[cfe-commits] r97672 - in /cfe/trunk: lib/CodeGen/CGExprScalar.cpp test/CodeGenObjC/id-isa-codegen.m

Fariborz Jahanian fjahanian at apple.com
Tue Mar 9 10:43:09 PST 2010


On Mar 8, 2010, at 11:28 AM, Daniel Dunbar wrote:

> On Mon, Mar 8, 2010 at 11:10 AM, John McCall <rjmccall at apple.com>  
> wrote:
>> On Mar 6, 2010, at 5:37 AM, Daniel Dunbar wrote:
>>> On Thu, Mar 4, 2010 at 11:19 AM, Fariborz Jahanian <fjahanian at apple.com 
>>> > wrote:
>>>>
>>>> On Mar 4, 2010, at 5:24 AM, Daniel Dunbar wrote:
>>>>
>>>>> Hi Fariborz,
>>>>>
>>>>> Is it possible to write this test to actually check the output  
>>>>> LLVM IR
>>>>> to make sure that whatever error was fixed, is fixed?
>>>>>
>>>>> - Daniel
>>>>
>>>> I am not sure how to write a test which checks for a pattern such  
>>>> as below
>>>> and passes
>>>> as we go through iterations of LLVM IR generations:
>>>>
>>>> %tmp2 = load i8** %resval                       ; <i8*> [#uses=1]
>>>>  %1 = bitcast i8* %tmp2 to i8**                  ; <i8**> [#uses=1]
>>>>  store i8* %tmp, i8** %1
>>>
>>> This doesn't seem like a pattern that would change much, as long as
>>> the test is written to not depend on the individual names (using
>>> FileCheck variables).
>>
>> Apropos, it would be really handy if FileCheck had a built-in  
>> pattern for LLVM register names;  in my experience, that is far-and- 
>> away the most common thing to match.
>
> Yeah, I still think we need better toolage in this direction.
>
> Another thing I think would help is an llvm-dis mode that would just
> print the LLVM IR out in a form that could be filecheckized.

Isn't it llc -march=c ?
- Fariborz

>
>
> - Daniel
>
>>
>> John.




More information about the cfe-commits mailing list