Should these x86 patches be integrated into the 3.6 release branch?
Andrea Di Biagio
andrea.dibiagio at gmail.com
Thu Mar 19 04:45:28 PDT 2015
Hi Quentin, Nadav and all,
I am not familiar with the release process so, sorry for the long post.
In the last few weeks I committed at least three patches to fix bad
codegen issues affecting x86 only. Here is the list of commits:
The first two commits fix problems in the new shuffle lowering, while
the last commit (r232046) fixes a problem in a target specific combine
on SETCC nodes.
There is also another important patch committed by Michael at revision
231219 which fixes a regression introduced at revision 224759.
Here is the link to his commit:
I haven't tested the 3.6 branch myself, however it is very likely that
these bugs are present in the 3.6 branch. Given that all these fixes
would resolve bad codegen issues, I suggest you consider whether these
should be integrated into the 3.6 branch.
I hope this information was useful. I don't know enough about the
release process, so I just wanted to make sure that people are aware
of these fixes.
More information about the llvm-commits