Jacob Lambert
Jacob Lambert
@atamazov Closing this for now. Feel free to re-open if you have any other multi-threading issues with Comgr! Hopefully we can move to actual parallel compilation with Comgr in the...
Good point. We can track the performance/parallel issue here: https://github.com/RadeonOpenCompute/ROCm-CompilerSupport/issues/57
Should be fixed by updating Comgr actions used. Let me know if you still have any issues with this!
After some discussion, we decided against incorporating this PR. This would require Comgr to make symbols outside its documented API public. We can't immediately expose LLVM symbols in the Comgr...
@candrews @littlewu2508 Do we know if this is still an issue for Gentoo? @littlewu2508 you've been building for Gentoo recently right?
PR no longer needed as mentioned above
@awehrfritz Thanks for bringing this up! In general the way Comgr interacts with the environment for Clang/ROCm/HIP Include Paths needs an overhaul. We're working on deprecating the AMD_COMGR_ACTION_COMPILE_SOURCE_TO_FATBIN action (https://github.com/RadeonOpenCompute/ROCm-CompilerSupport/commit/1a15280c08c73efb9e58f59ea98d9bbfc16789ad),...
This should be resolved by https://github.com/ROCm/llvm-project/commit/6eb9d123a905464a2c20a5605e5dbf599af17fde Please open a new issue at https://github.com/ROCm/llvm-project if you're still able to recreate this though! Thanks!
Looks this this was resolved, so closing this issue. Thanks @slinder1
We still don't have an action to deal with LLVM IR directly. Instead we mostly deal with things at the bitcode level. Let me know if this is still needed...