[llvm-commits] [llvm] r169459 - in /llvm/trunk: include/llvm/Target/TargetLowering.h lib/CodeGen/SelectionDAG/SelectionDAG.cpp lib/CodeGen/SelectionDAG/TargetLowering.cpp lib/Target/ARM/ARMISelLowering.cpp lib/Target/ARM/ARMISelLowering.h lib/Target/X86/X86ISelLowering.cpp lib/Target/X86/X86ISelLowering.h test/CodeGen/ARM/extload-knownzero.ll

[Evan Cheng]

I think you are right. It's not safe to compute known zero for anyext until just before selection time. I'll look into solving this deficiency another way. 

Thanks,

Evan

On Dec 6, 2012, at 1:00 AM, Duncan Sands <baldrick at free.fr> wrote:

> Hi Evan,
> 
> On 06/12/12 02:28, Evan Cheng wrote:
>> Author: evancheng
>> Date: Wed Dec  5 19:28:01 2012
>> New Revision: 169459
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=169459&view=rev
>> Log:
>> Let targets provide hooks that compute known zero and ones for any_extend
>> and extload's. If they are implemented as zero-extend, or implicitly
>> zero-extend, then this can enable more demanded bits optimizations. e.g.
> 
> is this correct?  All kinds of DAG combines exploit that anyext means "any",
> eg: any_extend(trunc X) -> X.  On a target where any_extend does zero-extend,
> won't your patch say that all the upper bits of "any_extend(trunc X)" are
> zero, causing all kinds of oddness when they turned later into the bits of X,
> which may not be zero.
> 
> Ciao, Duncan.
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits