Unify the output format of raw profiles for hybrid sample and LBR only sample.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/test/tools/llvm-profgen/inline-cs-pseudoprobe.test | ||
---|---|---|
1 | Add a test for show-relative-address=1? |
llvm/test/tools/llvm-profgen/inline-cs-pseudoprobe.test | ||
---|---|---|
1 | The default is true, so all others' test is for show-relative-address=1. Here is to test show-relative-address=0 |
llvm/test/tools/llvm-profgen/inline-cs-pseudoprobe.test | ||
---|---|---|
1 | I see. Makes sense. |
Looks like we're changing (begin, end): count to begin-end: count. Curious any restriction on the format, or reason why one is chosen over the other? Is that purely to align with internal raw output from fleet profiler?
llvm/tools/llvm-profgen/PerfReader.cpp | ||
---|---|---|
22 | nit: show-xx is often used to output xx as extra. how about something like [show-]address-as-offset or use-offset. I'd change the term/variable name from relative address to offset too. | |
595 | For CS raw profile (!CI.first.empty()), can we still keep the indentation like before for readability? Same for branch below. Otherwise it looks good. |
Add a test for show-relative-address=1?