[LLVMdev] why is coverage map and profile names mixed?

[Justin Bogner]

"Moshtaghi, Alireza" <Alireza.Moshtaghi at netapp.com> writes:
> My problem is that the instrumented code gets too big and I need to strip
> the coverage mapping.
> Would it make any functional difference if I save these variables in
> __llvm_covmap?

The llvm libraries that read the coverage data won't be able to handle
that.

>
> Thanks
> Ali
>
> On 5/19/15, 9:50 PM, "Justin Bogner" <mail at justinbogner.com> wrote:
>
>>"Moshtaghi, Alireza" <Alireza.Moshtaghi at netapp.com> writes:
>>> I¹m referencing the method: Lib/Transforms/Instrumentation/
>>> InstrProfiling.cpp:InstrProfiling::lowerCoverageData() At the end of
>>> the function, why is the variable being placed in __llvm_prf_names
>>> section?  Shouldn¹t it be placed in __llvm_covmap section?
>>
>>All of the names that coverage references are in the __llvm_prf_names
>>section today. This is shared with the profile data's references to
>>names: since a coverage mapping without profile data isn't useful, it
>>reasonable to expect these names are already there.
>>
>>The logic in lowerCoverageData moves any names that aren't found in the
>>profile data into the __llvm_prf_names, where the names that were found
>>are already expected to be. This is basically just a workaround for the
>>fact that functions that aren't emitted may still have a coverage
>>mapping, but there won't be any profile data for them.