Jakub Chlanda
Jakub Chlanda
I'm going to close this ticket, please feel free to re-open it or create a new one if there is any new information.
I've revisited this ticket and decided to close it. The problem is not with LLVM failing to unroll, but the quality of the source code and developers exploiting the peculiarity...
> @jchlanda My apologies, I accidentally rebased. I believe I've corrected my mistake, and you can see my changes addressing your feedback in [6b2ca14](https://github.com/intel/llvm/pull/12464/commits/6b2ca14b2793cf606790296a46318e2ac3ab3516). Not a problem, I've left one...
cc: @frasercrmck
@frasercrmck are you ok with this going in?
@intel/llvm-gatekeepers this is ready to go. Thank you.
Corresponding UR work: https://github.com/oneapi-src/unified-runtime/pull/954
Fixes: https://github.com/intel/llvm/issues/9353
> @jchlanda, could you please add FE test? @smanna12 I've updated the test to use the new metadata node here: 6e4364344269567bd78f712d4bd9ad2a2c3c66b3, do you think this gives us enough coverage?
> Regarding the naming, would this problem (and solution) not also apply to `work_group_size_hint` metadata? If so, could this metadata you're adding be made more general somehow? Maybe just trimming...