No API changes.
Security fixes:
- GHSA-ww64-v2h4-rvf9: Heap out-of-bounds write via SEI payload_size wrap-around in user-data SEI
- GHSA-x289-cp8m-g58r: Heap out-of-bounds read via collocated picture of different size in TMVP
- GHSA-g3vh-f4qw-m567: NULL pointer dereference of wrap-around reference buffer after mid-CLVS SPS change
- GHSA-pw57-fxvc-2qgr: Use-after-free write in Picture::fillGrey after a picture buffer is recreated as 4:0:0
- GHSA-qgqj-f693-2fqc: LMCS table out-of-bounds read after an SPS change within a CLVS
- GHSA-f7mj-r2mj-rfvh: PPS not re-validated after its SPS is replaced (NULL dereference, out-of-bounds read)
- GHSA-p27c-vmfj-jh2f: Heap out-of-bounds read in vvdec_get_nal_unit_type() for short access units
Conformance fixes:
- fix parsing of several SEI messages (scalable nesting OLS indices, omni_viewport_tilt_centre, payloadType > 255 reported as VVDEC_SEI_UNKNOWN)
- fix film grain synthesis for 4:2:2, 4:4:4 and 4:0:0, and for scaling factors of 128 and above
- disable conformance bitstream HRD_B_Fujitsu_2, which contains non-conformant SEI messages that are now rejected
Build:
- don't assume SSE4.1 for the whole project
- don't force -fPIE globally
- explicitly require CMake 3.15
- fix WASM build and test suite
Other changes:
- reject more invalid bitstreams: SPS changes within a CLVS, reference pictures that don't fit the current SPS, out-of-range SEI counts and payload sizes
- fix misaligned stores in the AVX2 4x4 interpolation filter (undefined behaviour reported by UBSan)
- performance: new or faster x86 SIMD for intra prediction (planar, angular PDPC, horizontal/vertical, MIP), CCLM, CIIP, LMCS chroma scaling, chroma deblocking, affine 4x4 filters, BDOF, buffer copies; CABAC speedups; reuse of CU maps across pictures
- add ARM NEON implementations for luma angular intra prediction, film grain synthesis and some inter-prediction/transform functions
- some general code cleanup