Takeshi Yamamuro

Good reference to http://code.google.com/p/poly-ir-toolkit/

During encoding, many memory allocations in std:vector makes encoding time longer, so these working space might need to be allocated only once for performance reasons.

There are many FIXME items in the sources of this library. These items should be modified in the future.

A newest coder was proposed in CIKM'11, which is titled 'SIMD-based Decoding of Posting Lists'. We would like the coder to be appended into this library.

Vectorization with SSE2 and AVX2 could improves their performance.

These codes are very similiar, so these should be merged while keeping their current performances.

When setting 128 to PFORDELTA_BLOCKSZ in 'include/compress/PForDelta.hpp', a error 'SEGV' happens.

See: https://lists.apache.org/thread.html/4727131e0334de2e0b37c1f2d6d907eabdac572d4b347baebd2f2388@%3Cdev.livy.apache.org%3E

To make coverage more correct, we should do this; https://github.com/maropu/spark-sql-server/blob/ee7855118eda5a30fd7b52cbf6fb955c55893c66/sql/sql-server/src/test/scala/org/apache/spark/sql/server/SparkPgServerTestUtils.scala#L70