https://github.com/zmodem commented:

Thanks for adding me.

I don't have the context of knowing anything about HIP or ROCm etc., so please 
correct me if something I say doesn't make sense.

Firstly, this is a large patch, covering Clang codegen, LLVM instrumentation, 
and significant changes to the profiling runtime library.

Could you split this up, e.g. establishing the runtime first in a separate PR? 
That would make reviewing it easier, and landing it potentially less disruptive.

---

My understanding is that all the issues we hit were due to merging parts of the 
sanitizer runtime (the interceptors) with the profiling library.

It seems the needs of ROCm are pretty different from the existing profiling 
library. Perhaps one good option, in addition to those you mention, would be to 
build a *separate* profiling library for ROCm? It could still re-use the 
existing sources, but you could combine them with the other parts you need 
(interceptors) and build it in the way that makes sense for your target (/MD) 
-- without disrupting existing users of the profile library. What do you think?

https://github.com/llvm/llvm-project/pull/201499
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to