emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-638620462
@wesm I expect your PR to get merged first but if it doesn't consider using
PopCount methods in this one.
Thi
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-638620176
@pitrou I think the C GLib build failure looks unrelated to me. I fixed the
s390x build.
This is an automat
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-637334193
OK build is green (side note I don't think this will compile on s390x, not
sure if that is a blocker).
This i
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-637187135
As long as @pitrou is happy with my workaround to avoid inflicting pain on
AMD processor users.
I think I need to fix one signed conversion bug to make CI happy. I shou
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-632468275
@pitrou I updated this PR to require AVX512 for the use of BMI2 codepath. I
think this should prevent issues for AMD architectures (I don't believe there
are any that support
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-630572790
> So the BMI2 instructions incur a large slowdown here (even catastrophic
with gcc).
I would expect bad numbers with GCC 7.5 since we lose autovectorization and
use BMI
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-629737181
@pitrou
https://github.com/apache/arrow/pull/6985/commits/a36dc55efe3def27eef413b8e17567ec400a01ff
adds a benchmark. Should be cherry-pickable to a clean branch for compariso
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-628755223
> As a sidenote, it seems BMI2 may be problematic on AMD CPUs:
https://twitter.com/trav_downs/status/1202793097962364928
Yea we need some benchmarks to see if the per
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-628370216
@pitrou Doing some investigation it appears GCC 7.5 fails to auto-vectorize
the comparison of levels. GCC 8.3 does autovectorize. Are we stuck with GCC
7.5 for releasing?
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-628364641
OK it seems like clang and GCC perform very differently.
Clang-8 new code vs master:
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-622266880
> One problem with introducing more SIMD code is that we won't yet have a
runtime dispatching strategy. We will need to go through all of our SIMD
accelerations in this librar
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-621974840
@wesm wanted to make sure this is still on your radar
This is an automated message from the Apache Git Service
emkornfield commented on pull request #6985:
URL: https://github.com/apache/arrow/pull/6985#issuecomment-619488093
@pitrou I think I addressed your comments. One of them that went stale was
the complexity for "AppendWord", I tried to remove parts that did not seem to
affect performance on
17 matches
Mail list logo
