OpenCL C "long" should always have 64 bits

[Michele Scandale]

On 09/03/2013 07:53 PM, Tom Stellard wrote:
> On Tue, Sep 03, 2013 at 01:34:50PM -0400, Erik Schnetter wrote:
>> Yes, R600 defines a "good" address space map.
>>
>> My patch currently overrides the target-specific address space maps...
>>
>> Instead of doing so, I think the right approach is to define a default address space map that already does the right thing for OpenCL and CUDA. This makes sense since address spaces seem currently defined for OpenCL and CUDA only, i.e. they won't be used by standard C/C++. The targets can then override the default (which they already do).
>>
> 
> Does the rest of this patch depend on resolving the mangling issues with
> address spaces?  If not, can we split the address space map out into a
> separate patch and commit the rest of the changes?  The OpenCL type
> changes are very useful, and I wouldn't want the address space mapping
> discussions to prevent them from being committed.

I agree. The problems of type size is orthogonal from mangling and address
spaces. I know that it's all related and to have everything working we would
need a global solution, but still being orthogonal they should be solved in
different patches.

This part related to type size is first step that fix a quite big lack in the
support of OpenCL.

Then the mangling can be fixed (I'm still waiting for feedback to know if the
last proposed patch can be fine or not to be committed).

The last part about address space information requires also modification in the
middle end (I am working on this... soon a proposed patch for metadata handling).

Regards,
-Michele