https://github.com/uweigand commented:
The code changes now look good to me, but the tests still need some work. In a nutshell, most of the tests don't actually seem to verify what you wanted to verify. These are IL tests, that only verify the translation from IL to assembler. You appear to be more interested in the translation from *source* to IL - but that isn't actually tested. Apparently you generated those tests from source code. But that means the test doesn't actually verify this - if common optimizers change in the future, your test will not notice this. Also, this leads to most of the tests actually not being interesting at the IL level - the IL was in some cases fully optimized away (due to tautological optimizations), or else many different tests are actually the same (or extremely similar) at the IL level. You should rather design a set of IL tests *starting* with IL, i.e. make sure that interesting combinations of IL are actually optimized by the back-end code you're adding. If you do want to add tests that certain source code is transformed to particular IL, these should be clang-level test cases. This may not be very useful if that ends up to verify just generic optimizations, but there may be some interesting cases remaining (e.g. because optimizers are able to exploit the < 4 assertion). https://github.com/llvm/llvm-project/pull/125970 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
