[compiler-rt] [profile] Use fprofile-continuous in compiler-rt tests (PR #126617)

[David Tellenbach via llvm-commits]

dtellenbach wrote:

> > does Darwin not relying on counter relocation change anything with respect to using -fprofile-continuous?
> 
> -fprofile-continuous does not enable counter relocation on Darwin. It [enables](https://github.com/llvm/llvm-project/blob/main/clang/lib/Driver/ToolChains/Clang.cpp#L765) it on targets using ELF, on AIX, and on Windows.

Thanks!

> > Also, I think we should definitely have some tests that are using %c since that is a workflow which is established for many years and is expected to be used much more than the explicit flag. I would actually prefer to have both versions tested.
> 
> -fprofile-continuous doesn't replace %c as a method of enabling continuous mode in the runtime. The option is a convenience option that appends '%c' to the default profile name. The runtime is unchanged. Given the above, I don't see the necessity to have both explicit filename setting and -fprofile-continuous in every continuous mode test. I think it's ok to have it in one test. What do you think?

Agreed, doesn't have to be in every test.

> After this PR, explicit filename setting with "%c" is used in `ContinuousSyncMode/online-merging.c`, `ContinuousSyncMode/set-filename.c`, `ContinuousSyncMode/multiple-DSOs.c`, and `ContinuousSyncMode/pid-substitution.c`. Also tests that were testing profile file name parsing obviously remain untouched, for example: `ContinuousSyncMode/reset-default-profile.c` and `ContinuousSyncMode/get-filename.c`.

Sounds good, as long as we have some tests that still use explicit `%c` that should be fine.

Thanks a lot!

https://github.com/llvm/llvm-project/pull/126617