[PATCH] D11774: [NVPTX] Use LDG for pointer induction variables

[Jingyue Wu]

On Wed, Aug 5, 2015 at 4:22 PM, Justin Holewinski <jholewinski at nvidia.com>
wrote:

> It’s not really a calling convention, and it’s not clear how to handle
> functions that are entry points *and* device-callable functions.
>

I can't have a __global__ and __device__ function in CUDA, can I? Or is
this for NVCL?


>
> If you have a strong use case for this, but otherwise I’d rather just have
> one way to annotate entry points instead of two.
>

No, go for it. I am just curious why you prefer one way vs the other.


>
> On Aug 5, 2015, at 7:11 PM, Jingyue Wu <jingyue at google.com> wrote:
>
> Hi Justin,
>
> Why are we deprecating ptx_kernel? What's bad about it?
>
> Jingyue
>
> On Wed, Aug 5, 2015 at 3:44 PM, Justin Holewinski <jholewinski at nvidia.com>
> wrote:
>
>> jholewinski added inline comments.
>>
>> ================
>> Comment at: test/CodeGen/NVPTX/load-with-non-coherent-cache.ll:243
>> @@ +242,3 @@
>> +  ret void
>> +}
>> +
>> ----------------
>> broune wrote:
>> > Thanks, that's good to know. I'll fix that in the whole file my
>> follow-up clean-up cl.
>> That works, but I would like to deprecate it.  The PTX calling
>> conventions are legacy cruft from the old back-end.
>>
>>
>> http://reviews.llvm.org/D11774
>>
>>
>>
>>
>
> ------------------------------
> This email message is for the sole use of the intended recipient(s) and
> may contain confidential information.  Any unauthorized review, use,
> disclosure or distribution is prohibited.  If you are not the intended
> recipient, please contact the sender by reply email and destroy all copies
> of the original message.
> ------------------------------
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150805/4453a801/attachment.html>