[llvm-dev] [RFC] Pass return status

Serge Guelton via llvm-dev llvm-dev at lists.llvm.org
Thu Jun 11 08:42:08 PDT 2020

Hi folks,

Per the documentation[0], whenever an LLVM pass doesn't modify the IR it's run on, it
should return `false`--it's okay to return `true` if no change happen, just less
optimal. In the New PM area, this is generally translated into a `PreservedAnalyses::all()`.

https://reviews.llvm.org/D80916 provides an `EXPENSIVE_CHECK` that computes a
hash of the IR before and after the pass, and checks that any change is
correctly reported. The hash is currently incomplete (on purpose, let's start
small), but it turns out a dozen of passes do not satisfy that

This could lead to various category of bugs (dangling references, inconsistent
state, etc). This affects both New and Legacy PM, as passes tend to wrap functions
that report their status.

I wrote a bunch of patches for all failure detected by this check, as I cannot land the
check now, it would break the buildbots :-) Any help to review the remaining
ones [1] is appreciated.

Once that check lands and we're relatively confident on the quality of the
return status, some more optimizations could be triggered, like

[0] https://llvm.org/docs/WritingAnLLVMPass.html#the-runonmodule-method
[1] https://reviews.llvm.org/D81230

More information about the llvm-dev mailing list