Dominique Leuenberger
16c7299bde
- Update to version 0.6.1+0: * Fix the capi deps * Revert "x86: Disable 8x16 16bpc inverse transforms for SSE4.1" * inverse_transform_add: Align to dav1d EOB convention * x86: Disable 8x16 16bpc inverse transforms for SSE4.1 * CI: Fix up deploy workflow * Add C API for level. * Add level to API and CLI. * CI: Correct format of target_cpu for x86_64 in deploy workflow * Produced optimized binary versions from CI * Write level stored in sequence struct. * Bump actions/download-artifact from 2 to 3 * Bump actions/upload-artifact from 2 to 3 * Bump actions/cache from 2 to 3 * Update deps * Update cargo-c binaries url * Add Dependabot configuration for GitHub Actions updates * tests: Add three-pass encode with target bitrate * Add SIMD helpers to speed up Rust get_sad (#3050) * Replace last 16x4 and 4x16 avx2 hbd satd intrinsics with asm (#3051) * x86: Add AVX2 HBD SATD asm for 8x8-transformed blocks (#3048) * Bump minimum rustc to 1.60, required by clap 4 * Update clap to 4.0 * Remove unused regex dependency * Cleanup log level parsing * Replace actions-rs/toolchain with dtolnay/rust-toolchain * Remove action-rs/cargo * rate: Retire and separate TwoPassOutParams * rate: Estimate target quantizer on first pass * segmentation: Reset min_segment on data update * Replace AVX2 HBD SATD intrinsics with ASM, for some block sizes (#3039) * Remove arrayref dependency * Fix new clippy lints * Move v_frame to its own repository * Add scales and segment thresholds to dump_lookahead_data * Fix UB in pred_max test * Delete build.sh * bin: Admit more than 2 encoder passes * Create new speed level 8, drop level 7 * Keep the version line compatible with 0.5 * Add back shorthand for `--verbose` and `--quiet` (#3022) * Mention cargo-criterion * Bump criterion version * Update to criterion non-deprecated api * Prepare for 0.6.0 release * bin: Bring --photon-noise out from under unstable feature * Update guide on how to produce target-specific builds * CI: Deploy rav1e-ch binary tuned for modern x86_64 * build: Handle absent CARGO_CFG_TARGET_FEATURE * Calculate geometric mean for DistortionScale::inv_mean() * Convert plane-level distortion scales to fixed-point * Expand precision of DistortionScale to Q14 * Preserve intermediate precision in get_weighted_sse * Add DistortionScale::blog64() * Bump several dependencies * Remove duplicated constants * Remove aq_strength configuration * logexp: Correct integer part in blog32_q11 * logexp: Correct rounding bias in blog32_q11 * Fix new clippy lints from Rust nightly (#3006) * Use approximate binary log in Q11 for segmentation_optimize * Add polynomial approximations of a binary exponential and logarithm. * segmentation: Avoid reallocation in collect::<Box<[_]>>() * logexp: Add a fixed-point binary log in Q24 for u32 to i32 * logexp: Rewrite blog64 so that it is const * rate: Pivot to util::logexp functions * logexp: Add assertion for q57 input range * logexp: Add binary exponential in Q24 format, saturated to 47 bits. * logexp: Add conversions between 64-bit Q57 and 32-bit Q24. * logexp: Add tests for bexp64 and blog64 * logexp: Add 64-bit binary exponentiation and logarithm in Q57. * kmeans: Split k-independent inner loop into its own function * kmeans: Tidy up all but the inner loop * segmentation: Change precision of log(scale) to Q23 * Add missing code for segmentation with --tune Psnr * Derive segments by k-means of spatiotemporal scales * Add an implementation for k-means * Remove vergen and use built to provide the same information * Updating QUALITY_&_SPEED_FEATURES document in regards to libaom * Adjust chroma offsets for frame-mean spatiotemporal scale * Compute segmentation map thresholds from the offsets * Implement From<f64> for DistortionScale * Use fixed-sized array in luma/chroma pred mode counts (#2989) * Add T.35 metadata C API. * Bump console + dav1d-sys + system-deps * CI: Pull in dev packages from Ubuntu Kinetic * Implement SegmentationLevel::Full again * segmentation: Limit minimum segment to avoid lossless mode * Use spatiotemporal scales relative to frame mean * Clean up encode_partition_topdown() (#2984) * CI: Fix clippy:borrow_deref_ref warnings * CI: Fix clippy::manual_range_contains warnings * CI: Fix