[LoongArch] Add codegen support for conditional branches

[wangleiat]

Summary

Setting ISD::BR_CC to Expand makes it much easier to deal with
matching the expanded form.

Depends on D128427

[wangleiat]

Update tests to use opaque pointers

[xen0n]

llvm/lib/Target/LoongArch/LoongArchInstrInfo.td

195

Could you explain a bit on why this is needed? I can't easily understand this as the only change is on the "guard" of said definition but the dropped check seems too important to ignore. (Maybe because I haven't looked at LLVM source code as hard as you did.)

llvm/test/CodeGen/LoongArch/ir-instruction/br.ll

22

Why this redundant unconditional jump? Is it because optimization is not enabled or other reason?

[wangleiat]

llvm/lib/Target/LoongArch/LoongArchInstrInfo.td

195

Sorry, I also do not understand.
I had to change simm26ls12, simm21ls12 and simm16lsl12_br to inherit from the Operand<OtherVT> class rather than Operand<GRLenVT> in order to keep tablegen happy.

[wangleiat]

llvm/test/CodeGen/LoongArch/ir-instruction/br.ll

22

We will implement overloading of the public function interface in later patches to eliminate unnecessary branches.
For example: TargetInstrInfo::analyzeBranch

[xen0n]

Someone else more familiar with LLVM internals could teach us about that particular tblgen behavior noted above, but the rest LGTM.

llvm/test/CodeGen/LoongArch/ir-instruction/br.ll

22

Fair enough.