starrocks
starrocks copied to clipboard
[Enhancement]add AVX512 implementation of bit unpack
What type of PR is this:
- [ ] BugFix
- [ ] Feature
- [x] Enhancement
- [ ] Refactor
- [ ] UT
- [ ] Doc
- [ ] Tool
Accelerate bit unpack with AVX512 intrinsic, get about 3~8X speedup in the microbenchmark
Hi, @imay, I have added the unit test of the simd bitunpack function and passed the CI, what can I do to go further? Thanks!
@wanweiqiangintel sorry for late reply, would you want to continue this PR?
@wanweiqiangintel
do you have any performance benchmark for this implementation? any improvement comparing to old implementation?
This routine is actually very critical for parquet reader. And if it improves a lot, then it will be huge help for parquet reader
@wanweiqiangintel do you have time to follow this pr?