clippy::derive_partial_eq_without_eq warnings * Fix aarch64 satd compilation on gcc * Remove always-true if statement in encode_partition_topdown() (#2980) * Add aarch64 NEON 8-bit SATD assembly for 4x4-transformed block sizes (#2972) * Retune constants for ssim_boost * Add enhanced version info * Add T.35 metadata support. * Simplify badges in README * Remove travis badge * CI: Switch to codecov * CI: Do not consider crates directory for code coverage * Stabilize the --film-grain-table parameter * Compute segmentation offsets dynamically * CI: macos-11 is the latest version * CI: Update checkout action to v3 * CI: Remove if construct to skip CI * Remove unused .travis CI * Remove unmaintained gitlab CI * CI: Remove default linker paths on Windows * CI: Deploy rav1e-ch in a new job * CI: Improve Windows toolchains * Optimize `BlockSize::from_width_and_height_opt` * Use external crate for grain table generation and parsing * Optimize write_uleb128 function * ci: Use release-no-lto profile for gnu deploy on Windows * Add a release-no-lto profile * Avoid taking ownership of Arc<Frame<T>> in scenechange API (#2958) * Disable undocumented_unsafe_blocks lint due to false positive regression * Rewrite argument parsing to use Clap's derives * Reuse plane allocations in estimate_intra_costs (#2957) * Use a RwLock in frame_me_stats * Replace as_tile_state_mut with a an apply function * Simplify full_pixel_me * Simplify the tile iterator * Simplify the lifetime * Move TileStateMut::enc_stats where is used * Do not double collect the stats * Simplify the tiler lifetimes * Reuse me stats for lookahead * Merge estimate_motion and motion_estimation into one function * x86: Fix relocations in `sad_plane` * Bump versions of aom and dav1d in CI (#2953) * x86: Reduce code size in 8-bit film grain AVX-512 asm * x86: Add high bit-depth film grain AVX-512 (Ice Lake) asm * arm: Only produce the PAC/BTI .note section when targeting ELF * arm: Add comments to #endif and #else in nonobvious cases * arm: itx: Do clipping in all narrowing downshifts * x86: Add 8-bit film grain AVX-512 (Ice Lake) asm * build: Make "film_grain" vs "filmgrain" DSP file names consistent * x86: Fix data being stored below the stack pointer in SSSE3 itx asm * x86: Remove redundant labels and undefs in SSSE3 itx asm * x86: Update x86inc.asm * arm64: Add Armv8.3-A PAC support to assembly files * x86: Add high bit-depth ipred filter AVX-512 (Ice Lake) asm * x86: Add high bit-depth pal_pred AVX-512 (Ice Lake) asm * x86: Add high bit-depth ipred smooth AVX-512 (Ice Lake) asm * x86: Add high bit-depth ipred paeth AVX-512 (Ice Lake) asm * x86/itx: Add 16x16 12bpc AVX2 transforms * x86: Add mc.resize AVX-512 (Ice Lake) asm * x86: Improve high bit-depth film grain AVX2 asm * x86: Improve 8-bit film grain AVX2 asm * x86: Fix overflow in 10-bit IDCT4 * arm32: mc16: Fix out of bounds reads/writes in 8tap/bilin w2/w4 for vertical OBMC * arm32: mc: Fix out of bounds reads/writes in 8tap/bilin w2/w4 for vertical OBMC * arm64: mc16: Fix out of bounds reads/writes in 8tap/bilin w2/w4 for vertical OBMC * arm64: mc: Fix out of bounds reads/writes in 8tap/bilin w2/w4 for vertical OBMC * x86: Add high bitdepth mc(t)_scaled SSSE3 asm * x86: Fix branch condition in high bit-depth put_bilin.h SSSE3 asm * x86: Improve AVX2 generate_grain asm * x86: Add 10-bit sgr AVX-512 (Ice Lake) asm * x86: Add 8-bit mc(t)_scaled SSSE3 32-bit asm * x86/itx: Add 16x8 12bpc AVX2 transforms * x86/itx: Add 8x16 12bpc AVX2 transforms * AArch64 Neon: Replace XTN, XTN2 pairs with single UZP1 * AArch64 Neon: Use CMLT instead of SSHR to compute sign * x86: Fix AVX/SSE state transitions in AVX-512 4x4 inverse transforms * x86: Add high bitdepth wiener AVX-512 (Ice Lake) asm * x86/itx: Add 16x4 12bpc AVX2 transforms * x86/itx: Add 4x16 12bpc AVX2 transforms * x86/itx: Convert 8bpc WHT to SSE2 * x86: Fix edge padding in wiener_filter7_16bpc_ssse3 * x86: Add high bitdepth mc blend AVX-512 (Ice Lake) asm * x86: Add high bitdepth mc warp_affine_8x8 AVX-512 (Ice Lake) asm * x86: Add high bitdepth mc bidir AVX-512 (Ice Lake) asm * x86: Add high bitdepth mc bilin/8-tap AVX-512 (Ice Lake) asm * x86/itx: Add 8x8 12bpc AVX2 transforms * x86/itx: Add 8x4 12bpc AVX2 transforms * x86/itx: Add 4x8 12bpc AVX2 