bodgit / rvz / 38185806013

63%

main: 74%

Build Type

Pull #184

github

Committed by bodgit

Commit Message

perf: Decompress raw data areas in parallel

Move the in-order background prefetching out of partReader into a
shared prefetcher, and use it for raw data areas too. Each 2 MiB block
of a raw area decompresses the groups it covers, with up to GOMAXPROCS
blocks running ahead of the reader.

Each group is now read with io.ReadFull for exactly the expected size,
so a group that decompresses short is an error rather than silently
misaligning the output.

GameCube images decompress about 4.6x faster, at the cost of higher
peak memory with many CPUs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Pull Request Pull Request #184: perf: Decompress raw data areas in parallel

Coverage Stats

96 of 106 new or added lines in 3 files covered. (90.57%)

1 existing line in 1 file now uncovered.

545 of 865 relevant lines covered (63.01%)

0.72 hits per line