fix: Remove same page hack for simdStrstr

[Yuhta]

Summary:
The hack is breaking ASAN check, removing it does not give huge
performance loss though so we remove it.

Benchmark results before vs after change:
https://gist.github.com/Yuhta/b3f66e85d52b62240c09c15d6e7941bb

Differential Revision: D66012678

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D66012678

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	dfd6085
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/67379973f8e936000873c0c5

[kevinwilfong]

LGTM , thanks JImmy!

[facebook-github-bot]

This pull request has been merged in 0a422f7.

[conbench-facebook]

Conbench analyzed the 1 benchmark run on commit 0a422f73.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.

[skadilover]

@Yuhta

This may not guarantee that simdstrstr will always be faster than std in most cases. Doing so will always lose the last batch.

tpchQuery13 54.72ms 18.28

tpchQuery13 27.57ms 36.27

[Yuhta]

@skadilover Are we slower than std in such cases? It probably should not be because the fallback is std.

I tried a few annotations to skip ASAN on the memory load without success, maybe you can give it a try when you have time.

[skadilover]

@skadilover Are we slower than std in such cases? It probably should not be because the fallback is std.

I tried a few annotations to skip ASAN on the memory load without success, maybe you can give it a try when you have time.

@Yuhta
I am trying to use simdstrstr to replace std::find in Like match funciton : matchSubstringsPattern, I found that the perfomance of TPCH-Q13 in LikeTpchBenchmark.cpp is not improved, so I tried revert the change of this MR

now :
tpchQuery13 54.72ms 18.28
revert :
tpchQuery13 27.57ms 36.27

Q13`s string-filter is the case that unmatch more frequently

[skadilover]

I tried a few annotations to skip ASAN on the memory load without success

So you mean the code before will trigger ASAN check failed ?

maybe you can give it a try when you have time

Could you detail how to reproduce it? Share the compile flag you used ?

I tried add -fsanitize=address， LikeTpchBenchmark.cpp will always failed (remove or not remove the hack code), seems nothing to do with this:

[skadilover]

@Yuhta
Do you see the error like this ?

[Yuhta]

@skadilover Yes it's the xsimd load_unaligned that triggers ASAN