Thanks for the quick review.

Add test description.

I'm still going through PGOInstrumentation.cpp ...

lgtm with a couple of minor comments.

PTAL, thanks!

Update binary names to be consistent.

In D126834#3551932, @tejohnson wrote:

I think the 2 changes are unrelated so maybe they should be committed separately?

Split out the segment information print in D126840.

PTAL, thanks!

Remove summary changes from this patch.
Add another field which tracks the number of mibs we have.

Thanks for the review.

PTAL, thanks!

Update comments.

Remove unnecessary blank lines.

lgtm

lgtm

lgtm

Thanks for the detailed review as always. :)

Address comments.

Thanks for the review, ptal!

Address comments.

PTAl, thanks!

Address comments.

With this patch on an internal binary, the time taken to merge indexed and raw profile increases to 85s (up from 48s in D120430). The zip profile size increases to ~520M (up from 247M in D118653). I'll look into de-duplicating the inline storage of callstacks and their frames in followup patches. Thanks for the review!

Update comment.

In D121830#3387825, @tejohnson wrote:

lgtm

Unfortunately the ordering is not documented on the DIInliningInfo class. And confusingly, one piece of code sets an Inlined flag by the check "I > 0" (https://github.com/llvm/llvm-project/blob/49c048add4c980936fc2918838288ae2d795587d/llvm/lib/DebugInfo/Symbolize/DIPrinter.cpp#L198).
However, if I look at how that flag is used, it is used to print "(inlined by)" *before* the current frame's function name, so it seems to apply to the I-1 frame, which actually means that all but the last frame are inlined - consistent with the change you have made here. Consider making a separate NFC change to add a comment to DIInliningInfo?

Updated the patch to track callsites for all functions in the frame. PTAL, thanks!

Fix lit test and some cleanup.

Address comments.

@tejohnson I've updated the logic to track and annotate additional functions with the allocation site profiles. Please take a look at the example in MemProfTest.cpp L176 to see if it matches your expectations. Thanks!

Update allocsite and callsite profile contents.

Updated the test documentation. PTAL, thanks!

Address comments.

Rebase, lint and fix test invocation.

Discussed with @tejohnson offline how we can make the matching more robust.

Thanks for the detailed review! PTAL.

Address comments.

Add a comment and remove some commented code.

PTAL, thanks!

Add TODO to update summary print.

Address comments.

Simplify discarded vaddrs check, rename method to indicate filtering and add comments.

Updated the patch with comments. PTAL, thanks!

Address comment.

Thanks for the reviews.

Move the InstrProfKind::MemProf comment above the decl to be consistent.

ping @ellis @davidxl :)

In D120147#3332688, @tejohnson wrote:

What about the other uses of PACKED e.g. in MemProfData.inc? Should/can they be removed?

This was reviewed separately in https://reviews.llvm.org/D120093 and https://reviews.llvm.org/D120103. Not sure why this draft revision was tagged. Sorry for the noise!