Skip to content

Video Decode (Cobra: .CB8 + .VDO)

The Cobra full-motion-video framework — EA's in-house movie player, not licensed middleware. A per-frame vector-quantization scheme: 2×2 codebooks rendered to 8/15/16/24 bpp with optional 2× pixel doubling. 0x456300–0x45CDA0.

Attribution corrected (#95): Cobra is first and foremost the CB8 player — InitCobra (0x46ae10) validates the DRBC container (rejecting the older ARBC/BRBC/CRBC generations), streams it through the engine's own LIB layer, and the 8-bit paletted path (DecodeFrame submode 5 → DecodeSVGA8Frame/DecodeDSVGA8Frame, ExpandDB/EDB, CopySB8/CopyDB8) is exactly the CB8 keyframe codec, with a 768-byte palette embedded per frame. The 15/16/24-bit submode-6 paths are the hi-color generalization. Note the shipped .VDO corpus does not run through this DecodeFrame dispatcher — those 320×200 8bpp movies decode via a separate, much smaller cluster (GetVDOFrame → UnRLE → DecompressVideo, 0x4C8Axx), documented in VDO.md (#138). DecodeFrame here is the CB8 player. The compiled-in engine structs, the LIB-layer I/O, and the private generation lineage mark the whole framework as homegrown (confirmed); the CB8 side is validated by fx_lib's pixel-exact codec (tests/test_cb8.cpp) and the on-disk layout is specified in CB8.md.

Provenance: Ghidra static analysis of the game executable with FA.SMS symbols applied; recorded in the symbol database and applied to the Ghidra project. Progress: reconstruction matrix. Markers follow spec-authoring.md: confirmed · inferred · unknown. Coordinates with the VDO format spec (epic #55).

Vector-quantized YUV, four output depths

Each frame carries a codebook of 256 entries, each 4 luma + 2 chroma decoded to four RGB pixels (a 2×2 block). DecodeDBook builds the book at 15/16-bit (5:5:5 or 5:6:5 chosen per frame), Decode24Book at 24-bit; the YUV→RGB conversion uses the per-movie scale/offset constants and saturates each channel through ClampU8 (the one recovered leaf in the range — the color clamp called from every book decoder). The 8-bit path (DecodeSVGA8Frame / DecodeDSVGA8Frame) renders paletted output with an interpolated 2×2 dither built by EDB (expand-book), optionally 2× doubled.

Frames are key (intra) or delta (inter) — the dispatch keys on the frame header, and the inter decoders are gated on the keyframe latch. .VDO audio is a separate .11K track, not interleaved in this cluster.

Cobra decode: per-frame codebook (DBook/24Book/EDB) decoded to RGB via ClampU8, rendered at 8/15/16/24 bpp with optional 2x doubling.

Functions

Full record: db/symbols/video.csv.

VA Symbol Role
0x457230 DecodeDBook decode the 15/16-bit YUV codebook
0x456300 DecodeDSVGA8Frame key frame → 8bpp paletted, 2× doubled
0x456EC0 DecodeSVGA8Frame key frame → 8bpp paletted
0x456AD0 EDB expand-book: interpolated 2×2 dither pattern
0x4575E0 ClampU8 saturate a channel to [0,255] (YUV→RGB leaf)

Open Questions

1. VDO.md corrections — resolved

The three VDO.md inaccuracies are now reconciled (#259): the DecodeFrame dispatch keys are documented on the FrameHeader (+8 kind, +9 submode), not the movie context; the frame-kind polarity is corrected (0 = key/intra, 1 = inter/delta); and the 0xC14E canvas pointer is attributed to GlobalData. The Cobra per-frame codec itself (the ~45 leaf decoders) remains the long pole, tracked under epic #55.

Status: resolved — re-static (#259; the codec long-pole stays under #55).