diff --git a/README.md b/README.md index dfd4ed6..40d74cd 100644 --- a/README.md +++ b/README.md @@ -5,9 +5,10 @@ TurboPFor: Fastest Integer Compression [ differential/Zigzag encoding/decoding for sorted/unsorted integer lists + - :sparkles: Full range 16/32, :new: 64 bits integer lists and :new: Floating point - :+1: Java Critical Native Interface. Access TurboPFor incl. SIMD from Java as fast as calling from C. + - Compress better and faster than special binary compressors like blosc
+ **Features** - :sparkles: Scalar **"Variable Byte"** faster and more efficient than any other implementation @@ -27,7 +28,7 @@ TurboPFor: Fastest Integer Compression [ scheme with **direct access** or bulk decoding. Outstanding compression and speed. More efficient than **ANY** other "integer compression" scheme.
- - :new: Scalar & SIMD Transform: Delta, Zigzag, Transpose/Shuffle, Floating point<->Integer + - :new: Scalar & SIMD Transform: Delta, Zigzag, Transpose/Shuffle, Floating point<->Integer conversion
+ **Inverted Index ...do less, go fast!**
- Direct Access to compressed *frequency* and *position* data in inverted index with zero decompression
@@ -276,4 +277,4 @@ header files to use with documentation:
- [On Inverted Index Compression for Search Engine Efficiency](http://www.dcs.gla.ac.uk/~craigm/publications/catena14compression.pdf)
- [Google's Group Varint Encoding](http://static.googleusercontent.com/media/research.google.com/de//people/jeff/WSDM09-keynote.pdf)
-Last update: 28 JUN 2015
+Last update: 8 JUL 2015