[LLVMdev] proposed patch to default to isl-only polly

[Tobias Grosser]

On 11/20/2013 05:37 PM, Jack Howarth wrote:
> On Wed, Nov 20, 2013 at 05:00:44PM +0100, Tobias Grosser wrote:
>> On 11/20/2013 04:50 PM, Jack Howarth wrote:
>>> On Tue, Nov 19, 2013 at 12:07:18PM +0100, Tobias Grosser wrote:
>>>> On 11/19/2013 08:50 PM, Jack Howarth wrote:
>>>>> Tobias,
>>>>>        Can we add something like the following to polly 3.4?
>>>>>
>>>>> Index: CMakeLists.txt
>>>>> ===================================================================
>>>>> --- CMakeLists.txt	(revision 195142)
>>>>> +++ CMakeLists.txt	(working copy)
>>>>> @@ -81,9 +81,14 @@ set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PAT
>>>>>
>>>>>     FIND_PACKAGE(Isl REQUIRED)
>>>>>     FIND_PACKAGE(Gmp REQUIRED)
>>>>> -FIND_PACKAGE(Cloog)
>>>>>     FIND_PACKAGE(Pluto)
>>>>>
>>>>> +option(POLLY_USE_CLOOG "Build Polly with Cloog support" OFF)
>>>>> +if (POLLY_USE_CLOOG)
>>>>> +  # Build Cloog support in Polly (default is ISL-only).
>>>>> +  FIND_PACKAGE(Cloog)
>>>>> +endif(POLLY_USE_CLOOG)
>>>>
>>>> Hi Jack,
>>>>
>>>> thanks for writing a patch for your feature request. The patch itself
>>>> looks good the way it is implemented. Though I wonder why it is needed.
>>>> Polly only uses CLooG if available, so I assume you have a CLooG version
>>>> installed, but you do not want to use it. What is the reason you do not
>>>> want to use the installed CLooG version. Is the CLooG version outdated
>>>> and Polly does not work with it? Or is there another reason for this?
>>>>
>>>> The reason the isl code generation is not yet default is because we do
>>>> not yet support openmp code generation support for it. Would you be OK
>>>> with installing Polly without OpenMP generation support?
>>>
>>> Tobias,
>>>       The idea is to have some control on whether Polly is built with
>>> cloog outside of crudely deinstalling the cloog headers and libraries
>>> each time Polly is built.
>>
>> Sure, I can see a need for this. The only reason I asked is if you want
>> to do this for package management reasons or if there is a problem with
>> the cloog parts of polly. In any case, I am happy to add this patch,
>> but if you use it to get around a bug, it would be good to also fix the
>> actual bug.
>
> Tobias,
>      No. As the original posting showed, there are no regression in Polly
> when built with the cloog support but the cmake build is missing fine
> control over how Polly is built. My understanding was the Polly developers
> were building isl-only but using configure for this since you have
> the --without-cloog option to use there. This patch provides the same
> functionality to the cmake build so that the isl-only build can be
> achieved without forcing the user to uninstall cloog.

This makes perfect sense. Committed in r195344 and r195348.

Thanks again for the patch. I really appreciate to get the fix as an 
actual patch, even though I unfortunately forgot to add your name in the 
commit message

Cheers,
Tobias