https://github.com/paulwalker-arm approved this pull request.
I'm more comfortable with the conservative approach now that we're not planning to error. I do think structurally there's more work needed to cleanly represent/separate cannot-inline-vl-dependent and cannot-inline-operation-no-supported, but I suspect that will fall out as the restrictions are relaxed. Things to consider: * Can we rely on the cost model to implement cannot-inline-operation-no-supported so the logic sits in one place? * Can we autogenerate something to identify mode-specific target intrinsics? * Or rather, is there a way to implement all the necessary relaxations so we don't have to revisit the implementation every time new intrinsics are added? Given this is WIP, I've not gone through the tests to verify complete coverage. We will need to remember to do this when it's closer to its final form. https://github.com/llvm/llvm-project/pull/223393 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
