[llvm-commits] [llvm] r48730 - /llvm/trunk/test/CFrontend/2008-03-24-BitFiel-And-Alloca.c

[Chris Lattner]

On Mar 24, 2008, at 2:12 PM, Devang Patel wrote:

>
> On Mar 24, 2008, at 2:04 PM, Chris Lattner wrote:
>
>>>> Are you testing the front-end here or the optimizer?  If you're
>>>> testing the front-end, don't pass -O2.  If you're testing the
>>>> optimizer, write this as a .ll file, not a .c file.
>>>
>>> I am testing FE and optimizer both here. One regression FE lowering
>>> caused optimizer to not eliminate unnecessary alloca instructions.
>>
>> I haven't seen all the patches yet, is the fix in the f.e. or the
>> optimizer?
>
> magic :) !
>
> I should have connected dots using appropriate commit messages. Sorry.
>
> http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20080324/060114.html
>
> This check-in updates lowering such that <i8, [7 x i8]> is not used to
> represent this struct, which in turns allows SRoA to eliminate  
> allocas.

Sounds good.  As we discussed, your existing test is a great  
"integration" test, which verifies that the optimizer + f.e. is  
working well together.  However, we still want a f.e. unit test that  
verifies it is producing what we expect :).  This could just be  
another run line in the same test.

Thanks Devang,

-Chris

>
>
> I added these lines earlier as an independent patch and the comments
> described its intent. Now, it is not clear, at least to me, whether
> bit-field types influence struct type alignment or not. When I added
> the original patch, I relied one test case which passes now. I am not
> super confident about this -- bit-field type's impact on struct
> alignment -- but now original behavior is restored.
> -
> Devang
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits