[llvm] [MachineVerifier] Query TargetInstrInfo for PHI nodes. (PR #110507)

[Vyacheslav Levytskyy via llvm-commits]

Nathan =?utf-8?q?Gauër?= <brioche at google.com>,
Nathan =?utf-8?q?Gauër?= <brioche at google.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/110507 at github.com>


VyacheslavLevytskyy wrote:

We should not say that `OpPhi` is not a `phi` just because it has one difference that is subtle enough to make us spend a lot of time trying to create a reproducer for it. This difference is buried deep and doesn’t interfere with anything. Just the opposite this is an argument against the suggested approach of delayed translation of phi to OpPhi, as it would make a price of delayed translation higher, along with SPIRV types and, in general, violation of the instruction selection notion.

We may describe the issue so that GenCode doesn’t expect that after lowering a target still supports a higher-level opcode representing a PHI node. SPIR-V has a PHI opcode, and it doesn’t match with what GenCode expects from a lowered code. I think that this PR presents a normal engineering practice of using defaults and overrides to resolve this issue.

https://github.com/llvm/llvm-project/pull/110507