[cfe-commits] r105824 - in /cfe/trunk: lib/AST/ExprConstant.cpp test/CodeGen/altivec.c

John McCall rjmccall at apple.com
Fri Jun 11 19:12:32 PDT 2010


On Jun 11, 2010, at 7:09 PM, Eli Friedman wrote:

> On Fri, Jun 11, 2010 at 6:38 PM, John McCall <rjmccall at apple.com> wrote:
>> 
>> On Jun 11, 2010, at 2:52 PM, Eli Friedman wrote:
>> 
>>> On Fri, Jun 11, 2010 at 10:54 AM, John McCall <rjmccall at apple.com> wrote:
>>>> Author: rjmccall
>>>> Date: Fri Jun 11 12:54:15 2010
>>>> New Revision: 105824
>>>> 
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=105824&view=rev
>>>> Log:
>>>> Fix the constant evaluator for AltiVec-style vector literals so that the
>>>> vector is filled with the given constant;  we were just initializing the
>>>> first element.
>>> 
>>> Are you sure this patch is correct for non-AltiVec cases?
>> 
>> Yes, but for terrible reasons:  all other InitListExprs seem to get normalized to
>> match the number of elements by padding with implicit zero expressions.
>> 
>> Applying a similar normalization to the altivec literal code would seem to be the
>> sensible solution, but unfortunately it's not sound because OpenCL lifts the AltiVec
>> restriction that element initializers be constant.  I should have a better patch
>> soon.
> 
> Does this mean you're going to fix the "unsupported GNU array range
> designator extension" errors as well?

I doubt it.

John.





More information about the cfe-commits mailing list