[llvm-commits] [llvm] r46387 - in /llvm/trunk: lib/Transforms/Scalar/SimplifyCFG.cpp test/CFrontend/2008-01-25-EmptyFunction.c

[Chris Lattner]

On Jan 25, 2008, at 5:43 PM, Bill Wendling wrote:
> URL: http://llvm.org/viewvc/llvm-project?rev=46387&view=rev
> Log:
> If we have a function like this:
>
> This is bad on some platforms (like PPC) because it will generate  
> the label for
> the function but no body. The label could end up being associated  
> with some
> non-code related stuff, like a section. This places a "trap"  
> instruction if the
> SimplifyCFG pass removed all code from the function leaving only one
> "unreachable" instruction.

This is a darwin-specific hack, so it needs to be conditionalized on  
the target.  Second, this should only be done if there are zero  
machineinstrs, not zero llvm instrs.

>   // If there are unreachable blocks in the CFG...
> -  if (Reachable.size() == F.size())
> +  if (Reachable.size() == F.size()) {
> +    if (F.size() == 1) {

F.size() is linear time, so you shouldn't call it here if you don't  
need it.  Why are you doing this in the Reachable.size() == F.size()  
case?

> +      // If the function has only one block with an "unreachable"  
> instruction,
> +      // then we should create *some* code for it. Issue a "trap"  
> instead.
> +      BasicBlock &BB = F.front();
> +
> +      if (BB.size() == 1 && dyn_cast<UnreachableInst>(&BB.front()))

Only use dyn_cast if you want the result: use isa in this case.   
BB.size() is also linear time.

Finally, the testcase you added was for llvm-gcc, this is a codegen  
thing so it should be in test/CodeGen/* and should only run llc.

-Chris