[llvm-commits] [llvm] r48680 - in /llvm/trunk: lib/Transforms/Scalar/GVN.cpp test/Transforms/GVN/form-memset.ll

[Chris Lattner]

On Mar 22, 2008, at 5:10 AM, Duncan Sands wrote:

> Hi Chris,
>
>> implement an initial hack at a straight-line store -> memset  
>> optimization.
>> This fires dozens of times across spec and multisource, but I don't  
>> know
>> if it actually speeds stuff up.  Hopefully the testers will show  
>> something
>> nice :)
>
> in llvm-gcc, small memset's are turned into stores.  Is this going  
> to turn
> the stores back into a memset?

I wouldn't suggest changing llvm-gcc.  We need to play with heuristics  
and beef up the optimizer's support for memcpy.  Doing the store- 
 >memcpy raising is useful in some code, but it restricts the ability  
to eliminate loads and stores.  For example, GVN won't currently turn  
memset + use(load) into memset + use(set value) etc.

-Chris