[PATCH] D76961: [BranchFolder] don't remove MBB's that have their address taken

Tue Mar 31 18:45:02 PDT 2020

nickdesaulniers added a comment.

In D76961#1951350 <https://reviews.llvm.org/D76961#1951350>, @nickdesaulniers wrote:

> Everything looks good.  I don't understand why `INLINEASM_BR` isn't the terminal instruction of `bb.0`.  Isn't it marked a terminator in `llvm/include/llvm/Target/Target.td` line 1021?  I also also don't understand why the `MachineBasicBlock` has a `terminators` method plural?  I thought all blocks have 1 and only 1 terminal instruction?

A colleague today explained to me that this happens with conditional jumps, and later I found such a case with multiple conditional jumps with an unconditional jump being the final instruction.

I have a fix for my test case, but it regresses one of the previous existing test cases.  I need to spend more time on it tomorrow.  Essentially, ISel is now emitting a messed up PHI node that's missing cases for some of the predecessors, which the machine verifier complains about.  I suppose this may be the `agita` (sic) referred to in the comment above `ScheduleDAGSDNodes::EmitSchedule`'s handling of `INLINEASM_BR. :P

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76961/new/

https://reviews.llvm.org/D76961