3 ms·
I believe LZ + Huffman hit a sweet spot in ratio/speed/complexity, and that's why it has remained very popular since the late 80s. It's only more recently that
by userbinator 2y ago
I believe LZ + Huffman hit a sweet spot in ratio/speed/complexity, and that's why it has remained very popular since the late 80s. It's only more recently that faster hardware made arithmetic coding with higher-order models fast enough to be practically usable.
- whizzter 2y agoMore importantly, there's been a huge shift in the cost of tables vs multiplications. Back in the 80s and early 90s multiplication was expensive (multiple cycles) while tables were more or less "free" in comparison, today cache-misses are super-expensive (100s of cycles) while multiplications can be run in parallel (MMX,SSE,etc). Sure a huffman table will probably mostly be in-cache but it'll still be at the cost of cache space. In addition to that various arithmetic encoding methods were patented and thus avoided.