[llvm] Introduction of typified section in ExtBinary format (PR #166553)

[Mingming Liu via llvm-commits]

mingmingl-llvm wrote:

>   Do you think there can be issues w/ the strightforward approach? I was thinking about adding a new map to FunctionSamples (for [anchor -> prof_value] correspondence).

I'm in favor of enriching profiles for new optimizations and onboard with the idea of pursuing forward compatibility to make new changes easier.

Could you illustrate on how a new type of profile is mapped onto the IR and how it will be represented in the both memory (class structures) and serialized format? 





https://github.com/llvm/llvm-project/pull/166553