Skip to content

Releases: fast-pack/streamvbyte

Version 0.2.0

25 Oct 22:38
Compare
Choose a tag to compare
Neon encoder/decoder (#10)

* poc for encoder, gets code and length so far

* added output shuffle

* decoder works but ugly

* use arm compiler-compatible load/stores

* more cleanup

* various cleanups

* fix build usage

* Making sure it builds on Aarch64

* encode integrated
passes unit

* bugfix for 0x00 source value
passes unit and perf validity checks

* remove conditional for 2nd shuffle (performance)

* add decoder with flexible entry/exit point for delta

* add decode
flattened streamvbyte_decode
passes unit/perf tests

* clean up warnings

* clean up const's

* clean up warnings
add restricts
decode is slower

* inlines make encode faster
move consts to globals

* decode perf back up
added restrict

* aarch64 vqtbl1q_u8 16x16 shuffles in encode and decode

* fix avx build

* 16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s16 byte shuffle in encoder lanebytes
376M/s

* ifdef aarch64 stuff

* Removing silly comment header (probably automagically genreated at some point in the past).

* Declaring a couple of ARM NEON functions "static". Making sure that the unit tests
correctly report that the testing code was vectorized. Removed one space.

* Updating the README to indicate that we now support ARM processors.

* Tuning the Makefile. Adding standard headers to one of our header file. Having one source
file point to correct header file, so that functions between defined have been correctly
declared. Adding several "missing" static keywords.

version 0.1.0

09 Oct 19:07
Compare
Choose a tag to compare
v0.1.0

Reformat.