R600 intrinsics patch

Matt Arsenault arsenm2 at gmail.com
Tue Mar 17 11:03:11 PDT 2015

> On Mar 17, 2015, at 10:09 AM, Tom Stellard <tom at stellard.net> wrote:
> On Tue, Mar 17, 2015 at 09:53:08AM -0700, Matt Arsenault wrote:
>> On 03/17/2015 08:11 AM, Tom Stellard wrote:
>>> The target prefix for all these has been changed from AMDGPU to amdgpu.
>>> I don't think this is necessary.  Matt may know more about this.
>> I don't remember all the details, but this is basically necessary for 
>> the automatic support from GCCBuiltin to work properly. The current set 
>> of builtins handled in clang need custom handling for things like 
>> conversion to/from i1, converting struct returns to out arguments, and 
>> handling overloads so they require custom code anyway. For simpler 
>> builtins, to not need custom code needs the target prefix to match.
> What does the target prefix need to match?  The triple?
> -Tom

Almost. It needs to match what getArchTypePrefix on the triple gives

More information about the llvm-commits mailing list