Thanks for putting this up! I'm trying to finish up a paper today on vectorizing VByte decompression, so have been thinking about some of these issues a lot recently.<p>It would be help orient people (like me) if you added more description of the data format you are using for each approach. It's difficult reading templatized C at a glance, so I'm not certain what's happening in each format. For example, since the average bit widths for your SimpleV are less than 8, it's clearly doing something different than traditional variable byte integer implementations. Nothing wrong with this, but would be helpful to know the details.<p>I'm mostly with you on the "Realistic and practical benchmark with large integer arrays" part. There are some arguments for variations, though. In particular, reading from memory but decoding to cache can be a more realistic benchmark for search engine applications. Feel free to write email to me or Daniel if you'd like to try to integrate your benchmarks with ours.