Chandler Carruth
2014-Jan-07 10:09 UTC
[LLVMdev] RFC: Does having a separate IPA library still make sense?
We have quite a few analyses that are in the Analysis library despite being module passes, and not very many interprocedural analysis anyways. Should we fold them together? If we do fold them together, should the source code still live in a subdirectory? If the code still lives in a subdirectory (regardless of the library separation) would it make sense to put the header files in a subdirectory as well? It would make it easier for my mental organization of things, but I don't care strongly either way. -Chandler -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140107/4d0c7af8/attachment.html>
Chris Lattner
2014-Jan-07 18:07 UTC
[LLVMdev] RFC: Does having a separate IPA library still make sense?
On Jan 7, 2014, at 2:09 AM, Chandler Carruth <chandlerc at google.com> wrote:> We have quite a few analyses that are in the Analysis library despite being module passes, and not very many interprocedural analysis anyways. Should we fold them together? > > If we do fold them together, should the source code still live in a subdirectory? > > If the code still lives in a subdirectory (regardless of the library separation) would it make sense to put the header files in a subdirectory as well? It would make it easier for my mental organization of things, but I don't care strongly either way.I’m in favor of having only a single analysis library. The only thing to watch out for is cyclic dependencies in the library graph. -Chris
Chandler Carruth
2014-Jan-07 18:22 UTC
[LLVMdev] RFC: Does having a separate IPA library still make sense?
Thanks! Will make it so. On Jan 7, 2014 10:07 AM, "Chris Lattner" <clattner at apple.com> wrote:> > On Jan 7, 2014, at 2:09 AM, Chandler Carruth <chandlerc at google.com> wrote: > > > We have quite a few analyses that are in the Analysis library despite > being module passes, and not very many interprocedural analysis anyways. > Should we fold them together? > > > > If we do fold them together, should the source code still live in a > subdirectory? > > > > If the code still lives in a subdirectory (regardless of the library > separation) would it make sense to put the header files in a subdirectory > as well? It would make it easier for my mental organization of things, but > I don't care strongly either way. > > I’m in favor of having only a single analysis library. The only thing to > watch out for is cyclic dependencies in the library graph. > > -Chris > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140107/435a2c28/attachment.html>
Reasonably Related Threads
- How to prevent optimizing away a call + its arguments
- [RFC] Heterogeneous LLVM-IR Modules
- cyclic dependency error
- How to prevent optimizing away a call + its arguments
- Possible soundness issue with available_externally (split from "RFC: Add guard intrinsics")