[PATCH RFC 1/1] utils: Fix segfault in flattencfg

[Matt Arsenault]

On Aug 4, 2014, at 3:24 PM, Jan Vesely <jan.vesely at rutgers.edu> wrote:

> ---
> 
> Not sure if it's the right fix. Looks more like ductaping over some bigger problem.
> Fixes segfault I hit with mad_sat libclc patch
> 
> lib/Transforms/Utils/FlattenCFG.cpp | 2 ++
> 1 file changed, 2 insertions(+)
> 
> diff --git a/lib/Transforms/Utils/FlattenCFG.cpp b/lib/Transforms/Utils/FlattenCFG.cpp
> index 51ead40..ce49a40 100644
> --- a/lib/Transforms/Utils/FlattenCFG.cpp
> +++ b/lib/Transforms/Utils/FlattenCFG.cpp
> @@ -241,6 +241,8 @@ bool FlattenCFGOpt::FlattenParallelAndOr(BasicBlock *BB, IRBuilder<> &Builder,
>     while (1) {
>       BranchInst *BI = dyn_cast<BranchInst>(CurrBlock->getTerminator());
>       CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition());
> +      if (!CI)
> +        break;

This needs a test. I’m guessing this is a branch on a constant? I’ve encountered problems like this from those in all of the GPU CFG passes


>       CmpInst::Predicate Predicate = CI->getPredicate();
>       // Canonicalize icmp_ne -> icmp_eq, fcmp_one -> fcmp_oeq
>       if ((Predicate == CmpInst::ICMP_NE) || (Predicate == CmpInst::FCMP_ONE)) {
> -- 
> 1.9.3
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits