[llvm-commits] [llvm] r85441 - in /llvm/trunk: lib/CodeGen/BranchFolding.cpp test/CodeGen/X86/2008-02-18-TailMergingBug.ll test/CodeGen/X86/2008-05-12-tailmerge-5.ll test/CodeGen/X86/convert-2-addr-3-addr-inc64.ll

[Chris Lattner]

On Oct 28, 2009, at 3:10 PM, Bob Wilson wrote:
> Author: bwilson
> Date: Wed Oct 28 17:10:20 2009
> New Revision: 85441
>
> URL: http://llvm.org/viewvc/llvm-project?rev=85441&view=rev
> Log:
> Reimplement BranchFolding change to avoid tail merging for a 1  
> instruction
> common tail, except when the OptimizeForSize function attribute is  
> present.
> Radar 7338114.

Hi Bob,

> +      // minCommonTailLength instructions in common.  Otherwise, if  
> we are
> +      // optimizing for code size, 1 instruction in common is  
> enough.  At
> +      // worst we will be replacing a fallthrough into the common  
> tail with a
> +      // branch, which at worst breaks even with falling through  
> into the
> +      // duplicated common tail.  We will always pick a block we do  
> not have
> +      // to split as the common tail if there is one.  (Empty  
> blocks will get
> +      // forwarded and need not be considered.)
> +      MachineFunction *MF = CurMPIter->second->getParent();
>       if (CommonTailLen >= minCommonTailLength ||
>           (CommonTailLen > 0 &&
> -           (TrialBBI1==CurMPIter->second->begin() ||
> -            TrialBBI2==I->second->begin()))) {
> +           MF->getFunction()->hasFnAttr(Attribute::OptimizeForSize)  
> &&
> +           (TrialBBI1 == CurMPIter->second->begin() ||
> +            TrialBBI2 == I->second->begin()))) {
>         if (CommonTailLen > maxCommonTailLength) {
>           SameTails.clear();
>           maxCommonTailLength = CommonTailLen;

I know that you're probably not the one to blame, but could you factor  
out this insane predicate into a static helper function that returns a  
bool?

-Chris