[PATCH] Allow CodeGen to take advantage of functions that always return arguments

Bill Wendling wendling at apple.com
Fri Apr 19 17:14:02 PDT 2013


On Apr 19, 2013, at 5:01 PM, Evan Cheng <evan.cheng at apple.com> wrote:

> 
> On Apr 19, 2013, at 4:53 PM, Bill Wendling <wendling at apple.com> wrote:
> 
>> On Apr 19, 2013, at 7:48 AM, Stephen Lin <swlin at post.harvard.edu> wrote:
>> 
>>>> OK, I removed some extraneous changes and updated the patch to be
>>>> based against a mostly non-recursive implementation of sameNoopInput:
>>>> 
>>>>  http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130415/172039.html
>>>> 
>>>> I've also added the restriction that 'returned' can only be on one
>>>> parameter to allow sameNoopInput to see through 'returned' without
>>>> recursion or extra bookkeeping. There's a minor tack-on bug fix;
>>>> apparently the verifier check that 'nest' is only on one parameter was
>>>> never working correctly...I'm willing to split that off to a separate
>>>> patch if that's preferable though.
>>>> 
>>>> The other patch listed is my original e-mail has been committed so is
>>>> no longer an uncommitted dependency.
>>>> 
>>>> Stephen
>>> 
>>> I fixed a small issue (an extraneous recursive call) and rebased against:
>>> 
>>>   http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130415/172067.html
>>> 
>>> (although the patch should merge cleanly against the previous version, too.)
>>> 
>> 
>> Hi Stephen,
>> 
>> The IR part of the patch looks okay to me.
>> 
>> Evan, Are you okay with the CodeGen changes?
> 
> LGTM

Thanks.

Stephen,

The patch didn't apply cleanly. Someone has updated lib/CodeGen/Analysis.cpp. Could you regen your patch? I'll apply it once you send it. :)

-bw




More information about the llvm-commits mailing list