Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
wrong when last block is full)
|
|
in the 32nd block prior to EOF
|
|
|
|
by new definition of xoff_t
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
use the average over 16 runs
|
|
reached and a copy reaches the end-of-source, causing unexpected request for second source block. New tests to cover this case. Removed the IOPT limit in xd3_encode_memory (improves compression). Doubles sizing of the large-checksum hash table (fewer collisions, better compression).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
the issue about updating the value of xd3_source_eof() in the same change).
|
|
|
|
xd3_source_eof () when eof is known in xd3_getblk. Now performance
is the same for FIFO and no.
|
|
it seems).
|
|
|
|
match-setup code, using source_eof instead of frontier position.
|
|
the encoder from seeking backwards further than this parameter. This
was supposed to be implemented long ago, but was especially problematic
in the recent release, which implements source-from-FIFO.
|
|
|
|
option_srcwinsz) until after decompression state is known. Working
but definitely needs more testing.
|
|
|