add FFI-friendly API and convenience methods for Chromium integration by @hjanuschka in #556
• Update to v0.1.5 by @veluca93 in #565

New Contributors

• @FooIbar made their first contribution in #551
• @Mrmaxmeier made their first contribution in #557

Full Changelog: v0.1.4...v0.1.5

This release brings significant binary size reductions.

What's Changed

• Do not build the simple pipeline outside of tests. by @veluca93 in #522
• Skipping zero size buffers by @zond in #523
• Improve weighted predictor cache locality by @hjanuschka in #526
• Precompute cosines in spline rendering by @hjanuschka in #524
• Made pipeline stages use flat buffers by @zond in #515
• Reduce code size with a generic transform impl for large transforms. by @veluca93 in #528
• Bump version to 0.1.4. by @veluca93 in #541

Full Changelog: v0.1.3...v0.1.4

What's Changed

• bt709 based on sRGB by @zond in #501
• Exclude test images on publish by @tirr-c in #502
• Symlink LICENSE files. by @veluca93 in #505
• Fewer allocations by @zond in #511
• Improve code size. by @veluca93 in #517
• Address comments from safety review. by @veluca93 in #518
• Bump version to 0.1.3. by @veluca93 in #519

Full Changelog: v0.1.2...v0.1.3

This release fixes a few decoding corner cases, removes unneeded dependencies, significantly speeds up decoding of many common images, and splits the jxl crate into a few supporting crates.

