On 29/01/2020 12:53, Tobias Burnus wrote:
Cf. PR93409 comments 4 and later. The comments 1–3 of the PR are covered
by patch https://gcc.gnu.org/ml/gcc-patches/2020-01/msg01663.html (skip
building libgomp's HSA/GCN plugin with -mx32).
For AMDGCN, the LLVM assembler is used. While for LLVM 7+8, GCC
generates code which is the same as LLVM, for LLVM 9 (and later), the
LLVM switched from Code Object V2 to V3.
With LLVM 9, the old variant is only accepted when also passing
"-mattr=-code-object-v3" to the compiler; that's a"-" after the "=" to
disable V3. (Cf. also
https://sources.debian.org/src/gcc-10/10-20200117-2/debian/patches/gcc-amdgcn-fix.diff/).
My initial idea was to add a configure check – which I also include as
attachment. However, it turned out that at last the assembler of LLVM
version 7.0.1 already supports this flag. Hence, we can go for the
simpler solution and use the assembler flag unconditionally.
OK for the trunk?
What testing have you done?
I've been holding off submitting what is essentially this patch myself
until I can run a full test with LLVM 8/9. It just hasn't got to the top
of my list to set that up yet.
It looks right, of course, but I don't want to just assume so.
Andrew
