[llvm-commits] [llvm] r89523 - /llvm/trunk/lib/Transforms/Scalar/SimplifyLibCalls.cpp

Chris Lattner clattner at apple.com
Fri Dec 4 22:24:07 PST 2009


On Dec 4, 2009, at 10:06 PM, Eric Christopher wrote:

>>> 
>>> The lower to "unknown" is merely a placeholder that keeps current behavior.
>> 
>> What do you mean?
>> 
> 
> We're lowering to unknown in simplifylibcall because I haven't had time to write the rest of the pass yet :)
> 
>>> We do lower to unknown at codegen time already if we didn't before.  It should be all lowered before we get to codegen in one way or another.
>> 
>> Is there code that lowers memcpy_chk(.. unknown)  -> memcpy that happens at codegen time?  If not, lowering the object size intrinsic only will pessimize real cases, no?
> 
> It would, but we should only get there unoptimized. I see what you mean though and I can get something to do the conversion in the unoptimized case if you'd like, but I probably wouldn't worry about it otherwise.

Many memfoo calls will have an unknown length.  We want to make sure to get a call to memfoo (instead of memfoo_chk) for both performance and code size reasons.

-Chris





More information about the llvm-commits mailing list