[PATCH] Add attributes for AMDGPU register limits.

[Matt Arsenault]

> On Dec 3, 2014, at 3:03 PM, Matt Arsenault <arsenm2 at gmail.com> wrote:
> 
> 
>> On Dec 3, 2014, at 2:55 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
>> 
>> On Wed, Dec 3, 2014 at 2:48 PM, Matt Arsenault <arsenm2 at gmail.com> wrote:
>>> 
>>> On Nov 17, 2014, at 1:32 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
>>> 
>>> 
>>> 
>>> +  let Args = [UnsignedArgument<"NumVGPR">];
>>> +  let Documentation = [AMDGPURegisterDocs];
>>> +  let Subjects = SubjectList<[KernelFunction], ErrorDiag,
>>> +                             "ExpectedKernelFunction">;
>>> +}
>>> +
>>> +def AMDGPUNumSGPR : InheritableAttr, TargetSpecificAttr<TargetAMDGPU> {
>>> +  let Spellings = [GCC<"amdgpu_num_sgpr">];
>>> 
>>> 
>>> Same comment here as above.
>>> 
>>> +  let Args = [UnsignedArgument<"NumSGPR">];
>>> +  let Documentation = [AMDGPURegisterDocs];
>>> 
>>> 
>>> This will render the same documentation a second time. The usual
>>> approach is to create a category in the docs, and then generate
>>> different docs in the same category. See the calling convention
>>> attributes as an example.
>>> 
>>> 
>>> It turns out they aren’t being rendered at all now. I’m not sure why, I
>>> don’t see what I’m doing differently from the interrupt attributes which I
>>> followed as an example. It seems many of the attributes documented here are
>>> not actually getting generated into the attribute documentation, such as
>>> align_value. This i when I attempt to generate the docs, the ones on the
>>> website seem to have them.
>> 
>> How are you attempting to generate the docs? Are you running the
>> tablegen pass first, and then sphinx-build?
>> 
>> ~Aaron
> 
> I’m just running make and looking at the output in the build output doc directory

The documentation only mentions something for the autotools build, so maybe it doesn’t work with cmake? There seems to be some kind of sphinx target in cmake though, so I would expect it to work