What's Changed

• Fix Cargo.toml for crates.io by @veluca93 in #377
• Use JxlOutputBuffer in SaveStage by @veluca93 in #378
• Improved render stage tests by @zond in #371
• Replaced cases of {x}.rs and {x}/ with a {x}/mod.rs file. by @zond in #379
• Removed a redundant TODO (the task is already done). by @zond in #382
• Made the API produce x/y size post transposing. by @zond in #381
• Split parts of frame/mod.rs into decode.rs and render.rs. by @zond in #380
• Updated jxlinspect to use the API. by @zond in #384
• Prepared for user-provided CMS integration. by @zond in #385
• make nearest_neighbor #[cfg(test)] by @mo271 in #386
• Expose intensity target in JxlBasicInfo by @sboukortt in #388
• Skip computing the output of the weighted predictor if unused by @sboukortt in #389
• Fix check by @sboukortt in #390
• Removed spurious debug output. by @zond in #392
• Added a test of the coeff order function. by @zond in #391
• Backport int_to_float fix from libjxl/libjxl#4461 by @sboukortt in #393
• Allow casting from a avx512 descriptor to a avx descriptor. by @veluca93 in #395
• Prepare for SIMD DCT by @sboukortt in #394
• Add a few inline annotations by @sboukortt in #397
• Less copying around in DCT by @sboukortt in #399
• DCT: fewer allocations by @sboukortt in #400
• To/FromLinear: for gamma, copy sign for negative values by @sboukortt in #401
• Add call method to SimdDescriptor trait for feature context pattern by @zond in #403
• Use padded frame size in PatchesDictionary::read by @mo271 in #405
• test image for patches out of bound by @mo271 in #406
• Cleaned up changes optimizing dct simd performance by @zond in #407
• Implement the low-memory render pipeline by @veluca93 in #408
• Unify the JxlOutputBuffer type with the Image type. by @veluca93 in #409
• Speed up Gaborish. by @veluca93 in #412
• Type-erased low_memory_pipeline and start SIMD save. by @veluca93 in #411
• SIMDfy dequant_lanes by @sboukortt in #398
• Make the encoders write to writers, instead of buffering bytes themselves by @sboukortt in #414
• Fix build with exr disabled by @sboukortt in #416
• Move SIMD to a new crate and set up better bench for dct. by @veluca93 in #413
• Fix modular buffer grid kind of unsqueeze output. by @szabadka in #415
• Simplify scalar SIMD implementation by @tirr-c in #417
• test image for alpha with squeeze by @mo271 in #410
• Fix assertions on group size for chroma subsampling. by @veluca93 in #419
• update readme by @mo271 in #420
• Correct Orientation enum naming and logic for 90-degree rotations by @mo271 in #427
• SIMDfy the gamma tf in From/ToLinear by @sboukortt in #418
• jxlinspect: milliseconds instead of seconds by @mo271 in #435
• Avoid .round() in PPM/PNG output by @sboukortt in #432
• SIMDfy sRGB by @sboukortt in #433
• add animation delay to apng encoder by @mo271 in #434
• Fix local MA tree node limit by @tirr-c in #439
• Add some #[inline] annotations. by @veluca93 in #440
• Fix max property for references by @LouisDISPA in #438
• Fix color channel count when calculating global MA tree limit by @tirr-c in #442
• Use benchmark groups to group by DCT sizes and compare implementations by @tirr-c in #443
• Fix panic when PassGroup comes before HfGlobal by @tirr-c in #444
• Added --num-reps to jxl_cli speedtest. by @zond in #446
• Fix padding rectangle calculation in extend stage. by @szabadka in #448
• add jxl art test image by @mo271 in #426
• Made frame indices seed the rng by @zond in #450
• Don't use the simple pipeline. by @veluca93 in #454
• Change IDCT impl and add a Sse4 SIMD descriptor. by @veluca93 in #453
• Prepare for generated reinterpreting_dct: change tests and bench. by @veluca93 in #455
• Generated reinterpreting_dct. by @veluca93 in #456
• Correct channel index for 420 squeeze by @zond in #457
• Fix odd-size images with upsampling and add a test. by @veluca93 in #458
• Get rid of SIGMA_PADDING. by @veluca93 in #459
• Fix images with empty squeeze residual channels. by @szabadka in #460
• Fix downsampling_bracket function. by @szabadka in #461
• Improve loading of sigma in epf. by @veluca93 in #462
• Tweak vardct decoding to improve speed. by @veluca93 in #463
• Add SIMD implementation for NEON by @tirr-c in #441
• Feature gate SIMD implementations by @tirr-c in #465
• Perform static dispatch when target feature is enabled at compile time by @tirr-c in #466
• Fix benchmarks not running except for the topmost instruction set by @tirr-c in #467
• SIMDify horizontal unsqueeze by @tirr-c in #468
• Let Images have a "default rect". by @veluca93 in #464
• Render groups eagerly (remove explicit do_render calls). by @veluca93 in #470
• Avoiding br.peek when unnecessary by @zond in #471
• Fix border handling for inner groups in the low memory pipeline. by @szabadka in #472
• SIMDify vertical unsqueeze by @tirr-c in #469
• Cleaned up speedtest by @zond in #474
• Made AnsHistogram::read branchless by @zond in #473
• Actually free buffers that will not be used again. by @veluca93 in #475
• Made bit reader refill conditional by @zond in #476
• Incremental TOC reader by @tirr-c in #477
• Speedup modular decoding. by @veluca93 in #478
• Add infrastructure for decoder special paths for Modular mode. by @veluca93 in #479
• Update and clean up dependencies by @sboukortt in #480
• Add RleSymbolReader by @tirr-c in #481
• Implement fast-lossless path by @tirr-c in #482
• Make SymbolReader optimistic by @tirr-c in #483
• Tighten the size of some buffers. by @veluca93 in #485
• SIMDify RCT by @tirr-c in #484
• Added criterion bench for decoding by @zond in #487
• Made the codestream parser retry when necessary by @zond in #488
• Reader instead of slice by @zond in #489
• Remove zerocopy from production dependencies by @hjanuschka in #492
• Add preview frame API support by @hjanuschka in #495
• Fix spline DISTANCE_EXP to match libjxl default by @hjanuschka in #494
• feat: Implement HDR tone mapping for ICC profile generation by @hjanuschka in #491
• Relax accuracy requirements for FMA tests. by @veluca93 in #498
• Fix rect bounds check for padded images + add alpha boundary test by @hjan...

What's Changed

Remove some unnecessary dependencies and clarify use of std::fs is for tests only.

Full Changelog: v0.1.0...v0.1.1

Very preliminary release. The code works on most images, but extensive testing is still needed and it is not especially fast.