Reimplementation of fqzcomp4 compression tool with caryless rangecoder replaced by fse codec & static mode, aiming at increased throughput for cases when speed is overly important, such as disk-based reads reordering and real-time compression.
git clone [email protected]:iam28th/fqcomp28.git
cd fqcomp28
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build
./fqcomp28 --help-all./fqcomp28 c --input reads.fastq --output archive.fqc --threads 4
./fqcomp28 d --input archive.fqc --output1 decompressed.fastq --threads 4fqcomp28 seemingly offers a better compression ratio/speed tradeoff than fqzcomp4 and fqzcomp5 with default mode (-5), but looks inferior to faster modes of fqzcomp5 (-3 and -1), particulary because of longer decompression times (the cause of which is under investigation).
All tools noticeably outperform gzip/pigz in both speed and compression ratio.
See Results.md for details.
- headers
- must all have same structure (same number of "fields" consisting of alphanumeric characters with non-alphanumeric separators between the "fields");
- a "field" can be no more than 255 characters long;
- each header must end with an alphanumeric character;
- read lengths must fit into uint32_t (that is, be <= 4294967296);
- bases must be one of
ACGTN; - quality headers (that is, repeated headers on the 3rd line of the record, after
+), if present, are discarded in decompressed data (but otherwise compression is lossless).