transforms * x86: Fix invalid memory access in cdef_filter_8x8_8bpc_avx512icl * x86/itx: Add clipping to iadst 4x16 * Remove lpf_stride parameter from LR filters * Allow CDEF and LR to run sbrows in parallel * arm64: Add Armv8.5-A BTI support to assembly files * arm64: Change br instructions to ret for function returns * x86/itx: Add 12-bit 4x4 transforms in AVX2 * x86/itx: Rename rax to r6 * x86/itx: Name constants more explicit * x86: Add deblock loop filters AVX-512 (Ice Lake) asm * x86: Add sgr AVX-512 (Ice Lake) asm * x86: Add wiener_filter AVX-512 (Ice Lake) asm * x86: Add ipred_filter AVX-512 (Ice Lake) asm * x86: Add ipred dc/h/v/paeth/smooth/pal_pred AVX-512 (Ice Lake) asm * x86: Add inverse transforms AVX-512 (Ice Lake) asm * x86: Add blend AVX-512 (Ice Lake) asm * x86: Add warp_affine_8x8 AVX-512 (Ice Lake) asm * x86: Add mc 8-tap AVX-512 (Ice Lake) asm * x86: Add mc put_bilin AVX-512 (Ice Lake) asm * x86: Remove the option to disable AVX-512 * x86: Add high bitdepth mc(t)_scaled AVX2 asm * Simplify sgr_x_by_x calculations * x86: Optimize shifts in 8-bit wiener_filter asm * Support distinct 10 and 12 bit inverse transform implementations * fix-up: x86: Add put/prep_bilin_scaled AVX2 asm * fix-up: x86: Add bpc suffix to mc functions * x86: Add high bitdepth cfl_ac_444 AVX2 asm * x86: Improve high bitdepth cfl_ac AVX2 asm * x86: Automatically convert SSE asm to AVX when compiling for AVX targets * x86: Add 8-bit w_mask_422 and w_mask_444 SSSE3 asm * x86: Improve high bitdepth cdef_filter AVX2 asm * x86: Prefer tzcnt over bsr in cdef sec_shift calculations * x86/itx: 64x64 inverse dct transforms hbd/sse4 * x86/itx: 64x32 inverse dct transforms hbd/sse4 * x86/itx: 64x16 inverse dct transforms hbd/sse4 * x86/itx: 32x64 inverse dct transforms hbd/sse4 * x86/itx: 16x64 inverse dct transforms hbd/sse4 * x86: Add high bitdepth cdef_filter SSSE3 asm * cdef: Remove redundant clipping * itx/x86: rewrite .transpose4x8packed so it uses only m0-3,4&6 * itx/x86: replace idct8x8.transpose with idct8x4.transpose4x8packed * x86/itx: add 1/sqrt(2) (rect2) multiply macro * x86/itx: share pass2 loop between {16,32}x32 dct^2 functions * x86/itx: combine .write_8x8 and .round{1,2,3,4} into a single function * x86/itx: combine .write_8x4 and .round{1,2} into a single function * x86/itx: split dct/adst/identity pass=2 implementations for 16x8 * x86/itx: 32x32 inverse dct transforms hbd/sse4 * x86/itx: 32x16 inverse dct transforms hbd/sse4 * x86/itx: 32x8 inverse dct transforms hbd/sse4 * x86/itx: 16x32 inverse dct transforms hbd/sse4 * x86/itx: 8x32 inverse dct transforms hbd/sse4 * x86: Add high bitdepth mc.resize SSSE3 asm * x86: Fix minor things in mc.resize_8bpc_ssse3 * x86: Add high bitdepth mc.resize AVX2 asm * x86: Add minor improvement to mc.resize_8bpc_avx2 * x86: Add bpc suffix to mc functions * x86/itx: merge pass=2 rounding and writing operations * x86/itx: 32x{8,16,32} & {8,16}x32 idtx transforms hbd/sse4 * x86/itx: replace .transpose8x8 with 2 calls to .transpose4x8packed * x86/itx: document third argument in INV_TXFM_WxH_FN macros * x86: Rewrite sgr8 SSSE3 asm * x86: Add minor improvements to sgr16 SSSE3 asm * x86/itx: 16x16 inverse transforms hbd/sse4 * x86/itx: 16x8 inverse transforms hbd/sse4 * x86/itx: 16x4 inverse transforms hbd/sse4 * x86inc: Support memory operands in src1 in 3-operand instructions * x86inc: Add stack probing on Windows * Properly fix LOAD_MM_PERMUTATION for AVX-512 * Replace abs() as unsigned with dedicated alternative. * v_frame: Disable missing const fn clippy lint * ci: Ignore tools for code coverage analysis * ci: Ignore asm for code coverage analysis * Reimplement sad_row as sad_plane (#2943) * Remove explicit VEX coding from sad ASM (#2941) * Rewrite x86 SAD row intrinsics in ASM * Split `write_coeffs_lv_map` into smaller functions * ci: Do not consider some directories for coverage * Remove unused lifetime. * ci: Enable source-based code coverage * Silence unused variable warning. * Use to_le_bytes in copy_to_raw_u8 (#2932) * Improve codegen of copy_from_raw_u8 (#2930) * Use const generics for downscale factor (#2927) * Implement grain synthesis via photon noise tables * Treat paths as OsString. * Change muxer to use AsRef<Path> arguments. * Remove a broken