[cfe-commits] r91437 - in /cfe/trunk: include/clang/Basic/DiagnosticSemaKinds.td lib/Sema/SemaExpr.cpp test/Sema/ext_vector_components.c

[Douglas Gregor]

On Dec 15, 2009, at 11:19 AM, Nate Begeman wrote:

>
> On Dec 15, 2009, at 10:33 AM, Douglas Gregor wrote:
>
>>
>> On Dec 15, 2009, at 10:13 AM, Nate Begeman wrote:
>>
>>> Author: sampo
>>> Date: Tue Dec 15 12:13:04 2009
>>> New Revision: 91437
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=91437&view=rev
>>> Log:
>>> Support OpenCL 1.1 odd-length vector component accessors.
>>>
>>> For hi/odd of an odd-length vector, the last component is  
>>> undefined.  Since
>>> we shuffle with an undef vector, no CodeGen needs to change to  
>>> support this.
>>
>> Do you want to differentiate between OpenCL 1.0 and 1.1? We could  
>> certainly have an OpenCL 1.1 dialect option, then diagnose this  
>> behavior as an extension if we're building for Open CL 1.0.
>
> I'm not aware of the need for such a thing, IIRC this behavior was  
> previously undefined and is now defined to do something in the 1.1  
> draft spec.  While having the option "couldn't hurt", I'm not sure  
> we'd use it for anything yet.


It's a policy decision that I'm happy to leave to you. Is it important  
to users to be able to differentiate between OpenCL 1.0 and 1.1? Is  
OpenCL 1.1 consider more of a "bug fix" release rather than a new  
standard?

Within Clang, we already have a couple different dialects with  
languages:
	- We differentiate between C++0x and C++03
	- We *don't* differentiate between C++98 and C++03, because we  
consider C++03 to be a "bug fix" standard
	- We differentiate between Objective-C 1.0 and Objective-C 2.0
	- We differentiate between C89 and C99 (and will differentiate C1x,  
once it matters)

Keep in mind that it's easier to differentiate on a patch-by-patch  
basis now than it will be to go back and figure out all of the OpenCL  
1.1-specific changes.

	- Doug