[llvm-dev] Is a PHI use of another PHI in the same block valid?

Sanjoy Das via llvm-dev llvm-dev at lists.llvm.org
Fri Feb 26 09:09:01 PST 2016

For people not following the PR, Joseph has pointed out some issues
with what the verifier currently accepts in an excellent writeup on
the bug: https://llvm.org/bugs/show_bug.cgi?id=26718#c9

On Fri, Feb 26, 2016 at 9:02 AM, Philip Reames via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Over in pr26718, we ran across a case where input IR had one PHI within a
> basic block using the value of another PHI within the same basic block
> (without a backedge).  There has been some disagreement as to whether this
> is valid IR.  I believe it is not.
> The verifier currently accepts the following code without error:
> define void @f() {
> entry:
>   br label %next
> next:
>   %y = phi i32 [ 0, %entry ]
>   %x = phi i32 [ %y, %entry ]
>   ret void
> }
> Looking at the code, this may be an oversight - due to a special casing of
> the dominance relation within a single basic block - but that's not
> obviously true.  Thus, we need to clarify what the intended semantics are.
> The lang ref seems pretty clear about this:
> "For the purposes of the SSA form, the use of each incoming value is deemed
> to occur on the edge from the corresponding predecessor block to the current
> block (but after any definition of an ‘invoke‘ instruction’s return value on
> the same edge)."
> But David pointed out that various bits of the optimizer appear to have code
> and test cases covering exactly this case.
> So, is this legal IR?  Or not?
> Philip
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Sanjoy Das

More information about the llvm-dev mailing list