[llvm] [BOLT][AArch64] Do not crash on authenticated branch instructions (PR #129898)

[Alexey Moksyakov via llvm-commits]

yavtuk wrote:




> > folks how about to use isIndirectBranchOpcode func from AArch64InstrInfo.h file?
> > Do we need own implementation ?
> 
> @yavtuk Do you suggest replacing `isBRA` with a call to `isIndirectBranchOpcode`? The difference is that `isBRA` only checks for `Branch to register, with pointer authentication` instructions defined in Armv8.3-a (`FEAT_PAUTH`), while `isIndirectBranchOpcode` checks for both `AArch64::BR` and its "authenticated" variants, as expected:
> 
> ```c++
> static inline bool isIndirectBranchOpcode(int Opc) {
>   switch (Opc) {
>   case AArch64::BR:
>   case AArch64::BRAA:
>   case AArch64::BRAB:
>   case AArch64::BRAAZ:
>   case AArch64::BRABZ:
>     return true;
>   }
>   return false;
> }
> ```
> 
> That is, `isIndirectBranchOpcode` intentionally **includes** `AArch64::BR` and `isBRA` intentionally **excludes** `AArch64::BR`.
> 
> Though, I agree that there may exist some redundancy in `AArch64MCPlusBuilder`: for example, `AArch64MCPlusBuilder::mayLoad` is [implemented](https://github.com/llvm/llvm-project/blob/29ff168363fa9a41f04b19a9afe5930330f35b4c/bolt/lib/Target/AArch64/AArch64MCPlusBuilder.cpp#L550) like this:
> 
> ```c++
>   bool mayLoad(const MCInst &Inst) const override {
>     return isLDRB(Inst) || isLDRH(Inst) || isLDRW(Inst) || isLDRX(Inst) ||
>            isLDRQ(Inst) || isLDRD(Inst) || isLDRS(Inst);
>   }
> ```
> 
> TableGen-erated descriptions from AArch64 backend could probably be re-used instead in `mayLoad`, but I'm not sure.

as example I think we can create the different function isIndirectAuthBranch or we can use  if (isIndirectBranchOpcode() & !Inst.getOpcode() != AArch64::BR) 


https://github.com/llvm/llvm-project/pull/129898