ampandey-1995 wrote:

> I think this is busted, compiler passes cannot reasonably emit calls to 
> internal, non-ABI functions. But I guess it probably works in practice if the 
> sanitizer lowering happens immediately before any linking. This is also 
> consistency broken with every function, and we're already doing special case 
> hacks for the sanitizer functions

Yes, This code will be  removed in following PR 
[179636](https://github.com/llvm/llvm-project/pull/179636.). 
For OpenMP, only a singular static device library exists i.e libompdevice.a 
which linked in lto phase for OpenMP compiled applications. 

>If we had functioning compiler libraries, this should be added once in the 
>final link invocation

Yes, I am working on downstream 
[PR](https://github.com/ROCm/llvm-project/pull/1334) to build an asan 
instrumented libompdevice.a. POST LTO link job happens to build asanified 
libompdevice.a here in this piece of downstream 
[code](https://github.com/ROCm/llvm-project/pull/1334/changes#diff-02cde0d1b09f1650f779f64f11032ff269d1015ad8a1e69c1bc0677ac35ec205R63)


https://github.com/llvm/llvm-project/pull/182825
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to