mmjjpp wrote:

> I started reviewing and have some high level questions and comments, and a 
> number of other initial comments sprinkled throughout. Can you split this PR 
> into 2? The first being the Utils support, llvm-split changes, and the 
> associated llvm-split based tests; the second being the LTO support with some 
> added LTO tests?
> 
> I know the callgraph based splitting is to avoid performance overhead from 
> loss of IPO, but have you evaluated this?

Thank you very much for the detailed review comments. I’ll carefully go through 
them and address the issues you raised.

I have now split the original PR into two, following your suggestion: one for 
the Utils support, llvm-split changes, and the associated llvm-split tests; and 
another for the LTO-related support and tests. The LTO backend related changes 
are available here: https://github.com/mmjjpp/llvm-project/pull/17

Regarding the potential IPO impact of callgraph-based splitting, we did 
evaluate it in our practical workload. The observed performance variation was 
within 1%. In this evaluation, PGO was enabled, while WPD, CFI, and MemProf 
were not exercised.

https://github.com/llvm/llvm-project/pull/198702
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to