Weekly pre-release

What's Changed

• Disable missing const fn clippy lint by @shssoichiro in #2923

Full Changelog: p20220405...p20220412

Weekly pre-release

What's Changed

• Use 64-bits for calculating 8x8 variance by @shssoichiro in #2919

Full Changelog: p20220329...p20220405

Weekly pre-release

What's Changed

• Remove unnecessary clones from metrics calculation by @shssoichiro in #2913
• Avoid temporary allocation of block importance costs by @shssoichiro in #2915

Full Changelog: p20220322...p20220329

Weekly pre-release

What's Changed

• Move complex pred modes for keyframes to s2 by @shssoichiro in #2906
• Change non-square partition to search below the threshold by @shssoichiro in #2905
• rate: add assertion for q57 input range by @tmatth in #2911

Full Changelog: p20220315...p20220322

Weekly pre-release

What's Changed

• Enable some more useful Clippy lints by @shssoichiro in #2904
• Also apply custom clippy lints to ivf and v_frame by @shssoichiro in #2907
• Enable suboptimal_flops clippy lint by @shssoichiro in #2908
• Fix memory leak when temporal RDO not in use by @shssoichiro in #2909

Full Changelog: p20220308...p20220315

Weekly pre-release

What's Changed

• Fix pyramid and frame type tests by @shssoichiro in #2901
• Improve storage of invalid and SEF frame invariants by @shssoichiro in #2902
• Avoid multiple inlining of residual loop in SAD intrinsics by @redzic in #2903

Full Changelog: p20220301...p20220308

Weekly pre-release

What's Changed

• New clippy fixes from Rust 1.59 by @shssoichiro in #2899
• Fix unnecessary auto-vectorization in residual loop of SAD intrinsics by @redzic in #2897

Full Changelog: p20220222...p20220301

Weekly pre-release

What's Changed

• Fix new deprecation warnings from Clap 3.1.0 by @shssoichiro in #2896

Full Changelog: p20220215...p20220222

Weekly pre-release

What's Changed

• Cache and reuse intra costs from scenechange by @shssoichiro in #2895

Since intra costs are computed completely independently of any information outside the frame itself, these costs are guaranteed not to change for a given input frame.

This also contains what is technically a bugfix in passing the proper input frame number to the run_comparison method in the scenecut detector. However, this bug was unlikely to affect any real-world use cases prior to the introduction of this caching.

Full Changelog: p20220208...p20220215

Weekly pre-release

What's Changed

• Init Motion Est: Refine mvs when subsampleing reduced by @KyleSiefring in #2892

After doubling the sample rate of motion estimation (e.g. quarter-res to half-res), refine the search performed at the lower supsampling level by searching a 4x4 region around the previous mv.

https://beta.arewecompressedyet.com/?job=crop_edge_init_me%402022-01-27T19%3A13%3A43.336Zjob=refine_me_ssdec3%402022-02-03T16%3A23%3A23.724Z

PSNR Y	PSNR Cb	PSNR Cr	CIEDE2000	SSIM	MS-SSIM
-0.0595	-0.2343	0.0195	-0.1685	-0.0728	-0.0919

PSNR-HVS Y	PSNR-HVS Cb	PSNR-HVS Cr	PSNR-HVS	VMAF	VMAF-NEG
-0.0768	-0.2228	0.0468	-0.0770	-0.1966	-0.2165

Full Changelog: p20220201...p20220208