JonPsson1 wrote: Think I found the problem I ran into -- see commit message for details.
I have with this been able to take two .ll files, replace all uses of 'double' with 'half' (thousands of them), and it *compiles*. I have gone over the handlings of v4f32 and duplicated/handled for v8f16 similarly. I hope there are no generic cases in the backend that have some assumption about float/double that would go wrong with 'half'. Is there any code with _Float16 that could be compiled and run, for further testing? https://github.com/llvm/llvm-project/pull/171066 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
