https://github.com/jhuber6 commented:
IIRC host allocations are a little strange so I'll need to double check. HSA does its own pooling, but allocations that pin memory use full page granularity so they're not as easily split. E.g. if you want to pin 4 bytes you round down to the page size and pin the whole page. If you then unpin that and free it you could unpin in-flight memory that shares the page, so I think HSA just gives you an entire page. https://github.com/llvm/llvm-project/pull/214754 _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
