[llvm-commits] [llvm] r72507 - in /llvm/trunk: include/llvm/Target/TargetLowering.h lib/CodeGen/SelectionDAG/DAGCombiner.cpp lib/Target/X86/X86ISelLowering.cpp lib/Target/X86/X86ISelLowering.h test/CodeGen/X86/narrow_op-1.ll test/CodeGen/X86/narrow_op-2.ll

[Sebastian Redl]

Evan Cheng wrote:
> On May 27, 2009, at 11:38 PM, Duncan Sands wrote:
>
>   
>> Hi Evan, nice transform!  But I think you shouldn't do it if the
>> original store was volatile.
>>     
>
> Why not? This would just reduce the store width.
>   

Wouldn't it be better to default the "is profitable" check to false for
platforms that don't override it? It seems to me that the problems from
incorrectly allowing the narrowing outweigh the problems from
incorrectly disallowing it. I think there are more platforms that only
support full-word stores efficiently (e.g. the Alpha) than there are
architectures where every narrowing is profitable.

Sebastian