[llvm-commits] [llvm] r130428 - /llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

[Benjamin Kramer]

On 28.04.2011, at 22:20, Duncan Sands wrote:

> Hi Benjamin,
> 
>> -  // (trunc x) == C1&  (and x, CA) == C2 ->  (and CA|CMAX) == C1|C2
>> +  // (trunc x) == C1&  (and x, CA) == C2 ->  (and x, CA|CMAX) == C1|C2
>>    // where CMAX is the all ones value for the truncated type,
>>    // iff the lower bits of CA are zero.
> 
> don't you need the lower bits of C2 to also be zero?
> 
> Ciao,
> 
> Duncan.
> 
> PS: Of course, if they are not zero then "(and x, CA) == C2" can be folded to
> false, so probably you will never hit this in practice.

We do require the bits of C2 to be zero. This should be folded away earlier, however I was bitten
by this before, it's possible to create test cases where SimplifyDemandedBits hits the recursion
limit and doesn't fold things that should be folded, I'll add a check to be safe.