[llvm] [MemProf] Optionally save context size info on largest cold allocations (PR #142837)

[Teresa Johnson via llvm-commits]

teresajohnson wrote:

> > > Reapply PR142507 with fix for test: add in the same x86_64-linux
> > > requirement as other tests as the stack ids are currently computed
> > > differently on big endian systems. This will be investigated separately.
> > 
> > 
> > Hi @teresajohnson, is there an open issue that tracks this investigation? What is the expected impact of the difference?
> 
> Hey Hubert, we don't plan to investigate the underlying endian-ness issue in the near future. We will update the tests to other run on other architectures (e.g. ARM) while excluding big endian targets.
> 
> Since the differences show up in stack IDs I'd expect that memprof simply won't work on big endian systems at this time.

I'm not convinced that it won't work, because the stack ids are all internal to the profile-use. I think the tests just might be failing because the stack ids don't match what we have hardcoded.

https://github.com/llvm/llvm-project/pull/142837