https://github.com/serge-sans-paille approved this pull request.
That's a great use case of the `target` attribute, the cost of resolving the target is handled by the compiler runtime and only paid once. the only limitation seems to be that it prevents inlining, but the benchmark speaks for itself. I think it's fine not to add regression tests for the i686 path without sse4.2 path, as it relies on compiler support that *should* be working correctly, right? https://github.com/llvm/llvm-project/pull/171914 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
