# Historical RK3588 DDR blob snapshots

Extracted from the rkbin git log for cross-version analysis. v1.19 is
the deployed version; v1.16-v1.18 are the closest-in-time prior revisions.

| file | size | commit msg |
|------|------|------------|
| v1.16.bin | 75480 B | "update ddrbin to v1.16" — 2736 MHz LP5 dropped |
| v1.17.bin | 73456 B | "update ddrbin to v1.17" — smaller, feature cull |
| v1.18.bin | 75320 B | "update ddrbin to v1.18" — single-rank LPDDR5 derate crash per BUG_ANALYSIS.md |
| v1.19     | 76704 B | current deployed (in `rkbin/bin/rk35/`) |

## Byte-level divergence

Each release is essentially a near-rewrite, NOT a patch:

  v1.17 vs v1.18: ~88% bytes differ
  v1.18 vs v1.19: ~88% bytes differ
  v1.16 vs v1.19: even higher

This explains why cross-version symbol porting (via Polypyus or BinDiff)
has low yield on this target — not enough preserved byte sequences.
Function-level fuzzy matching (i.e., reduce each function to an opcode
silhouette with wildcards for branch targets and immediates) would help,
but requires Ghidra-analysed baselines in each version, not just
one-sided v1.19 annotations.

## Tooling note (2026-04-15)

Attempted Polypyus → blocked by Python 3.14 / pony 0.7 decompiler
incompatibility. See `project_rk3588_ddr.md` memory for retry options
(pyenv Python 3.11, or switch to BinDiff CLI which has no Python dep).