[llvm] [RISCV] Add OR/XOR/SUB to RISCVInstrInfo::isCopyInstrImpl (PR #132002)

[Philip Reames via llvm-commits]

preames wrote:

> welcome discussion on to what degree we want isCopyInstrImpl to be exhaustive.

In general, sticking to those that we can write a test case from IR (or otherwise not over reduced) for seems like a reasonable line.  That's not quite the same as causing codegen differences in spec, but if we can't test something, having code seems undesirable.   If we can test something, but it doesn't show up, then I'd say that's largely at the discretion of the author.  (i.e. if they find the time worthwhile)

>     * The case of OR with indentical register operands isn't covered. Although this causes no codegen difference, possibly I should update this PR to handle it? That way at least the opcodes that isCopyInstrImpl recognises are covered exhaustively.

This might be a reasonable exception to the "only with high level tests" policy I suggest above.  I could go either way, and am happy to defer to your judgement.  

https://github.com/llvm/llvm-project/pull/132002