Bolo -- Josef Burger

I found that this or a similar behavior also exists in ROCm itself. I filed a ROCm issue on it, and refer to it here to link them together: https://github.com/RadeonOpenCompute/ROCm/issues/1778

> > > @@ -60,6 +60,7 @@ namespace Dyninst > Arch_amdgpu_vega = 0x84000000, > Arch_cuda = 0x88000000, > Arch_amdgpu_cdna2 = 0x94000000, //future support for cdna2 > + Arch_amdgpu_gfx908 = 0x98000000,...

> So every architecture might have > [variants](https://llvm.org/docs/AMDGPUUsage.html#instructions). > > > Each variant might > [map](https://llvm.org/docs/AMDGPUUsage.html#processors) to more than > one product (mi-xxx). > > > I think in one...

> The thing with the libabigail ABI checks is, you can just increment > the library age and revision and it should be fine. That would allow > backward compatible...

> @kupsch @bigtrak > There are at least two 'debug' libraries: one in libcommon and one > in libsymtabapi. I've never understood why the ones in libsymtabapi > are not...

I am looking at this in with respect for test failures on mahir right now. News at 11.

The open of a archive library from symtab fails both on mahir and on coriander opening the libc.a archive. The Object() constructor says that the elfhdr is not valid, and...

The crash only occurs when using the libc.a with debug symbols. If the normal libc.a is used, /usr/lib64/libc.a there is no crash and it completes successfully. The crash occurs because...

Yes, this is fixed definitely in rocm-6.0. IIRC correctly it may have been fixed in a late rocm-5.x release; I no longer have them available to test when it was...

This pull request Resolves #1380.