mylai-mtk wrote: > A high level suggestion is don't hash that until MC layer, so that we can > easier debug and observe that from the IR level, so that means we can drop > the hash part from this PR.
@kito-cheng I prefer to keep the function signature as an auxiliary metadata with the name `riscv_lpad_func_sig`, so future schemes and other language frontends can reuse the `riscv_lpad_label` metadata flow and have the opportunity of not being forced to adopt the idea of function signatures. Besides, this is in fact what we currently have in our internal branch, since a later development of the Zicfilp psABI spec introduced the lpad mapping symbol, which writes function signatures into ELF files, and thus the internal branch had already been modified to pass function signature as an extra metadata. Regarding the testability issue, I think later I'll update this PR so clang attaches both the `riscv_lpad_label` and `riscv_lpad_func_sig` metadata to Functions, and unittests can be added specifically for the `zicfilpFuncSigHash()` function. https://github.com/llvm/llvm-project/pull/111661 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
