[llvm-commits] [PATCH] Defer some shl transformations from InstCombine to DAGCombine

[Jim Grosbach]

On Apr 18, 2012, at 4:58 PM, Jakob Stoklund Olesen wrote:

> 
> On Apr 18, 2012, at 4:51 PM, Chandler Carruth <chandlerc at google.com> wrote:
> 
>> Cool, one somewhat unrelated question...
>> 
>> On Wed, Apr 18, 2012 at 4:48 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk> wrote:
>> > Can we add a test for the LSR usage as well? Maybe using the same pattern as above?
>> 
>> LSR is using SCEV, so I am not so worried about that. LSR tests are also tricky because it needs target information, so it can only be run from llc.
>> 
>> Does the newly added '-mtriple' flag to opt allow us to give LSR the targetinfo it needs? If not, could we add an opt flag to make this easier? Just a random thought.
> 
> I don't think it works, unfortunately.
> 
> LSR needs a TargetLoweringInfo pointer, and opt -mtriple only sets the Module triple. I think it might be a library dependency issue?

That's my understanding. As things are currently organized, at minimum, TLI would bring in the codegen layer, which opt very specifically doesn't want to do.

> 
> It is possible to run llc -print-lsr-output, though. You'll get IR followed by assembly.
> 
> /jakob
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20120418/2396a503/attachment.html>