efriedma-quic wrote: My impression is that the treatment of address-spaces varies so much across different languages and different targets that it doesn't make sense to encode the entire address-space layout the datalayout. Narrowly encoding just the specific properties optimizations need means we don't need to rework optimizations and the datalayout for each new generation of GPU/accelerator. And developers working on IR optimizations don't need to try to decipher the implications of different address-spaces.
https://github.com/llvm/llvm-project/pull/108786 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
