v2.1.0

[2.1.0] — 2026-04-05

Upgraded — VaptVupt 1.4.0 Codec

• Cross-block dictionary carry — hash chain now spans block boundaries. The encoder passes absolute positions to compress_block() so matches can reference data from previous blocks. Large structured files (7MB logs) compress 5.73:1 instead of per-block independent ratios. The decoder accepts cross-block offsets via a dst_base parameter threaded through all decode functions.
• Context model decode prefetch — __builtin_prefetch in the order-1 context ANS decode loop hides L2/L3 latency for the 4MB context tables. Extreme-mode decode throughput improved significantly on cache-constrained systems.
• Faster adaptive window trial — greedy depth=4 on 256KB sample instead of full lazy parse on entire first block. Encode speed improved 2.6× with same ratio decisions.
• Zupt integration API — new vvz_compress/vvz_decompress/vvz_compress_bound wrappers (vaptvupt_api.h/vaptvupt_api.c) simplify codec dispatch with backup-optimized defaults.

Changed

• zupt_format.c compress paths (normal, solid) now use vvz_compress() API instead of raw vv_compress() with manual option setup.
• zupt_format.c decompress path now uses vvz_decompress() API.
• Version bumped to 2.1.0.