Hi Rong, I tested this series by following the documentation added in the series using Clang 19 and my standard distribution configuration on an Intel platform with the combinations of * AutoFDO * AutoFDO + ThinLTO * AutoFDO + Propeller * AutoFDO + ThinLTO + Propeller and I noticed no issues (it would be great to see create_llvm_prof somewhere in LLVM upstream for ease of access but that's a small complaint). I did not do any real benchmarking to see if those combinations were actually quicker but I think it is pretty clear from the cover letter that any sort of gains are going to depend on the profiling and test case so for time's sake, I did not bother. Tested-by: Nathan Chancellor <nathan@xxxxxxxxxx> Cheers, Nathan