hawktracer annotation * Break the scenechange module into smaller parts * Add a mock around rayon for disabling threading * Disable missing const fn clippy lint * Use 64-bits for calculating 8x8 variance * Avoid temporary allocation of block importance costs * Remove unnecessary clones from metrics calculation * rate: add assertion for q57 input range (#2911) * Change non-square partition to search below the threshold * Move complex pred modes for keyframes to s2 * Fix memory leak when temporal RDO not in use * Enable suboptimal_flops clippy lint (#2908) * Also apply custom clippy lints to ivf and v_frame * Fixes to existing clippy lints * Enable linting functions that could be const * Enable extra documentation lints * Organize existing clippy lints * Avoid multiple inlining of residual loop in SAD intrinsics (#2903) * Improve storage of invalid and SEF frame invariants * Fix pyramid and frame type tests * Fix unnecessary auto-vectorization in residual loop of SAD intrinsics (#2897) * Use a debian mirror with https support * Update aom version and shasums for CI * New clippy fixes from Rust 1.59 * Fix new deprecation warnings from Clap 3.1.0 * Cache and reuse intra costs from scenechange (#2895) * Init Motion Est: Refine mvs when subsampling reduced * Fix confusing scale factor (#2891) * Speed up estimate_inter_costs again (#2890) * Speed up estimate_inter_costs (#2889) * Handle edges by cropping in initial motion estimation * Simplify importance block sum in `estimate_importance_block_difference` (#2885) * Add safety asserts for mc assembly callers * Update to Edition 2021 * Bump minimum Rustc to 1.56 * Omit loop restoration on fully skipped areas * Fix new clippy lints * Improve codegen of `Plane::downsampled` * Fix scale of motion estimation threshold * Use const generics for stride_sse * Make the doctests compile only * Put doctests code in a separate test * Bump minimum rust version to 1.54.0 * Update to clap3 * Refactor TxfmType to not have an invalid variant * Improve codegen of av1_round_shift_array * Use const generics for hadamard transforms (#2877) * Add x86 assembly for cdef distortion * Fix underflow that can occur in cdef distortion * Use cdef_dist on smaller block and edges * Display info for all y4m error types * Allow passing cropped blocks sad and satd * Fix overflow in cdef distortion * Refactor `BlockSize` to not have an invalid variant (#2866) * Skip transform mode search based on transform cost estimate * Improve autovectorization of `downscale_in_place` * Fix doc comments in speedsettings.rs * Change non_square_partition feature to a threshold * Silence repetitive warning on Rust nightly * Do not call put/prep avx2 functions for ssse3 * Organize speed settings into sub-structs * Mark SpeedSettings struct as non-exhaustive * Refactor the SpeedSettings impl * Merge `no_scene_detection` and `fast_scene_detection` into one field * Move rdo_lookahead_frames to SpeedSettings * CI: Update to libaom to 3.2.0-2 * Update a help message to be more relevant (#2853) * Fix stable clippy lints for Rust 1.57 * Use nasm-rs for nasm version check * Better autovectorization of `spatiotemporal_scale` * CI: Run cargo-c in offline mode * Replace float sqrt+divide with fixed point rsqrt * Perform UMH before resorting to full search * CI: Always use latest grcov version * CI: Always use latest cargo-c version * Extend the rayon API mock to include scope() * CI: Move to Debian Unstable for aom and dav1d OBS-URL: https://build.opensuse.org/request/show/1039292 OBS-URL: https://build.opensuse.org/package/show/openSUSE:Factory/rav1e?expand=0&rev=15
# How to update the rav1e package
## Prerequisites:
You need the download_files and cargo_vendor obs services installed:
zypper in obs-service-download_files obs-service-cargo_vendor
## Updating to a new version from upstream
Edit the spec file and update the version variable.
Download the new source file by running:
osc service ra download_files
Uncompress the rav1e-%{version}.tar.gz file:
tar xf rav1e-*.tar.gz
This will create a rav1e-%{version} directory
Edit the _service file and set the srcdir param
to the extracted directory.
Run the cargo_vendor service with:
osc service disabledrun
Update the changelog file with the upstream release notes.