[llvm-commits] [llvm] r44860 - /llvm/trunk/docs/LangRef.html

[Chris Lattner]

On Dec 11, 2007, at 5:56 PM, Christopher Lamb wrote:

>
> On Dec 11, 2007, at 3:54 PM, Chris Lattner wrote:
>
>> On Dec 11, 2007, at 1:31 AM, Christopher Lamb wrote:
>>> Add information on address space qualifiers for pointer types and
>>> global
>>> declarations to the LangRef.
>>
>> Nice!
>>
>>>  <div class="doc_code">
>>>  <pre>
>>> + at G = constant float 1.0 addrspace(5), section "foo", align 4
>>
>> Is there a comma after 1.0?
>
> Nope. It's not treated like other attributes as it affects the type  
> of the GlobalVariable that's created, rather than modifying it after  
> the fact. Ideally I would have liked it to be between the type name  
> and the initializer value, but the current Assembly parser made this  
> very yucky. The other option would be to support the addrspace  
> attribute on types other than pointers in the Assembly syntax, but  
> discard them in all cases except  when declaring a global variable,  
> also yucky.
>
> The key here is that the order of asterisks and addrspace()  
> qualifiers are consistent in the global variable declaration and the  
> type name:
>
> @foo = constant float addrspace(1)* 1.0 addrspace(2)
>
> @foo has a type of
>
> float addrspace(1)* addrspace(2)*

That is a very strong argument to me, but it seems to argue even more  
strongly for:

@G = constant float addrspace(5) 1.0, section "foo", align 4

Your example above would then be:

> @foo = constant float addrspace(1)* addrspace(2) 1.0

which has type:

> float addrspace(1)* addrspace(2)*

What do you think?  the downside is that this may cause bison to have  
issues :)

-Chris