tmatheson-arm wrote: The autogenerated checks for `clang/test/CodeGen/aarch64-fmv-dependencies.c` did not actually pass, due to changes in function ordering, so I have manually written the check lines, which makes the intent clearer anyway.
Some things look odd to me: - The "aes", "dgh", "fp", "ls64", "ls64_v", "memtag", "rpres", "sha1", "simd", "ssbs" functions all have the same set of `target_features`: `"+fp-armv8,+neon,+outline-atomics,+v8a"` - `bf16` and `ebf16` seem to be equivalent, and `+ebf16` doesn't get added to the `target-features` - `memtag2` and `memtag3` both have the same `target-features` (which has only `+mte`) - `+fp-armv8,+neon` are present in all cases. I'm guessing this is due to the base architecture enabling them. However, disabling them with `-march=+nofp` adds `-fp-armv8,-neon` to all `target-features`, even where the `target_version` should re-enable it. Are any of these cause for concern? https://github.com/llvm/llvm-project/pull/91490 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
