[PATCH] D72538: [ThinLTO] Add additional ThinLTO pipeline testing with new PM

[Teresa Johnson via Phabricator via cfe-commits]

tejohnson marked 2 inline comments as done.
tejohnson added a comment.

In D72538#1815074 <https://reviews.llvm.org/D72538#1815074>, @wmi wrote:

> The additional pipeline testing will catch any future pass change to the pipeline. A related but separate question is do we have a way to check whether there is any other missing pass in thinlto newpm similar as that in D72386 <https://reviews.llvm.org/D72386>?


Unfortunately not. That will take some painstaking comparisons between the pass debug output between the old and new PMs, which is going to be manual given the different format and structuring of the dependences. I am going to try to do that at least for a few important cases. As you noted, this is to help flag any future changes (although the last one did actually result in a test being changed, but it looks like no one realized the implications). Hopefully by having additional testing it will be more likely to trigger concerns if it happens again.



================
Comment at: clang/test/CodeGen/thinlto-distributed-newpm.ll:11
+
+; RUN: %clang_cc1 -triple x86_64-grtev4-linux-gnu \
+; RUN:   -O2 -fexperimental-new-pass-manager -fdebug-pass-manager \
----------------
wmi wrote:
> Can we test %clang intead of %clang_cc1? so we don't have to add flags like -vectorize-slp and -vectorize-loops. clang `O2/O3` pipeline is something we care about. Those flags added by clang to cc1 options may be subject to change.  
> 
Good idea, done


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72538/new/

https://reviews.llvm.org/D72538