[llvm-dev] Update control flow graph when splitting a machine basic block?

[Friedman, Eli via llvm-dev]

On 11/10/2017 6:00 PM, 章明 wrote:
>> The right way to update the CFG very much depends on how you're
>> transforming it.
> I would like to export the CFG for control flow checking.
> Theoretically, it should be possible for a compiler to know every target of every control flow instruction, except for computed targets that are not known at compile-time.
Every MachineBasicBlock has a list of successors; you can access it with 
the successors() accessor.  That's what you should be using for any CFG 
analysis.

> When a machine basic block is split between two branches, as shown below:
>
> BB:                 ; Old machine basic block shortened by the split
>      ...
>      branch to A
> ------------------- ; Split here
> BB':                ; New machine basic block created by the split
>      branch to B
>      ...
>
> BB' is a successor of BB (if there is no control flow barrier near the end of BB).
> In addition, targets of terminators like A should be added to the set of successors of BB and possibly removed from the set of successors of BB'.
> It is not always safe to remove B from the successors of BB', because A and B may be the same machine basic block.
>
> If edges in the CFG are interpreted as possible control flow transfers, rather than definite control flow transfers, the most conservative way to update the CFG might be:
> 1) Let BB' have all successors of the original machine basic block before the split.
> 2) Add BB' and all successors of BB' to the set of successors of BB.
> However, this introduce false control flow transfers into the CFG. False control flow transfers are harmful to control flow checking, because different basic blocks have to be assigned the same signature, which is called aliasing.
>
> To update the CFG without introducing any false control flow transfers, one needs to determine the target(s) of each terminator.
> That is what I desire. I am not quite familiar with LLVM, so I am asking whether it is possible to achieve that with the current LLVM API.

I don't think this actually has any impact in practice; I mean, I guess 
it's an issue in theory, but in practice we don't stick branches into 
the middle of basic blocks.

-Eli

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project